Underwater fish species recognition: A case study on QUT fish dataset

This study presents a deep learning framework for underwater fish species recognition using the QUT Fish Dataset, refined to 592 images across 21 species. We address underwater imaging challenges–such as noise, poor lighting, and variable orientations–where some images appear upside down. A preprocessing pipeline enhances image quality, while YOLOv8, trained via Roboflow, corrects orientation. Inception V3 classifies species with 86.52% test accuracy. A Streamlit interface enables interactive single-image processing, integrating orientation correction and color detection, supporting marine biodiversity monitoring and fisheries management.