Feature Extraction and Classification of Three Indian Major Carps Using Morphological and Texture Features

Appropriate identification of fish species is very challenging for the people with finite ichthyological knowledge owing to a huge diversity of fishes with varying taxonomical features. The automatic identification of fish species is gaining its momentum in fish identification because of its swift response time, user-friendly nature, and low cost. In the present work, we have proposed a machine learning-based approach for identification of three widely cultured Indian major carps, namely Labeo rohita, Labeo catla, and Cirrhinus mrigala belonging to the family Cyprinidae having some common morphological characteristics. A new image dataset of fish species consisting of 1500 samples has been collected from different fish markets in Kolkata, India. After doing necessary preprocessing, each sample is used for extracting different morphological and texture-based features. Various combinations of extracted features are fused and employed to classify the fish species using SVM classifiers. The results are also compared with five other popular classifiers. The maximum recognition accuracy of 93.5% has been achieved on test dataset using the best approach.