Fish Detection and Classification Using SOM

Abstract

Fish species detection, classification and study have been an integral part of marine science research. Processing of fish images require methods having various visual techniques, one of the implementation is pattern recognition. The best technique used are the neural networks. Neural network is a “connectionist” computational system. Using these methods an effective system is built for recognition and classification of fish. The goal of the project is to recognize fish based on the texture. There are mainly four features taken into consideration for detection and they are contrast, energy, homogeneity, correlation. For classification unsupervised neural network self-organized map is used, which gives the output as different classes of fish. The system accepts input as a set of images. The output of the system is SOM-plots. The SOM plots are hexagonal neurons. Each neuron consists of the number of images belonging to their respective class. The system can be used by marine experts and for fish research to distinguish between fish species and also has medicinal benefits, considering the medicinal values of fish.