YSBE-Based Target Detection via Multibeam Sonar on Velocity-Constrained AUVs: Toward Port Inspection

Abstract

Translate Article

Port underwater inspection is essential for ensuring maritime safety and operational continuity. However, low visibility and complex environments make it challenging to achieve reliable detection. This study addresses the issue of underwater target recognition based on multibeam sonar imagery. Specifically, we propose YOLO-ShuffleNet-BiFPN-EIOU, an enhanced YOLOv5 detection framework with three key improvements: first, the adoption of the lightweight backbone network ShuffleNetv2, second, the incorporation of a bidirectional feature pyramid network, and third, the optimization of the enhanced intersection over union loss function. Furthermore, we theoretically derive the maximum permissible velocity threshold for autonomous underwater vehicle (AUV), leading to a novel velocity-constrained controller that improves sonar imaging quality. A general transformation function is applied, and a functional dependence with quasi-linear characteristics between the independent and dependent variables is established, converting the partially constrained AUV system into an unconstrained one. Finally, experiments conducted in a pool and a real-world port demonstrate that the proposed method achieves significant improvements in accuracy and efficiency compared to YOLOv5m, with mAP@0.5 increasing by 3.4%, mAP@0.5:0.95 improving by 5.3%, giga floating-point operations per second reduced by 89.8%, and the velocity-constrained AUV operation effectively enhances detection performance.