Nir 2013 Review

NIR 2013 is held under the auspices of the International Council for Near Infrared Spectroscopy
Proceedings
NIR 2013 - 16th International Conference on Near Infrared Spectroscopy
2 - 7 June 2013, la Grande-Motte, France
Pic
k
i
n
g

u
p

g
o
o
d

v
i
b
ration
s

Copyright 2013 IRSTEA France

Published by
IRSTEA France Institut National de recherche en sciences et technologies pour lenvironnement et lagriculture.

Address :
IRSTEA NIR2013 Organization Committee 361, Rue Jean-Franois Breton
34196 Montpellier Cedex 5
France

Edited by
Vronique Bellon Maurel, NIR2013 Chair, Editor
Phil Williams, Editor
Gerard Downey, Editor
Rbecca Kabor, NIR2013 Event Coordinator

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or
by any means, electronic, mechanical, photocopying or otherwise, without the prior permission of the copyright owner.
Editorial

NIR 2013 was held from 2
nd
to 7th June, 2013 in La Grande Motte, France. It provided the
NIR community with tremendous opportunities for scientific and social gatherings, with
different social events provided by the organising committee and the Platinum sponsors:
these included happy hours, cocktails around the posters, barbecues and beach parties as
well as the gala dinner, all of them imbued with a very relaxed atmosphere. More than 380
abstracts had been selected by the scientific committee, giving rise to more than 100 oral
presentations, 75 flash presentations and 260 posters (one quarter being involved in a flash
presentation). In addition, an internationally-renowned guest speaker (a total of six) opened the
methodology session on each morning of the conference. Application sessions provided the
opportunity to discover numerous and not only classical uses of NIR spectrometry: greater
attention was given to the most original applications including soil analysis, biomedical use,
unusual applications in ecology, archeology or industry, etc.

In order to allow the NIR community to have access to new scientific knowledge as quickly as
possible, the timetable for production of the conference proceedings was shorter than has been
the norm for previous NIR conferences: 160 full papers were collected before the conference
and submitted to a tutor-type review i.e. advice was given to the author to improve his/her
paper. Editing was carried out during the summer of 2013 and proceedings were made available
on the internet no later than 5 months after the conference. The format has been designed so
that one can download either the whole book (757 pages!) or only particular chapters.

We hope you will enjoy reading these proceedings as much as we did!

Lets meet again in Brazil for NIR2015

Vronique Bellon Maurel, NIR2013 Chair, Editor
Phil Williams, Editor
Gerard Downey, Editor

Proceedings

A1 Agriculture, Environment

Fruits

S. Barzaghi. A portable microNIR instrument as a useful screening tool in discriminating between conservation
conditions applied to preserve pear samples cv Conference. P.135
M. T. Blanco-Diaz. Use of near-infrared reflectance spectroscopy for predicting total phenolic content in zucchini fruits. P.142
S. Bureau. FRUITGRADING: Development of a fruit sorting technology based on internal quality parameters. P.145
M. J. De la Haba. On-tree monitoring of quality evolution in citrus by using a mems-based NIR spectro-photometer. P.149
P. P. Subedi. A bad apple in the barrel? Apple internal defect detection using NIR transmittance spectroscopy. P.153
D. Martinez-Valdiviesoa. Prediction by infrared reflectance spectroscopy of the lutein content in summer squash fruits. P.156
P. Rungpichayapichet. Application of near-infrared spectroscopy to determine the -carotene content of mango. P.161
M. T. Sanchez. In situ monitoring of fruit ripening during posthaverst storage using a miniature handheld NIR sensor. P.167
Z. Schmilovitch. Forecast of apple internal quality indices during storage by spectroscopy. P.171
M. Vanoli. Quality of Brazilian mango fruit in relation to optical properties non-destructively measured by
time-resolved reflectance spectroscopy. P.177
T. zum Felde. Screening for pro-vitamin A components in Cassava (Manihot esculenta) using NIR to support
bio-fortification. P.182

NIR2013 Proceedings, 2-7 June, La Grande-Motte, France. A1 Agriculture and Environment
B e l l o n - M a u r e l V . , W i l l i a m s P . , D o w n e y G . , E d s 1 3 5

A portable microNIR instrument as a
useful screening tool in discriminating
between conservation conditions applied
to preserve pear samples cv Conference.
S.Barzaghi
a
, M. Grassi
b
, F. Beccari
c
, T. M.P. Cattaneo
b

a
Consiglio per la Ricerca e la Sperimentazione in Agricoltura - CRA-FLC, Lodi, Italy
b
Consiglio per la Ricerca e la Sperimentazione in Agricoltura - CRA-IAA, Milan, Italy
c
Diessechem S.r.l., Milan, Italy
Corresponding author: stefania.barzaghi@entecra.it

Introduction
One of the most important fruit characteristics to be
maintained during storage is taste. Furthermore, consumers
usually choose and purchase food, particularly fruit and
vegetables, on the basis of sensory parameters, such as
appearance, colour and glossiness, which are often closely-
associated with sweetness, firmness and nutritional
properties. The use of NIR spectroscopy to measure internal
quality attributes of fruit has been investigated extensively,
even during shelf-life (Cavaco et al., 2010; Prez-Marn et al.,
2010; Bobelyn et al., 2010; Prez-Marn et al., 2011), and the
availability of a new generation of instruments could allow
monitoring and control of fruit quality and the efficiency of
different preservation procedures with simple and fast tools.
The aim of this work was to verify the ability of a diode
array microNIR to discriminate between preservation
procedures applied to preserve fresh pear samples. Different
practices were considered.
Controlled Atmosphere (CA) storage at low
temperatures is the common practice to delay pear ripening
and prolong its commercial availability. As an alternative,
pears can be treated by the ethylene inhibitor 1-
MethylCycloPropene (1-MCP) which has been intensively
studied over recent decades (Watkins, 2006). Since the
discovery of 1-methylcyclopropene (1-MCP) as an inhibitor
of ethylene action, over 100 studies have examined details
of its action, application and effects on ethylene inhibition.
1-MCP prevents ethylene effects in a broad range of fruits,
vegetables and floriculture crops. However, pear is highly
sensitive to 1-MCP (Cucchi and Regiroli, 2011) and many
researchers reported that 1-MCP concentrations which are
able to inhibit scald development can also inhibit fruit
ripening; the final effect can have a negative impact on
consumers because pears are firm, hard and green even
after the end of normal shelf-life (Vanoli et al., 2010;
Villalobos-Acua et al., 2011a ,b). Finally Dynamically

Controlled Atmosphere (DCA) is another innovative and
alternative technique for fruit storage based on non-
destructive chlorophyll fluorescence measurement to
dynamically set O
2
levels slightly above the tolerance level of
the fruit (Zanella et al., 2008). Quality preservation and
control of physiological disorders obtained with DCA in
apples have been very similar to those obtained by 1-MCP
treatment.
Material and Methods
Sample collection
One hundred samples of pears cv Conference were
partitioned into five groups of 20 samples each. All samples
were stored for four weeks at 0.5C. Then, three groups
(60 samples) were treated with 300 ppb of Smart Fresh
SM
1-
MCP (Rohm & Hass, USA) in order to maintain optimal pear
texture; two remaining groups (40 samples) were stored
untreated.
The second step of the storage was thirteen weeks in
air at 0.5C (1-MCP 0.5C ), 1C (1-MCP 1C ) or under
Dynamically Controlled Atmosphere (1-MCP DCA; DCA
conditions of 1%O
2
and 0.3%CO
2
). The untreated pears
were, for the same time, stored in air at 0.5C (U-air) or
under DCA (U-DCA). Before the measurements, all pear
samples had been kept at room temperature for 24 h.

Acquisition of spectra
NIR spectra were recorded by a diode array microNIR
(MicroNIR1700, Diesseinstrument, Italy) over the range 950

to 1650 nm in reflectance mode. Measurements were
performed both on the surface (S) and, after peeling, on the
pear pulp (P).
Data processing
Spectral data of the S and P sample sets were used to
calculate calibration models by applying a Partial Least
Squares Discriminant Analysis (PLSDA) method using the
PLS_Toolbox (Eigenvector Research, Inc USA) software. Only
autoscaling (i.e. mean-centering and scaling of columns to
unit variance) was used as a spectral pre-treatment.
Different classifications were considered in order to
emphasize the contribution of each practice in delaying pear
ripening and spectra were grouped accordingly, as reported
in Table 1.

Table 1. Sample grouping
Treatment Group 1 Group2
A Temperature 1-MCP 0.5C + U-air + U-DCA + 1-MCP DCA 1-MCP 1C
B 1-MCP 1-MCP 0.5C + 1-MCP 1C + 1-MCP DCA U-air + U-DCA
C DCA 1-MCP DCA + U-DCA 1-MCP 0.5C +1-MCP 1C + U-air
D Treated 1-MCP 0.5C + 1-MCP 1C + 1-MCP DCA + U-DCA U-air

The S and P sample sets were divided into two subsets: a calibration sample set (67 pears) and a validation sample set (33
pears) which were used to develop and validate the models. The Kennard-Stone algorithm was applied to select samples to be
included into the calibration and validation sets. Models were developed for both intact (S) and peeled (P) pear samples.

Results and Discussion
Before the measurements, pears had been kept at 20C for
24h so that the temperature inside the fruit could reach the
optimum value at which measurements of quality (hardness,
colour, sensory analysis, chemical analysis) should be
performed. Moreover, this allowed evaluation, by NIR of the
effect of atmosphere on the quality of stored fruits at the
beginning of shelf-life when the differences in appearance
between fruits, stored in DCA or CA, are not so marked.
Upon removal, 1-MCP-treated fruits maintained similar

firmness levels to those at harvest, while a slight softening
happened later; on the contrary, fruit used as a control
group softened quickly, reaching the minimal acceptable
quality level after 7 days of storage at 20C according to
literature data (Chiriboga, et al., 2013).
Partial least squares discriminant analysis (PLSDA) was
applied to discriminate fruits stored under different
conditions. Spectra recorded on intact fruits were quite
different from spectra recorded on pear pulp samples
(Figure 1), so calibrations were calculated separately.


Figure 1. Spectra recorded on intact fruits (red lines) and on pear pulp (blue lines)

Classifying each conservation condition simultaneously, best
performances were obtained analysing the peeled fruits
(Table 2).

Considering the pear samples stored at different
temperatures (Table 1, A), the calculated model allowed
discrimination at the two different temperatures (i.e. -0.5C
and 1C) independently if they had been treated or not with
1-CMP. Classification errors in prediction found for S and P
sets were 0.225 and 0.226 respectively.
Even if the errors in prediction were very close, spectra
recorded on peeled pear samples (P) showed better
performance in calibration. In fact, higher values of
sensitivity and specificity were found for spectra recorded
for P samples than for spectra obtained from intact fruits (S).
Sensitivity is defined as the number of samples predicted as
elements of one class divided by the samples actually
belonging to that class while specificity is defined as the
number of samples predicted as different and not included
into one class divided by the actual samples out of the same
class. Statistical figures are reported in Table 3.

Conversely, applying PLSDA in order to discriminate
between pear samples treated or not with MCP (Table 1, B),
best results were obtained using spectra of the skin of intact
samples (S). Prediction errors of the models calculated for
discriminating pears treated with MCP from the untreated
samples were 0.182 in the case of intact fruits and 0.323 for
the pulp sample set, as shown in Table 4. These results could
be partly explained by the fact that MCP decreased colour
development in the fruit (Vanoli et al., 2010).


Table 2. PLS-DA performances in discriminating pears stored under different conditions

Intact pears (S) Pear pulp (P)
8LV 10 LV
Sensitivity Specificity Class Err RMSE Sensitivity Specificity Class Err RMSE
1-MCP
1C
cal 0.846 0.704 0.225 0.357 1.000 0.927 0.036 0.229
CV 0.692 0.704 0.302 0.416 0.833 0.873 0.147 0.323
pred 0.571 0.769 0.330 0.383 0.625 0.885 0.245 0.331
1-MCP
0.5C
cal 0.933 0.788 0.139 0.307 0.857 0.925 0.109 0.288
CV 0.800 0.788 0.206 0.402 0.500 0.755 0.373 0.424
pred 1.000 0.786 0.107 0.372 0.857 0.926 0.108 0.301
U-air
cal 1.000 0.963 0.019 0.259 1.000 0.911 0.045 0.241
CV 0.846 0.852 0.151 0.315 0.636 0.875 0.244 0.353
pred 0.286 0.808 0.453 0.384 1.000 0.833 0.083 0.341
1-MCP
DCA
cal 0.867 0.885 0.124 0.331 1.000 0.900 0.050 0.282
CV 0.667 0.788 0.272 0.463 0.882 0.900 0.109 0.343
pred 0.400 0.821 0.389 0.348 0.000 0.938 0.531 0.290
U-DCA
cal 0.818 0.750 0.216 0.305 0.923 0.889 0.094 0.272
CV 0.636 0.696 0.334 0.377 0.615 0.778 0.303 0.380
pred 0.778 0.792 0.215 0.357 0.714 0.815 0.235 0.361

Table 3. PLS-DA performances in discriminating pears stored at different temperatures (Table 1, A)

Intact pears (S) Pear Pulp (P)

P
r
e
p
r
o
c
e
s
s

L
V

-
0
.
5
C

1
C

P
r
e
p
r
o
c
e
s
s

L
V

-
0
.
5
C

1
C

Sensitivity/Specificity
(Cal)
a
u
t
o
s
c
a
l
i
n
g

6
0.867/0.827 0.827/0.867
a
u
t
o
s
c
a
l
i
n
g

5

1.000/0.891 0.891/1.000
(CV)
0.800/0.827 0.827/0.800 0.917/0.891 0.891/0.917
(Pred)
0.800/0.750 0.750/0.800 0.625/0.923 0.923/0.625
Class. Err (Cal) 0.153 0.153 0.054 0.054
Class. Err (CV) 0. 186 0. 186 0.096 0.096
Class. Err (Pred) 0. 225 0. 225 0.226 0.226
RMSEC 0.308 0.308 0.270 0.270
RMSECV 0.378 0.378 0.311 0.311
RMSEP 0.397 0.397 0.357 0.357


Table 4.PLS-DA performances in discriminating pears treated with 1-MCP (Table 1, B)

Intact pears (S) Pears Pulp (P)

P
r
e
p
r
o
c
e
s
s

L
V

1
-
M
C
P

u
n
t
r
e
a
t
e
d

P
r
e
p
r
o
c
e
s
s

L
V

1
-
M
C
P

u
n
t
r
e
a
t
e
d

Sensitivity/Specificity (Cal)
a
u
t
o
s
c
a
l
i
n
g

7
0.958/0.907 0.907/0.958
a
u
t
o
s
c
a
l
i
n
g

9
0.917/0.884 0.884/0.917
Sensitivity/Specificity (CV) 0.792/0.860 0.860/0.792 0.500/0.721 0.721/0.500
Sensitivity/Specificity (Pred) 0.813/0.824 0.824/0.813 0.765/0.588 0.588/0.765
Class. Err (Cal) 0.067 0.067 0.100 0.100
Class. Err (CV) 0.174 0.174
0.389 0.389
Class. Err (Pred) 0.182 0.182 0.323 0.323
RMSEC 0.308 0.308 0.315 0.315
RMSECV 0.440 0.440 0.492 0.492
RMSEP 0.403 0.403 0.544 0.544

The use of DCA (Table 1, C) in preserving fruits during
shelf-life showed, in general, the best possibility to be
discriminated. Very satisfactory models were obtained. The
first model, calculated by processing spectra recorded on the
surface (S), showed a CV sensitivity of 0.805, a CV specificity
of0.808 and a RMSECV equal to 0.414 using 9 latent
variables (LVs). The second, calculated on the peeled sample
set (P), showed a CV sensitivity of 0.889, a CV specificity of
0.871 and a RMSECVof 0.315 using 7 LVs. Classification
errors in prediction obtained considering the independent
test sets were 0.16 and 0.02 respectively. Statistical figures
are reported in Table 5.

The best results to evaluate NIR ability for the
discrimination of pear samples stored in air at 0.5 C (Table
1, D), were obtained by analysing intact pears (S). In fact, the
best calibration performances were found for this set of
samples, as shown in Table 6. Higher values of sensitivity
and specificity, both in calibration and cross- validation,
were achieved in comparison with the values obtained
analysing the pear pulp set of spectra (P). Moreover, even if
the classification error in prediction was lower for the
spectra recorded on the pear pulp set (0.044 vs 0.061), the
RMSEP value was higher than that obtained for spectra
recorded on the fruit surface (S).

0.389 0.389

Table 5. PLS-DA performances in discriminating pears stored in DCA (Table 1, C)


P
r
e
p
r
o
c
e
s
s

L
V

D
C
A

A
i
r

P
r
e
p
r
o
c
e
s
s

L
V

D
C
A

A
i
r

a
u
t
o
s
c
a
l
i
n
g

9
0.927/1.000 1.000/0.927
a
u
t
o
s
c
a
l
i
n
g

7
0.944/1.000 1.000/0.944
Class. Err (Cal) 0.036 0.036 0.028 0.028
Class. Err (CV) 0.194 0.194 0.136 0.136
Class. Err (Pred) 0.160 0.160 0.020 0.020
RMSEC 0.276 0.276 0.247 0.247
RMSECV 0.413 0.413 0.314 0.314
RMSEP 0.339 0.339 0.253 0.253

Table 6. PLS-DA performances in discriminating pears stored in air without any treatment (Table 1, D)


P
r
e
p
r
o
c
e
s
s

L
V

t
r
e
a
t
e
d

u
n
t
r
e
a
t
e
d

P
r
e
p
r
o
c
e
s
s

L
V

t
r
e
a
t
e
d

u
n
t
r
e
a
t
e
d

a
u
t
o
s
c
a
l
i
n
g

8
0.900/0.915 0.915/0.900
a
u
t
o
s
c
a
l
i
n
g

6
0.762/0.891 0.891/0.762
Class. Err (Cal) 0.092 0.092 0.173 0.173
Class. Err (CV) 0.164 0.164 0.217 0.217
Class. Err (Pred) 0.061 0.061 0.044 0.044
RMSEC 0.281 0.281 0.351 0.351
RMSECV 0.345 0.345 0.429 0.429
RMSEP 0.272 0.272 0.295 0.295

Conclusion

NIR was demonstrated to be a useful tool in
discriminating between different fruit storage treatments. In
particular, very satisfactory results were obtained for the
discrimination of fruits stored under DCA. The best
performances were obtained by analysing peeled fruits but
measurements made on the fruit surface also gave
satisfactory statistical figures. This potentially suggests the
possibility of using NIR for non-invasive fruit quality
screening. NIR was also able to discriminate between
different preservation temperatures, and to identify when
the 1-MCP treatments were applied, even though if
predictive performances were less satisfactory than was the
case for the use of DCA.
These results were obtained using a portable
instrument of new generation, the smallest one now
available on the market. It is characterised by some key
features: ultra-compact, lightweight, robust with no moving
parts, USB powered, and highly customisable. Less precision
is in general expected from portable instruments but the
results obtained in this study suggested the use of microNIR
apparatus for fruit quality screening during storage at
different stages of the post-harvest chain. Further studies
can confirm the results obtained, suggesting the potential
use of NIR for the easy control of fruit quality also at the
GDO level.

Acknowledgements

Authors thank the CRA-IAA team of Dr.ssa Anna Rizzolo
for the collection, treatment and storage of pear samples.

References

Bobelyn, E., Nicolai, B.M., Saeys, W., Lammertyn, J., Serban,
A.S. and Nicu, M. (2010). Postharvest quality of apple
predicted by NIR-spectroscopy: Study of the effect of
biological variability on spectra and model
performance. Postharvest Biol. Tech., 55, (3), 133-143.
Cavaco, A.M., Guerra, R., Marques da Silva, J. and Antunes
M.D.C. (2010). A preliminary approach to the
prediction of 'Rocha' pear skin pigments by Vis/NIR
Reflectance Spectroscopy. Acta Hortic. , 858, 373-377.
Chiriboga, M.-A., Schotsmans, W. C., Larrigaudire, C.,
Dupille, E. and Recasens, I. (2013). Responsiveness of
Conference pears to 1-methylcyclopropene: the role
of harvest date, orchard location and year. J. Sci. Food
Agric., 93, 619625.
Cucchi, A. and Regiroli, G. (2011). Temperature and
ethylene: two useful tools to be used in combination
with smartfreshsm (1-MCP) for delivering optimal
quality pears. Acta Hortic. , 909, 679-686.
Prez-Marn, D., Garrido-Varo, A., Snchez, M.T., Paz, P. and
Guerrero, J.E. (2010). Miniature handheld NIR sensor
for the on-site non-destructive assessment of post-
harvest quality and refrigerated storage behavior in
plums. J. Food Eng., 99(3), 294-302.
Prez-Marn, D., Snchez, M.T., Paz, P., Gonzlez-Dugo, V.
and Soriano, M.A. (2011). Postharvest shelf-life
discrimination of nectarines produced under different
irrigation strategies using NIR-spectroscopy. LWT -
Food Science & Technology, 44,(6), 1405-1414.
Villalobos-Acua, M. G., Biasi, W.V., Flores, S., Jiang, C.Z.,
Reid, M.S., Willits, N.H. and Mitcham, E.J. (2011a).
Effect of maturity and cold storage on ethylene
biosynthesis and ripening in Bartlett pears treated
after harvest with 1-MCP. Postharvest Biol. Techl.,
59(1), 1-9.
Villalobos Acua, M.G., Biasi, W. V., Mitcham, E. J. and
Holcroft, D. (2011b). Fruit temperature and ethylene
modulate 1-MCP response in Bartlett pears.
Postharvest Biol. Tech., 60(1), 17-23.
Vanoli, M., Rizzolo, A., Grassi, M. and Eccher Zerbini, P.
(2010). Ethylene Production and Quality in 1-
Methylcyclopropene Treated Abb Ftel Pears after
Storage in Dynamically Controlled Atmosphere. Acta
Hortic., 876, 31-38.
Watkins, C. B. (2006). The use of 1-methylcyclopropene (1-
MCP) on fruits and vegetables. Biotechnol. Adv., 24(4),
389-409.
Zanella, A., Cazzanelli, P. and Rossi, O. (2008). Dynamic
controller atmosphere (DCA) storage by the means of
chlorophyll fluorescence response for firmness
retention in apple. Acta Hortic., 796, 77-82.

Use of near-infrared reflectance
spectroscopy for predicting total phenolic
content in zucchini fruits.

M. T. Blanco-Daz
a
, M. Del Ro-Celestino
b
, D. Martnez-Valdivieso
b
, J. M.
Moreno
c
, F. Pea-Rodriguez
c
, R. Font
a
a
Department of Postharvest Technology of Horticultural Crops, Centro IFAPA Centro La Mojonera. La Mojonera, Almera, Spain.

b
Department of Plant Breeding and Biotechnology, Centro IFAPA La Mojonera, , La Mojonera, Almera, Spain
c
Department of Postharvest Technology of Horticultural Crops, Centro IFAPA Alameda del Obispo Avda. Crdoba, Spain.

Corresponding author: g32bldim@hotmail.com

Introduction

Nutritional characterisation of fruits and vegetables is a
subject of great importance in food industries and also for plant
breeders. Zucchini is an important vegetable in the main area
of production of vegetables in the EU - Almera province in
southern Spain. Total phenolic content (TPC) is a time-
consuming analytical technique due to the multiple steps
needed. In addition, many fruits have to be analysed in plant
breeding programmes to select those showing the required
TPC. For decades, near-infrared (NIR) reflectance spectroscopy
has been used as a rapid and cost-effective technique for
determining multiple components of organic matrices.
However, no references have been found in the scientific
literature to NIR for determining TPC in zucchini fruits, in spite
of this being an important issue for breeding programmes and
the food industry. In this work, we present the potential of this
technique for predicting the total phenolic content in zucchini
fruit (Cucurbita pepo ssp. pepo).

Material and method

In this study, a total of 499 samples of zucchini fruit (237 for
exocarp and 262 for mesocarp) were collected from 14
accessions of the Centre IFAPA La Mojonera (Almera, Spain)
together with 17 commercial cultivars. Fruits were harvested
from consecutive seasons (2009 to 2012).

Plant material

Summer squash seeds were germinated and transplanted into
rockwool cubes (Grodan BV, 6040KD Roermond, NL) in a
greenhouse (36 46' N, 2 48' O). Plants were grown in a
greenhouse in Centro IFAPA La Mojonera (Almera, Spain)
following standard local cultural practices for plant nutrition,
insect pests and disease control. Each cultivar was
represented by at least three fruits and three samples were
taken from each fruit. About 100 g of fresh summer squash
samples was freeze-dried (Telstar, Terrasa, Spain) until constant
weight. Then, samples were ground with an industrial mill.

