This study assessed the use of animal-borne video systems (AVEDs) and deep learning models to detect and classify shark species in a coral reef environment. Cameras were attached to the dorsal fins of four Carcharhinus amblyrhynchos individuals in the Fakarava Channel, French Polynesia. Two YOLO11 models were trained: a Shark Detection Model, focused on identifying shark presence, and the Shark Species Classification Model, trained to distinguish among four shark species. The models achieved F1 scores of 0.91 and 0.84, respectively.
Manual MaxN counts were used to validate model predictions and showed strong alignment. MaxN values were also compared across environmental variables, including depth, habitat type, and tidal phase.
The results demonstrate that combining AVEDs with targeted deep learning models is an effective and scalable approach for shark detection and behavioral analysis. Future work should prioritize increasing annotated data, applying data augmentation techniques to enhance training diversity, and integrating environmental sensor data to improve ecological insight and model interpretation.
promotor/supervisor feedback
nothing yet
