Ship Detection Methods Based on Improved YOLOv8 in Surveillance Video

To effectively address the special challenges of ship detection in key maritime surveillance areas such as ports and waterways—including large scale variation, dynamic background interference, and drastic lighting changes—this paper proposes a ship detection method based on an improved YOLOv8 model. By introducing a MSPA module and an AFPN, the model enhances its ability to perceive key features across scales and optimizes multilevel feature fusion. Experimental results show that the mAP@0.5:0.95 of the improved YOLOv8 reaches 60.31%, which is 2.69% higher than the baseline model. The Rec. for small targets is notably improved to 46.95%. While maintaining high accuracy, the model achieves an inference speed of 12.84 FPS, meeting real-time processing requirements. This demonstrates that the improved YOLOv8 offers a favorable balance between detection performance and computational efficiency, making it highly suitable for deployment in practical maritime monitoring systems.