DEEP LEARNING APPROACHES FOR AUTOMATED FISH SPECIES IDENTIFICATION

Abstract

The classification of fish species is crucial for effectively tracking and categorizing marine life within intricate underwater environments, alleviating the need for manual intervention. Existing methods for species recognition encounter various limitations, including restricted scalability, accuracy in identification, and the ability to discern multiple species simultaneously. This paper aims to address these challenges by leveraging various deep learning and transformer models, such as CCT, MobileNetV2, and ResNet50 for the automatic classification of 10 Mediterranean fish species. This study aims to reduce manual intervention in species identification, enhance identification accuracy, and automate species recognition by comparing the performance of these three different deep learning and Transfer Learning algorithms. Among the three algorithms evaluated, MobileNetV2 emerged as a top performer, achieving accuracies of 99%  respectively.