Total phenolics analysis

Reference values for TPC were obtained by the Folin-Ciocalteu
reagent method (Sellappan et al., 2002 with some
modifications) based on the standard calibration curve of gallic
acid using a UV-Visible spectrometer (Perkin Elmer).

NIRS analysis

A total of 237 exocarp and 262 mesocarp samples of zucchini
were first freeze-dried and then ground in a mill to a fine
powder. NIR spectra were obtained using a near-infrared
spectrophotometer (NIRSystems model 6500, Foss NIR
Systems, Inc., Silver Spring, MD, USA) in reflectance mode over
a wavelength range from 400 to 2500 nm (visible and near-
infrared regions). Absorbance values (log 1/R, where R is
reflectance) were registered at 2 nm intervals. Calibration
equations for total phenolic content in exocarp and mesocarp
were developed using GLOBAL programme (WINISI II, v. 1.50;
Infrasoft International, LLC, Port Matilda, PA, USA). In a second
step, 137 samples for exocarp and 158 for mesocarp were
selected for calibration; selection was made following a PCA
analysis on the basis of spectral variability (Mahalanobis
distance) representing the whole range in the spectral
collection. Thus, exocarp and mesocarp samples did not
necessarily belong to the same fruits, as they were considered
as independent populations. Calibration equations were
computed using the raw optical data (log 1/R, where R is
reflectance) or first or second derivatives of the log 1/R data,
with several combinations of derivative (gap) sizes and
smoothing (i.e. (1,4,4,1; derivative order, segment of the
derivative, first smooth, second smooth); (1,5,5,1);(1,10,5,1);
(2,5,5,1); (2,5,5,2); (2,10,5,1)). Wavelengths from 400 to 2500
nm every 8 nm were used to develop the calibration equations.
Modified partial least square (MPLS) regression was used to
correlate spectral information and TPC content in the samples.
Those samples identified as outliers were removed in two
elimination passes (Shenk and Westerhaus, 1991).
The final objective of the mathematical procedure
was to reduce the high number of spectral data points
(absorbance values from 400 to 2500 nm every 2 nm, i.e. 1050
data) and to eliminate the correlation of absorbance values
presented by neighbouring wavelengths (Shenk and
Westerhaus, 1994). Standard normal variate and detrend
transformations (SNV-DT; Barnes et al., 1989) were used to
correct baseline offset due to scattering effects (differences in
particle size between samples). Cross-validation was performed
on the calibration set for determining the appropriate number
of terms to use in the equation, as well as to determine the
ability of each equation to predict unknown samples (Shenk et
al., 2001). Prediction error shown by the different equations in
normal cross-validation was computed as the standard error of
cross-validation (SECV). This statistic was also used for
computing the optimum number of MPLS terms to be
supported in the calibration model (Shenk and Westerhaus,
1991). Additional discussion on the SECV statistic has been
reported by Shenk and Westerhaus (1991). Cross-validation
was also used to identify samples which were chemical (t) or
spectral (H) outliers. t outliers are samples that have a
relationship between their reference values and spectra that is
different from the relationship of the other samples in the set
and have large residuals (t values > 2.5). Samples with a large t
statistic often place the reference chemistry value in doubt. A H
outlier identifies a sample that is spectroscopically different
from other samples in the population and has a standardised H
value > 3.0.

Results and discussion

Table 1 shows the composition in the TPC of the
samples used in the calibration. Mean and S.D. (standard
deviation) values of the TPC were similar for exocarp and
mesocarp of the fruits. The range of TPC was similar in
mesocarp to that in exocarp which showed values from 1.34 to
7.11 mg/g dw and from 0.97 to 8.29 mg/g dw,
respectively.These results are similar to those observed in
pumpkin (Oloyede et al., 2012). However, other authors have
reported slightly higher total phenolic content in fruits of the
Cucurbitaceae family (Mongkolsilp et al., 2004).

Table 1. Descriptive statistics for total phenolic content (mg/ g dry weight basis) in calibration samples .
Sample n Mean S.D. SEL Min Max
Exocarp 122 3.97 1.34 0.20 0.97 8.29
Mesocarp 150 3.54 1.37 0.03 1.34 7.11
S.D.: standard deviation; SEL.: standard error of laboratory; Min.: minimum value; Max.: maximum value

Equations selected to have the best predictive
ability were those showing the highest coefficient of
determination in cross-validation (R
2
cval
) and the
highest ratio of standard deviation to standard error
of cross-validation (RPD; Williams, 1987). These
latter two statistics enable SECV to be standardised,
facilitating a comparison of results obtained with
sets of different means.
Table 2 shows the prediction abilities of the
equations developed for predicting TPC in exocarp
and mesocarp zucchini fruits. On the basis of the
above statistics, prediction of TPC in the mesocarp of
the fruit was better than in the exocarp. For exocarp
and mesocarp samples, the use of the second
derivative transformation (2,10,5,1) yielded the
highest prediction ability equation in mesocarp, with
an R
2
cval
of 0.76 and RPD of 2.07. NIR can be used for
predicting low, medium and high values of TPC in the
exocarp with an R
2
cval
of 0.61 and RPD of 1.60 in
second derivative transformation.

Table 2. Statistics of best calibration for total phenolic content in exocarp and mesocarp of zucchini
squash using MPLS regression.
Sample Mathematic
treatment
Calibration Cross-Validation
R
2
cal
SEC MPLS
terms
R
2
cval
SECV RPD
Exocarp
(n=122)
2,10,5,1 0.72 0.70 10 0.61 0.83 1.60
Mesocarp
(n= 150)
2,10,5,1 0.83 0.56 11 0.76 0.66 2.07
R
2
cal
: coefficient of determination in the calibration; SEC: standard error of calibration; R
2
cval
:
coefficient of determination in the cross-validation; SECV: standard error of cross-validation; RPD: ratio
of the standard deviation of the laboratory data SD to SECV.

Scientific literature (Williams and Sobering, 1996)
reported interpretation of RPD values as follows: values
below 1.5 are considered unusable, values of between
1.5-2.0 can be used for rough predictions, those
between 2.0-2.5 allow approximate quantitative
predictions to be made, while values above 2.5-3.0 are
considered to be good and excellent predictive models.
On the basis of such considerations, in our study, these
ratios confirm that accuracies of the developed Vis/NIR
models are suitable for approximate, quantitative
prediction screening of TPC content in mesocarp (RPD
equal to 2.07) and for the rough screening of TPC
content in exocarp (RPD equal to 1.60). RPD for TPC in
zucchini mesocarp was similar to values obtained in
blueberries (RPD of 2.05; Sinelli et al., 2008) while, for
exocarp, the RPD obtained was lower than that shown
by the mesocarp in our study, which could be due to the
characteristics of the exocarp matrix. Considering the
R
2
cval
obtained for TPC in the mesocarp, the equations
obtained showed good correlation (R
2
cval
= 0.76) and a
correct separation of low, medium and high values in
exocarp zucchini samples (Shenk and Westerhaus, 1996).

Conclusion

NIR is a technique highly-suited for the
measurement of quality attributes in fruit products as it
is rapid and enables the simultaneous determination of
several parameters using a single measurement. In this
work, we studied the potential of NIR for predicting total
phenolic content (TPC) in zucchini fruits. Equations
obtained for the prediction of TPC in the mesocarp of
zucchini showed good accuracy (Shenk and Westerhaus,
1996). For exocarp, although the RPD value for the
equation was near to the limit reported by Williams and
Sobering (1996), the R
2
value shown in the cross-
validation was included in those values reported by
Shenk and Westerhaus (1996) for the correct separation
of samples into low, medium and high values. On the
basis of these results, NIR can be used in plant breeding
programmes and the food industry to minimise labour
input in the laboratory. Once selected, those samples
showing the required TPC can be analysed by standard
techniques.

Acknowledgements

This work was supported by project RTA2009-
00036-00-00 (INIA) and FEDER funds. M.T. Blanco-Daz
acknowledges financial support from the Spanish
Ministry of Science and Innovation as a fellow of the
Program Training to Personal of Research (Formacin
de Personal Investigador, FPI-INIA).

References

Barnes, R.J., Dhanoa, M.S. and Lister, S.J. (1989).
Standard normal variate transformation and
detrending of near-infrared diffuse reflectance
spectra. Appl. Spectrosc., 43, 772-777.
Mongkolsilp, S., Pongbupakit, I., Sae-Lee, N. and
Sitthithaworn, W. (2004). Radical Scavenging
Activity and Total Phenolic Content of Medicinal
Plants Used in Primary Health Care SWU. J.
Pharm.Sci., 9(1), November.
Oloyede, F.M., Agbaje, G.O., Obuotor, E.M. and
Obisesan, I.O. (2012). Nutritional and antioxidant
profiles of pumpkin (Cucurbita pepo Linn.)
immature and mature fruits as influenced by NPK
fertiliser. Food Chem., 135, 460463.
Sellappan, S., Akoh, C.C. and Krewer, G. (2002). Phenolic
compounds and antioxidant capacity of Georgia-
grown blueberries and blackberries. J. Ag. Food
Chem., 50, 24322438.
Shenk, J.S. and Westerhaus, M.O. (1991). Population
structuring of near infrared spectra and modified
partial least squares regression. Crop Sci., 31,
1548-1555.
Shenk, J.S and Westerhaus M.O. (1994). The application
of near infrared reflectance spectroscopy (NIRS) to
forage analysis, in Forage Quality, Evaluation and
Utilization, ed. by Fahey GC, Collins M, Mertens DR
and Moser LE, American Society of Agronomy,
Crop Science Society of America, Soil Science
Society of America, Madison, WI, USA. pp. 406
450.
Shenk, J.S. and Westerhaus, M.O. (1996). Calibration the
ISI way. In: Near Infrared Reflectance
Spectroscopy. The Future Waves, Davies, A.M.C.
and Williams, Ph. (eds). NIR Publications,
Chichester, UK, pp. 198-202.
Shenk J.S, Workman, J. and Westerhaus, M. (2001).
Application of NIR spectroscopy to agricultural
products, in Handbook of Near Infrared Analysis,
2nd Edition, ed. by Burns DA and Ciurczac EW.
Marcel Dekker, Nueva York, USA, pp. 419-474.
Sinelli, N., Spinardi, A., Di Egidio, V., Mignani, I. and
Casiraghi, E. (2008). Evaluation of quality and
nutraceutical content of blueberries (Vaccinium
corymbosum L.) by near and mid-infrared
spectroscopy. Postharvest Biol. Technol., 50, 31
36.
Williams, P.C. (1987). Variables affecting Near-Infrared
Reflectance Spectroscopy Analysis. In: Near
Infrared Technology in the Agricultural and Food
Industries. AACC Inc., St. Paul, MN, pp. 145169.
Williams, P.C. and Sobering, D.C. (1996). How do we do
it: a brief summary of the methods we use in
developing near infrared calibrations. In: Near
Infrared Spectroscopy: The Future Waves; Davies,
A.M.C., Williams, P.C., (Eds).; IM Publications,
Chichester, U.K. pp. 185-188.

NIR2013 Proceedings, 2-7 June, La Grande-Motte, France. A1- Agriculture and Environment

FRUITGRADING: Development of a fruit
sorting technology based on internal
quality parameters.
Bureau S.
a
, Bertrand D.
b
, Jaillais B.
c
, Reling P.
a
, Gouble B.
a
, Renard
C.M.G.C.
a
, Dekdouk B.
d
, Marsh L.A.
d
, OToole M.D.
d
, Armitage, D.W.
d
,
Peyton A.J.
d
, Alvarez-Garcia J.
e

a
INRA, UMR408 Scurit et Qualit des Produits d'Origine Vgtale, F-84000 Avignon, France.
a
Universit d'Avignon et des Pays de Vaucluse, UMR408, F-84000 Avignon, France.
b
Data frame, 44300 Nantes, France.
c
INRA, UR1268 Biopolymres Interactions et Assemblages, F-44000 Nantes, France.
d
The University of Manchester, School of EEE, Manchester, M13 9PL, UK.
e
CRIC, S.A, Lenz-Instruments S.L. Barcelona, Espaa.

Corresponding author: Sylvie.bureau@avignon.inra.fr
Introduction
Fruit sorting is currently based on morphological and basic
physical characteristics (size, shape, and weight), and external
appearance (colour, presence of surface defects, etc.). These
parameters are not totally representative of the product quality,
which encompasses not only external appearance, but also other
sensory properties (taste and aroma), nutritional value, chemical
constituents, mechanical properties, functional properties and
defects.
The European project, Fruitgrading (FP7-SME, 2011-2014),
presented in this paper, aims to develop a new technology to
grade fruits based on internal quality. The proposed technology is
based on the combination of two non-contact techniques:
1) Near-infrared spectroscopy (NIRS): One of the most
widely studied quality assessment tools for the last twenty years
has been NIRS. This technology is based on the analysis of the
optical properties of fruits, within the NIR spectral range (750 nm-
2500 nm). Due to the need of very high intensities required in
transmittance mode for large objects such as fruits, we decided to
work in reflectance mode. Indeed the latter is commonly used to
estimate soluble solids contents (SSC) in fruits (Nicola et al., 2007)
and it can be easily integrated into existing fruit sorting lines. Over
the last twenty years, several NIR investigations were performed
on a wide range of fruits, including apple, apricot, cherry, citrus,
grape, guava, kiwifruit, mandarin, mango, melon, nectarine,
papaya, peach, pear, pineapple and plum (Nicola et al., 2007 ;
Woodcock et al., 2008, Bureau, 2009). In general, for SSC, good
correlation indicators are found (RMSEP is around 0.5%), though in
external validation (other orchards and/or years) the RMSEP is not
so good (1-1.5%).
2) Magnetic Induction Spectroscopy (MIS): Internal
changes in fruits are always accompanied by physico-chemical
changes in the fruit tissue, which result in variations of the water
content, organic acids, sugars, etc. These internal changes modify
the electrical conductivity of the fruits. Several published studies

have proved that electrical conductivity can be used for identifying
internal changes related to the ripening process in a variety of
fruits including avocados, watermelon, strawberry, apricots,
bananas, and apples (Barai et al., 2012 ; Bauchot et al., 1999 ;
Rehman et al., 2011 ; Harker and Maindonald, 1994). Most
recently, improvements in the instrumentation have allowed
attaining correlation coefficients for firmness and TA in Gala apple
as high as r = 0.986 and r = 0.934, respectively. Nonetheless, a
major limitation of conventional methods for measuring the
conductivity of fruit deals with the variability of the measurements
introduced by using electrodes, and the fact that inserted
electrodes damage the fruit. The contactless MIS technique
measures the electrical properties of the sample by exploiting the
principles of electromagnetic induction. No published data exists
referring to use of MIS for the analysis of fruits or vegetables.
This paper presents the first results concerning the prediction of
several quality traits such as soluble solids content (SSC), titratable
acidity (TA) and flesh penetrometry value (FP) by NIRS and MIS
obtained on four different fruits, apple, pear, peach, and kiwi.
Materials and methods
Plant materials
Fruit from 4 different species were individually analyzed by
NIRS, MIS and reference measurements, representing in total 135
kiwis, 236 pears, 350 apples and 378 peaches and nectarines. The
different samples were provided by the SME-AGs, AJAP in
Portugal, SOTO in Spain and TRSK in Poland; the origins, varieties,
maturities and quality classes are detailed in Table 1.

Table 1: Description of the characterized fruit samples.

Non-destructive characterization of fruits
NIRS
NIRS spectra were recorded on a multi-purpose analyzer
(MPA) spectrometer (Bruker Optics, Wissembourg, France)
equipped with an integrating sphere to provide diffuse reflectance
measurements and a TE-InGaAs detector. The MPA was
completely software-controlled by the OPUS software Version 5.0
which was provided by Bruker Optics. The NIR spectra were
obtained by taking the average of 32 scans. They were acquired
between 800-2500 nm at 2 nm spectral resolution, with scanner
velocity of 10 KHz and a background of 32 scans. The time
required to achieve a spectral measurement was 30 s. The intact
fruits were placed on an automated 30-position sample wheel,
each position corresponding to an 18 mm diameter hole. Two
diffuse reflectance spectra were measured on the opposite sides
of each fruit piece.

MIS
MIS uses a magnetic field to scan fruit samples. A transmit
coil produces a primary magnetic field,
, to which the fruit

sample is exposed. This field induces circulating eddying currents
within the sample which in turn produce a secondary field,
.
This secondary field is dependent on the conductivity of the
sample, which is suspected to vary as fruit ripens. A second coil
can be used to measure the field,
, and thus is able to

observe the effect of
.
Two different MIS-based systems were used to record
spectral data for the fruit samples. The first system consisted of a
conveyor belt passing through two cascaded gradiometers
(Mettler Toledo-Safeline Limited, Salford, UK); the second
consisted of a Solartron 1260a impedance analyzer (Solartron
Group Limited, Farnborough, UK) which was connected to a pair of
coils which formed a solenoid-style arrangement into which the
fruit samples could be placed.
In the case of the first system, each gradiometer was
operated at a single frequency, and fruit samples were passed
backwards and forwards through the pair of machines providing a
total of four passes. After each pass the frequency of the machines
was updated, thereby providing a spectrum consisting of eight
different frequencies in the range of 10kHz-1MHz.
For the second system fruit samples were placed into the
centre of a pair of coils, one of which was wound as a solenoid,
and the other of which was wound with a similar geometry, but
had a single crossover point at the half-way position along the
overall length of the solenoid, so as to cancel the induced
background signal produced by the first coil. The impedance
analyzer was operated at a total of 11 frequencies, thereby
producing three extra spectral points when compared with the
conveyor system.

Reference measurements
The skin color (un-blushed) was determined using a CR-400
chromameter (Minolta, Osaka, Japan) and expressed in the CIE
1976 L*a*b* color space (illuminant D65, 0 view angle,
illumination area diameter 8 mm). After the non-destructive
measurements, the flesh firmness was determined by a
penetrometry test, using a TAPlus analyzer (LLOYD Instruments)
equipped with a flat tipped probe (2 mm diameter, 10 mm
penetration depth) after removing the skin. Two measurements
per fruit were realized on the two opposite sides. Flesh firmness is
expressed as the work to limit parameter in J. Fruits were then
cut and frozen at 20C for a few days until biochemical analyses.
Fruit pieces were homogenised with an Ultraturrax T25 equipment
(Ika Labortechnik, Staufen, Germany) to obtain a puree. Soluble
solids content was determined with a digital refractometer (PR-
101 ATAGO, Norfolk, VA) and expressed in Brix at 20C. Titratable
acidity was determined by titration up to pH 8.1 with 0.1N NaOH
and expressed in meq/100g of fresh weight (= mmol H
+
/100g of
Species Origins Varieties Date of
reception
Remarks
13-sept
18-sept
08-oct
Fuji 29-oct
SOTO Gala 28-sept 2 ripening stages
Gala 15-oct
Ligol 15-oct 2 quality classes
Idared 15-oct
(Class I and
Premium)
Sampion 15-oct
Earligreen,
Verde-
Precoce 20-sept
Soreli,
Amarela 20-sept
Hayward 09-nov
Blanquilla 24-avr
Two ripening
stages
24-avr
except for
Moretini
10-mai
Moretini 27-juil
Limonera 03-aot
Williams 22-aot
AJAP Rochas 11-mai
Louise
Bonne 12-oct
Red
Williams 12-oct
Abat 12-oct
Rochas 12-oct
03-aot
22-aot
Sweet
Dreams 22-aot
Rommer
Star 22-aot
Alba Red 25-sept
Miraflores 04-oct
Tardibell 04-oct
27-juil
03-aot
03-aot
22-aot
Venus 22-aot
2 dates of analysis
after storage at
10C
Nectarine SOTO
Big Top
Luciana
Pear
SOTO
Conference
Supermarket
, near
Avignon
Peach SOTO
Summer
Rich
Apple
AJAP
Bravo de
Esmolfe
TRSK
Kiwi AJAP
3 dates of analysis
after storage at
10C

fresh weight) using an autotitrator (Methrom, Herisau,
Switzerland).
Data treatment
During the building up of models, signals of the same fruit were
averaged to take into account the fruit heterogeneity. Calibration
models were calculated for each variety of fruit and each quality
trait. Due to the possibility of a small number of spectra being
incorrectly recorded, data were pre-processed by the application
of Principal Component Analysis (PCA) to the NIR spectra. A
procedure of outlier elimination was applied, whereby up to 1% of
the total number of spectra can be removed, as a function of the
distance between them and the other spectra. Predictive models
were carried out using PLS1 regression with the spectral data as
predictive variables and the individual fruit quality parameters as
predicted variable. To appreciate the accuracy of the results, the
following calibration procedure was developed: for each fruit, the
data collection was randomly divided into 3 sets: i) dataset for
training (50% of spectra), ii) dataset for tuning (25% of spectra)
and iii) dataset for validation (25% of spectra).
The observed physical and biochemical traits, skin color, flesh
firmness, SSC and TA, are presented (Table 2).
The samples covered a large range of characteristics for each
of the four species, apple, kiwi, pear and peach and nectarine.
Fruits had been chosen to cover most of the possible range of
quality traits, by analyzing different varieties, and for several
varieties, different ripening stages or different quality classes. For
example the ratio between minimum and maximum was higher
than 2 for SSC in all species, and covered most of the acceptable
commercial range for these products, while for TA the ratio
between minimum and maximum was higher than 5 for apple and
peach/nectarine.
Prediction models were established between NIR spectra and
quality parameters (flesh firmness, soluble solids content,
titratable acidity), and the results are given in Table 3. These
results confirmed that NIRS is a useful technique to estimate SSC
in different fruits and particularly in apple, kiwi, pear and peach
and nectarine (R
2
0.72 and RMSE1.30). Expressed in %, the error
of prediction represents 5.6% for apple, 6.3% for pear, 8.1% for
peach and 10.3% for kiwi, even when very different fruits
(different origins and cultivation methods) are analyzed together.

However, concerning the other quality parameters, the results
are dependent on fruit. In fact, the prediction results are
acceptable for TA only in pear (R
2
=0.74 and RMSE=0.84) and for
firmness only in apple (R
2
=0.76 and RMSE=0.0034).
For samples scanned by the MIS systems it has been possible
to identify the expected response of the different types of fruit.
Figure 1 shows the mean and standard deviation for 319 scanned
apples. It is possible to observe a negative trend with respect to
increasing frequency for both measurement systems, which
indicates a possible dispersion in the conductivity of the fruit.
Figure 2 shows a similar response for peaches. For both fruits it is
possible to identify that the results from the solenoid-based
system contain less noise; as a result of this the trend of the data
is much easier to see.
Further measurements and development of the system is on-going
with a view to improving the signal-to-noise ratio of
measurements. It is believed that by understanding the trends of
fruit samples it will be possible to identify changes in the
conductivity of the samples as the fruit matures. It should then be
possible to sort the fruit according to expected quality using this
method.

a

b

Figure 1. Resistive components for apples for (a) the conveyor-
based system and (b) the solenoid based system (n=319)

a

b

Figure 2. Resistive components for peaches for (a) the conveyor-
based system (n=105) and (b) the solenoid based system (n=97)

Mean
+/- 1 Standard Deviation


Table 3. Results of prediction models for SSC, TA and flesh
firmness in peach and nectarine, pear, apple and kiwi.

*Means of 20 times
RMSE: root mean square error; LV: latent variable

Conclusion

With the aim of developing a new technology to grade fruits based
on internal quality, two contactless techniques, near-infrared
spectroscopy (NIRS) and magnetic induction spectroscopy (MIS),
have been tested on four fruit species. A large number of fruits,
135 kiwis, 236 pears, 350 apples and 378 peaches and nectarines
were individually analyzed by NIRS, MIS and classical reference
techniques to determine flesh firmness, soluble solids content and
titratable acidity. Whereas, NIRS was confirmed to be a useful
technique to estimate SSC in different fruits (R20.72 and
RMSE1.30), NIRS allowed marginally acceptable predictions of TA
only in pear (R2=0.74 and RMSE=0.84) and of flesh firmness only
in apple (R2=0.76 and RMSE=0.0034).
Concerning MIS, the system was generally not successful, and
improvement is on-going to increase the signal-to-noise ratio.

