With the continuous increase in global maritime operations, to prevent marine accidents, unmanned surface vessels (USVs) need to perform real-time target detection of moving maritime targets. Taking ships as an example, this paper proposes an improved ship target detection algorithm based on YOLOv8 to achieve more accurate and faster ship detection capabilities for USVs. The YOLOv8n network is selected, and the lightweight GhostNetV2 bottleneck module is integrated into the backbone network to enhance detection speed and long-distance ship detection accuracy by leveraging its long-range mechanism and lightweight architecture. The SEAM (Spatially Enhanced Attention Module) attention mechanism is introduced in the Head layer to mitigate the impact of ship occlusion on detection. The results show that the improved model's mAP increased by 7.6%, 5.8%, 4.5%, and 3.9% compared to YOLOv4, v5, v7, and YOLOv8, respectively. The improved algorithm provides an effective solution for more accurate and faster ship detection by USVs, advancing the development of maritime monitoring and management technologies.
Marine moving target detection algorithm based on improved YOLOv8
Second International Conference on Image Processing and Artificial Intelligence (ICIPAI 2025) ; 2025 ; Changchun, China
Proc. SPIE ; 13780 ; 137801M
2025-08-11
Conference paper
Electronic Resource
English
Ship Detection Based on Improved YOLOv8 Algorithm
IEEE | 2024
|Ship target detection based on CBAM-YOLOv8
SPIE | 2024
|