A Hybrid Deep Learning Framework for Fish Species Classification using Vision Transformers (ViT) and Visual Geometry Group - 16 (VGG-16)

This paper introduces a hybrid model that integrates VGG-16 and Vision Transformers (ViT) for computer-aided fish species identification, an integral part of monitoring biodiversity, fisheries management, and assessing marine ecosystems. VGG-16 effectively extracts high-level local features, and ViT extracts global context relations with self-attention mechanisms. With the integration of both, the model is well-balanced in computational complexity and classification performance and performs better than individual models. The assessment shows significant improvements in accuracy, recall, and F1-score at 92.22% accuracy. It provides scalability and is therefore most suitable for large-scale biodiversity assessments, fisheries surveillance, and rapid ecological intervention. With the development of deep learning, automated classification plays an increasingly important role in marine science, allowing for timely ecological intervention. This study naturally integrates new approaches to enhance precision in image categorization and offers an affordable and effective fish species identification tool and ultimately aids marine conservation and sustainable fisheries.