Acknowledgements

Financial support from the EU (FP7-SME) is gratefully
acknowledged.
References
Barai A, Watson S, Griffiths H, Patz R (2012). Magnetic induction
spectroscopy: non-contact measurement of the
electricalconductivity spectra of biological samples.
Measurement Science and Technology 23, 085501 (11pp).
Bauchot AD, Harker FR, Arnold WM (1999). The use of electrical
impedance spectroscopy to assess the physiological condition
of kiwifruit. Postharvest Biology and Technology 18, 9-18.
Bureau S (2009). The use of non-destructive methods to analyse
fruit quality. Global Science Books, Fresh produce, 2009,
Volume 3 Special issue 1, 23-34.
Harker FR, Maindonald JH (1994). Ripening of Nectarine Fruit.
Changes in the Cell Wall, Vacuole, and Membranes Detected
using Electrical Impedance Measurements. Journal of Plant
Physiology 106, 165-171.
Nicolai BM, Beullens K, Bobelyn E, Peirs A, Saeys W, Theron KI,
Lammertyn J (2007). Nondestructive measurement of fruit and
vegetable quality by means of NIR spectroscopy: A review.
Postharvest Biology and Technology 46, 99-118.
Rehman M, Basem AJA, Izneid A, Abdullah MZ, Arshad MR (2011).
Assessment of quality of fruits using impedance spectroscopy.
International Journal of Food Science and Technology 46,
1303-1309
Woodcock T, Downey G, O'Donnell CP (2008). Better quality food
and beverages: the role of near infrared spectroscopy. Journal of
Near Infrared Spectroscopy 16, 1-29.

SSC TA Flesh firmness
Mean
values* (Brix)
(meq/100g
FW) (Work to limit, J)
LV = 11 LV = 6 LV = 5
R = 0.78 R = 0.30 R = 0.32
RMSE = 0.99 RMSE = 2.66 RMSE = 0.007
LV = 8 LV = 11 LV = 5
R = 0.76 R = 0.74 R = 0.40
RMSE = 0.84 RMSE = 0.84 RMSE = 0.007
LV = 8 LV = 9 LV = 9
R = 0.82 R = 0.65 R = 0.76
RMSE = 0.73 RMSE = 1.30 RMSE = 0.0034
LV = 8 LV = 3 LV = 6
R = 0.72 R = 0.20 R = 0.51
RMSE = 1.30 RMSE = 1.75 RMSE = 0.0056
Apple
Kiwi
Peach and
Nectarine
Pear

Fruit species Flesh firmness SSC TA
L* a* b* Work to Limit (J) (Brix) (meq/100g)
Apple mean 72.9 -0.8 43.1 0.0172 13.5 4.6
standard deviation 4.4 8.8 4.8 0.0068 1.7 2.2
minimum 45.2 -18.7 23.1 0.0070 8.8 1.4
maximum 83.6 43.5 61.5 0.0400 18.1 11.7
Kiwi mean 46.4 3.1 21.8 0.0079 12.3 18.8
minimum 37.9 -1.3 17.9 0.0009 7.2 15.2
maximum 51.8 7.7 26.8 0.0411 18.3 42.7
Pear mean 64.9 -7.8 42.0 0.0155 13.2 3.6
minimum 49.5 -17.6 32.1 0.0022 8.1 0.8
maximum 78.7 4.8 51.1 0.0412 18.4 7.0
Peach/Nectarine mean 58.6 18.6 38.8 0.0104 12.2 9.1
minimum 33.7 -7.9 10.6 0.0012 6.6 2.6
maximum 80.1 44.0 65.4 0.0363 19.7 18.6
Colour
Table 2. Data of color, flesh firmness, SSC and TA of fruits for apple, kiwi, pear and peach and nectarine.

On-tree monitoring of quality evolution
in citrus by using a mems-based NIR
spectrometer.
De la Haba M J.; Garrido-Varo A.
b
; Sanchez M T.; Guerrero Ginel J.E;
Prez-Marin D.
b

a: Department of Bromatology and Food Technology, University of Cordoba, Spain
b: Department of Animal Production, University of Cordoba, Spain

Corresponding author: bt1hacem@uco.es (M.-J. De la Haba)
Introduction
Citrus growers are increasingly demanding rapid, cost-
effective and non-destructive methods for monitoring changes
during on-tree ripening, with a view to establishing the optimum
harvest date.
Near-infrared spectroscopy (NIRS) may provide an ideal way
of meeting growers needs, since it is a non-destructive
technique that combines swift and precise measurement with
considerable versatility, high throughput and low cost.
Moreover, the current trend towards the use of NIR
spectroscopy in situ has led to the miniaturization of optical
components and the application of new techniques such as
Digital Transformation Spectroscopy (DTS), which have enabled
the development of compact, portable NIRS instruments, ideal
for use in the field for the prediction of various external and
internal fruit quality parameters measured simultaneously
(Prez-Marn et al., 2009; 2010; Snchez et al., 2013).

The main aim of this study was to evaluate the performance
of a handheld MEMS-based NIR spectrophotometer, combined
with chemometric techniques, for predicting quality parameters
(soluble solids content, pH, titratable acidity, maturity index) in
mandarins and oranges during on-tree ripening, as a means of
monitoring the ripening process and optimizing the harvest
date.
Material and methods
Samples
The initial samples sets comprised 228 mandarins (Citrus
reticulate,cv. Clemevilla) and 192 oranges (Citrus sinensis, (L.)
Osbeck cv. Powell Summer Navel) grown on a commercial
plantation near the village of La Campana (Seville, Spain).
However, because during the study 22 ripe mandarins and 2 ripe
oranges dropped off the tree and were excluded, the final
sample sets comprised 206 mandarins and 190 oranges.
Both, mandarins and oranges were harvested on six
different dates. Mandarins were harvested from October to
December 2010, whereas oranges were harvested from
February to March 2011.
Harvested citrus fruits were kept in refrigerated storage at
5C and 90% RH until the laboratory testing was performed.

Prior to each test, samples were allowed to reach room
temperature.
Spectra collection
NIR spectra of intact fruits were collected in reflectance
mode (log 1/R) using a handheld micro-electromechanical
system (MEMS) instrument (Phazir 2400, Polychromix, Inc.,
Wilmington, MA, USA). The instrument scans at intervals of
approximately 8 nm (pixel resolution 8 nm, optical resolution 12
nm), across the range of NIR wavelengths between 1600 and
2400 nm.
Four spectral measurements were made on each citrus
fruit whilst on the tree, taking orientation (north, south, east,
and west) into account. The four spectra were averaged to
provide a mean spectrum for each fruit.
Spectra were converted into WinISI format using the
Unscrambler software package version 8.0 (CAMO ASA, Norwey)
and the File Conversion option included in the WinISI II software
version 1.50 (Infrasoft International LLC, Port Matilda, PA, USA).
Determination of quality parameters
Soluble solids content (SSC), pH and titratable acidity
(TA) were measured following Obenland et al. (2008). The
maturity index (SSC:TA ratio) was also calculated. Each sample
was analysed in duplicate.
Quantitative calibrations: sets and data processing
Samples for the calibration and validation sets were
selected using the CENTER algorithm included in the WinISI II
software package version 1.50 (Infrasoft International LLC, Port
Matilda, PA, USA). Having ordered the samples of both sets by
spectral distances from the centre, 3 and 2 anomalous spectra
or outliers were identified in the mandarins and oranges sets,
respectively (Shenk and Westerhaus, 1991). Of the remaining
overall sets (N = 203 and 188 samples for mandarins and
oranges, respectively), one of every 3 samples was removed to
build the validation set, leaving a calibration set comprising 137
mandarins and 124 oranges.
Data were thus subjected to chemometric treatment using
the WinISI II software package (ISI, 2000). Prediction equations
were obtained using the MPLS regression method with cross-
validation. Six cross-validation steps were included in the
process in order to avoid overfitting (Shenk and Westerhaus,
1995).
For each analytical parameter, a number of different pre-
processing combinations were evaluated for scatter correction,
including the Standard Normal Variate (SNV) and Detrending
(DT) (Barnes et al., 1989). Additionally, four derivative
mathematical treatments were tested in the development of

NIRS calibrations: 1,5,5,1; 2,5,5,1; 1,10,5,1 and 2,10,5,1 (Shenk
and Westerhaus, 1995).
The statistics used to select the best equations were:
standard error of calibration (SEC), coefficient of determination
of calibration (R
2
), standard error of cross-validation (SECV),
coefficient of determination for cross-validation (r
2
), RPD or
ratio of the standard deviation of the original data (SD) to SECV,
and the coefficient of variation (CV) or ratio of the SECV to the
mean value of the reference data for the calibration set.
The best-fitting equations obtained for the calibration
procedure were subsequently evaluated by external validation
following the protocol outlined by ISI (2000) and Shenk et al.,
(2001).
Descriptive data for NIR calibrations and validations sets
Values for range, mean, standard deviation and
coefficient of variation (CV) for each of the parameters analysed,
using the calibration and validation sets after application of the
CENTER algorithm, are shown in Table 1, together with the
number of samples on which each parameter was measured.
All parameters, excepting the pH parameter in both fruits,
displayed considerable variation, mainly titratable acidity and
maturity index, as evidenced by the CV values.

Calibration for predicting quality parameters
Cross-validation statistics for the best models
obtained for the on-tree prediction of SSC, pH, TA and maturity
index (MI), in both intact mandarins and oranges using the MPLS
algorithm are shown in Table 2.
The best results for predicting SSC in mandarins, TA in oranges
and pH in both fruits were reached using the first derivative of
the spectrum, whereas for the remaining parameters (SSC in
oranges, TA in mandarins and MI in both fruits) the second
derivative provided the best results.

The predictive capacity of models developed for predicting
pH, TA and MI in intact mandarins (r
2
= 0.69, 0.60 and 0.53;
SECV = 0.10, 0.15% and 1.64, Table 2), may be considered
acceptable for rough screening, discriminating samples between
high, medium and low values. However, the precision of models
for predicting SSC (r
2
= 0.44; SECV = 0.82) suggests that this
model would only enable discrimination between high and low
values of this parameter (Williams, 2001).
The predictive capacity of the SSC model was poorer than
that reported for intact mandarins by Hernndez-Gmez et al.
(2006) (RPD = 4.00; CV = 1.67%) and by Liu et al. (2010) (RPD =
2.90; CV = 4.17%), although these authors used a
monochromator and a diode-array instrument, respectively.
However, values of the statistics for the models constructed in
this study were better than those reported by Hernndez-
Gomz et al. (2006) for pH (RPD = 1.73; CV = 3.25%) and by Liu
et al. (2010) for TA (RPD = 0.8; CV = 22.72%).
By contrast, the model obtained for predicting SSC in
oranges may be considered acceptable(r
2
= 0.84; SECV = 0.55%),
being similar to that obtained by Cayuela and Weiland (2010)
(RPD = 2.13; CV = 6.12 %).
Models developed for predicting pH and MI (r
2
= 0.64 and
0.53, respectively) would allow the discrimination between high,
medium and low values, whereas, for TA the value of the r
2
for
the best model obtained (r
2
= 0.46) suggests again that this
model would only allow the classification of the samples as high
and low values of this parameter (Williams, 2001).

Similar findings although using diode array instruments have
been reported by a number of authors attempting to measure
acidity in oranges using NIRS technology (Cayuela, 2008; Cayuela
and Weiland, 2010), displaying limited predictive capacity.

Parameter Fruit Set N Range Mean SD CV (%)
SSC (%) Mandarin Calibration 137 9.95-15.65 12.48 1.09 8.73
Validation 66 10.10-15.00 12.59 1.21 9.61
Orange Calibration 124 6.80-14.14 9.34 1.37 14.67
Validation 62 7.40-11.45 9.26 1.08 11.66
pH Mandarin Calibration 137 2.08-3.80 3.25 0.21 6.46
Validation 66 2.83-3.69 3.24 0.20 6.17
Validation 62 3.28-4.07 3.76 0.18 4.79
Mandarin
Calibration
137 0.68-2.15 1.20 0.29 24.16
Validation 66 0.80-1.75 1.20 0.24 20.00
Validation 62 0.46-1.02 0.64 0.12 18.75
Mandarin Calibration 137 5.41-17.37 10.88 2.43 22.33
Validation 66 6.66-15.06 10.82 2.17 20.05
Validation 62 8.55-21.05 14.88 2.83 19.02
Titratable
acidity (%
citric acid)
Maturity
index
Table 1.Statistical analysis of calibration and prediction sample sets, i.e., number of samples (N), data ranges, means, standard
deviations (SD) and coefficients of variation (CV), for mandarins and oranges.


Validation statistics for the prediction of quality

Validation statistics for the prediction of quality
parameters in intact oranges and mandarins, during on-tree
ripening, are shown in Table 3.
The models constructed for predicting SSC in both fruits
and pH in mandarins, using a portable MEMS-NIR instrument,
met the validation requirements in terms of r
2
(r
2
> 0.6) (ISI,
2000; Shenk et al., 2001) and both the SEP(c) and the bias were
within confidence limits. However, for the remaining
parameters in both citrus fruits, it should be stressed that
although SEP(c) and bias lay within confidence limits, with the
exception of SEP(c) for the pH parameter in oranges samples,
r
2
values did not attain the recommended minimum value. This
indicated that the NIRS equations constructed here can be
considered as a first step in the fine-tuning of NIRS technology
for the on-tree monitoring of quality parameters in oranges and
mandarins.

Table 3. Validation statistics for the best models for predicting
quality parameters in citrus fruits.

Conclusions
These results confirm that NIRS technology, using a hand-
held compact MEMS instrument, is a viable option for the
simultaneous and non-destructive measurement of quality
parameters in intact mandarins and oranges during on-tree
ripening. Nevertheless, the results should be considered to be
the first step in the development of a NIR application for in-situ
monitoring of ripening in citrus fruits, which will provide to
farmers, over the coming years, a precise and accurate
indication of the fruits internal quality and the optimum harvest
date.
Acknowledgements
This research was funded by the Andalusian Regional
Government under the Research Excellence Program (Project
no. P09-AGR-5129 MEMS and NIRS-image sensors for the in
situ non-destructive analysis of food and feed). The authors are
grateful to Mr. Antonio Lpez and Mrs. M Carmen Fernndez of
the Animal Production Department (ETSIAM-UCO) for technical
assistance.
References
Barnes, R.J., Dhanoa, M.S., Lister, 170 S.J. 1989. Standard normal
variate transformation and detrending of near infrared
diffuse reflectance spectra. Applied Spectroscopy 43,
772777.
Cayuela, J.A. 2008.Vis/NIR soluble solids prediction in intact
oranges(Citrus sinensis L.) cv. Valencia Late by
reflectance. PostharvestBiology and Technology 47,
7580.
Cayuela, J.A., Weiland, C. 2010. Intact orange quality prediction
with two portable NIR spectrometers. Postharvest
Biology and Technology 58, 113120.
Hernndez-Gmez, A.H., He, Y., Pereira, A.G. 2006. Non-
destructive measurementof acidity, soluble solids and
firmness of Satsuma mandarin using
Vis/NIRspectroscopytechniques. Journal of Food
Engineering 77, 313319.
ISI, 2000. The Complete Software Solution Using a Single Screen
for Routine Analysis, Robust Calibrations and
Networking. Manual. Foss NIRSystems/Tecator. LLC,
Silver Spring, MD: Infrasoft International.
Liu, Y., Sun, X., Zhang, H., Aiuguo, O. 2010. Nondestructive
measurement of internal quality of Nanfeng mandarin
fruit by charge coupled device near infrared
spectroscopy. Computers and Electronics in Agriculture
715, S10S14.

Parameter Fruit Mathematic
treatment
N Mean SD SEC
R
2
SECV
r
2
RPD CV (%)
SSC (%) Mandarin 1,5,5,1 136 12.47 1.08 0.67 0.61 0.82 0.44 1.32 6.57
Orange 2,10,5,1 117 9.33 1.35 0.47 0.88 0.55 0.84 2.45 5.89
pH Mandarin 1,5,5,1 134 3.26 0.18 0.09 0.71 0.10 0.69 1.80 3.07
Orange 1,5,5,1 116 3.75 0.17 0.10 0.68 0.10 0.64 1.70 2.67
Mandarin 2,5,5,1 128 1.16 0.24 0.14 0.64 0.15 0.60 1.60 12.93
Orange 1,10,5,1 120 0.64 0.11 0.08 0.49 0.08 0.46 1.60 12.50
Mandarin 2,5,5,1 134 10.97 2.38 1.51 0.60 1.64 0.53 1.45 14.95
Orange 2,10,5,1 118 14.53 2.40 1.42 0.65 1.67 0.51 1.44 11.49
Titratable
acidity (%
citric acid)
Maturity
index
Table 2. MPLS regression statistics for NIR-based models for predicting quality parameters in citrus fruits.

Parameter Fruit SEP SEP(c) Bias r
2
Slope
SSC (%) Mandarin 0.78 0.75 0.23 0.62 1.09
Orange 0.72 0.71 -0.16 0.65 0.75*
pH Mandarin 0.12 0.12 0.01 0.66 0.99
Orange 0.13 0.13* -0.02 0.46* 1.01
Titratable
acidity (%
citric acid) Mandarin 0.19 0.19 0.01 0.41* 0.72*
Orange 0.09 0.09 0.01 0.48* 1.07
Maturity
index Mandarin 1.57 1.58 .016 0.55* 0.72*
Orange 1.91 1.93 -0.11 0.54* 1.12

Obenland, D., Collin, S., Sievert, J., Fjeld, K., Doctor, J., Arpaia,
M.L. 2008. Commercialpacking and storage of navel
oranges alters aroma volatiles and reduces flavour
quality. Postharvest Biology and Technology 47, 159
167.
PrezMarn, D., Snchez, M.T., Paz, P., Soriano, M.A., Guerrero,
J.E., GarridoVaro, A. 2009. Nondestructive
determination of quality parameters in nectarines
during ontree ripening and postharvest storage.
Postharvest Biology and Technology 52, 180188.
PrezMarn, D., Paz, P., Guerrero, J.E., GarridoVaro, A.,
Snchez, M.T. 2010. Miniature handheld NIR sensor
for the onsite nondestructive assessment of
postharvest quality and refrigerated storage behavior
in plums. Journal of Food Engineering 99, 294302.
Snchez, M.T., De la Haba, M.J., Prez-Marn, D. 2013. Internal
and external quality assessment of mandarins on-tree
and at harvest using a portable NIR
spectrophotometer. Computers and Electronics in
Agriculture 92, 66-74.
Shenk, J.S., Westerhaus, M.O. 1991. Population definition
sample selection and calibration procedures for near
infrared spectra and modified partial least squares
regression. Crop Science 31, 469474.
Shenk, J.S., Westerhaus, M.O. 1995. Analysis of Agriculture and
Food Products by Near Infrared Reflectance
Spectroscopy. Monograph. Silver Spring, MD:
NIRSystems Inc.
Shenk, J.S., Workman, J., Westerhaus, M.O. 2001. Application of
NIR spectroscopy to agricultural products. In: Burns,
D.A., Ciurczak, E.W. (Eds.), Handbook of Near Infrared
Analysis 2
nd
Edition, pp. 419474. New York: Marcel
Dekker.
Williams, P.C. 2001. Implementation of nearinfrared
technology. In: Williams, P.C., Norris, K.H. (Eds.),
Nearinfrared Technology in the Agricultural and Food
Industries, pp. 145169. St. Paul, MN: AACC, Inc.

NIR2013 Proceedings, 2-7 June, La Grande-Motte, France. A1 - Agriculture and Environment
B e l l o n - M a u r e l V . , W i l l i a m s P . , D o w n e y G . , E d s
1 5 3
A bad apple in the barrel? Apple internal
defect detection using NIR transmittance
spectroscopy.
P.P. Subedi
a
, B.P. Khatiwada
a
, K.B. Walsh
a
and P. Plagnol
b

a
Centre for Plant and Water Sciences, Central Queensland University, Rockhampton, Australia
b
MAF RODA AGROBOTIC Montauban, Cedex, France

Corresponding author: k.walsh@cqu.edu.au
Introduction
Fresh fruit are heterogeneous in terms of quality. Current
practice for quality assessment involves lot sampling of fruit, with
rejection of whole consignments if a threshold level of defect
incidence is exceeded. For example, apple internal defects include
the physiological disorders of internal browning, water core and
bitter pit. Internal browning is a disorder induced during prolonged
controlled-atmosphere storage, with the incidence and severity of
this disorder loosely dependent on fruit variety, maturity, nutrition,
growing conditions and CO
2
/O
2
levels during storage. For example,
the problem is manifest in the varieties Pink Lady, Golden Delicious,
Red Delicious, Sundowner, Kenzi, Rubens, Fuji etc. with different
degrees of susceptibility to this disorder. Internal browning generally
develops in fruit after three or more months in cold storage.
Australian retailers treat this disorder as a major defect, with a
maximum tolerable limit of affected fruit set as less than 2% of a
given consignment. The disorder is considered to be a result of
membrane dysfunction at low O
2
levels, with release of polyphenol
oxidase resulting in the formation of quinone compounds (deCastro
et al., 1996).

Logically, a full transmission optical geometry, using visible or
shortwave near infrared spectroscopy (Vis SWNIRS) should have
potential to screen fruit for this disorder (Abbott, 1999).
Transmission geometry is sensible for assessment of an internal
disorder and, while the browning disorder obviously has a visible
region spectral signature, the lower absorptivity of light in the SWNIR
region may hold advantages for transmission studies of whole fruit
(McGlone et al., 2005). A number of authors have considered the
application of such technology. For example, Upchurch et al. (1997)
acquired transmission spectra of Red Delicious apples affected with
internal browning, reporting affected fruit to have lower transmission
in wavelengths below 750 nm and less absorption at wavelengths
above 750 nm than unaffected fruit. A ratio between the
transmittance at 720 and 810 nm was recommended for
discrimination between good and defective apples, with a coefficient
of determination (R
2
)of 0.71 reported between browning score (0 =
no browning, to 3 = severe browning). In another study by Clark et al.
(2003) involving Braeburn apples, significant changes in
transmittance over the range of 700-900 nm was noted with degree
of browning. A R
2
of 0.90 was reported for models based on the ratio
of transmittance at 815 and 823 nm (for 180
o
geometry) and an R
2

value of 0.91 was reported for a PLS model based on absorbance data
(90
o
geometry) over the wavelength range 697-861 nm.

However there has been no subsequent report on the
commercial application of this technology. Here we report on a
commercially available on-line transmission SWNIRS technology
(MAF Roda, Australia) in the context of sorting of apple fruit with
internal browning defect, relative to a partial transmission laboratory
based system.
Plant material
Apple fruit (cv. Pink Lady) were sourced from a commercial
packing shed in Stanthorpe (southern Queensland, Australia)
following fruit release from cold store for commercial grading prior to
dispatch to retailers. Fruit had been harvested in May 2010 and
stored for 8 months at 2
o
C. One population (n = 69) of fruit was
transported overnight at room temperature into cold store at 2
o
C for
another month at the University. Two further populations (n = 125
and 100, respectively) were assessed using commercial
instrumentation in the Stanthorpe pack-house.

On the day of assessment, apples were removed from cold store
and allowed to warm to a minimum of 20
o
C. Each fruit was
numbered and marked at four equidistant positions around the
equator of the fruit.
Fruit measurements
The fruit held at the University were used for assessment of
chlorophyll fluorescence after dark adaptation. Each apple was
wrapped with aluminium foil for 24 h prior to measurement of
photosynthetic efficiency using an OptiSciences 30p (through
Bioscientific, Australia) at a modulation intensity of 4. This unit
employs a PIN photodiode with a 700 -750 nm bandpass filter. One
reading was taken per fruit within 2 s of removing the aluminium foil.
Vis-SWNIR spectra of these fruit were collected using an in-house
bench top unit consisting of one 300W halogen lamp and a MMS1
Zeiss Si diode array detector (310-1100 nm range, pixel resolution of
3.3 nm and repeatability of 0.1 mAbs). A partial transmittance

1 5 4
geometry was employed wherein the light source was directed onto
the equator of the fruit while the detector was positioned at 90
o
to
the previous position. Four spectra were acquired and averaged per
fruit (with marked positions facing the detector).

Fruit spectra were also acquired using a MAF-Roda IDD-1 unit
installed in a commercial packing shed in Stanthorpe, Queensland
(wavelength range 572-1002 nm, pixel resolution 0.42 nm and
repeatability - mean/SD of maximum count of twenty repeated
measurements - of 16 mAbs). This system employed a transmittance
optical configuration with 15 tungsten halogen lamps used as the
light source. Spectra of fruit moving at 375 cups/min (i.e. 6 fruit/sec.)
were acquired. Four spectra were acquired and averaged per fruit
(with marked positions facing the detector). Following spectral
acquisition, fruit were cut transversely in three locations. A
subjective score (based on visual ranking) of fruit internal browning
(using a 0 to 9 scale, with increasing severity of defect) was assessed
by a panel of two to three people.
Data pre-processing and analysis
Partial least squares regression (PLSR) of attribute against
spectra was undertaken using Unscrambler X software (Camo, Oslo,
Norway). Second derivative (second order Savitzky Golay, based on 9
data points) and standard normal variate (SNV) were used as data
pre-processing options.

