Você está na página 1de 4

2011 Third International Conference on Intelligent Human-Machine Systems and Cybernetics

A SOA-based Fish Recognition System Prototype


Xitao Zheng, Junsheng Zhong, Yongwei Zhang
College of IT, College of Fisheries and Life Science
Shanghai Ocean University, Shanghai 201306, China
xtzheng@shou.edu.cn, Zhangyongwei_108@163.com

Abstract - This paper presents a new system to automatically recognition experiences. To enable the quick deployment and
recognize fishes based their digital pictures. The system follows global participation of the project, SOA architecture are
the requirement of SOA (Service Oriented Architecture) in order adopted, the services and database can be deployed on multiple
to provide a self-service platform for public. The system provides servers around the world and service repository can be enriched
services like the training mode, the recognition mode, and the by all the participants.
consulting mode. These services can be deployed onto various
application servers at different locations in the world and serve as Before elaborating our job, we will have a short review of
the interface system of fish knowledge retrieving and our existing jobs and related works in the area.
collaboration. The system services utilize the previous works on
fish morphological ontology construction and digital picture II. RALATED WORKS
recognition for carps and other species. Both user and expert can
upload a training fish picture and the name of the fish, the system Service-oriented architecture (SOA) is a flexible structure
will decompose the picture into properties and compose them into to design system architecture and integration. A system based
owl with the given fish name. In recognition mode the fish picture on a SOA will package functionality as a suite of interoperable
will follow the same procedure and be compared with stored services that can be used within multiple, separate systems
features to get the fish name and then the other properties and from several business domains. The geologically diversified
relations. The system prototype is implemented using web location of fishes and users require a service based, not just
services from Tomcat. The fish images and owl repository are computer based platform. There are a lot SOA system
saved in MySql database and ontologies are stored in OWL files. deployed around the world, here we just list a few.
Keywords: SOA, ontology, fish recognition, Web services Reference [1] presents the Living Human Digital Library
(LHDL) project, which takes advantage of Web Services to
I. INTRODUCTION build a biomedical digital library infrastructure that allows
clinicians and researchers to preserve, trace, share data
In order to find a fish property over the internet, the fish
resources, and to collaborate at the data-processing level. The
name is necessary. A vague fish name will lead to uncertain
paper focuses on transferring existing systems into services.
results. For an expert, it is possible to search fish pictures and
descriptions over the Wikipedia or fishnet, read over a lot of To utilize the divertive and collaboration feature of SOA,
pages of words and picture to determine what a given fish will reference [2] describes a SOA-based precision irrigation
belongs to; for a normal user, he/she may not have the expertise decision-making system which can quickly build a decision
to read through all the links and keywords to reach the model to meet the needs of different users.
destination. Our solution is to develop a picture based Apart from the architecture side, the service
searching system which can recognize a fish by its looking, and implementation takes advantage of the previous jobs on fish
then translate the recognition into a few keywords, and search ontology and picture recognition [3, 4], and the illumination of
the existing knowledge database with the keywords to get other similar systems [6-8].
expected information. For the fish image recognition, there are some papers that
use pattern recognition to do fish identification from a given
As we know, the recognition of a digital image is highly
dependent on its shooting quality. For a given fish, different set, fish counting, ocean color analysis to find fishing area by
shooting angles, illumination condition and camera resolution the satellite images, etc. Because of the huge amount of fish
will produce a picture with great difference in its pixel density species, there is no new definition of fish ontology so it can be
and pixel grayscales. When a low quality photo is uploaded to a used for pattern recognition and classification system.
system, the system can check the quality first and issue Reference [3] presents a new method to define and obtain
instructions to retake the photo if minimum requirement is not morphological ontology features for carp using its digital
met. Also, if the architecture is flexible, the database can be images. The method will be based on digital photograph
enriched and weight can be given on different image qualities. processing to analyze the pictures of the fish and a number of
So the fuzzy neural network system can be applied to organize geometrical features of the fish are defined as its ontological
the fish picture database. Here we assume all the fish pictures features.
are prescreened by computer and stored in a database. The Reference [4] moves further on [3] to make use of these
failed pictures that have passed the prescreening but are morphological features. The ontologies based on the
rejected by feature retrieving and/or matching system will also morphological characteristics of the silver carp are defined,
be stored in the database and can be periodically reviewed and obtained, and compared with those of its family. The
maintained by system administrator or an expert. This kind of constraint, which is the existence of a certain features or the
prototype can be helpful to build up fish pictures and

