Cyber physical solutions for aquatic monitoring using YOLO with BCP loss for intelligent underwater camouflaged object detection

Efficient fish detection in underwater environments is crucial for monitoring marine biodiversity, yet it poses significant challenges due to low visibility, complex backdrops, and diverse fish morphologies. This study presents a vision-based cyber-physical solution leveraging an enhanced YOLO architecture with a novel Balanced Coverage and Penalization (BCP) loss function. Our key contributions included a novel multi-head detection strategy. This unique approach not only improved the model's adaptability to diverse fish morphologies but also led to the development of the BCP metric, which reduced prediction-ground truth discrepancies, thereby enhancing detection accuracy and robustness. Experiments on the COD10K and Halibut datasets show consistent gains in precision and stability compared with existing methods. These results demonstrate the effectiveness of the BCP loss in refining spatial coverage and highlight its potential for scalable applications in smart aquaculture, automated fisheries management, and marine environmental monitoring.