Chlorophyll fluorescence
Chlorophyll florescence is a fluorescence emission caused by the
radiative de-excitation of excited chlorophyll molecules. The
photochemical efficiency (Fv/Fm) decreased with increased severity
of browning (Figure 1). This result supports consideration of
chlorophyll-associated wavelengths in the Vis-SWNIRS Study.

Vis-SWNIR spectroscopy
A clear distinction was found between mean absorbance spectra
of good, moderate and severely affected fruit in absorbance and
normalised spectra, for both the bench top (Fig. 2A, B) and on-line
(data not shown) instruments. Absorbance at single wavelengths
was correlated to defect score with an R
2
value of up to 0.75 (Figure
2C).

Figure 2. (A) Mean absorbance and (B) mean d
2
Absorbance spectra
of 69 Pink Lady apples by score category (average of more than 10
fruit in each category). (C) Univariate correlation coefficients of
determination between second derivative of absorbance and
browning score at each wavelength for 69 Pink Lady apple fruit.
Spectra were collected using the bench top unit.
PLS regression models were developed over a series of
wavelength regions, with the importance of the chlorophyll peak
around 680 nm to the model demonstrated in cross-validation and
prediction of independent sets. A correlation cross-validation
coefficient of prediction (R
2
cv
) of >0.87 was achieved using the 600-
800 and 540-810 nm regions.

Figure 1. Photochemical efficiency (Fv/Fm) of photosynthesis in fruit
relative to degree of internal browning (score 0-2, no browning; 3-5,
moderate browning; and 6-10, severe browning).

1 5 5

Table 1. PLS regression calibration model statistics for internal
browning (visual score) in Pink Lady apples (69 fruit, mean score 5.01,
SD 3.69) using partial transmission spectra (four spectra averaged per
fruit). No data (outliers) were removed.

A similar exercise was undertaken for the on-line IDD system. Good
PLSR model statistics of R
2
cv
> 0.77 were achieved using either the
window of 640 860 nm or 725 975 nm for both populations
assessed; a poorer result was obtained for spectral windows that did
not encompass the chlorophyll peak (Table 2).

Table 2. PLS regression calibration model statistics for internal
browning (visual score) in Pink Lady apples using transmission spectra
(IDD unit). Population 1 consisted of 125 fruit (500 spectra, mean
score 3.05, SD 3.34) and population 2 consisted of 100 fruit (400
spectra, mean score 3.10, SD 2.89).

In prediction of an independent population, a R
2
p
value of 0.66 was
achieved, with a relatively high bias (to 1 unit on a 0-9 scale) (Table
3). Accepting inclusion of 2% of fruit with browning scores above 2,
this model allowed separation of fruit with score levels below 2,
although with the loss of 7.4 % of acceptable fruit through false
positive identification of defects (Fig. 3).
Conclusion
The commercially available IDD-1 unit was demonstrated to achieve a
sorting accuracy adequate for practical application.

Table 3. PLSR model statistics for prediction of an independent
population using models developed over the 790-840 nm regions.

Figure 3. Scatter plot of predicted against actual internal browning
score for a PLSR model based on partial transmission spectra over the
range 725-975 nm for Pink Lady population 1 (125 fruit). The
reference method (a subjective visual score) requires improvement
and is likely to have contributed to an observed bias in model
predictions of independent populations.
Acknowledgements
Funding support was received from Horticulture Australia Ltd
and CVS-MAF P/L. The support of Stanthorpe, Queensland based
apple growers Savio and Fast Fruit is appreciated.
References
Abbott, J. A. (1999). Quality measurement of fruits and vegetables.
Postharvest Biol. Tec., 15, 207-225.
Clark, C. J., McGlone, V.A. and Jordan, R.B. (2003). Detection of Brownheart in
Braeburn apple by transmission NIR spectroscopy. Postharvest Biol.
Tec., 28, 87-96.
de Castro, E., Barrett, D.M., Jobling, J. and Mitcham, E.J. (2008). Biochemical
factors associated with a CO2-induced flesh browning disorder of Pink
Lady apples. Postharvest Biol. Tec., 48,182-191.
McGlone, V. A., Martinsen, P.J., Clark, C.J. and and Jordan, R.B. (2005). On-line
detection of Brownheart in Braeburn apples using near infrared
transmission measurements. Postharvest Biol. Tec., 37, 142-151.
Upchurch, B.L., Throop, J.A. and Aneshansley, D.J. (1997). Detecting internal
breakdown in apples using interactance measurements. Postharvest
Biol. Tec., 10,15-19.
Calibration
population
Pink Lady 1 Pink Lady 2 0.66 3.32 -2.48 0.56
Pink Lady 2 Pink Lady 1 0.67 3.44 2.23 0.738
Prediction
population
R
2
p
RMSEP Bias Slope
Range (nm) # PC R
2
cv
RMSECV SDR
500-700 4 0.85 1.45 2.54
600-800 4 0.87 1.32 2.79
790-840 4 0.83 1.50 2.46
800-1000 2 0.80 1.63 2.26
600-700 2 0.84 1.45 2.54
540- 810 4 0.88 1.31 2.81
939-1094 3 0.82 1.60 2.30

Population
(fruit #)
Spectral
window
(nm)
PC (#) R
2
cv
RMSECV SDR
Pop 1 (125) 725-975 5 0.74 1.54 2.16
Pop 1 (125) 640-860 6 0.74 1.55 2.15
Pop 1 (125) 790-840 3 0.68 1.71 1.95
Pop 2 (100) 725-975 6 0.77 1.35 2.14
Pop 2 (100) 640-860 9 0.77 1.34 2.15
Pop 2 (100) 790-840 4 0.70 1.56 1.85

(Visual Score (0-10 scale)
0 2 4 6 8 10
N
I
R

P
r
e
d
i
c
t
e
d

V
a
l
u
e
-4
-2
0
2
4
6
8
10
Bellon-Maurel V., Williams P., Downey G., Eds 1 5 6
Prediction by infrared reflectance
spectroscopy of the lutein content in
summer squash fruits.

D. Martnez-Valdiviesoa, M.T. Blanco-Daza, P. Gmeza, J. M. Moreno-
Rojasb, R. Fonta, M. Del Ro-Celestinoa

a
IFAPA Centro La Mojonera. (Almera, Spain)

b
IFAPA Alameda del Obispo (Crdoba, Spain)

Corresponding autor: mercedes.rio.celestino@juntadeandalucia.es

Introduction

Summer squash fruits (Cucurbita pepo subsp.
pepo) are widely-consumed around the world and are
cultivated intensively in southeastern Spain, concentrating
mostly in Almeria where it is one of the main crops
produced in greenhouses reaching 354.156 tonnes in 2012
(Anuario Agricultura Almeriense, 2012). Fruits of this species
are important in human nutrition due to their phytochemical
composition: carotenoids, vitamins, phenolic compounds,
minerals, etc. Animals cannot synthesise carotenoids so they
must eat vegetables. Previous studies showed that lutein
was the major carotenoid present in the mesocarp and
epicarp of the fruit of C. pepo (Tadmor et al., 2005; Del Ro
et al., 2012). Associations between the consumption of
lutein and prevention of age-related macular degeneration
(Khachik et al., 1999) and reduced risk of atherosclerosis
(Khachik and Beecher, 1988) have been documented.
However, no information is available about the genetic
variability of lutein composition of summer squash fruit and
the breeding potential for improvement of this trait.
HPLC methods are useful to determine carotenoids
in different foods. These methods require the previous
extraction of the analyte from the matrix, so many
difficulties may arise regarding their stability over the whole
procedure. In fact, carotenoids and xanthophylls are very
sensitive to heat and acids, which may cause trans-cis
isomerisation and structural changes; these problems are
aggravated by light and/or oxygen. Near-infrared reflectance
(NIR) spectroscopy has been widely used as a fast and cost-
effective method within the food industry, including, for
example, the control of fruit and vegetable quality. This
technique has been applied to the analysis of carotenoid
contents in Tritordeum (Atienza et al., 2005) or banana
(Davey et al., 2009). No studies have been done to explore
the capabilities of NIR to determine carotenoid in squash
fruits.
The objective of this work was to develop and
evaluate a methodology based on near infrared
spectroscopy (NIR) for determining the lutein content in
epicarp and mesocarp of the fruit for use in germplasm
selection for breeding programmes.


Seeds of 14 accessions of landraces and
commercial C. pepo maintained in the Center IFAPA La
Mojonera, Almera (Spain) were germinated; plants were
transplanted to a greenhouse and grown following standard
local cultural practices for both plant nutrition and insect
pest and disease control. Summer squash (Cucurbita pepo
sub. pepo) fruit (80 epicarps and 80 mesocarp samples) from
different morphotypes (vegetable marrow, zucchini,
pumpkin and cocozelle) were collected and harvested at the
immature stage (commercial size). Epicarp and mesocarp of
fruit were frozen in sealed polyethylene bags immediately
after harvest and subsequently lyophilised and ground for
analysis.
Analysis of lutein content was performed by HPLC
using a Waters 1525 HPLC and Waters Symmetry C18
column controlled by a Breeze workstation. All solvents and
chemicals were supplied by Sigma Chemical Co., St. Louis,
MO. Finely-ground sample (400 mg) was used following the
method described by Tadmor et al. (2000) with slight
modifications to adjust the protocol to C. pepo lutein
contents. All steps were carried out in darkness or under
fluorescent light to prevent photodegradation of analytes.
Lutein was identified by comparison of retention time, co-
injection with a known standard, and comparison of its UV-
visible spectra with an authentic lutein standard.
Quantification was carried out by external standardisation; a
full standard curve was constructed using five different
concentrations for lutein in triplicate.
The spectra of the samples were recorded (400
2500 nm) in a Vis/NIR spectrometer model 6500 (Foss-
NIRSystems, Inc., Silver Spring, MD, USA). The calibrations
were performed using GLOBAL routine (WinISI II v.1.50;
Infrasoft International, LLC, Port Matilda, PA, USA).
Lyophilised samples were analysed in triplicate as log 1/R (R
= reflectance). Calibration equations were computed using
raw optical data (log 1/R, where R is reflectance), or first or
second derivatives of the log 1/R data, with several
combinations of derivative (gap) sizes and smoothing [i.e. (0,
0, 1, 1; derivative order, segment of the derivative, first
smooth, second smooth); (1, 4, 4, 1); (1, 10, 10, 1); (2, 5, 5,
2); (2, 20, 20, 2)]. Wavelengths from 400 to 2500 nm ( 8 nm
step) were used to perform the different calibration
equations. The regression method employed to correlate
spectral information and lutein content in the samples was
modified partial least squares (MPLS). The final objective of
the mathematical procedure is to reduce the high number of
spectral data points (absorbance values from 400 to 2500
nm every 2 nm, i.e. 1050 data) and to eliminate the
correlation of absorbance values presented by neighbouring
wavelengths (Shenk and Westerhaus, 1994). Standard
normal variate and detrend transformations (SNV-DT)
(Barnes et al., 1989) were used to correct baseline offsets
due to scattering effects (differences in particle size
between samples). Cross-validation was performed on the
calibration set for determining the best number of terms to
use in the equation, as well as to determine the ability of
each equation to predict on unknown samples (Shenk et al.,
2001). The predictive ability of calibrations was assessed
from the r
2
, the standard error of prediction (SEP) and the
ratio of standard error of prediction to standard deviation
(RPD value; SD/SEP).


Results revealed variations in the lutein content of
summer squash depending on variety and the part of the
fruit studied, with epicarp showing higher lutein content
than mesocarp. HPLC results (expressed in mg kg
-1
dry
weight) were: in epicarp, the lutein content ranged from 4.3
to 963 mg kg
-1
; in mesocarp, the lutein content ranged from
4.1 to 240 mg kg
-1
. The samples showed a wide variation in
lutein content since the analysed samples covered most of
the variability cited in the literature for C. pepo (Ben-Amotz
and Fishler, 1998; Rodriguez-Amaya et al., 2008; El-Qudah,
2009). The maximum result obtained for lutein content in
mesocarp in this study (240 mg kg
-1
dry weight) was lower
than previous reports of contents of up to 332 mg kg
-1
(dry
weight) (Ben-Amotz and Fishler, 1998).
Lutein is a xanthophyll which can be found in the
retina of the eye, where its main function is protecting the
photoreceptor cells from free radicals generated in the
photochemical processes. It has a major role in the
prevention of age-related macular degeneration, as
evidenced by different studies (Beatty, 1999; Seddon, 1994).
It has been suggested that 6 mg of lutein per day can reduce
the risk of age-related macular degeneration by 43 %
(Seddon 1994). This concentration is equivalent to the daily
consumption of approximately 1 kg of corn, 7 kg of
tomatoes, 2 bowls of spinach salad, a bowl of coleslaw, and
,as revealed in this study, could achieved with 209 g of fresh
pulp of C . pepo. Although lutein is not a provitamin A, is a
more effective antioxidant than many other carotenoids (in
vitro inhibition of the oxidation of lipids in a more efficient
manner than -carotene, -carotene or lycopene). It should
be noted also that the bioavailability of lutein from plant
sources is much higher than -carotene (Erdman, 1999).
Figure 1 shows typical NIR spectra obtained for the
different C. pepo samples analysed. It can be clearly seen
that a considerable contribution arises from the visible
wavelength range (400-700 nm). Figure 2 shows a mean
near infrared spectrum of Cucurbita samples of epicarp (a)
and mesocarp (b) fruits using SNVD and second derivative as
data pre-treatments. In the VIS region, spectra displayed
two peaks characteristic of fruit and vegetable produce i.e.
at 532 nm and 676 nm, which correspond to electronic
transitions in the green and red, respectively. The band at
670 nm has been assigned to chlorophyll (Tkachuk and
Kuzina, 1982), which near 680 nm has a strong inverse
correlation with sugar content (Stchur et al., 2002). The NIR
region of the spectrum (Figure 2) showed characteristic
absorption bands at 1432 and 1916 nm related to O-H
stretch second and first overtone of water, respectively;
1726, 2310 and 2348 nm related to C-H stretch first
overtone and combination bands of lipids (Murray, 1986). A
band at 2274 nm was related to O-H+C-O deformation, O-H
stretch plus deformation, and O-H+C-C stretch of starch
(Osborne et al., 1993). These absorption bands in the NIR
region of the spectrum agree with previous studies for
carotenoid analysis in other species such as maize and
banana (Brenna and Berardo, 2004; Davey et al., 2009).
The usefulness and accuracy of the models
developed were evaluated on the basis of the R
2
VAL and
RPD values (Williams and Sobering, 1996). The r
2
values give
an indication of the percentage variation in the Y variable
that is accounted for by the X variable. In our study, cross-
validation resulted in coefficients of determination (1-VR) of
0.68 and 0.66 for lutein in epicarp and mesocarp
respectively, indicating that 68 % and 66 % of the variability
present in the data was explained by the respective
calibration equations. The r
2
values obtained in the cross-
validation for the equations for lutein content were
characteristic of equations that can be used for a good
separation of the samples in the validation set into high,
medium and low contents (Figure 3; Shenk and Westerhaus,
1996).
Although r
2
values give good information about the
quality of the calibration, they do not provide information
on the prediction accuracy. For this, we used the RPD
classifications defined by Williams and Sobering (1996).
Here, RPD values below 1.5 are considered unusable, values
of between 1.5-2.0 may be used for rough predictions, those
between 2.0 and 2.5 allow approximate quantitative
predictions to be made, while values above 2.5 and 3.0 are
considered to be good and excellent predictive models
respectively. In our study, these ratios confirm that the
accuracies of the Vis/NIRS models developed are suitable for
the rough screening of lutein content in peel and pulp with
values of RPD between 1.5 > RPD < 2.0. In general terms,
RPDs reported in our study for lutein in C. pepo fruit (1.61
and 1.62 in epicarp and mesocarp, respectively) were higher
than those obtained for fruit of banana (1.16; Davey et al.,
2009).

Conclusion

In conclusion, our results show that this use of
Vis/NIR spectroscopy can be routinely applied to determine
lutein content in summer squash fruits. This method could
greatly simplify the analysis of this compound, because no
extraction step with organic solvents (without using
hazardous chemicals) is required and samples are easily
analysed in minutes. The only required step is grinding,
representing an important reduction of the analysis time.
This is of particular importance in nutritional quality
evaluation and quality plant-breeding programmes in which
a large number of samples must be analysed and used to
select the best genotypes after screening thousands of fruits
in a breeding programme of Cucurbita pepo.

Acknowledgements

This work was supported by FEDER funds and the
project INIA RTA2009-00039.

References

Anuario Agricultura Almeriense. (2012). La voz de Almera
(Ed.), pp 180. Almera (Spain).
Atienza, S.G., vila, C. M., Ramrez, M. C. & Mart, A. (2005).
Application of near infrared reflectance spectroscopy
to the determination of carotenoid content in
tritordeum for breeding purposes. Aust. J. Agr. Res., 56
,85-89.
Barnes, R.J., Dhanoa, M.S. and Lister, S.J. (1989). Standard
normal variate transformation and de-trending of
near-infrared diffuse reflectance spectra. Appl.
Spectrosc., 43, 772777.
Beatty, S., Boulton, M., Henson, D., Koh, H. H. and Murray, I.
J. (1999). Macular pigment and age related macular
degeneration. Brit.J. Ophthalmol., 83, 867-877.
Ben-Amotz, A. and Fishler, R. (1998). Analysis of carotenoids
with emphasis on 9-cis -carotene in vegetables and
fruits commonly consumed in Israel. Food Chem., 62,
515-520.
Brenna, O. V. and Berardo, N. (2004). Application of Near-
Infrared Reflectance Spectroscopy (NIRS) to the
Evaluation of Carotenoids Content in Maize. J.
Ag. Food Chem., 52, 5577-5582.
Davey, M.W., Saeys, W., Hof, E., Ramon, H., Swennen, R. L.
and Keulemans, J. (2009). Application of Visible and
Near-Infrared Reflectance Spectroscopy (Vis/NIRS) to
determine carotenoid contents in banana (Musa spp.)
fruit pulp. J. Ag. Food Chem., 57, 17421751.
El-Qudah, J.M. (2009). Identification and quantification of
major carotenoids in some Vegetables. Am. J. Appl.
Sci., 6, 492-497.
Erdman, J.W.J.R. (1999). Variable bioavailability of
carotenoids from vegetables. Am. J. Clin. Nutr., 70,
179-180.
Khachik F. and Beecher, G.R. (1988). Separation and
identification of carotenoids and carotenol fatty acid
esters in some squash products by liquid
chromatography. I. Quantification of carotenoids and
related esters by HPLC. J. J. Ag..Food Chem., 36, 929-
937.
Khachik, F., Steck, A. and Pfander, H. (1999). Isolation and
structural elucidation of (13Z,13'Z,3'R,6'R)-lutein from
marigold flowers, kale, and human plasma. J.Ag. Food
Chem., 47, 455-461
Murray, I. (1986). The NIR spectra of homologous series of
organic compounds. In J. Hollo, K.J. Kaffka & J.L.
Gonczy (Eds.), Proceedings of the International NIR/NIT
Conference (pp. 13-28). Budapest: Akademiai Kiado.
Osborne, B.G., Fearn, T. and Hindle, P. (1993). Practical NIR
Spectroscopy with Applications in Food and Beverage
Analysis. London: Longman Scientific and Technical.
Rodrguez-Amaya, D.B., Kimura, M., Godoy, H.T. and Amaya-
Farfan, J. (2008). Update Brazilian database on food
carotenoids: Factors affecting carotenoid composition.
J.Food Compos.Anal., 21, 445-463.
Seddon, J. M., Ajani, U. A., Sperduto, R. D., Hiller, R., Blair,
N., Burton, T. C., Farber, M. D., Gragoudas, E. S., Haller,
J., Miller, D. T., Yannuzzi, L. A. and Willet, W. (1994).
Dietary carotenoids, vitamins A, C, and E, and
advanced age-related macular degeneration. J. Am.
Med. Assoc., 272, 1413-1420
Shenk, J.S., & Westerhaus, M.O. (1994). The application of
near infrared reflectance spectroscopy (NIRS) to forage
analysis. In Fahey Jr., G.C. (Ed.), Forage Quality,
Evaluation and Utilization (pp. 406- 449). Madison: Soil
Science Society of America/American Society of
Agronomy/Crop Science Society of America.
Shenk, J. S. and Westerhaus, M. O. (1996). Calibration the ISI
way. In Davies, A.M.C., & Williams, P.C. (Eds.), Near
Infrared Spectroscopy: The Future Waves (pp. 198-
202). Chichester: NIR Publications.
Shenk, J.S., Workman J. and Westerhaus, M. (2001).
Application of NIR spectroscopy to agricultural
products. In Burns, D.A., & Ciurczac, E.W. (Eds.),
Handbook of Near Infrared Analysis (pp. 419-
474).New York: Marcel Dekker.
Stchur, P., Cleveland, D., Zhou, J. and Michel, R.G. (2002). A
review of recent applications of near infrared
spectroscopy, and of the characteristics of a novel Pb
SCCD array-based near-infrared spectrometer. Appl.
Spectrosc. Rev., 37, 383-428.
Tadmor, Y., Larkov, O., Meir, A., Minkoff, M., Lastochkin, E.,
Edelstein, M., Levin, S., Wong, J., Rocheford, T. and
Lewinsohn, E. (2000). Reversed-phase high
performance liquid chromatographic determination of
vitamin E components in maize kernels.
Phytochem.l Anal., 11, 370-374.
Tkachuk, R. and Kuzina F.D. (1982). Chlorophyll analysis of
whole rape-seed kernels by near infrared reflectance.
Can. J. Plant Sci., 62, 875-884.
Williams, P.C. and Sobering, D.C. (1996). How do we do it: a
brief summary of the methods we use in developing
near infrared calibrations. In Davies, A.M.C., &
Williams, P.C. (Eds.), Near Infrared Spectroscopy: The
Future Waves (pp. 185-188). Chichester: NIR
Publications

Tables and Figures

414 935 1456 1977 2498
414 935 1456 1977 2498
0.987
0.758
0.530
0.301
0.073
0.775
0.597
0.420
0.243
0.065
Wavelength
L
o
g

(
1
/
R
)
L
o
g

(
1
/
R
)

Figure 1. Typical NIR spectra obtained for the different samples of epicarp (a) and mesocarp (b) of summer squash fruits analysed.

Figure 2. Mean near infrared spectrum of Cucurbita samples of epicarp (a) and mesocarp (b) fruits using SNVD and second
derivative pre-treatment.


Figure 3. NIR predicted data vs. chemical reference data for lutein content (mg kg
-1
dry weigth) in epicarp (a) and mesocarp (b) of
summer squash fruit.

Bellon-Maurel V., Williams P., Downey G., Eds 161
Application of near-infrared spectroscopy
to determine the -carotene content of
mango.
P. Rungpichayapichet
a
, B. Mahayothee
b
, M. Nagle
a
and J.Mller
a
a
Institute of Agricultural Engineering in the Tropics and Subtropics (440e), University of Hohenheim, 70599, Stuttgart, Germany
b
Department of Food Technology, Faculty of Engineering and Industrial Technology, Silpakorn University, 73000, Nakhon Pathom,
Thailand

Corresponding author: parika_rung@yahoo.com

Introduction

Vitamin A, or retinol, is one of the essential vitamins for
human nutrition. The function of vitamin A is to help the visual
system work normally, support growth and development and
maintain integrity of epithelial cells, immune system function,
and reproductive health (Chen et al., 2009; Rodriguez-Amaya,
2010; Orenalas-Paz et al., 2007). In plants, vitamin A is
produced as pro-vitamin A in the form of carotenoids that can
be converted to vitamin A during the digestion process. Mango
fruit is an excellent source of carotenoids, most of which are
found in the form of -carotene (Ajila et al., 2007; Chiumarelli
et al., 2010; Ma et al., 2011; Schieber et al., 2000; Sivakumar et
al., 2011). General laboratory methods for determining -
carotene content are usually destructive, expensive, time-
consuming, and require expertise. Recently, near-infrared
spectroscopy (NIRS) has been used as a non-destructive
technique for determination of carotenoids contents in
agricultural products. Many studies showed the potential of
NIRS to determine carotenoids contents, with R
2
of more than
0.7 for tomato (Baranska et al., 2006; Clment et al., 2008;
Nardo et al., 2009; Rubio-Diaz et al., 2011), potato (Boneirbale
et al., 2009), maize (Brenna and Berardo, 2004), Chinese kale
(Chen et al., 2009), banana (Davey et al., 2009), and apricot
(Ruiz et al., 2008). In this work, the feasibility of NIRS as a rapid
and non-destructive technique to evaluate -carotene content
in mango was studied.

