Fish Species Classification Using Probabilistic Neural Network

Abstract

Abstract The number of varieties of fish species in the same family causes difficulties in classifying fish species directly. Currently, the process of fish species classification accomplished in the fisheries section uses direct eye observations and knowledge assumption and then compares the existing characteristics with reference books. Therefore, an image processing and neural network approach are needed to classify fish species effectively and efficiently. In this study, there are three fish species classified in the Scombridae family, those are skipjack tuna, tongkol, and tuna with ‘out of water’ conditions. The proposed approach utilizes ROI (Region of Interest) in the form of a fish belly. The combination of GIM (Geometric Invariant Moment) feature extraction, GLCM (Grey Level Co-occurrence Matrix) texture feature extraction, and HSV (Hue Saturation Value) colour feature extraction was used to extract features in the image. For the process of determining the type of fish species, the method used is Probabilistic Neural Network. Based on the results of research on 112 images of data training and 29 images of data testing, an accuracy rate of 89.65% was obtained.