Fish Image Species Classification Using Conventional Neural Network

Abstract

The classification of fish species is a crucial issue for many disciplines, including marine biology and fisheries management. Conventional Neural Network (CNN) in deep learning has demonstrated promising outcomes in image classification applications in recent years. Using three cutting-edge CNN architectures ResNet50, EfficientNetB0, and InceptionV3 and two different classifiers Support Vector Machine (SVM) and Multilayer Perceptron (MLP), this work offers a study on the classification of fish image species. A dataset of fish photos featuring 10 different species is used in the study. According to the experimental findings, EfficientNetB0 with SVM classifier achieves the maximum accuracy, 97.53%, while EfficientNetB0 with MLP classifier reaches 96.92% accuracy. With 95.00% accuracy with SVM and 95.47% accuracy with MLP, ResNet50 also performs admirably. However, MLP with InceptionV3 only achieve 79.86% accuracy, while InceptionV3 and SVM classifier get 96.78% accuracy. We compared the three CNN architectures with regard to their accuracy and training duration in order to further analysis the results. In comparison to ResNet50 and InceptionV3, the analysis reveals that EfficientNetB0 not only gets the highest accuracy but also has the quickest training time .According to the results, combining CNN with EfficientNetB0 and SVM classifier.