Lightweight Ship Target Detection based on CA-Yolov5

Abstract The models of existing ship target detection algorithms are large and not suitable for deployment at the edge devices, therefore, a lightweight ship target detection algorithm named CA-Yolov5 based on coordinate attention mechanism is proposed. Firstly, CA-Yolov5 takes YOLOv5s as the baseline and replaces the backbone with ShuffleNet v2 which effectively reduces the number of parameters and the amount of calculation. Secondly, the coordinate attention module is introduced to improve the cross-channel positioning ability of the model. Finally, the experimental results on the Seaships dataset show that compared with YOLOv5s, the size of the CA-Yolov5 model is reduced by 90.8% to only 1.3M, and the accuracy can reach 96.9%, which is only 0.2%. CA-Yolov5 is small and accurate, making it more suitable and feasible for deployment on mobile and edge devices.