Detection and Enhancement of Aqua Images Using Haze and YOLO Algorithms

The study addresses the challenge of fish classification in marine ecology using underwater cameras. Utilizing a two-stage deep learning approach, we employ YOLO object detection for recognizing each fish in the image without pre-filtering, ensuring inclusivity regardless of gender or species. The second stage employs a convolutional neural network (CNN) with click-and-excite (SE) architecture for classification without filtering. To enhance classification accuracy for temperate fish species, we propose curricular adjustments addressing low educational standards. The model, trained on public ImageNet data, achieves a remarkable 99.27% accuracy, with post-training values of 83.68% and 87.74%, with and without image enhancement, respectively. This approach maximizes data sets generated by underwater cameras, contributing to effective marine ecology research.