978-0-7695-4444-1/11 $26.00
$25.00 © 2011 IEEE 66
DOI 10.1109/IHMSC.2011.86
range of a certain features, is defined for silver carp, common able to expose and share functions provided by such
carp and snakehead fish. The digital geometric calculation is applications.
formalized with pattern recognition and other digital image (3) Providing efficient and flexible tools for transforming
processing technique; shared and distinguished features are digital pictures into various set of morphological ontologies,
discussed for three kinds of samples. include definition of new features. This will avoid the problem
Based on the works on [3] and [4], the author proposed a when a lot of fish species are input, the existing features will
method [5] based on ontology formation and fuzzy recognition overlap each other. So new rule and features can be added to
of digital pictures. The method provides automatic creation of transform the pixel data into meaningful data for pattern
the fish geometric ontology and automatic indexing to existing matching. The services of such kinds of transaction are also
Semantic Web. Fuzzy set and fuzzy recognition are used to open to be added.
decide if a new fish picture belongs to an existing training set (4) Making the good use of our services. The business
or not. Carp is used as an example. layer is open to be re-organized so the services can be reusable
Another example [6] is the works from AOS over the internet. There will be technical concerns like the
(Agricultural Ontology Services) project. The project is operations are located in one place and it is operated in another
constructing a primitive ontology in fishery area. The ontology locations with a completely different hardware/software
consults ASFA (Thesaurus Aquatic Sciences Thesaurus), platform.
ARGOVOC thesaurus with multiple languages, oneFish list of (5) Storing and possibly transferring a large amount of
keywords, oneFish Topic headings/titles and FIGIS, and other data. This is because all the pictures the system handled will
thesaurus of fishery or fishery related resources. Plenty of be stored in a database, so the data will be a benefit and a
existing jobs for fish information system are organized by problem, if the database storage is distributed over the internet,
AOS projects. the data synchronization can be complicated. This, as we
Reference [7] uses a computer vision machine (The Catch mentioned before, will be left for future. The usual picture size
Meter) to identify and measure different species of fish on the is about 3Mb, if only a few hundred users visit the system each
conveyor, which provide a comparable prototype for the day, the daily storage increase can be Gb size. So future work
business layer definition. will be focused on database management, data re-organizing
So our system will be the combination of the SOA and indexing.
architecture, the service collaboration and integration of
A. System design with SOA
previous software modules.
The SOA structure is usually composed of four layers:
III. OUR METHOD the presentation, the business, the service and the data layer.
Web Services make it possible to interoperate fish This is the detailed illustration of these layers:
ontology services across the Internet, though it still remains a 噝 Presentation layer: Provides user-friendly interaction
challenge to build a cooperative database environment so the interface, takes user’s input, and displays the feedback from
data can be stored and analyzed in multiple networked the server. Here the presentation layer is accomplished by
locations. There are ways to handle this issue manually, but Dreamweaver with JSP
automatic network database updating and synchronizing is still 噝 Business layer: Processes business logic, composes
a problem for a lot of companies, especially when we are business with several business processes, business function
using MySql as our server database engine. There will be that calls granularity services.
future issues like too many pictures or possible internet attacks 噝 Service layer: granularity services layout, basic
may make the database too big to handle. So the web service functioning unit, can pass and receive message from all
enabled SOA system has more benefit than problems, here are directions (upper levels, same levels, or lower levels). Service
a list of benefits: repository is stored in the database so the business layer can
visit and get the calling details.
(1) Searching and retrieving resources effectively. The 噝 Data layer: Uses a relational database handling, saving
system is server based and all the image preprocessing and data including multimedia data and repository data.
ontology retrieving are performed on the server, only the Fig 1 shows the fish recognition system architecture,
searching results are send back, and the searching using where the presentation layer is the web browser, and the
ontology properties and relations are more effective than harsh business layer is the EJB that interoperate with service
table or other database searching techniques. This is a benefit repository, and the service listed all the services. Data layer
of semantics web and also the benefit of SOA structure. contain the database management and database storage.
(2) Sharing the algorithm, not just the data. While the
current morphological ontologies are defined and implemented B. Service Analysis
by our group, the system is open to all contributors so all the As we know, the fish image needs to be preprocessed:
internet users may have the chance to participate in the service remove noises, adjust direction, turn to gray-scale and edge
development and upload their services. This is a typical sharpening. There are two kinds of services we are currently
feature of our system because it targets the public service and using: the preprocessing service, which include the processes
information sharing and technology reuse. It is crucial to be that are described in the above description; and the ontology

