Sea Surface Small Object Detection for Unmanned Surface Vessels

Small-scale targets on the sea surface are difficult to be accurately detected in unmanned surface vessels (USVs) operating at high speeds in the presence of ocean current interference. To solve this problem, this paper proposes an improved method based on Mobile YOLOv5 network to enhance the detection accuracy under high-speed operation conditions. By introducing the MobileNetv3 network structure as well as the coordinate attention (CA Attention) mechanism, the proposed network has more advantages in generalisation performance compared to the traditional methods. Specifically, the overall accuracy is improved by 1.6%, and the detection accuracy for small targets is improved by up to 6.4%. In addition, this paper is trained by collecting real maritime USV images. Finally, field experiments based on the USV platform further validate the real-time performance of this method.