Cost-effective ecological monitoring in shallow waters using amphibious unmanned aerial vehicles (AUAV) and deep learning-based computer vision.

Traditional marine monitoring techniques, like line intercept transects, are inefficient. Modern methods, including autonomous underwater vehicles (AUVs) and Portable Speedy Sea Scanner (P-SSS), face high costs and labor demands, with limited data quality. This study proposes an amphibious unmanned aerial vehicle (AUAV) integrated with deep learning, enabling both aerial and underwater imaging with real-time GNSS information. The AUAV operates continuously for approximately 30 min and achieves high survey efficiency in shallow waters (0.5-10 m), providing a practical alternative to conventional underwater survey platforms. YOLOv8 is integrated with the AUAV for underwater litter detection and sea cucumber instance segmentation. For underwater litter detection, YOLOv8 achieved a mean average precision (mAP) of 0.764, representing an absolute improvement of 0.144 (approximately 23.2% relative gain) compared with a previously reported Mask R-CNN benchmark (mAP = 0.62). For sea cucumber monitoring, YOLOv8 achieved mAPs of 0.792 for object detection and 0.798 for instance segmentation across two classes, enabling robust pixel-level delineation of individual organisms. Based on the instance segmentation results, pixel-level size estimation of sea cucumbers was achieved by converting skeleton-edge distances into physical dimensions using the mapped spatial resolution (0.21 cm), demonstrating the capability of the AUAV-YOLOv8 framework for quantitative and efficient shallow-water ecological monitoring.