Monitoring Tunas and Sharks Using YOLO Models in the Galápagos Islands

Sharks and tunas play a pivotal role in marine ecosystems, yet their populations are declining due to overfishing and habitat loss. Accurate, non-invasive monitoring methods are urgently needed to guide effective conservation strategies. In this study, we propose a YOLO-based automated detection system designed to accurately identify sharks (specifically silky and tiger sharks) and tunas in underwater videos recorded in the Galápagos Islands. Our training dataset was constructed from two one-minute video clips-one focusing on silky sharks and the other on tiger sharks and tunas-yielding 229 annotated images. We used 90 % of these images for training and 10 % for testing, applying a 5-fold cross-validation procedure. Each model was trained for 30 epochs, and multiple YOLO architectures such as YOLOv8 and YOLOv9 were evaluated based on mean Average Precision (mAP@50) and inference speed. Among the tested configurations, YOLOv9 Medium achieved the highest mAP@50 (95.83 %), while YOLOv8 Medium provided a strong balance between accuracy and computational efficiency, attaining a mAP@50 of 94.20 %. By adjusting the frame processing rate, the system can be optimized for real-time or near real-time monitoring. To avoid data contamination, training and evaluation were conducted on distinct video clips. Results indicate that YOLO-based detection frameworks can facilitate efficient, reliable monitoring of sharks and tunas, providing a powerful tool for informed conservation efforts and sustainable management of marine protected areas.