BEHAVIOUR ANALYSIS OF FISHES USING COMPUTER VISION

Fishes are sensitive to changes in their environment and they respond to these changes with diverse behaviour. Studying these behavioural changes such as feeding habit, swimming speed, swimming pattern, resting period etc. would give a better understanding of their requisite environment, thereby aiding in management to enhance productivity. There is a need for applying economic and rapid methods to analyse fish behaviour in order to prevent any stress and diseases during culture period. The use of sensors, Wireless Sensor Network (WSN) and Machine Vision System (MVS) in aquaculture opens up a new possibility of evolving an innovative, rapid and inexpensive technique for monitoring fishes. in aquaculture.

Wireless sensor network consist of several Sensor nodes connected to each other in the form of a wireless network. Each sensor node has a controller and water quality parameter measuring sensors. To monitor WSN, a Field Programmable Gate Array (FPGA) based control system is developed. FPGA's are extremely flexible since it can be reprogrammed to do any logical function that can be fit into the number of gates the board contains. FPGA's are much faster, has higher computation power and high parallelism presents a good opportunity to collect WSN data concurrently which makes the study more accurate.

Machine Vision System (MVS) is used for analysing fish behaviour to identify stress factors. Farmers need to keep track of the growth rate of fish in order to calculate the right amount of feed to be given. The traditional practice of measuring fish size is laborious and stressful. Image processing can be employed for observing the growth rate and size without causing any stress to fish. The movement pattern is a direct indication of any stress and thus fish behaviour can be predicted using the pattern. Cameras with high resolution are used to capture the video data and the fish are detected in a given frame of video using computer vision techniques. An investigation   can be    performed        using Background subtraction,         Convolution Neural Network,  Watershed algorithms for the detection of fish. Following detection, tracking is performed to obtain the movement pattern. Many tracking algorithms such as Kalman Filter, KLT tracker and Optical Flow can be investigated. Movement pattern of fish is then analysed along with water quality parameter to distinguish a normal pattern from an abnormal one. To classify the pattern as normal and abnormal, algorithms such as Support Vector Machine (SVM) and Histogram of Oriented Gradient with SVM can be explored. In the present work, an automated, non-invasive and smart system for behaviour analysis of fishes is proposed. The outline of the proposed system is shown in figure 1.

Thus with the application of sensors, WSN and MVS in monitoring fishes, a better understanding of the factors required for culturing fishes for commercial purposes can be obtained. Also texture and colour features of fish can be used to identify different species of fish.