Automated Fish Recognition and Monitoring System

Project Title

Automated Fish Recognition and Monitoring System

Project Area of Specialization Artificial IntelligenceProject Summary

Measurements of the abundance, distribution, and movement of fish are critical to fishery management. The information of migration of fishes is an important information to know about the behavioral responses of fishes to environmental conditions and weather variations, so for this purpose migration of fishes has been observed in all seasons. Currently, the fish monitoring has been done manually, i.e. humans do it live on the site or by recording videos. But this method of monitoring fish is not accurate and possibility of errors remain in observations. The solution to errors by recording videos and then reviewing them manually takes a lot of time to ensure accurate data and reduction of cost. The video quality sometimes is not good for identification of species by the environmentalists as which is also the reason of inaccurate data. Equipment maintenance for video recording is also a concern, as analog video recording technology has become outdated. The manual data collection of wildlife requires  a lot of labor and assistance of many technologies. In most of the cases, the benefits have resulted from better sensing or sensor deployment (for example, aerial thermal imaging), rather than from automated processing of the observation data.

An automated fish recognition and monitoring system will perform more efficiently and effectively and the results will be more accurate. This type of system automatically captures fish image , extracts contour and categorizes fishes. A fully automated fish recognition system will perform this function more accurately and give reliable data and with reduction in efforts as compared to current manual process. Our research will lead to an affordable vision system for environmental studies and commercial fishery.

Project Objectives

Our goal is to distinguish between the fish species, since the images’ color information is dominated by the water’s hue, color is not useful to differentiate these fish types. Shape also provides little discernment, so we will focus on texture-based classification. We use deformable template matching to align template images and query images in an attempt to improve the performance of such a texture-based classi?er, whose results are sensitive to pixel alignment. The following sections describe the details of this approach.

The overall object is to develop an automated system which can detect a fish, identify its species, and keep a count of each species which has gone past the system. The purpose is to monitor fish migration and to determine the quantity of each species at a particular location.

Project Implementation Method

The system will work according to the key stages of image processing which includes acquisition, pre-processing, segmentation, extraction of information, interpretation and recognition. The proposed method uses image input from a single camera at a resolution of 750x576 pixels.

The entire system can be divided into three components;

• Segmentation,
• Training and recognition using an SVM,
• Displaying the result.

Image set classification involves the comparison of: (i) a set of images containing a single but unknown species of fish (in the analysis that follows, the test set), with: (ii) multiple sets of images each containing a single known species of fish (the training sets). The goal is to determine the closest match between a training set of images and a test set of images, in order to establish the species of the test set. Importantly, both the test and training sets contain multiple images of fish of a single species, encompassing variation in image characteristics such as pose, lighting, background, etc. This fact makes the technique particularly suitable for identifying fish in unconstrained environments.

Benefits of the Project

In Pakistan, there is no awareness to monitor the fish migration using any system. It is all done manually by fisherman and others, they are experienced enough to monitoring the fish because they are working from number of years over it to provide the source of food and taste to other people. But sometimes their prediction can be wrong. Hence this issue motivates the project to be developed for Pakistan specifically to gain efficient results to predict and count the fish migration using the pattern matching through image processing.  An automatic fish recognition and migration monitoring system is more efficient and can provide more accurate data.

Technical Details of Final Deliverable

the final deliverable will be a desktop application. the working is shown in the following diagram

Final Deliverable of the Project Software SystemType of Industry IT , Others Technologies Artificial Intelligence(AI), OthersSustainable Development Goals Decent Work and Economic Growth, Life Below WaterRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	16000
waterproof camera	Equipment	1	15000	15000
printing	Miscellaneous	1	1000	1000