Machine-Vision-Based Extinct Fish Recognition: Bangladeshi Freshwater Case

Abstract

The biodiversity of fish in Bangladesh is at great risk because of overfishing, poaching, destruction of habitats, pollution, and increased market demands. The safeguarding of such species requires the growth of rare native fish to be recognized immediately and precisely to ensure conservation projects can intervene in time. We concentrated on 6 rare species in this study, which are Bain, Deshi Magur, Kajoli, Khoilsha, Koi, and Tangra. A total of 8,676 high-resolution images were obtained in various districts. The pictures include real-life situations such as different backgrounds, light, and single and multiple fish per picture. Five deep learning models in Convolutional Neural Network (CNN), ResNet-50, VGG-16, Inception V3, and EfficientNet-B0 were tested and trained on the basis of the precision, the accuracy, the F -score and the recall. The most promising one was the Custom CNN, which 1 had an overall accuracy of 97 percent and had good and stable performance with all the species in all conditions. The method offers an efficient and effective method to track fish populations in real-time and assist in the conservation of biodiversity. This approach offers a robust and practical solution for tracking fish populations and supporting biodiversity conservation efforts.