Abstract: Aiming at target tracking and trajectory drawing of ships in satellite remote sensing images, a method combining the global attention mechanism (GAM) module improved YOLO (you only look once) v8 algorithm (GAM-YOLOv8) and the DeepSORT algorithm is proposed. The GAM module is added to the YOLOv8 network structure to enhance the model's ability to extract satellite remote sensing image features and improve the accuracy and stability of ship target tracking. The data enhancement technology based on the RGB (red, green, blue)-HSV (hue, saturation, value) fusion color space conversion convolution module is implemented to expand the data set, helping the model capture more dimensional feature information and further improve the recognition accuracy. The DeepSORT algorithm is used to enhance the stability and accuracy of the tracking process by combining the target's feature appearance and motion information, thereby effectively reducing identity switching and target loss. Experimental results show that the proposed method of combining GAM-YOLOv8 with DeepSORT shows significant performance improvement in remote sensing image ship target detection and tracking tasks compared with the original YOLOv8 model. The accuracy, recall rate, and average precision are increased by 7.6%, 7.9%, and 6.0%, respectively. Meanwhile, the frame rate, multi-target tracking accuracy, and multi-target tracking precision are improved by 17.7%, 6.9%, and 1.9%, respectively, and the number of identity switches is decreased by 10.0%.

Key words: satellite remote sensing, deep learning, target tracking, YOLOv8