Maritime Ship Tracking Algorithm Based on Improved YOLOv8 and StrongSORT

Maritime targets detection and tracking play an important role in the development of intelligent maritime technology, but the current research on maritime ship tracking is still deficient. For the maritime tracking process, the ship movement, UAV shooting camera jitter and other factors lead to the target recognition fuzzy or lost and aerial photography detection of the existence of small targets difficult to recognize and other issues, this paper proposes a combination of improved YOLOv8 and Strongsort based on maritime ship tracking method. Firstly, an improved GhostConv module is used to replace some of the convolutional layers in the YOLOv8 backbone network, and BiFormer attention is added to the header, which improves the detection accuracy and computational efficiency of the model, and also improves the detection ability of small targets to a certain extent; secondly, the original optimizer of the model is replaced with Lion, which makes the training model lighter and the training speed is faster; and finally. WIOU loss function is used instead of the original CIOU to improve the generalization ability and overall performance of the model. The improved YOLOv8 model is validated on the VOC2007 dataset, and the mAP50 of the improved algorithm is improved by 1.3%. The training is also performed on the produced unmanned boat dataset, and the trained weights are used in the Strongsort algorithm to track the ship, and the results show that the improved tracking algorithm combining YOLOv8 and Strongsort has a better performance effect in ship tracking.