LAUSR

Underwater target detection is the foundation and guarantee for the autonomous operation of underwater vehicles and is one of the key technologies in marine exploration. Due to the complex and special underwater environment, the detection effect is poor, and the detection precision is not high. In this paper, YOLOv5 (You Only Look Once v5) is used as the overall structural framework of the target detection algorithm, and improvement is made on the basis of its detection precision in the underwater environment. Specifically, an attention mechanism (Channel and Spatial Fusion Attention, CSFA) that fuses the channel attention and spatial attention is proposed and added to the YOLOv5 network framework, enabling the network to focus on both the prominent features of the detected object and the spatial information of the detected object. The proposed method was tested on the underwater target detection dataset provided by the China Underwater Robot Professional Competition. The experimental detection precision (P) reached 85%, the recall (R) reached 82.2%, and the mean average precision (mAP) reached 87.5%. The effectiveness of the proposed method was verified, and its underwater target detection performance was better than that of ordinary models.