Anti-Interference and Multi-Scale Ship Detection Algorithm Based on YOLOv10

Aiming at the challenges of significant scale and perspective variations of targets, marine environmental variability, and dataset scarcity in ship detection under drone vision, this paper proposes the EPA-YOLOv10 ship detection model to effectively suppress wave interference and enhance small target detection accuracy. At the algorithmic level: (1) the C2f-EMSC multi-scale feature enhancement module is designed to achieve local and global feature fusion of ships and multi-scale integration of distant and nearby ship targets; (2)the PTSSA spatiotemporal attention mechanism is proposed to suppress sea surface interference through reverse spatial attention. Additionally, to address the scarcity of ship data, a multi-source ship dataset containing four major ship categories is constructed. Experiments show that the detection accuracy of the proposed model on the self-built dataset Myship is significantly improved with mAP0.5.