Hybrid Deep Learning Approach for Accurate Fish Species Recognition from Bangladeshi Markets

Fish play a crucial role in global diets, economies, and cultural traditions. In Bangladesh, where over 260 native freshwater species have been identified, fish contribute significantly to nutrition and the national economy. However, species identification in local markets remains challenging due to visual similarities between species, inconsistent market conditions, and the absence of standardized labeling. Misidentifications can lead to fraud, consumer misinformation, and inaccuracies in biodiversity records, emphasizing the need for an efficient classification system. This study proposes a hybrid deep learning model integrating EfficientNetB0 and ResNet50 architectures to enhance fish species recognition. A dataset of 27,106 high-resolution images representing 20 fish species was meticulously annotated to ensure precise labeling. Unlike many prior studies, our approach achieves a classification accuracy of 98.52% without requiring data augmentation, demonstrating its robustness in handling real-world variations. The proposed model offers a scalable and efficient solution for fish species classification, with potential applications in market regulation, fisheries management, and biodiversity conservation.