Ship-YoloDeep: an algorithm for multitarget ship recognition and tracking in the maritime

In an attempt to surmount the challenges of suboptimal recognition and tracking precision for multi-target vessels in intricate maritime milieu, easy switching of target identity numbers, ineffective occlusion association, and insufficient robustness of cross-domain collaboration mechanisms, this article puts forward a multi-target ship recognition and tracking algorithm Ship-YoloDeep that integrates cross-domain collaboration mechanisms. In the recognition stage, an enhanced ship object detection algorithm based on YOLOv8 algorithm, Ship-YOLO, was proposed. Add GAM after the convolutional layer to improve recognition accuracy. In the tracking stage, an improved ship target tracking algorithm Ship-DeepSort was proposed, which integrates the LSTM-AKF network structure and SIoU loss function into DeepSort, replacing the original Kalman filter and IoU matching, to achieve tracking of ship targets. The experimental outcomes indicate that compared with traditional methods, the precision of ship target detection using the Ship-YoloDeep algorithm has been enhanced precision by 2.48%, mAP by 2.41%, MOTA by 3.47%, and MOTP by 5.21%. The results indicate that Ship-YoloDeep exhibits superior comprehensive performance in cross domain collaborative multi-target ship recognition and tracking at sea.