An Efficient Deep Learning Model for Whale Shark Detection

Abstrak

Whale sharks (Rhincodon typus) play an essential ecological role as plankton feeders and serve as valuable assets for marine biodiversity and ecotourism. Effective monitoring of their presence and behavior is crucial for conservation and sustainable management; however, conventional observation techniques are often expensive, invasive, and limited in scalability. With the advancement of deep learning-based vision systems, real-time and automated detection has become increasingly feasible. This study employs the lightweight YOLOv10 architecture to develop an efficient whale shark detection system capable of accurate and rapid inference. The model was trained on a curated dataset of underwater images under diverse illumination and visibility conditions. Experimental results show that the proposed YOLOv10-based model achieved a mAP@50 of 97.2% and a mAP@50–95 of 85.5%, while maintaining computational efficiency with only 2,707,430 parameters and 8.4 GFLOPs. These findings highlight the strong balance between accuracy and model compactness, demonstrating that YOLOv10 offers a promising solution for real-time, resource-efficient whale shark detection in marine monitoring applications.