67
retrieving services, which are digital measurement actions to request is recognition, the action will be ontology retrieving
get the size, angle, texture, and a lot of other features that are from OWL file and feature matching will be applied. Then the
described in [3-5]. These features are not mandatory and new image will be stored in database with serial number and fish
features can be added anytime on the basis that the feature can name (recognized or given), necessary information will be sent
be automatically retrieved by computer, so the related back to the browser. If a recognition request failed, the picture
automatic retrieving algorithms need to be wrapped with will be tagged with fail and be left for expert to process, which
services and deployed to our application server. is the consulting mode.
There are five kinds of services in our system as shown in
Fig 1: IV. SYSYTEM IMPLEMENTATION
Image preprocessing service, remove the background The system is developed with Tomcat 6.0 on the server
noises, adjust the fish direction, change the color images to side and Dreamweaver + JSP on the client side. The system
gray-scale and finally take the edge detection using Canny has processed a dozen fish species and the results are satisfied.
operator The current processing limit the fish image to side picture only,
Image training service, image saved to database with with clean background and good illumination. The system user
serial number and annotation field, morphological ontology needs to register in order to prevent abuse.
generation service will be called, OWL service will be called Fig 3 shows the return information when a silver carp is
to validate and save ontologies. uploaded to be recognized and the recognition is successful,
Image recognition service, same as image training, OWL The picture on the top right corner is the trained fish picture
service will be called to retrieve ontologies to perform feature and the picture on the top left is the picture is one that we
matching. upload. Although the two fish have different colors, the
Morphological ontology generation service, global and matching rate is 90%. We will classify a matching if 75%
local sizes will be measured and stored for OWL processing. percent of more of all the 15 features fall in 10% tolerance
OWL service, saving or retrieving of morphological area. In our current models color is not a feature, so the
ontologies and feature validation and cooperation with lexicon matching is unbelievingly higher than expected. In some cases
ontologies. when the color is the only factor to differentiate two samples,
color need to be added as an ontology, currently we can not
C. System Flowchart
handle these kinds of samples.
From Fig 2 we can see the system flowchart of our fish As we mentioned just now, the current fish recognition
recognition system. The user will choose from the browser to system can only process the side image of fish with a clean
decide if to send a known fish picture for training (usually background. This is because that for most fishes their side
expert or faculty) or for recognition, the JSP will call a system pictures provides a better view of their geometrical features.
service to upload the picture. The picture will be prescreened As the system improves, we need to add multiple view of a
and then preprocessed to meet the requirement of fish, like the top view (over the back), the bottom view (over
morphological ontology generation, if the request is training, the stomach), and the head view (over the head). Also, the
the ontologies will be validated and incooperated with the imaging instruction is required and can be intelligently popped
ontology of the same fish name and stored in OWL file. If the

Fig1 Fish recognition system architecture

68
up if the uploaded image failed the prescreening. This is why a click of the picture. Also, for those that can be efficiently
we call it a prototype because the current work is still at the identified by pictures, suggestions can be presented to help
beginning of the fish recognition, a lot of new algorithm and user to do further identification job.
services need to be developed to handle the hundreds of
thousands of fish species. Also, chemical, cell and genetic CONCLUSION
features can be added or linked into the system so the One of the major advantages of our Fish Recognition
researchers can share the very expensive feature extraction by System over the other fish ontology systems is that it
preserves and shares not only original pictures of the fish but
also the data processing and visualization algorithms. The
adoption of Web Service technology in the system has
successfully contributed to this aspect. Breaking a recognition
system into loosely coupled services and enable the
redistribution of the data will enable the participation of the
project and the public fish information service. This will
minimize the possible infrastructure requirement for an image
servicing system.
The implementation of morphological and lexicon
ontology services will help the building up of the knowledge
base and also provide the possibility of automatic importing
from existing fish knowledge base.
A published service structure is important prerequisite of
the global participation and contribution to the project.
So the SOA-based fish recognition system can
successfully handle the image uploading, preprocessing,
feature extraction and matching, the knowledge correlation
and presentation. The system will help the global sharing of
the fish information.
REFERENCES
[1] Xia Zhao, Enjie Liu, Gordon J. Clapworthy, Marco
Viceconti, Debora Testi. SOA-based digital library
services and composition in biomedical applications.
Computer Methods and Programs in Biomedicine, In
Press. Available online 17 September 2010.
Fig2 System flow chart [2] Liyuan Xu, Liping Chen, Tianen Chen, Yunbing Gao.
SOA-based precision irrigation decision support system.
Mathematical and Computer Modeling, In Press,
Available online 4 December 2010.
[3] Yongwei Zhang,Xitao Zheng, Junsheng Zhong. A New
Morphological Ontology Method for Carp Based on
Digital Image Measurement. Web Information Systems
and Mining, 2010, page23-27.
[4] Xitao Zheng, Yongwei Zhang, Junsheng Zhong. An
Ontology Method for Silver Carp Auto-Recognization
Based on Digital Image. Web Information Systems and
Mining, 2010, page331-334.
[5] Xitao Zheng, Yongwei Zhang. A Fuzzy Recognition
Method for Fish Ontology Retrieving. Journal of
Measurement Science and Instrumentation, 2010 Vol.1
Supplement
[6] Qian ping, Zheng Yelu. The Research and Application of
Agricultural Ontology. China agricultural science and
technology press, 2006, Pages 51-52.
[7] D.J. White, C. Svellingen, N.J.C. Strachan. Automated
measurement of species and length of fish by computer
vision. Fisheries Research, Volume 80, Issues 2-3,
Fig 3 The returned information when a match is found September 2006, Pages 203-210.

69

Você também pode gostar