Material and methods

Mature, unripe mango fruits (cv. Nam Dok Mai Si Thong)
were used. A total of 120 fruits were obtained from a local
export company. Fruits were washed and air-dried after
checking for external injuries and blemishes. The fruits were
ripened in baskets covered with paper at 33.0 1.2C and 63.9
6.9% relative humidity for up to seven days. Five fruits were
randomly chosen daily for the acquisition of NIR spectra,
determination of total soluble solids (TSS) content, titratable
acidity (TA) and -carotene content. Three replications were
carried out in this study
Sample temperature was controlled at 25C prior to
spectral acquisition. A scanning FT-NIR spectrometer (Tango,
Bruker, USA.) was used to acquire spectra of mango. Spectra
were measured at 10000-4000 cm
-1
. The resolution was set at 8
cm
-1
and spectra were obtained by averaging 32 scans in 3
positions and were recorded by OPUS software (version 6.5,
Bruker, USA.).

The Unscrambler software (version 9.8, Camo, Norway)
was used for chemometric analyses. Calibration was developed
by partial least squares regression (PLS) using full cross
validation. Savitzky-Golay smoothing (S), Savitzky- Golay second
derivative (D
2
) (varying number of averaging points at 15, 29
and 43 fixed at second order polynomial), standard normal
variate (SNV), multiplicative scatter correction (MSC) and a
combination of the two treatments were applied for
pretreating spectral data. The performance of the calibration
was evaluated using the coefficient of determination (R
2
),
standard error of cross validation (SECV), bias, and the ratio of
the standard deviation to SECV (RPD). The best equation was
chosen from the higher R
2
and RPD combined with the lower
SECV and bias values. The differences between actual and
predicted -carotene values were analyzed by paired t-tests
using the SAS program (version 9.0, USA.)
Extraction of -carotene was done using a method
adapted from Marx et al. (2000) and Pott et al. (2003). A
portion of flesh which had been scanned during the NIR
measurement (5-10 g) was removed and finely macerated.
Methanol (Merck, Germany) was used as a solvent for
extraction of the -carotene from the flesh and then the
mixture was filtered through a glass suction funnel.
The yellow-clear fluid was extracted in an amber glass
separation funnel using a mixture of acetone (Merck, Germany)
and hexane (Macron chemicals, USA) (1:1, v:v). After
separation, the hexane layer was washed with water (40 ml) to
remove acetone. The extract was dried with sodium sulfate (2
g) (Ajax Finechem, Australia). Hexane was evaporated at 25C
and 150 mbar using a rotary evaporator. The residue was
dissolved in isopropanol (Merck, Germany) and accurately
adjusted to a volume of 10 ml. Extracts were kept at -20C until
HPLC analysis.
Twenty microliters of extract were analyzed in a Shimadzu
HPLC system (LC20AD pump, CTO-10Asvp column oven, CBM-
20A system controller, Shimadzu, Japan) consisting of a UV-VIS
detector (SPD-M20A) set at 450 nm and a column C
18
(Inertsil
ODS-3 m 4.6 I.D x 150 mm) (GL sciences Inc, Japan). Methanol
: acetronitrile (7:3) was used as the mobile phase at a flow rate
3.0 mlmin
-1
. HPLC grade -carotene standard (type II, Sigma-
Aldrich, U.S.A) was used for the qualitative and quantitative
analysis.
Total soluble solids, TSS (Brix) and titratable acidity, TA
(mg citric acid100 g
-1
sample) contents were determined using
a digital refractometer (Atago, Japan) and titration with 0.1N
NaOH, respectively.


The -carotene content of mangoes ranged from 1.60-
926.40 g100 g
-1
of fresh edible parts (Table 1). Wide
variations in -carotene levels were found to be affected by
ripening stage of the samples (Figure 1). Changes in the -
carotene content of fruits showed exponential behavior,
slightly increasing for the first two days of ripening, then rising
rapidly on the third day and reaching the highest level on the
sixth day. This behavior is similar to nine Thai mango cultivars
previously studied (Vsquez-Caicedo et al., 2005) as well as
two Mexican cultivars Manila and Ataulfo (Ornelas-Paz et al.,
2008). From this study, the results showed Nam Dok Mai Si
Thong had lower -carotene contents than Indian mango
culitvars (Badami (3210 g100 g-1 of flesh pulp ), Mallika
(2800 g100 g
-1
of flesh pulp), Raspuri (1830 g100 g
-1
of
flesh pulp), Neelam (1450 g100 g
-1
of flesh pulp) and
Totapuri (1270 g100 g
-1
of flesh pulp) but higher than Malgoa
(550 g100 g
-1
of flesh pulp) (Veda et al., 2007) at the ripened
stage.
The original spectra of mangoes are illustrated in Figure 2.
Three overlapping band peaks at 8800-8000 cm
-1
, 7300-6100
cm
-1
and 5300-4500 cm
-1
were observed, which were assigned
to the absorptions of -CH second overtone, -OH first overtone
and first overtone of CH combination and -OH combination,
respectively (Osborne et al., 1993). The presence of these three
overlapping band showed the relationship between NIR
absorption at specific wavelengths to functional chemical
composition which included moisture, carbohydrates and
pigment contents in mango.
Statistics for the full cross-validation PLS model developed
from different mathematically pretreated spectra are
presented in Table 2. R
2
for calibration ranged from 0.77 - 0.84
and SECV was between 116.25 145.15 g100 g
-1
of flesh
edible parts (wet basis). The best calibration was obtained
when using Savitzky-Golay second derivative with a second
order polynomial at 43 averaging points for the pretreatment
of spectra. Application of the derivatives had a positive effect
on baseline shifts and superposed peak spectra, due to the fact
that it could correct for both additive and multiplicative effects
(Nicola et al., 2007) (Figure 3). This result was consistent with
those of Brenna and Berardo (2004) and Chen et al. (2009),
who also used second derivative pretreatments, which
provided the best calibration for predicting -carotene in
maize and Chinese kale, respectively.
A scatter plot between measured -carotene content and
NIRS-predicted values is shown in Figure 4. Error in prediction
was found for low -carotene concentrations, which may
indicate limited sensitivity of the spectrometer. However no
statistically significant differences (p0.05) were found
between actual and predicted -carotene values. The RPD
ranged from 1.64 to 2.07, indicating that the model could be
used for rough quantitative predictions (Nicola et al., 2007;
Clment et al., 2008; Davey et al., 2009; Penchaiya et al., 2009).

Conclusion

The feasibility of NIRS for predicting -carotene content in
mango was investigated. Full cross validation of the PLS model
from Savitzky-Golay second derivative pretreated spectra
showed that NIRS could be used as a non-destructive technique
for predicting -carotene content in mango cv. Nam Dok Mai
Si Thong. However, model robustness should be improved by
using a larger number of samples with a wider variation of -
carotene content.

Acknowledgement

This study is a part of the transfer project SFB 564 T4
funded by Deutsche Forchungsgemeinschaft (DFG), Germany.
We gratefully acknowledge their financial contributions. In
addition, the authors wish to express their appreciation to the
students and staff of the Silpakorn University as well as the kind
support of Mr. Paichayon Uathaveekul and the employees at
Swift Co., Ltd., Thailand.

Table 1. Statistical characteristics of -carotene analysis and chemical parameters from 120 mango fruits during ripening at ambient
temperature
Items -carotene TSS
TA
g per 100 g edible part (wb)
(Bri
x)
(mg100g
-1
sample)
Maximum 926.40 26.3 3.01
Minimum 1.60 8.7 0.07
Mean 249.40 16.9 1.04
Standard deviation 237.56 4.7 0.87

Table 2. Calibration and statistical results for -carotene prediction by NIR spectroscopy
Pretreatment R
2
r
2
SEC SECV Bias F RPD
S 15 0.81 0.71 103.92 129.11 -2.74 10 1.84
S 29 0.79 0.70 107.78 131.37 -2.85 10 1.81
S 43 0.78 0.69 110.92 134.19 -2.91 10 1.77
D2 15 0.80 0.63 105.04 145.15 -4.56 4 1.64
D2 29 0.84 0.72 96.58 127.28 -2.76 6 1.87
D2 43 0.86 0.77 89.40 114.82 -0.01 8 2.07
SNV 0.83 0.73 98.89 123.70 -1.33 10 1.92
SNV& S 15 0.81 0.73 102.32 123.92 -0.98 10 1.92
SNV & S 29 0.80 0.72 106.50 127.54 -1.37 10 1.86
SNV & S 43 0.77 0.70 113.96 132.21 -0.25 9 1.80
SNV & D2 15 0.80 0.65 106.38 141.21 -2.36 4 1.68
SNV & D2 29 0.80 0.74 107.21 122.92 0.47 5 1.93
SNV & D2 43 0.83 0.76 98.68 116.38 -0.69 6 2.04
MSC 0.83 0.74 98.52 122.75 -1.27 10 1.94
MSC & S 15 0.82 0.74 102.02 123.19 -1.02 10 1.93
MSC & S 29 0.80 0.72 106.31 126.84 -1.50 10 1.87
MSC & S 43 0.77 0.70 114.53 132.35 -0.45 9 1.79
MSC & D2 15 0.80 0.65 106.36 141.14 -2.35 4 1.68
MSC & D2 29 0.80 0.74 107.16 122.87 0.47 5 1.93
MSC & D2 43 0.83 0.76 98.57 116.25 -0.71 6 2.04
R
2
, coefficient of determination for calibration; r
2
, coefficient of determination for prediction; SEC, standard error of calibration; SECV, standard error
of cross validation; F, number of factor; RPD, the ratio of the standard deviation of sample to SECV; 15, 29, 43, number of averaging point

Figure 1. Increase of -carotene content in mango flesh (g100 g
-1
of edible part) during ripening.


Figure 2. Raw NIR spectra of 120 mango samples used in the study.

Figure 3. Second derivative spectra of 120 mango fruits.


Figure 4. Scatter plot showing prediction of -carotene content in mango by NIR spectroscopy. ( cross-validation set,
calibration set)

References

Ajila, C. M., Bhat, S.G., Prasada Rao, U.J.S. (2007). Valuable
components of raw and ripe peels from two Indian
mango varieties. Food Chemistry, 102, 1006-1011.
Baranska, M., Schtze, W., Schulz, H. (2006). Determination of
lycopene and -carotene content in tomato fruits and
related products: comparison of FT-Raman, ATR-IR, and
NIR spectroscopy. Analytical Chemistry, 78, 8456-8461.
Bonierbale, M., Grneberg, W., Amoros, W., Burgos, G., Salas,
E., Porras, E., Felde, T.Z. (2009). Total and individual
carotenoid profiles in Solanum phureja cultivated
potatoes: II. Development and application of near-
infrared reflectance spectroscopy (NIRS) calibrations for
germplasm characterization. Journal of Food
Composition and Analysis, 22, 509-516.
Brenna, O. V., Berardo, N. (2004). Application of near-infrared
reflectance spectroscopy (NIRS) to the evaluation of
carotenoids content in maize. Journal of Agricultural
and Food Chemistry, 52, 5577-5582.
Chen, X., Wu, J., Zhou, S., Yang, Y., Ni, X., Yang, J., Zhu, Z., Shi, C.
(2009). Application of near-infrared reflectance
spectroscopy to evaluate the lutein and -carotene in
Chinese kale. Journal of Food Composition and Analysis,
22, 148-153.
Chiumarelli, M., Pereira, L.M., Ferrari, C.C., Sarantpoulos,
C.I.G.L., Hubinger, M.D. (2010). Cassava starch coating
and citric acid to preserve quality parameters of fresh-
cut Tommy Atkins mango. Journal of Food Science,
75(5), 297-304.
Clment, A., Dorais, M., Vernon, M. (2008). Nondestructive
measurement of fresh tomato lycopene content and
other physicochemical characteristics using visible-NIR
spectroscopy. Journal of Agricultural and Food
Chemistry, 56, 9813-9818.
Davey, M. W., Saeys, W., Hof, E., Ramon, H., Swennen, R.L.,
Keulemans, J. (2009). Application of visible and near-
infrared reflectance spectroscopy (vis/NIRS) to
determine carotenoid contents in banana (Musa spp.)
fruit pulp. Journal of Agricultural and Food Chemistry,
57, 17421751.
Ma, X., Wu, H., Liu, L., Yao, Q., Wang, S., Zhan, R., Xing, S., Zhou,
Y. (2011). Polyphenolic compounds and antioxidant
properties in mango fruits. Scientia Horticulturae, 129,
102-107.
Marx, M., Schieber, A., Carle, R. (2000). Quantitative
determination of carotene stereoisomers in carrot
juices and vitamin supplemented (ATBC) drinks. Food
Chemistry, 70, 403-408.
Nardo, T. D., Shiroma-Kian, C., Halim, Y., Francis, D., Rodriguez-
Saona, L.E. (2009). Rapid and simultaneous
determination of lycopene and -carotene contents in
tomato juice by infrared spectroscopy. Journal of
Agricultural and Food Chemistry, 57, 1105-1112.
Nicola, B. M., Beullens, K., Bobelyn, E., Peirs, A., Saeys, W.,
Theron, K.I., Lammertyn, J. (2007). Nondestructive
measurement of fruit and vegetable quality by means
of NIR spectroscopy: A review. Postharvest Biology and
Technology, 46, 99-118.
Ornelas-Paz, J. D. J., Yahia, E.M., Gardea-bejar, A. (2007).
Identification and quantification of xanthophyll esters,
carotenes, and tocopherols in the fruit of seven
mexican mango cultivars by Liquid Chromatography-
Atmospheric Pressure Chemical Ionization-Time-of-
Flight Mass Spectrometry [LC-(APcI
+
)-MS]. Journal of
Osborne, B.G., Fearn, T., Hindle, P.H. (1993).Practical NIR
spectroscopy : with applications in food and beverage
analysis. Singapore: Longman Singapore Publishers
(Pte) Ltd.
Penchaiya, P., Bobelyn, E., Verlinden, B.E., Nicola, B.M., Saeys,
W. (2009). Non-destructive measurement of firmness
and soluble solids content in bell pepper using NIR
spectroscopy. Journal of Food Engineering, 94, 267-273.
Pott, I., Marx, M., Neidhart, S., Mhlbauer, W., Carle, R. (2003).
Quantitative determination of -carotene
stereoisomers in fresh, dried, and solar-dried mangoes
(Mangifera indica L.). Journal of Agricultural and Food
Chemistry, 51, 4527-4531.
Rodriguez-Amaya, D. B. (2010). Quantitative analysis, in vitro
assessment of bioavailability and antioxidant activity of
food carotenoidsA review. Journal of Food
Composition and Analysis, 23, 726-740.
Rubio-Diaz, D. E., Francis, D.M., Rodriguez-Saona, L.E. (2011).
External calibration models for the measurement of
tomato carotenoids by infrared spectroscopy. Journal of
Food Composition and Analysis, 24, 121-126.
Ruiz, D., Reich, M., Bureau, S., Renard, C.M.G.C., Audergon,
J.M. (2008). Application of reflectance colorimeter
measurements and infrared spectroscopy methods to
rapid and nondestructive evaluation of carotenoids
content in apricot (Prunus armeniaca L.). Journal of
Schieber, A., Ullrich, W., Carle, R. (2000). Characterization of
polyphenols in mango puree concentrate by HPLC with
diode array and mass spectrometric detection.
Innovative Food Science and Emerging Technologies, 1,
161-166.
Sivakumar, D., Jiang, Y., Yahia, E.M. (2011). Maintaining mango
(Mangifera indica L.) fruit quality during the export
chain. Food Research International, 44, 1254-1263.
Veda, S., Platel, K., Srinivasan, K. (2007). Varietal differences in
the bioaccessibility of -carotene from mango
(Mangifera indica) and papaya (Carica papaya) fruits.
Journal of Agricultural and Food Chemistry, 55, 7931-
7935.
Vsquez-Caicedo, A. L., Sruamsiri, P., Carle, R., Neidhart, S.
(2005). Accumulation of all-trans--carotene and its 9-
cis and 13-cis stereoisomers during postharvest
ripening of nine Thai mango cultivars. Journal of

B e l l o n - M a u r e l V . , W i l l i a m s P . , D o w n e y G . , E d s 167
In situ monitoring of fruit ripening during
postharvest storage using a miniature
handheld NIR sensor.
Sanchez M T.
a
; Garrido-Varo A.
b
; De la Haba M J
a
; Fernandez-Novales
J.
a
; Guerrero-Ginel J E.
b
; Perez-Marin D.
b

a
Department of Bromatology and Food Technology, University of Cordoba, Spain
b
Department of Animal Production, University of Cordoba, Spain

Corresponding author: teresa.sanchez@uco.es
Introduction
The introduction of effective analytical techniques for
measuring in-situ quality evaluation of fruit during postharvest
storage, and establishing shelf-life remains something of a
challenge for the food industry, which requires non-destructive
methods that are sufficiently accurate, not only to identify
potential quality differences during the storage, but also to fix
the limit of palatability. Near-infrared reflectance (NIR)
spectroscopy is a promising analytical technique likely to meet
many of the fruit industrys requirements (Saranwong and
Kawano, 2007; Snchez and Prez-Marn, 2011). Moreover, in
recent years NIR instruments have undergone radical changes,
becoming much more versatile, more compact and portable,
cheaper and better adapted to hostile working areas.
In this study, a MEMS (micro-electro-mechanical
system) spectrometer was assessed for the determination in-
situ of quality (soluble solid content, firmness) of intact plums
and strawberries during postharvest storage under refrigeration
conditions.
Materials and Methods
Sampling
A total of 264 plums (Prunus salicina L.) belonging to
six cultivars and harvested at commercial stage in the Province
of Seville (Spain) were evaluated. All varieties were received in 6
batches of 44 plums. Batches were stored at 0C and 95% RH for
9 days, with samples drawn for analysis at 6 and 9 day intervals.
The raw material samples (0 days storage) served as control.
Strawberries (Fragaria x ananassa Duch.) belonging to
nine cultivars, were picked at Gibralen (Huelva, Spain) at
commercial ripeness stage. A total of 189 punnets were
evaluated; 108 punnets were processed as soon as they reached
the laboratory, and 81 punnets were placed in cold storage (0C,
95% RH) and sampled at 3 and 6 days. From each of the punnets
received and processed, 5 strawberries were selected to form
the sampling unit.
Spectra collection
Spectra were collected on all fruit in reflectance mode
(Log 1/R) using a handheld MEMS spectrometer (Phazir 2400,
Polychromix, Inc., Wilmington, MA, USA). The instrument scans
at intervals of approximately 8 nm (pixel resolution 8 nm, optical
resolution 12 nm), across the range of NIR wavelengths (1600-
2400 nm).
For plum, two spectral measurements were made with
this instrument from both sides of the equator on each intact
fruit. The two spectra were averaged to provide a mean
spectrum for each fruit.
Using this instrument, four spectral measurements
were made on each strawberry. The first was made on the
widest part of the shoulder, and the fruit was then turned
through 90 for each successive measurement. To obtain a
mean spectrum for each punnet, a mean spectrum was first
obtained for the four spectra for each fruit, and then a mean
spectrum was obtained from the five mean spectra for each
punnet.
Determination of physical-chemical quality parameters
The Magness-Taylor technique was used to measure
the maximum force required to pierce the plum to a depth of 10
mm with an 8 mm diameter probe. For strawberry the
puncturing depth was of 5 mm measured with a 3 mm diameter
probe. The measurement was carried out in a Universal Testing
Machine (Model 3343, Single Column, Instron Corporation,
Norwood, MA, USA). In plum the firmness of each individual
fruit was measured at two positions around the equator,
approximately 180 apart, and perpendicular to the stem-calyx
axis, and flesh firmness was calculated as the average of two
measurements per fruit. For strawberry, texture was measured
on all individual fruits at only one position on the widest part of
the shoulder and perpendicular to the stem-calyx axis. Values
for each of the five fruits were averaged to provide a punnet
value.
For plums two longitudinal (from stem end to calyx
end) wedges were removed from each fruit, pressed through
cheesecloth, and the soluble solids content (SSC) of the juice
was measured with a temperature-compensated refractometer
(model ATC-1, Atago Co., Tokyo, Japan). The SSC of strawberries
was determined using five fruits per punnet those used for
firmness measurements, using a traditional destructive test
(Aguayo et al., 2006). Each sample was analyzed in duplicate.
Quantitative calibrations: sets and data processing
The WinISI software package was used for the
chemometric treatment of data (ISI, 2000).
Next, and prior to carrying out NIRS calibrations, the
CENTER algorithm included in the WinISI II software package
was applied to ensure a structured population selection based

solely on spectral information for the establishment of
calibration and validation sets (Shenk and Westerhaus, 1995).
This algorithm performs an initial principal component analysis
(PCA) to calculate the center of the population and the distance
of samples (spectra) from that center in n-dimensional space,
using the Mahalanobis distance (GH); samples with a statistical
value of greater than GH3 were considered to be outliers or
anomalous spectra.
The CENTER algorithm was applied in the spectral
region 16002400 nm. Mathematical treatments SNV (Standard
Normal Variate) and DT (De-trending) were applied for scatter
correction (Barnes et al., 1989), together with the mathematical
derivation treatment 1,5,5,1, where the first digit is the
number of the derivative, the second is the gap over which the
derivative is calculated, the third is the number of data points in
a running average or smoothing, and the fourth is the second
smoothing (Shenk and Westerhaus, 1995; ISI, 2000).
After the outlier spectra samples were eliminated (N =
4 for plums; N = 2 strawberries), approximately 15% of samples
forming the validation sets were selected by taking one of every
seven samples in the initial set for plum and one of every six for
strawberry.
Prediction equations were obtained using the MPLS
regression method with cross-validation (Shenk and
Westerhaus, 1995). Different mathematical treatments were
evaluated for scatter correction, including SNV and DT methods
(Barnes et al., 1989). Two derivative treatments: 1,5,5,1; 2,5,5,1
(Shenk and Westerhaus, 1995) were found to be the most
effective.
The statistics used to select the best equations were:
SEC, R
2
, SECV, r
2
, RPD, and CV. These latter two statistics enable
SECV to be standardized, facilitating the comparison of the
results obtained with sets of different means (Williams, 2001).
The best-fitting equations obtained for the calibration
set were subsequently evaluated by external validation
following the protocol outlined by ISI (2000) and Shenk et al.,
(2001).
Descriptive data for NIR calibrations and validations
Values for range, mean, standard deviation (SD) and
coefficient of variation (CV) for each of the parameters analyzed
using the calibration and validation sets after application of the
CENTER algorithm and removal of spectral outliers are shown in
Table 1. The number of samples for each parameter measured
for the two fruits tested was also included.
All parameters displayed considerable variation,
mainly of the firmness parameter, as evidenced by the CV
values.

Predictive models for quality parameters in plum and
strawberry
Table 2 provides statistics for the best-fitting
equations obtained for the prediction of SSC and firmness, and
for each of the two fruits tested.
For plum, the best results were obtained using D1
Log(1/R), for both parameters while with strawberry the best
results were obtained using D2 Log(1/R), for both parameters.
As Table 2 shows, the effectiveness of the models
developed for SSC (r
2
= 0.57 and 0.69; SECV = 1.39% and 0.79%
for plum and strawberry, respectively) may be considered
acceptable for very rough screening, enabling samples to be
classified as high, medium or low. It is important to highlight
that this was achieved using a unique, rapid, non-destructive,
handheld sensor, which would provide the fruit industry with an
instant response and enable plum and strawberry harvesting to
be started at the optimum time, and also allow the end-point of
the fruits commercial life to be established.
Validations of the best calibration models obtained for
SSC were performed (Figure 1). In plum, five samples were
excluded from the validation set because they displayed
Mahalanobis H values greater than 4, and were thus considered
as outliers, meaning that those samples could not be reliably
predicted using the models developed. No samples were
excluded from the validation set in strawberry.
Examination of the validation statistics shown in Figure
1 and the calibration statistics (Table 2) revealed similar SECV
and SEP values for NIRS prediction of SSC, in both fruits,
suggesting that application of these models was feasible. Using
the monitoring procedure proposed by Shenk et al., (2001), for
this quality parameter, both the SEP(c) and the bias were below
the confidence limits except for SEP(c) in strawberry. The r
2
and
slope results attained recommended minimum values except for
r
2
in strawberry (ISI, 2000).
However, for firmness prediction, the low predictive
capacity of the models performed with the MEMS instrument
(r2 = 0.29 and 0.38; SECV = 2.76 N and 0.18 N), may be because
the Phazir 2400 does not cover the first infrared region, which
includes bands strongly related to C-H bonds, associated with
constituents such as cellulose; it may also be due to the
measurement area involved (around 2 mm at each
measurement). Although Prez-Marn et al., (2009) report that
the Phazir 2400 instrument yielded high-precision calibrations
for predicting SSC and firmness in nectarines, it should be
stressed that plum spectra are noticeably different from those
of other stone fruits, such as peach and nectarine; the flesh is
more translucent, and thus absorbs less light (Golic and Walsh,
2006). Moreover, the lower predictive ability obtained for this
texture-related parameter for both fruits may be due to the
difficulty in predicting this parameter in certain fruits using NIRS
technology with MPLS regression (Prez-Marn et al., 2010).
Models were not externally validated due to their low predictive
capacity.


