Published in: 2024 6th International Conference on Electrical Engineering and Information & Communication Technology (ICEEICT)
Abstract:
As a riverine country like Bangladesh, where fish play an indispensable role in both cultural practices and dietary habits, they serve as a primary source of protein in rural households. However, the disheartening reality of the gradual disappearance of numerous traditional indigenous fish species poses an imminent threat to their survival. Compounded by the younger generation’s waning familiarity with these culturally significant fish, urgent measures are imperative. Addressing this pressing concern, we propose the implementation of an automatic fish classification system, not just as a timely necessity but as a proactive initiative. Introducing BanFish, the largest fish classification dataset designed for commonly available Bangladeshi fish. Comprising 2,506 images spanning 18 distinct fish species, it aims to fill a critical void in preserving the diversity of indigenous fish. To assess the dataset’s quality, we employ a Vision Transformer with a transfer learning model, achieving an impressive overall accuracy of 83%. Beyond contributing a valuable resource for fish classification, our work marks a significant step toward preserving the rich tapestry of traditional fish species in Bangladesh, envisioning a mobile application to complement our system and bridge generational knowledge gaps.
Paper Link : 10.1109/ICEEICT62016.2024.10534462