Figure 1. Reference measured data versus NIRS predicted data
for the validation sets for SSC (%), for the two fruits tested.

This research was funded by the Andalusian Regional
Government under the Research Excellence Program (project
no. P09-AGR-5129 MEMS and NIRS-image sensors for the in
situ nondestructive analysis of food and feed. The authors are

Acknowledgements

grateful to Mr. Antonio Lpez and Mrs. M Carmen
Fernndez of the Animal Production Department (ETSIAM-UCO)
for technical assistance.
References
Aguayo, E., Jansasithorn, R., Kader, A.A. 2006. Combined effects
of 1-methylcyclopropene, calcium chloride dip, and/or
atmospheric modification on quality changes in fresh-
cut strawberries. Postharvest Biology and Technology
40, 269278.
Barnes, R.J., Dhanoa, M.S., Lister, S.J. 1989. Standard normal
variate transformation and de-trending of near
infrared diffuse reflectance spectra. Applied
Spectroscopy 43, 772777.
Golic, M., Walsh, K.B. 2006. Robustness of calibration models
based on near infrared spectroscopy for the in-line
grading of stone fruit for total soluble solids content.
Analytical Chimica Acta 555, 286291.
ISI, 2000. The Complete Software Solution Using a Single Screen
for Routine Analysis, Robust Calibrations and
Networking. Manual. Foss NIRSystems/Tecator. LLC,
Silver Spring, MD: Infrasoft International.
Prez-Marn, D., Snchez, M.T., Paz, P., Soriano, M.A.,
Guerrero,J.E., Garrido-Varo, A. 2009. Non-destructive

Table 1
Statistical analysis of calibration and prediction sample sets, i.e., number of samples (N),
data ranges, means, standard deviations (SD) and coefficients of variation (CV),
for plums and strawberries
Parameter Fruit Set N Range Mean SD CV (%)
SSC (%)
Plum
Calibration
220 8.30-
19.60
13.34 2.20 16.50
Validation 40 9.95-
17.60
13.13 2.14 16.30
Strawberry
Calibration
157 5.40-
11.55 7.48 1.46 19.52
Validation 30 5.85-
11.15 7.84 1.53 19.52
Plum
Calibration
220 1.93-
16.12 6.16 3.35 54.38
Validation 40 2.03-
15.25
6.48 3.48 53.70
Strawberry
Calibration
157 0.44-
1.69 0.94 0.28 29.79
Validation 30 0.56-
1.46
0.90 0.24 26.67
Firmness
(N)

Table 2.
Calibration statistics for internal quality parameters (SSC and firmness) in plums and strawberries
using MPLS regression.
Parameter Frui t Mathemati c
treatment
Mean SD SEC
R
2
SECV
r
2
RPD CV (%)
SSC (%) Pl um 1,5,5,1 13.44 2.12 1.28 0.57 1.39 0.57 1.52 15.77
Strawberry 2,5,5,1 7.42 1.42 0.72 0.74 0.79 0.69 1.79 10.64
Pl um 1,5,5,1 6.12 3.25 2.52 0.29 2.76 0.29 1.17 53.10
Strawberry 2,5,5,1 0.89 0.23 0.17 0.43 0.18 0.38 1.27 20.22
Fi rmness (N)

determination of quality parameters in nectarines
during on-tree ripening and postharvest storage.
Prez-Marn, D., Paz, P., Guerrero, J.E., Garrido-Varo, A.,
Snchez, M.T. 2010. Miniature handheld NIR sensor
for the on-site non-destructive assessment of post-
harvest quality and refrigerated storage behavior in
plums. Journal of Food Engineering 99, 294302.
Snchez, M.T., Prez-Marn, D. 2011. Nondestructive
measurement of fruit quality by NIR spectroscopy. In:
Vzquez, M., Ramrez, J.A. (Eds.), Advances in Post-
Harvest Treatments and Fruit Quality and Safety, pp.
101163. Hauppauge, NY: Nova Science Publishers, Inc.
Saranwong, S., Kawano, S. 2007. Applications to agricultural and
marine products: Fruits and vegetables. In: Ozaki, Y.,
McClure, W.F., Christy, A.A. (Eds.), Near-Infrared
Spectroscopy in Food Science and Technology, pp.
219242. New Jersey: John Wiley & Sons, Inc.
Shenk, J.S., Westerhaus, M.O. 1995. Analysis of Agriculture and
Food Products by Near Infrared Reflectance
Spectroscopy. Monograph. Silver Spring, MD:
NIRSystems Inc.
Shenk, J.S., Workman, J., Westerhaus, M. 2001. Application of
NIR spectroscopy to agricultural products. In: Burns,
D.A., Ciurczac, E.W. (Eds.), Handbook of Near Infrared
Analysis, 2
nd
Edition, pp. 419474. New York: Marcel
Dekker.
Williams, P.C. 2001. Implementation of near-infrared
technology. In: Williams, P.C., Norris, K.H. (Eds.), Near-
infrared Technology in the Agricultural and Food
Industries, pp. 145169. St. Paul, MN: AACC, Inc.
Forecast of apple internal quality indices
during storage by spectroscopy.
Z. Schmilovitch
a
, T. Ignat
a
, J. Nyasordzi
b
, V. Ostrovsky
a
, H. Egozi
a
, A.
Hoffman
a
, H. Friedman
b
, A. Weksler
b
, I. Rot
b
, S. Lurie
b

a
Department of Agricultural Engineering, Volcani Center, Agricultural Research Organization, Bet Dagan 50250, Israel
b
Department of Postharvest Science, Volcani Center, Agricultural Research Organization, Bet Dagan 50250, Israel

Corresponding author: veshmilo@volcani.agri.gov.il

Introduction

Prediction of the optimal harvest date of apples is of high
importance as it controls the fruit quality after long-term
storage in cooled and controlled atmosphere conditions.
Currently, most farmers rely on pre-harvest sampling of a
limited number of fruit and examination of their internal
characteristics by destructive measurements (Nyasordzi et al.,
2013). Parameters measured include fruit firmness, starch
and soluble solids contents (SSC). Visible/near infrared
spectroscopy (VIS/NIR) is developing as a major technique to
determine internal quality including SSC and firmness in a
number of fruit and vegetables (Nicolai et al., 2007). This
method has proven to be useful for the measurement of
apple SSC, acidity and firmness (Lovasz et al., 1994; Ventura et
al., 1998; Piers et al., 2000; McGlone et al., 2002). A number
of papers have shown the relevance of spectral data to
determination of time of harvest (Piers et al., 2000, 2003,
2005; Liu and Ying, 2005; Qing et al., 2008; Fan et al., 2009;
Bertone et al., 2012) although few have attempted to predict
post-storage apple quality from measurements taken at
harvest.
Firmness and SSC are the properties of apples which
determine consumer acceptance although the sugar/acid
ratio can also be important. Changes in these parameters
during storage (decrease in firmness, change in sugar/acid
due to increase in SSC and decrease in acids) can affect apple
quality during post-storage marketing. The prediction of shelf
life has rarely been addressed using NIR technology.

Camps et al. (2007) used NIR measurements to distinguish
between apple varieties and between storage type and
storage duration. Rizzolo et al. (2010) used time-resolved
spectroscopy (TRS) to examine eating and sensory quality of
apples after six months of air or controlled atmosphere
storage. However, neither of these studies were predictive;
they examined the apples after storage.

In recent years, NIR instruments have undergone radical
changes. They are much more versatile in terms of the infra-
red region in which measurements can be made, more
portable, cheaper and better adapted to hostile working areas
(e.g. high temperatures, vibrations). They are thus better
suited to use on the processing line and do not necessitate a
product being taken into a laboratory setting for
measurements. For estimation of internal quality in fruit, the
penetration depth of the light into the fruit is another factor
to be considered. In reflectance mode, Lammertyn et al.
(2000) found a penetration depth for apple of up to 4 mm in
the 700900 nm range and between 2 and 3mm in the 900
1900 nm range. For wavelengths in the visible range,
penetration was deeper.
The present study sought to assess the feasibility of
rapidly measuring apple quality using two commercial
spectrophotometers: USB 2000 (VIS-NIR, 340-1014 nm) and
Liga (SWIR, 850-1888 nm). The specific objectives were (1) to
compare both spectrophotometers on apples in a static mode
and on a moving conveyer, (2) to use both instruments to
measure three apple cultivars at harvest and after different
storage durations, and (3) to use the harvest determinations
to predict apple quality after storage.

Fruit material and their handling
Apples from three cultivars (Starking, Granny Smith and
Pink Lady) were collected from three orchards, 200 fruit from
each orchard (600 fruit per cultivar). Harvest dates were
September 19 for Starking, October 10 for Granny Smith and
November 11 for Pink Lady. All the samples were collected in
5 kg crates and held at 10
o
C, 95% relative humidity (RH) on
arrival for 1 day

before measurements commenced. The
following day, fruits from each orchard were numbered,
weighed and measured spectrally on each individual fruit.
Measurements were taken at two equatorial points at 90
o
to
the location of the number on each apple. Samples were then
measured destructively as detailed below. Fifty fruits from
each orchard were used for harvest measurements and the
remainder were placed in storage at 0C, 95% RH for 2, 4 and
6 months. At each removal time-point, 50 fruits from each
orchard were taken from storage and spectra were collected.
Then fruits were examined destructively.
Total soluble solids (TSS), titratable acidity (TA), starch
content and firmness were measured destructively at harvest
and removal. TSS and TA were measured from freshly
prepared juice from each individual fruit. An automatic digital
refractometer (Atago, Tokyo, Japan) was used to measure the
TSS readings and results were expressed as Brix %. TA was
determined by titrating 1 ml of apple juice with 50 ml of
distilled water and 0.1M NaOH to pH 8.2 using an automatic
titrator (Metrohm, Geneva, Switzerland). Results were
expressed as % of malic acid/volume juice. Firmness was
measured using a penetrometer (Penefil, Lyons, France) with
an 11 mm tip on opposite pared sides of each apple. At
harvest (only) starch content was determined. A slice of apple
taken from the equatorial area of the fruit was dipped into an
I/KI solution and the visible pattern of starch rated on a scale
from 1 to 8, with 1 showing starch in all the flesh and 8
showing no starch left in the flesh.

Spectral measurement methods

USB 2000 (VIS-NIR) and Liga (SWIR) miniature
spectrophotometers were used for taking spectral
measurements in two different modes; static and in motion.
The USB2000 (Ocean Optics, Dunedin, FL, USA) mini
spectrometer had spectral range of 340-1014 nm; grating: 600
lines blazed at 750 nm; optical spectral resolution at full-width
at half-maximum 1.2 nm. A bidirectional reflection probe
(BIF600-UVeVIS, Ocean Optics, Dunedin, FL, USA) was used in
static mode with one fibre to collect light reflected towards
the spectrometer and a bundle of six fibres to carry incident
light from the LS-1 Tungsten Halogen Light Source (Ocean
Optics, Dunedin, FL, USA). The spectrometer yielded 2048
data points with a spectral sampling interval of 0.5 nm. The
bidirectional reflection probe did not contact the sample and
was mounted on the top of a hollow cone (25 mm in diameter
at the base, 15 mm in height).
Spectral measurements were also taken with a Liga SWIR
spectrophotometer (STEAG Micro Parts, Dortmund, Germany)
based on a 128-diode InGaAs diode array. It has a tungsten
halogen light as a light source. A single-directional optical
fibre was mounted on the top of a hollow cone (30 mm in
diameter, black Delrin plastic) attachment as described above.
The incident beam from the light source fell perpendicularly
onto the fruit sample and radiation reflected at 45 angles
was collected. Altogether, 128 data points within an 850 to
1888 nm sampling interval were acquired in each scan at an
optical spectral resolution of 8.1 nm. Both configurations
were calibrated with a Spectralon, WS- 1-SL standard white
ceramic background disc (Ocean Optics, Dunedin, FL, USA).
Measurements in motion were conducted with a
conveyer (Figure 1) with 24 fruit holder cells at a speed of one
sample/second. The cell holder was designed to convey the
fruit through a measurement compartment. The
measurement compartment includes a 50 W halogen light
source and a connector for the fibre optic of the mini-
spectrometer, suitable to project the reflected light from the
upper surface of the apple to both USB 2000 and Liga
spectrophotometers. The light source illuminated the sample
vertically while the fibre optic collected the reflected light at a
45 inclination. In both spectral ranges and both
measurement modes three scans were averaged for each
apple sample.

Figure 1. SWIR/VIS-NIR measurement conveyor

In the chemometric data analysis, spectral information
from the apples at harvest were related to the TSS, TA and
starch content and firmness of the fruit after storage. This
enabled the prediction of change in TSS, TA and starch
content and firmness of apples from the time of harvest.
Spectral pre-treatments of the relative reflectance (R) such as
first derivative (D
1
R), log(1/R), D
1
log(1/R) and D
2
log(1/R) were
applied. Partial least squares regression (PLSR) (PLS;
Eigenvector Research, Wenatchee, WA, USA), run under
Matlab software R2012a (MathWorks, Natick, MA, USA) was
used for model development. Comparisons were made
betwen PLSR models based on spectra recorded on the two
spectrophotometers (USB 2000 and Liga) and two
measurement modes (static and motion). The following
statistical parameters are shown for each model: number of
latent variables (LV), coefficient of determination (R
2
), root-
mean-square error of calibration (RMSEC), root-mean-square
error of cross-validation (RMSECV), root-mean-square error of
prediction (RMSEP), residual predictive deviation (RPD), ratio
of RMSEP and RMSEC, and standardised weighted sum index
(SWS). Comparisons of the regression models were conducted
by the standardised weighted sum index (SWS) (Ignat et al.,
2012). PLSR was performed on 224 samples consisting of the
three apple cultivars (Starking, Granny Smith and Pink Lady)
which were randomly divided into a calibration cross-
validation (150 samples) and prediction (74) set.
Results of apple TSS, TA, firmness and starch predictions,
using data from the VIS-NIR and SWIR spectral measurements
are presented in Table 1 (for D
1
log(1/R)). Models were
established using all three apple cultivars (Starking, Pink Lady
and Granny Smith) at harvest in both static and motion
measurement modes. For TSS, TA, firmness and starch, best
results were found by VIS-NIR in motion, VIS-NIR static, SWIR
static and SWIR static, respectively. For motion mode
measurements, both spectrometers resulted in high
correlations for TSS and starch. The model of VIS-NIR in
motion for TSS with 4 LV yielded an RMSEP of 0.8 Brix% ,RPD
equal to 2.6 and SWS of 0.57.
Results of apple TSS, TA contents and firmness forecast at
harvest for two, four and six months storage are presented in
Table 2. For four months storage, the best models were
achieved by the spectral pre-treatments of log(1/R) and
D
1
log(1/R). Comparing the two spectrophotometers, based on
the SWS index in both cases predicting TSS and TA, the SWIR
spectrophotometer resulted in better models both in static
and in motion measurements than the VIS-NIR instrument.
Comparison of the measurement arrangements resulted in
slightly better prediction models in motion for TSS and TA for
both spectrophotometers. The overall comparison of the PLSR
models resulted in the best prediction of TSS, TA and firmness
being produced by SWIR in motion. TSS had prediction R
2

equal to 0.88, RMSEP of 0.83, RPD of 2.6 and SWS equal to
0.73; similarly, TA had an R
2
value of 0.84, RMSEP equal to
0.064, RPD of 2.3 and SWS equal to 0.88. Corresponding
values for firmness were 0.60, 1.2, 1.7 and 0.58. The VIS-NIR
bands have lower resolution and shallower penetration depth
but, at the same time, the relative intensity of the chemical
component bonds in these bands are weaker, meaning that,
in the SWIR region, the lower overtones of absorbance are
detected. This explains the better results obtained using SWIR
bands for prediction of TSS and TA components.
It is interesting that the motion mode give higher SWS
values than static measurements. One explanation is that, in
motion, the scanned area of the samples is greater and
therefore reflects more accurately the individual apples
compared to the static mode in which the fibre optic observes
a relatively smaller area.

Table 1. Performance measurements of PLSR models for the prediction of TSS, TA, firmness and starch, using data from the VIS-NIR
(USB 2000) and SWIR (Liga) spectral measurements. Models were established using the three apple cultivars (Starking, Pink Lady and
Granny Smith) at harvest.

Component
Measurement
arrangement
Spectrometer LV
Cal R
2
Pred R
2
RMSEC RMSECV RMSEP RPD
RMSEP/
RMSEC
SWS
USB 2000 6 0.89 0.84 0.43 0.59 0.87 2.4 2.0 0.40
Liga 5 0.90 0.82 0.56 0.63 0.79 2.6 1.4 0.53
USB 2000 4 0.78 0.86 0.56 0.70 0.80 2.6 1.4 0.57
Liga 5 0.90 0.86 0.54 0.60 0.81 2.6 1.5 0.53
USB 2000 5 0.76 0.77 0.03 0.03 0.03 2.1 1.3 0.62
Liga 5 0.78 0.59 0.03 0.03 0.04 1.7 1.6 0.39
USB 2000 4 0.59 0.66 0.03 0.04 0.04 1.7 1.4 0.45
Liga 7 0.78 0.53 0.03 0.03 0.05 1.6 1.7 0.29
USB 2000 6 0.70 0.68 1.11 1.35 1.51 1.9 1.4 0.35
Liga 3 0.69 0.65 1.16 1.25 1.40 2.1 1.2 0.47
USB 2000 3 0.65 0.76 1.25 1.43 1.48 1.9 1.2 0.44
Liga 6 0.72 0.66 1.08 1.28 1.49 1.9 1.4 0.36
USB 2000 6 0.92 0.89 0.48 0.60 0.89 2.6 1.8 0.21
Liga 6 0.93 0.87 0.51 0.62 0.80 2.9 1.6 0.33
USB 2000 5 0.85 0.88 0.53 0.81 0.87 2.6 1.6 0.26
Liga 6 0.91 0.86 0.60 0.68 0.86 2.7 1.4 0.25
Total soluble solid,
Brix%
Static
Motion
Titrable acidity, %
Static
Motion
Firmness, lb
Static
Motion
Starch, (1-8)
Static
Motion
Table 2. Performance measurement of PLSR models for forecasting (for 2, 4 and 6 months of storage) TSS, TA and firmness using data
from the VIS-NIR (USB 2000) and SWIR (Liga) spectral measurements. Models were established using the three apple cultivars
(Starking, Pink Lady and Granny Smith).

An example of the prediction results for TSS for all three
varieties after 4 months storage by a D
1
log(1/R) PLS
regression model developed on spectra recorded at harvest
are presented as a scatter plot in Figure 2. A high correlation
was found (R
2
= 0.91) together with an RMSEP value of 0.83
using 6 LVs. Calibration samples are marked with black circles
and prediction samples with red triangles (Figure 2). For the
same model, VIP scores are presented on Figure 3; the most
significant wavelengths were found to be 880, 980, 1077,
1174, 1215, 1452, 1707, 1814 nm.
Component Storage time
Measurement
arrangement
Spectrometer
Spectral
processing
LV Cal R
2
Pred R
2
RMSEC RMSECV RMSEP RPD
RMSEP/
RMSEC
SWS
USB 2000 log(1/R) 7 0.89 0.85 0.56 0.75 0.83 2.6 1.5 0.46
Liga D
1
log(1/R) 5 0.90 0.90 0.61 0.70 0.68 3.2 1.1 0.77
USB 2000 D
1
log(1/R) 4 0.84 0.82 0.41 0.84 0.79 2.7 1.9 0.54
Liga D
1
log(1/R) 8 0.90 0.86 0.55 0.69 0.89 2.4 1.6 0.37
USB 2000 log(1/R) 3 0.80 0.76 0.75 0.97 1.02 2.1 1.4 0.18
Liga D
1
log(1/R) 5 0.88 0.88 0.61 0.69 0.86 2.5 1.4 0.48
USB 2000 D
1
log(1/R) 3 0.98 0.83 0.63 0.94 0.91 2.3 1.4 0.40
Liga D
1
log(1/R) 6 0.90 0.91 0.59 0.67 0.83 2.6 1.4 0.53
USB 2000 log(1/R) 5 0.87 0.88 0.85 0.96 0.99 2.8 1.2 0.31
Liga D
1
log(1/R) 4 0.93 0.92 0.59 0.72 0.78 3.5 1.3 0.69
USB 2000 D
1
log(1/R) 5 0.90 0.88 0.25 0.89 0.89 3.1 3.5 0.39
Liga D
1
log(1/R) 5 0.92 0.94 0.59 0.70 0.80 3.4 1.3 0.66
USB 2000 log(1/R) 7 0.89 0.91 0.04 0.06 0.06 3.5 1.5 0.94
Liga D
1
log(1/R) 5 0.89 0.87 0.05 0.06 0.07 2.9 1.4 0.90
USB 2000 D
1
log(1/R) 4 0.87 0.87 0.03 0.07 0.07 2.9 2.1 0.87
Liga D
1
log(1/R) 8 0.86 0.80 0.06 0.07 0.10 2.1 1.8 0.79
USB 2000 log(1/R) 5 0.83 0.76 0.03 0.06 0.07 2.1 2.1 0.78
Liga D
1
log(1/R) 5 0.86 0.81 0.05 0.06 0.07 2.2 1.4 0.84
USB 2000 D
1
log(1/R) 3 0.84 0.82 0.05 0.06 0.06 2.4 1.4 0.87
Liga D
1
log(1/R) 4 0.88 0.84 0.05 0.05 0.06 2.3 1.3 0.88
USB 2000 log(1/R) 6 0.36 0.33 0.04 0.05 0.06 1.5 1.5 0.63
Liga D
1
log(1/R) 4 0.24 0.06 0.05 0.06 0.07 1.4 1.3 0.54
USB 2000 D
1
log(1/R) - - - - - - - - -
Liga D
1
log(1/R) 4 0.34 0.25 0.05 0.06 0.06 1.7 1.2 0.64
USB 2000 log(1/R) 6 0.71 0.64 0.74 0.88 1.16 2.2 1.6 0.67
Liga D
1
log(1/R) 8 0.78 0.66 0.51 0.70 1.21 2.1 2.4 0.57
USB 2000 D
1
log(1/R) 4 0.64 0.73 0.61 0.97 1.03 2.5 1.7 0.85
Liga D
1
log(1/R) 8 0.73 0.72 0.71 0.87 1.17 2.2 1.6 0.68
USB 2000 log(1/R) 4 0.56 0.56 1.19 1.27 1.30 1.6 1.1 0.48
Liga D
1
log(1/R) 7 0.59 0.56 0.82 1.12 1.46 1.4 1.8 0.29
USB 2000 D
1
log(1/R) 5 0.64 0.53 0.54 0.11 1.36 1.5 2.5 0.35
Liga D
1
log(1/R) 4 0.62 0.60 1.03 1.11 1.20 1.7 1.2 0.58
USB 2000 log(1/R) 5 0.32 0.18 0.87 0.95 1.26 1.1 1.4 0.43
Liga D
1
log(1/R) 5 0.34 0.19 0.70 0.98 1.25 1.1 1.8 0.42
USB 2000 D
1
log(1/R) - - - - - - - - -
Liga D
1
log(1/R) - - - - - - - - -
Total soluble
solid, Brix%
Titrable
acidity, %
Firmness, lb 4 months
Static
Motion
6 months
Static
Motion
Static
Motion
2 months
Static
Motion
Static
Motion
Static
Motion
2 months
4 months
6 months
Motion
Static
Motion
2 months
4 months
6 months
Static
Motion
Static

Figure 2. Scatter plot of TSS prediction for all apple varieties predicted by a D
1
log(1/R) PLS regression model on spectra recorded at
harvest and chemical measurements made after four months storage.

Figure 3. VIP scores of the TSS prediction model for for all apple varieties predicted by a D
1
log(1/R) PLS regression model on spectra
recorded at harvest and chemical measurements made after four months storage.

Conclusions
The findings of this study indicate that the method offers
potential for non-destructive prediction of TSS, TA and starch
content and firmness of different cultivars of apples
originating from different orchards. Moreover, the method
enables the forecast of apple internal composition after
storage based on the spectral signature at the time of harvest.
Application of the results of this study could serve as a basis
for the development of an automatic system for forecasting of
apple internal composition and of a sorting system. Highest
prediction correlation results were found in most cases for
static measurements; however, the motion measurement
mode yielded sufficient correlation to be considered for
practical application. For the storage model, in most cases the
motion measurement mode yielded better correlations which
was explained by the scanning of larger areas of the apple
surface by this meathod. Comparing the two spectral ranges,
it was found that in most cases the SWIR spectral
measurements yielded more efficient prediction models
(RMSEP range was found to be 0.68-1.02 Brix%, 0.06-0.10 %
and 1.03-1.46 lb for TSS, TA and firmness, respectively). Since
the measured components are known to have their signatures
in this range, this was expected. However, the VIS-NIR
spectrometers are in general priced lower, operate faster and,
based on the results presented, are also applicable for online
sorting. RMSEP range was found to be 0.68-1.02 Brix%, 0.06-
0.10 % and 1.03-1.46 lb for TSS, TA and firmness, respectively.
No significant changes were found in RMSEP for TSS, TA and
firmness related to storage duration, showing the prospect
for future implementation.
Further research involving a larger number of varieties
should be conducted over consecutive years in order to
establish models for field and online application. The
presented study offers a non-destructive approach for
forecasting the changes in internal composition during
storage of apple based on the harvested samples spectral
signature.

References

Ignat, T., Schmilovitch, Z., Fefoldi, J., Steiner, B. and Alkalai-
Tuvia, S. (2012). Non-destructive measurement of
ascorbic acid content in bell peppers by VIS-NIR and
SWIR spectrometry. Postharvest Biol. Technol. 74, 91-
99.
Bertone, E., Venturello, A., Leardi, R. and Geobaldo, F. (2012).
Prediction of the optimum harvest time of 'Scarlet'
apples using DR-UV-Vis and NIR spectroscopy.
Postharvest Biol. Technol. 69, 15-23.
Camps, C., Guillermin, P., Mauget, J.C. and Bertrand, D.
(2007). Discrimination of storage duration of apples
stored in a cooled room and shelf life by visible-near
infrared spectroscopy. J. Near Infrared Spectrosc. 15,
169-177.
Fan, G., Zha, J., Du. and Gao, L. (2009). Determination of soluble
solids and firmness of apples by Vis/NIR transmittance. J. Food
Engin. 93, 416-420.
Lammertyn, J., Piers, A., de Baerdemaeker, J. and Nicolai, B.M.
(2000). Light penetration properties of NIR radiation in fruit
with respect to non-destructive quality assessment.
Liu, Y. and Ying, Y. (2005). Use of FT-NIR spectrometry in non-
invasive measurements of internal quality of 'Fiji' apples.
Lovasz, T., Meresz, P. and Salgo, A. (1994). Application of near
infrared transmission spectroscopy for the
determination of some quality parameters of apples. J.
Near Infrared Spec. 2, 213-221.
McGlone, V.A., Jordan, R.B. and Martinsen, P.J. (2002).
Vis/NIR estimation at harvest of pre- and post-storage
quality indices for 'Royal Gala' apple. Postharvest Biol.
Technol. 25, 135-144.
Nicola, B.M., Beullens, K., Bobelyn, E., Peirs, A., Saeys, W.,
Theron, K.I. and Lammertyn. J. (2007). Nondestructive
measurement of fruit and vegetable quality by means of
NIR spectroscopy: A review. Postharvest Biol. Technol.
46, 99118.
Nyasordzi, N., Friedman, H., Schmilovitch, Z., Ignat, T.,
Weksler, A, Rot, I. and Lurie, S. (2013). Utilizing the IAD
index to determine internal quality attributes of apples
at harvest and after storage. Postharvest Biol. Technol.
77, 8086
Piers, A., Ooms, K., Lammertyn, J. and Nicolai, B. (2000).
Prediction of the optimal picking date of different apple
cultivars by means of VIS/NIR-spectroscopy. Postharvest
Biol. Technol. 21, 189-199.
Piers, A., Tirry, J., Verlinden, B., Darius, P. and Nicolai, B.M.
(2003). Effect of biological variability on the robustness
of NIR models for soluble solids content of apples.
Piers, A., Schenk, A. and Nicolai, B.M. (2005). Effect of natural
variability among apples on the accuracy of VIS-NIR
calibration models for optimal harvest date predictions.
Qing, Z., Ji, B. and Zude, M. (2008). Non-destructive analyses
of apple quality parameters by means of laser-induced
light backscattering imaging. Postharvest Biol. Technol.
48, 215-222.
Rizzolo, A., Vanoli, M., Spinelli, L. and Torricelli, A. (2010).
Sensory characteristics, quality and optical properties
measured by time-resolved reflectance spectroscopy in
stored apples. Postharvest Biol. Technol. 58, 1-12.
Ventura, M., de Jager, A., de Putter, H. and Roelofs, F.P.M.M.
(1998). Non-destructive determination of soluble solids
in apple fruit by near infrared spectroscopy (NIRS).
Postharvest
Biol. Technol. 14, 21-27.

Quality of Brazilian mango fruit in relation to
optical properties non-destructively
measured by time-resolved reflectance
spectroscopy.
M. Vanoli
a,b
, A. Rizzolo
b
, M. Grassi
b
, L. Spinelli
c
, P. Eccher Zerbini
d
, R.
Meirelles de Azevedo Pimentel
e
and A. Torricelli
a

a
Politecnico di Milano, Dipartimento di Fisica, Milan, Italy.
b
Consiglio per la Ricerca e Sperimentazione in Agricoltura Unit di ricerca per i processi dellindustria agroalimentare (CRA-IAA),
Milan, Italy.
c
Istituto di Fotonica e Nanotecnologie, CNR, Milan, Italy.
d
Horticultural Supply Chains, Wageningen University, Wageningen, The Netherlands.
e
Empresa de Pesquisa Agropecuria de Minas Gerais (EPAMIG), Minas Gerais, Brazil.

Corresponding author: alessandro.torricelli@polimi.it
Introduction
Mango is one of the most important fruit crops in Brazil and
about 25% of the production is exported, mainly to USA and
Europe (Pinto et al., 2004). A major challenge is to minimise
quantitative and qualitative losses during the supply chain of
fresh mangoes while ensuring an acceptable ripening. Quality of
mangoes is strictly dependent on proper maturity at harvest as
fruit picked too early are more sensitive to chilling injury and
may fail to ripen, while fruit harvested at a late maturity stage
have a reduced shelf-life and greater susceptibility to disease
(Sivakumar et al., 2011). Differences among cultivars and
growing conditions have precluded universal maturity indices.
Recently, Padda et al. (2011) showed that firmness, flesh colour
and total soluble solids content are the best parameters to
assess ripening of mangoes but these factors have the
disadvantage of being destructive.
Time-resolved reflectance spectroscopy (TRS) is a non-
destructive technique which simultaneously quantifies the
internal optical properties of fruit through the absorption (i.e.
pigments) and the scattering (i.e. physical structure) coefficients
by probing the pulp at a depth of 1-2 cm with no or limited
influence from the skin (Torricelli et al., 2008). The absorption
coefficient (a) measured at harvest at 630-690 nm, near the
chlorophyll peak, allowed selection of fruit of different maturity
showing different quality attributes as regards chemical
composition and sensory characteristics during shelf-life and
after storage (Torricelli et a., 2008). In nectarines, the
absorption coefficient measured at 670 nm (a670) can be
considered an effective maturity index as it is linked to the
biological age of the fruit; in fact, it has been successfully used
to predict the softening rate of nectarines during shelf-life
allowing selection of fruit for different market destinations
(Eccher Zerbini et al., 2009). On the other hand, the scattering
coefficient (s) gave an insight into the textural properties of
the fruit pulp. In apples, Vanoli et al. (2009) and Rizzolo et al.
(2010) found that s measured in the range between 750 and
790 nm were related to the mechanical properties of fruit
(firmness, stiffness, intercellular spaces) as well as to pectin
composition and sensory attributes related to structure (firm,
crispy, mealy and juicy).
The possibility of using the TRS technique to assess
maturity in mangoes has been previously explored but the
results were not always completely satisfactory. Pereira et al.
(2010) measured Brazilian mangoes at 630 nm and, by
converting a630 into the biological shift factor, it was only
possible to explain about 70% of the variation in softening
behaviour. When a630 was employed to sort mangoes into
different maturity classes, it was possible to obtain fruit with
different pulp colour characteristics, showing different colour
rate changes during postharvest ripening; however, in the last
ripening period, the increase in pulp yellowness interfered with
chlorophyll measurements (Vanoli et al., 2011a).
Two problems have to be considered when measuring
mangoes: the presence of a green layer (2-3 mm) under the peel
which attenuates the TRS signal intensity mainly in the
chlorophyll range (630-690 nm), and the carotenoid
accumulation in the pulp during mango ripening, especially in
more mature fruit, which affect the chlorophyll estimate
(Spinelli et al., 2012; Vanoli et al., 2011a).
In order to overcome these problems, in this work we
tested the TRS technique in the 540-900 nm range with the aim
of assessing the quality of Brazilian mangoes during shelf-life.
One hundred and twenty mangoes (cv Tommy Atkins,
harvested in Pernambuco, Brazil and immediately transported
by plane to Milan, Italy) were individually measured by TRS at
540 nm on two opposite sides in the fruit equatorial region and
ranked by decreasing a540, averaged over the two sides, that
is from more (high a540) to less (low a540) mature fruit.
Then, mangoes were randomised into 4 batches of 30 fruits and
analysed after 0, 1, 2 and 5 days of shelf-life at 20C (d0, d1, d2
and d5). At each time of analysis, fruit were measured by TRS in

the 540-900 nm range on two opposite sides and analysed at
the same points for flesh firmness (Instron UTM model 4301,
crosshead speed 200 mm/min, 8 mm diameter plunger) and
pulp colour (Spectrophotometer CM-2600d, Minolta; L*,a*,b*,
C*, H). Fruit of d5 (batch 4) were measured by TRS in the 540-
900 nm range after 0, 1, 2 and 5 days at 20C in order to follow
the changes of the optical properties during the shelf-life period
on the same fruit.
A TRS setup developed at Politecnico di Milano was used.
The light source is a fibre laser (SC450-6W, Fianium) providing
white-light picosecond pulses, adjustable in power by a variable
neutral-density attenuator. A filter wheel loaded with 14 band-
pass interference filters is used for spectral selection in the
range 540-900 nm. Light is delivered to the sample by a
multimode graded-index fibre. Diffuse remitted light is detected
by 1 mm fibre. The light is detected with a photomultiplier and a
time-correlated single-photon counting board (HPM-100-50,
SPC-130; Becker & Hickl). A model for photon diffusion in turbid
media was used to analyse TRS data to assess the bulk optical
properties (absorption coefficient, a, and reduced scattering
coefficient, s) of samples. Convolution of the photon diffusion
model with the instrument response function (measured facing
the injection and the collection fibres) is performed before
fitting the experimental data. An approximation to the Mie
theory is used to relate the reduced scattering coefficient to the
structural properties of the diffusive sample: s = A (/0)-B,
where is the wavelength, A is the scattering coefficient at
wavelength 0 = 600 nm, and B is a parameter related to the
size of scatterers.
TRS absorption and scattering spectra were elaborated by
The Unscrambler X (v. 10.0.1; Camo, Norway) in order to build
partial least square (PLS) regression models for firmness and
pulp colour. TRS spectra were also analysed by linear
discriminant analysis to classify the samples on the basis of the
day of shelf-life (Statgraphics, v. 7; Manugistic Inc., Rockville
MD, USA). Firmness and pulp colour data were processed by
analysis of variance and means were compared by Tukeys test.
Fruit characteristics
Flesh firmness (Table 1) significantly decreased during shelf-
life. At d0, 70% of mangoes had firmness greater than 20N, at d1
56% of fruit softened below 20N, while at d5 all fruit had
firmness lower than 20N with 89% of mangoes softening below
10N (Fig. 1). This means that at d0 only 30% of fruit are ready to
eat and that at d5 all mangoes are ready for consumption.
Pulp colour turned from greenish-pale yellow to yellow-
orange during the shelf-life period: a*, b* and C* parameters
significantly increased while L* and H significantly decreased
from d0 to d5 at 20C (Table 1).

TRS spectra characteristics
Figure 2 shows the absorption and scattering coefficients in
the 540-900 nm spectral range measured at d0 and d5, and the
absorption coefficient measured at 540 nm and at 650 nm
during the 5 days of shelf-life of mangoes belonging to batch 4.
Absorption spectra showed two maxima: the first at 540
nm and the second at 670 nm. On average, the a540 values
increased while those at 670 nm decreased with shelf-life: fruit
at d0 were characterised by the highest values of a670 and the
lowest for a540; the opposite was true for mangoes at d5. The
increase of a540 values during shelf-life could be due to
carotenoid biosynthesis as reported by Vasquez-Caicedo et al.
(2006) which found in Tommy Atkins mangoes an 18.9 %
increase in the amount of total -carotenes during postharvest
fruit ripening together with a dynamic interconversion of the
plastid structures. The Tommy Atkins cultivar is also
particularly rich in all-trans-violaxanthin and 9-cis-violaxanthin
(Ornelas-Paz et al., 2007). In various mango cultivars, it was
found that the concentrations of these carotenoids increased in
an exponential manner during fruit ripening and were highly
correlated with the mesocarp a* (positive) and H (negative)
values (Ornelas-Paz et al., 2008). In our work, TRS absorption
spectra reflected the changes in pulp colour. High and negative
correlations were found between H and a540 while lower but
positive correlations were observed between H and a670
(Table 2), confirming our previous results on Tommy Atkins and
Palmer mangoes (Vanoli et al., 2011a; Spinelli et al., 2012).
Optical properties measured by TRS also include the
estimate of scattering phenomena. The scattering spectra of
Tommy Atkins mangoes were flat and slightly decreased with
increasing wavelength (Fig. 2). On average, scattering spectra
decreased with shelf-life. The density of the scatterers
(parameter A) significantly decreased during the shelf-life period
while the size of the scatterers (parameter B) was constant up to
d1, significantly increased at d2, and remained constant until d5
(Fig. 3). Changes in scattering parameters reflected changes
Table 1. Firmness and pulp colour data of Tommy Atkins mangoes during shelf-life at 20C (mean standard error).
days at 20C firmness (N) L* a* b* C* H
0 46.1 7.2 82.7 0.3 -0.32 0.56 52.3 0.9 52.4 0.9 90.7 0.6
1 27.8 4.6 81.5 0.6 1.53 0.54 54.0 0.8 54.1 0.8 88.6 0.5
2 16.9 1.8 81.0 0.3 2.58 0.49 56.5 0.8 56.6 0.8 87.6 0.5
5 8.0 0.4 78.8 0.3 5.41 0.36 60.5 0.7 60.7 0.7 84.9 0.3
0
20
40
60
80
100
120
d0 d1 d2 d5
%
>30N
20-30N
10-20N
<10N
Figure 1. Frequency of Tommy Atkins mangoes according 1
to firmness classes at d0, d1, d2 and d5 of shelf-life 2


which occurred in the pulp structure due to mangoe softening
during shelf-life. Softening is accompanied by the enzymatic cell
wall breakdown and so the density of the scattering particles in
the fruit flesh decreased, as reported for apples and tomatoes
(Qin and Lu, 2008; Vanoli et al., 2011b). As a consequence of this
phenomenon, the light scattering at the cell interfaces is
reduced leading to fewer scattering events in the tissue
(Bobelyn et al., 2008). Actually, the situation in fruit pulp is
much more complex than just described, as the scattering
centres of a fruit are not expected to be homogeneous spheres
and the parameters A and B do not assess the real size of

scattering centres in the tissue; they are average equivalent
parameters that could be related to physical characteristic of
fruit. In our work, considering the days of shelf- life together, a
significant positive correlation between firmness and s 880
was found, but the highest r was observed at d1 when firmness
classes were evenly represented within the sample (Table 2; Fig.
3). Significant correlations were also found between firmness
and a measured at 650 and 670 nm (Table 2).
Different PLS models were built to predict pulp colour and
firmness of Tommy Atkins mangoes considering either
scattering spectra alone, absorption spectra alone, or
absorption spectra coupled with scattering spectra or with A
and B parameters of the scatterers (Table 3). Poor results were
obtained for all the quality characteristics using scattering
coefficients alone: the highest R2cv was 0.40 for a*pulp. On the
contrary, good correlations (R2cv = 0.9) were obtained between
TRS absorption spectra and pulp colour (a*, H). However, the
performance of the models was improved adding scattering
spectra or A and B parameters of the scatterers to absorption
coefficients, reaching R2cv =0.90 and 0.93 for a* and H,
respectively. PLS regression models for pulp b* and C*
presented lower coefficients of determination (0.70 - 0.74) in
contrast to Subedi et al. (2007) who predicted flesh b* values
with R2=0.88 - 94; however, these workers found that the
optimal wavelengths to model pulp b* did not include the
visible wavelengths, upon which b* is based, but the longer
1
2
3
4
Figure 2. Top row: absorption coefficient (left) and reduced scattering coefficient (right) in the 540-900 nm 5
spectral range at d0 for 30 fruits (batch 4). Middle row: absorption coefficient (left) and reduced scattering 6
coefficient (right) in the 540-880 nm spectral range at d5 for 30 fruits (batch 4). Bottom row: absorption 7
coefficient at 540 nm (left) and 650 nm (right) during shelf-life for 30 fruits (batch 4). Fruits are ranked from 1 8
to 30 on the basis of decreasing absorption coefficient at 540 nm corresponding to decreasing maturity. 9
a
540
a
650 or
a
670
s
880
H pulp d0 -0.799*** 0.798*** 0.138
d1 -0.890*** 0.396* 0.595***
d2 -0.894*** 0.529** 0.243
d5 -0.902*** -0.588*** 0.337
d0-d5 -0.858*** 0.537*** 0.551***
firmness d0 -0.470** 0.759*** 0.412*
d1 -0.413* 0.604*** 0.701***
d2 -0.470** 0.435* 0.391*
d5 -0.595*** -0.329 0.457*
d0-d5 -0.487*** 0.677*** 0.609***
Table 2. Correlation coefficients (r) between
a
540,
a
650or
a
670,
s
880
and firmness and pulp H value of Tommy Atkins mangoes at d0, d1, d2 and
d5 of shelf-life.


wavelengths. This result was attributed to low penetration
through the fruit tissue. In contrast, our results confirmed that
the TRS technique actually measures the optical properties of
the fruit pulp even in the visible range, as shown by the
relationships between pulp colour and the absorption
coefficients measured at 540, 650 and 670 nm (Table 2). As for
firmness, the best prediction model (R2cv=0.73) was achieved
coupling absorption coefficients to the A and B parameters of
the scattering centers, even if RMSECV values remained high
(Table 3). Many authors explored the possibility of using VIS/NIR
spectroscopy to predict firmness of different mango cultivars
with opposite results. Valente et al. (2009) found that good
prediction results for firmness depend on maturity degree and
range of firmness of mangoes as the best results (R2=0.62)
were obtained for ripe fruit with firmness lower than 20N.
Better results (R2=0.77) were found by Jha et al. (2006) by using
a handheld colorimeter.

TRS absorption and scattering coefficients and A and B
parameters of the scatterers were also used as explanatory
variables in the discriminant analysis in order to classify
mangoes according to the day of shelf-life. Better classification
performance was obtained by including both absorption and
scattering spectra (Table 4). In Fig. 4 the values of the two
discriminant functions (Table 4) for every fruit and the group
(day of shelf-life) centroids are plotted. Function D1
discriminated mangoes at d0 from the others while function D2
discriminated d1 fruits from the others (Fig. 4). Both functions
allowed segregation of immature from mature fruit as most of
the mangoes at d0 and at d1 were not ready for consumption.
Mangoes at d0 shelf-life were all correctly classified while fruit
belonging to d1, d2 and d5 were correctly classified in about
90% of the cases (Table 4).
Conclusion
The results of this research show that the TRS technique in
the range of 540-900 nm can be used to non-destructively
estimate the quality of mango fruit imported from Brazil. The
absorption coefficient measured in the visible range is sensitive
to carotenoid and chlorophyll contents of the pulp, pigments
which have been shown to be essential in determining the
maturity of mango fruit. TRS is also able to assess the properties
related to scattering, allowing evaluation of this type of
variability and hence the generation of more robust calibrations.
By using both absorption and scattering optical properties, TRS
is able to predict with good accuracy the pulp colour of mangoes
and to sort fruit with different maturity degrees.
Acknowledgements
We acknowledge partial support from Project TROPICO
(n.17077, Rif. AGRO16; Regione Lombardia, Italy).

Dependent TRS N of
variable parameters factors
R
2
cal
RMSEC
R
2
cv
RMSECV
a* polpa
a
540-880 3 0.91 1.02 0.89 1.13
s
540-880 5 0.47 2.46 0.40 2.63
a
540-880+
s
540-880
5 0.92 0.95 0.90 1.09
a
540-
880+MIE A B
5 0.92 0.94 0.90 1.07
b* polpa
a
540-880 2 0.70 2.87 0.63 3.17
s
540-880 3 0.35 4.21 0.32 4.32
a
540-880+
s
540-880
4 0.73 2.70 0.65 3.18
a
540-
880+MIE A B
3 0.73 2.70 0.68 3.01
C* polpa
a
540-880 2 0.71 2.85 0.64 3.18
s
540-880 3 0.35 4.27 0.31 4.45
a
540-880+
s
540-880
4 0.75 2.68 0.67 3.07
a
540-
880+MIE A B
3 0.74 2.69 0.68 3.09
H polpa
a
540-880 3 0.91 1.01 0.90 1.07
s
540-880 5 0.47 2.47 0.39 2.66
a
540-880+
s
540-880
7 0.94 0.80 0.93 0.94
a
540-
880+MIE A B
7 0.94 0.79 0.93 0.91
firmness
a
540-880 7 0.71 14.9 0.65 16.5
s
540-880 2 0.41 21.1 0.39 21.6
a
540-880+
s
540-880
3 0.73 14.2 0.70 15.3
a
540-
880+MIE A B
4 0.76 13.6 0.73 14.5
Calibration Cross-validation
Table 3. Performance of different regression models on original TRS spectral data for prediction of pulp colour
and firmness of Tommy Atkins mangoes.
0.0
0.1
0.2
0.3
0.4
0.5
0 1 2 5
day at 20C
B
15
16
17
18
19
20
0 1 2 5
days at 20C
A
A B
Figure 3. Density (A) and size (B) of the scatterers of Tommy Atkins 1
mangoes during shelf-life at 20C (bars refer to standard error) 2
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
-15 -10 -5 0 5 10
D
2

(
6
.
3
%
)
D1 (92.2%)
d0
d1
d2
d5
centroids
d0
d1
d2
d5
Figure 4. Classification of Tommy Atkins mangoes as a function 1
of day of shelf-life using discriminant analysis (see Table 3) 2
DA function Relative
percentage
Canonical
correlation
P -value
Actual group
d0 d1 d2 d5 model
1 92.21 0.988 0.0000 d0 100.0 0.0 0.0 0.0
2 6.32 0.855 0.0000 d1 0.0 90.0 3.3 6.7
3 1.48 0.623 0.0002 d2 0.0 0.0 90.0 10.0
d5 0.0 0.0 10.7 89.3 91.8
Classification results
Predicted group (percentage)
Table 4. Results of discriminant analysis according to the day of shelf-life using TRS absorption and scattering spectra measured at 540-880 nm and classification
results (column: actual group; row: predicted class).

References
Bobelyn, E., Serban, A.S., Nicu, M., Lammertyn, J., Nicola, B.M.,
Saeys, W. (2010). Postharvest quality of apple predicted by
NIR spectroscopy: study on the effect of biological
variability on spectra and model performance. Postharvest
Biology and Technology, 55, 133143.
Eccher Zerbini, P., Vanoli, M., Rizzolo, A., Jacob, S., Torricelli, A.,
Spinelli, L. and Schouten, R.E. (2009). Time-resolved
reflectance spectroscopy as a management tool in the fruit
supply chain: an export trial with nectarines. Biosystems
Engineering, 102, 360-363.
Jha, S.N., Kingsly, A.R.P. and Chopra, S. (2006). Non-destructive
determination of firmness and yellowness of mango during
growth and storage using visual spectroscopy. Biosystems
Engineering, 94, 397402.
Ornelas-Paz, J. de J., Yahia, E. M. and Gardea, A. A. (2007).
Identification and quantification of xanthophyll esters,
carotenes and tocopherols in the fruit of seven Mexican
mango cultivars by liquid chromatography-APcI+-time-of-
flight mass spectrometry. J. Ag. Food Chem., 55, 6628
6635.
Ornelas-Paz, J. de J., Yahia, E. M. and Gardea, A. A. (2008).
Changes in external and internal colour during postharvest
ripening of Manila and Ataulfo mango fruit and
relationship with carotenoid content determined by liquid
chromatography. APcI+-time-of-flight mass spectrometry.
Postharvest Biology and Technology, 50, 145-152.
Padda, M.S., do Amarante, C.V.T., Garcia, R.M., Slaughter, D.C.
and Mitcham E.J. (2011). Methods to analyse physico-
chemical changes during mango ripening: a multivariate
approach. Postharvest Biology and Technology, 62, 267-
274.
Pereira, T., Tijskens, L.M.M., Vanoli, M., Rizzolo, A., Eccher
Zerbini, P., Torricelli, A., Spinelli, L. and Filgueiras, H.
(2010). Assessing the harvest maturity of Brazilian mangos.
Acta Horticulturae, 880, 269-276.
Pinto, A.C.Q., Andrade, S.R.M., Amaro, A.A. and Gomes, U.
(2002). Mango industry in Brazil. Acta Horticulturae, 645,
37-50.
Qin, J. and Lu, R. (2008). Measurement of the optical properties
of fruits and vegetables using spatially resolved
hyperspectral diffuse reflectance imaging technique.
Rizzolo, A., Vanoli, M., Spinelli, L. and Torricelli, A. (2010).
Sensory characteristics, quality and optical properties
measured by time-resolved reflectance spectroscopy in
stored apples. Postharvest Biology and Technology 58, 1-12.
Sivakumar, D., Jiang, Y. and Yahia, E.M. (2011). Maintaining
mango (Mangifera indica L.) fruit quality during the export
chain. Food Research International, 44, 1254-1263.
Spinelli, L., Rizzolo, A., Vanoli, M., Grassi, M., Eccher Zerbini, P.,
Pimentel, R. and Torricelli, A. (2012). Optical properties of
pulp and skin in Brazilian mangoes in the 540-900 nm
spectral range: implication for non-destructive maturity
assessment by time-resolved reflectance spectroscopy.
International Conference of Agricultural Engineering, CIGR
Ageng 2012, Valencia (Spain), July 8-12, 2012. ISBN 10 84-
615-9928-4
Subedi, P.P., Walsh, K.B. and Owens, G. (2007). Prediction of
mango eating quality at harvest using short-wave near
infrared spectrometry. Postharvest Biology and Technology,
43, 326-334.
Torricelli, A., Spinelli, L., Contini, D., Vanoli, M., Rizzolo, A. and
Eccher Zerbini, P. (2008). Time-resolved reflectance
spectroscopy for nondestructive assessment of food
quality. Sensing and Instrumentation for Food Quality and
Safety, 2, 82-89.
Valente, M., Leardi, R., Self, G., Luciano, G. and Pain, J.P. (2009).
Multivariate calibration of mango firmness using vis/NIR
spectroscopy and acoustic impulse method. Journal of Food
Engineering, 94, 7-13.
Vanoli, M., Eccher Zerbini, P., Spinelli, L., Torricelli, A. and
Rizzolo, A. (2009). Polyuronide content and correlation to
optical properties measured by time-resolved reflectance
spectroscopy in Jonagored apples stored in normal and
controlled atmosphere. Food Chemistry, 115, 1450-1457.
Vanoli, M., Pereira, T., Grassi, M., Spinelli, L., Filgueiras, H.,
Tijskens, L.L.L., Rizzolo, A. and Torricelli, A. (2011a).
Changes in pulp colour during postharvest ripening of
Tommy Atkins mangoes and relationship with optical
properties measured by Time-resolved Reflectance
Spectroscopy. 6th CIGR, Section VI, International
Symposium Towards a Sustainable Food Chain-Food
Process, Bioprocessing and Food Quality Management,
Nantes, (France) - April 18-20, 2011. CD-ROM Proceedings,
ISBN 978-2-7466-3203-5.
Vanoli, M., Rizzolo, A., Grassi, M., Farina, A., Pifferi, A., Spinelli,
L. and Torricelli, A. (2011b). Time-resolved reflectance
spectroscopy nondestructively reveals structural changes in
Pink Lady apples during storage. Procedia Food Science,
81-89.
Vsquez-Caicedo, A.L., Heller, A., Neidhart, S. and Carle, R.
(2006). Chromoplast morphology and -carotene
accumulation during postharvest ripening of mango cv.
Tommy Atkins. J Ag.Food Chem., 54, 5769-5776.

Screening for pro-vitamin A components
in Cassava (Manihot esculenta) using NIR
to support bio-fortification.

T. zum Felde
a
, Oladeji E. Alamu
b
, E. Porras
a
, N. Maroya
b
and
B. Maziya-Dixon
b

aInternational Potato Centre (CIP), Lima, Peru
b
International Institute of Tropical Agriculture (IITA), Ibadan, Nigeria

Corresponding author: t.zumfelde@cgiar.org

Introduction

Vitamin A deficiency is the most important cause of
total blindness in developing countries. It is estimated that
25% of pre-school age children of the worlds population
have vitamin A deficiency. Worldwide around 250 million
children have subclinical vitamin A deficiency; 3 million have
clinical vitamin A deficiency (1). It is estimated that 20
million pregnant women have subclinical vitamin A
deficiency; 7 million have clinical vitamin A deficiency and 6
million are night blind (2). Other health consequences of
vitamin A deficiency are reduced immune defence including
increased risk of illness or death from childhood diseases.
Risk factors for vitamin A deficiencies are monotonous
plant-based diets containing little vitamin A, aggravated by
low intake of animal source foods, orange and yellow fruits
and vegetables.

Plant breeding and genetic engineering techniques for
development of new staple food crop varieties with
increased vitamin A content and other deficient
micronutrients is known as biofortification. Several
initiatives (HarvestPlus, Agrosalud, SASHA) have been set up
to increase the vitamin A concentration in staple food crops
to help improve human nutrition status in developing
countries. Biofortification is complementary to other
strategies for reducing malnutrition such as
supplementation, fortification and diversification;
nutritional benefits come directly from the biofortified crops
with no or little additional costs for consumers. To support
biofortification programmes, there is a need for high
throughput techniques to screen macro- and micronutrient
concentrations of germplasm and breeding populations in
tens of thousands of genotypes in short time frames. High
performance liquid chromatography (HPLC) is the common
method to determine vitamin A concentration in food crop
samples. Although HPLC is very accurate, the high cost of
this method and the time required for analysis limits their
use to small numbers of samples relative to those required
in extensive screening and biofortification programs.
Requiring only simple sample preparation methods, near
infrared (NIR) spectroscopy was selected to facilitate the
analysis of several traits simultaneously. The HarvestPlus
Challenge Program (www.harvestplus.org) committed to
evaluate in a feasibility study the potential of NIR for
vitamin A in sweet potato, cassava, maize and potato.

Cassava (Manihot esculenta) is a tropical root crop that
serves as a food security and income generation crop for
many millions of people in the developing world (3). Cassava
is considered as a key food security staple because of its
tolerance to abiotic stresses like drought and heat (4), its
capacity to grow in poor and degraded soils (5), tolerance or
resistance to major biotic stress factors (6), its capacity to
interact positively with mycorrhiza (7) and the flexibility for
farmers that cassava roots can be stored in the field and
harvested when needed. Cassava also has the advantage of
a variety of potential uses which make it an important
source of raw material for starch, animal feed and ethanol
production. The cassava root comprises peel (10-20%) and
the edible fleshy portion (80-90%). Before human
consumption, the cassava root is peeled to obtain the fleshy
portion, which is used for food or industrial applications.
The limitations of cassava utilisation include its relatively-
low nutritional quality like low protein and micronutrient
contents of the roots which basically contribute energy to
the human diet, poor post-harvest shelf-life, roots are bulty
to transport and toxic cyanogenic glycosides (8) are
present. However, cassava is the third-largest source of food
carbohydrates in the tropics and is a major staple food in
the developing world, providing a basic diet for around 500
million people. With provitamin A components, biofortified
cassava can have a significant positive impact on nutrition
and overall human health, especially among poorer
communities. Significant progress has already been
achieved in the last ten years with almost a tripling of the
amount of provitamin A components in cassava roots (9,
10). In this study, we report the development of NIR
calibrations to enable the determination of pro-vitamin A
components (converted in the human body to vitamin A) in
cassava.


Samples, sampling and sample preparation

A total of 105 freshly-harvested cassava genotypes with
increased pro-vitamin A components were obtained from
the experimental fields of International Institute of Tropical
Agriculture (IITA) in Ibadan, Nigeria. At time of harvest, the
cassava plants were about 10-12 months old. The
HarvestPlus Standard method of sampling (11) was
employed with a selection of five roots sampled from the
harvested roots of each genotype in at least two
replications. Three (large, medium and small size roots) of
the five roots were washed, air-dried, peeled and again
washed and dried with soft paper. During sample
preparation, special care was taken to avoid direct exposure
of the storage roots to sunlight and the lights in the
laboratory were covered with UV filters.

Each peeled root was cut in half longitudinally and the
two halves were again cut longitudinally into quarters. Two
opposite quarters of the three roots were pooled for total
carotene quantification. The six quarters were cut into small
pieces of about 1 cm
3
and mixed together. After many
subdivisions, a sample of approximately 1015 g of small
pieces of root was taken as uniform and representative of a
genotype (Figure 1). This 10-15 g sample material of pieces
of about 1 cm
3
was scanned twice by NIR using a coarse cell.
After that, the same material was blended with a standard
kitchen blender and scanned again twice by NIR (Figure 1)
before HPLC analysis was performed.

Spectrophotometric and HPLC reference analysis,
instrumentation and conditions

Directly after scanning by NIR, the raw cassava samples
were transferred to a mortar and ground with refrigerated
cold acetone (30ml) and celite (3g) using a pestle and
filtered through Whatman no 1 paper. This process was
repeated until the residue was colourless. Twenty (20) ml of
petroleum ether (PE, boiling point 35-60
o
C) was put into a
500 ml separator funnel with a Teflon stop-cock and added
to the acetone extracts. Distilled water was slowly added
(~300 ml), flowing along the walls of the funnel to avoid
formation of an emulsion. After adding the distilled water,
two separated phases were identified (the aqueous and
ether phases) and the lower, aqueous phase was discarded.
Washing with distilled water was repeated 3 to 4 times to
remove residual acetone. In the last washing, the totality of
the lower phase was discarded. The PE phase was then
collected in a volumetric flask through a small funnel
containing anhydrous sodium sulphate to remove any
residual water.
The absorbance of the ethereal phase was read at 450
nm in a photo spectrometer after dilution to 25ml and the
total carotenoid concentration was calculated. Fifteen ml of
the PE extract was then removed and dried under nitrogen
gas; the dried extract residue was re-dissolved in 500 L of
HPLC grade, 50:50 mixture of methanol/dichloroethane
solution and 10 L of this solution was injected into the
HPLC (Waters Corporation, USA). Separation of carotenoids
was achieved using a Waters YMCTM Carotenoid 4.6mm
x250mm column with a Waters Nova-Pak C30 guard Column
and isocratic eluted with a mobile phase 50:50 of 100%
MeOH (Solvent A) and 100% MTBE (solvent B) for 15
minutes. The flow rate was 1.0ml/min (12, 13). All chemicals
used were of analytical/HPLC grade and sourced from
Sigma-Aldrich, UK.

=>NIRS 1 => NIRS 2 +
HPLC
Figure 1. Sample preparation workflow for NIR analysis in fresh cassava root samples


NIR analysis, calibration development and validation

Each cassava genotype was scanned four times (twice
chopped and twice blended) by NIR within the range of 400
to 2498 nm, registering the absorbance values log (1/R) at
0.5nm intervals for each sample using a NIR
monochromator (model FOSS XDS, solid module) and using
coarse cell cups. Calibration equations for b-cryptoxanthin,
13-cis -carotene, trans -carotene, 9-cis -carotene, total
-carotene and total carotenoids were developed under
WinISI 4 Project Manager software, with spectral
information from 400 to 2498 nm and using modified partial
least squares (MPLS) regression and cross-validation
techniques. The derivative and mathematical treatments
were 2, 5, 5 and 1 for total carotenoids and 1, 4, 4 and 1 for
-cryptoxanthin, 13-cis -carotene, trans -carotene, 9-cis
-carotene and total -carotene. The first number is the
derivative, the second the gap, and the third and fourth
numbers are the smooth. The results of the calibration
calculation were checked observing the t-outliers with t >
2.0 and GH-outliers >4.0. The number of outlier elimination
passes was two. Samples with t > 2.0 were deleted from the
calibration file.
One hundred and five samples used for calibration
development and validation of the calibration models for -
cryptoxanthin, 13-cis -carotene, trans -carotene, 9-cis -
carotene, total -carotene and total carotenoids,
respectively; they were separated into different sample sets
several times. Each time, the samples used for validation
were deleted from the calibration data file. Each new
validation file consisted of ca. 15% samples of the complete
dataset. New calibration equations were developed which
showed only small deviations compared to the calibration
equations developed for the complete sample set (results
not shown). These calibration equations were used to
predict the -cryptoxanthin, 13-cis -carotene, trans -
carotene, 9-cis -carotene, total -carotene and total
carotenoid concentrations of the samples of the genotypes
not represented in the calibration file.


Mean values, standard deviations and ranges of the
reference values and the statistics of the NIR calibration and
of the cross-validation are shown in Tables 1 and 2. The NIR
calibration equations for 13-cis BC, trans BC, 9-cis BC, Total
BC (analysed by HPLC) and Total Carotenoids (analysed by
spectrophotometer) showed high coefficients of
determination for the calibration curve for chopped samples
(0.95, 0.98, 0.88, 0.96 and 0.74, respectively) and medium
to high coefficients of determination in cross-validation
(0.76, 0.90, 0.69, 0.88 and 0.67, respectively; Table 1).
Results for the same genotypes as blended samples were
nearly equal (Table 2). The standard errors of calibration
(SEC) and the standard errors in cross-validation (SECV)
were low for all traits. The NIR calibration equations for -
cryptoxanthin showed very low coefficients of
determination in calibration for chopped and blended
samples, 0.45 and 0.43, and consequently very low
coefficients of determination in cross-validation, 0.38 and
0.36 (Tables 1 and 2). However this was not surprising
because concentrations and available ranges measured by
HPLC were very low; only between 0.04 and 0.55 g/g -
cryptoxanthin based on fresh weight base was measured.

Table 1: Variation of concentrations, NIR-calibration and cross-validation statistics for -cryptoxanthin, 13-cis BC, trans BC, 9-cis
BC, Total BC and Total Carotenoids concentrations in cassava in complete calibration sets (105 samples). Chopped samples,
scanned by NIR as cubes.

Trait
Reference Values Calibration
Cross
Validation

Range
a

Me
an
a

SD

a

R
c
SEC
a
R
cv

SECV

a

-cryptoxanthin 0.04 0.55 0.24 0.09 0.45 0.04 0.38 0.05
13-cis BC 0.11 2.04 0.90 0.41 0.95 0.08 0.76 0.19
trans BC 0.20 8.98 3.26 1.79 0.98 0.26 0.90 0.50
9-cis BC 0.10 2.72 1.09 0.55 0.88 0.18 0.69 0.29
Total BC 0.41 11.73 5.26 2.40 0.96 0.42 0.88 0.77
Total Carotenoids 0.46 10.55 5.92 2.42 0.74 1.11 0.67 1.25
SD = standard deviation, Rc = coefficient of determination in calibration, SEC = standard error of calibration, Rcv = coefficient of determination
in cross-validation, SECV = standard error of cross-validation,
a
= g g
-1
in fresh weight, BC=-carotene

Table 2: Variation of concentrations, NIR-calibration and cross-validation statistics for b-cryptoxanthin, 13-cis BC, trans BC, 9-cis
BC, Total BC and Total Carotenoids concentrations in cassava in complete calibration sets (105 samples). Blended samples,
scanned by NIR as mash.

Trait
Reference Values Calibration
Cross
Validation

Range
a

Me
an
a

SD

a

R
c
SEC
a
R
cv

SECV

a

-cryptoxanthin 0.04 0.55 0.24 0.09 0.43 0.04 0.36 0.05
13-cis BC 0.11 2.04 0.90 0.41 0.83 0.15 0.72 0.20
trans BC 0.20 8.98 3.26 1.79 0.98 0.21 0.92 0.44
9-cis BC 0.10 2.72 1.09 0.55 0.90 0.17 0.75 0.26
Total BC 0.41 11.73 5.26 2.40 0.94 0.54 0.86 0.84
Total Carotenoids 0.46 10.55 5.92 2.42 0.92 0.70 0.81 1.05
SD = standard deviation, Rc = coefficient of determination in calibration, SEC = standard error of calibration, Rcv = coefficient of
determination in cross-validation, SECV = standard error of cross-validation,
a
= g g
-1
in fresh weight, BC=-carotene

Validations of the NIR calibrations with 14 to 16
samples randomly selected and not represented in the
calibration set confirmed the results obtained in cross-
validation. The coefficients of determination in validation
(R2
v
) were comparable to those found for the calibration
and cross-validation. The standard errors of prediction
corrected for bias [SEP(C)] showed only minor increased
values compared to the SECV for all traits (detailed results
not shown). Figures 1 and 2 showed exemplary validation
plots for the most important trait for cassava biofortification
(total beta-carotene components) in chopped and blended
fresh harvested cassava samples. Excellent validation results
proved that NIR can serve to estimate concentrations for all
traits investigated here, except b-cryptoxanthin.

Figure 1: Validation, chopped samples, n=14, total BC by HPLC estimated by NIR in fresh cassava samples

Figure 2: Validation, blended samples, n=14, total BC by HPLC estimated by NIR in fresh cassava samples

The potential to estimate vitamin A carotenoid
concentrations by NIR has been demonstrated and applied
for example for maize, potato, sweet potato and Chinese
kale (14-17). At least in cassava it is an alternative to use NIR
in fresh material for carotenoids. It is not clear if this would
also work in combination with other micro- and
macronutrients.

Conclusion

This study showed that NIR technology can be helpful
for fast and cost-effective cassava carotenoid quality
assessment in large sample sets based on fresh material.
Applying NIR technology for carotenoid screening was
reported for milled maize, freeze-dried potato and sweet
potato sample sets but rarely on fresh root or tuber crop
samples. However, NIR detection limits are often not
compatible with the concentrations of micronutrients in
food crops but we have included the visible wavelength
range in calibration development which is more sensitive
than NIR. Taking into account that plant breeders work with
several thousand breeding lines each year, developed
calibrations provide a useful tool for fast selection of
cassava candidates for further testing by HPLC. They help to
characterise carotenoid profiles to orient nutritional
enhancement of cassava in West Africa and in a worldwide
network in future. The cassava breeding programme at IITA
is working to develop genotypes with good yield, resistance
to biotic (virus and pest) and abiotic (drought, heat) stresses
under African conditions and higher nutritional value. The
developed calibrations have been used to screen cassava
clones from the breeding program at IITA and sets of
advanced clones with high total BC concentrations as well as
good yield have been identified. To date, more than 3000
cassava samples have been analysed by NIR technology at
IITA.

Acknowledgement

This research was largely supported by the HarvestPlus
Challenge Program. This study was conducted in
collaboration with the International Potato Centre (CIP) in
Lima/Peru and the International Institute of Tropical
Agriculture (IITA) in Ibadan/Nigeria.

References

FAO/WHO. (2002). Human vitamin and mineral
requirements. Report of a joint FAO/WHO expert
consultation. Bangkok, Thailand.
Hotz, C. and Brown, K.H. (2004). (Eds.), In Assessment of the
risk of zinc deficiency in populations. Food and
Nutrition Bulletin, 25: 130-162.
Scott, G. J., Rosegrant, M. W. and Ringler, M. W. (2000).
Roots and Tuber for the 21st Century: Trends,
Projections and Policy Options. Food, Agriculture and
the Environment. Discussion Paper 31, International
Food Policy Research institute.
El-Sharkawy, M.A. and de Tafur, S.M. (2010). Comparative
photosynthesis, growth, productivity and nutrient use
efficiency among tall- and short-stemmed raid-fed
cassava cultivars. Photosynthetica, 48,173-188.
Howeler, R.H. (2002). Cassava mineral nutrition and
fertilization. p. 115-147. In: R.J. Hillocks, J.M. Tresh y
A.C. Bellotti. (Eds.). Cassava: biology, production and
utilization. CABI Publ. Wallingford, United Kingdom. p.
115-148.
Alvarez, E., Llano, G.A. and Meja, J.F. (2011). Cassava
diseases in Latin America, Africa and Asia. In: R.H.
Howeler (ed.) The Cassava Handbook: a reference
manual based on the Asian Regional Training Course
held in Thailand. CIAT, Chatuchak, Bangkok, Thailand.
p. 258-304.
Howeler, R. (2011). Importance of Mycorrhiza for
phosphorous absorption by cassava. In: R.H. Howeler
(ed.) The Cassava Handbook: a reference manual
based on the Asian Regional Training Course held in
Thailand. Centro Internacional de Agricultura Tropical,
Chatuchak, Bangkok, Thailand. p. 497-523.
Bourdoux, P., Mutta, M., Hanson, A. and Ermans, A. M.
(1980). Cassava toxicity: the role of Linamarin In: The
role of cassava in the etiology of endemic goitre and
cretinism. Eds. Ermans, A. M. and Mbulamoko, N. M.,
Delange, F. and Ahluwalia, R., IDRC - 136 ottawa pp 15
- 27.
Ceballos, H., E. Okogbenin, J.C. Prez, L.A. Becerra and D.
Debouck. 2010. Cassava. In: J. Bradshaw (ed.). Root
and tuber crops. Springer Publishers. p. 53-96.
Ceballos, H., Hershey, C. and Becerra-Lpez-Lavalle, L.A.
(2012a). New approaches to cassava breeding. Plant
Breed.Rev., 36 (in press).
Delia, B., Rodriguez-Amaya and Kimura, M. (2004).
HarvestPlus Handbook for Carotenoid Analysis:
HarvestPlus Technical Monograph series 2.
Howe, J.A. and Tanumihardjo, S.A. (2006).Evaluation of
Analytical Methods For Carotenoid Extraction from
Biofortified Maize (Zea mays sp.). J. Ag. Food Chem.,
54(21), 7992-7997.
Kimura, M., Kobori, C.K., Rodriguez-Amaya, D.B. and Nestel,
P. (2007). Screening and HPLC methods for
carotenoids in sweetpotato, cassava and maize for
plant breeding trials. Food Chem., 100, 1734-1746.
Brenna, O.V. and Berardo, N. (2004). Application of near-
Infrared refectance spectroscopy (NIRS) to the
evaluation of carotenoids in maize. J. Ag. Food Chem.,
52, 5577 5582.
Berardo, N., Brenna, O.V., Amato, A., Valoti, P., Pisacane, V.
and Motto, M. (2004). Carotenoids concentration
among Maize genotypes measured by near infrared
reflectance spectroscopy (NIRS). Innov.Food Sci.
Emerg. Technol.. 5, 393-398.
Bonierbale, M., Grneberg, W., Amoros, W., Burgos, G.,
Salas, E., Porras, E. and zum Felde, T. (2008). Total
and individual carotenoid profiles in the Phureja group
of cultivated potatoes: II. Development and
application of near-infrared reflectance spectroscopy
(NIRS) calibrations for germplasm characterization. J.
Food Compos. Anal., doi:10.1016/j.jfca.2008.08.009.
Chen, X., Wu, J., Zhou, S., Yang, Y., Ni, X., Yang, J., Zhu, Z.
and Chunhai, S. (2009). Application of Near-infrared
reflectance spectroscopy to evaluate the lutein and -
carotene in Chinese kale. J. Food Compos. Anal., 22,
148-153.
http://www.nir2013.org
AN MKS COMPANY
G
o
l
d

S
p
o
n
s
o
r
s
G F S V
P
i
c
k
i
n
g

u
p

g
o
o
d

v
i
b
r
a
tions
Please do not litter - Ne pas jeter sur la voie publique

Nir 2013 Review

Enviado por

Dados do documento

Descrição original:

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Nir 2013 Review

Enviado por

Direitos autorais:

Formatos disponíveis

NIR 2013 is held under the auspices of the International Council for Near Infrared Spectroscopy

, to which the fruit

, and thus is able to

Você também pode gostar