YOLO-HPSD: A high-precision ship target detection model based on YOLOv10

Area of Science:

• Computer Vision
• Maritime Technology
• Artificial Intelligence

Background:

• Accurate ship target detection is vital for maritime traffic management, smart ports, and autonomous systems.
• Current methods require enhancement for high-precision applications in complex maritime environments.

Purpose of the Study:

• To develop a high-precision ship target detection algorithm (YOLO-HPSD) based on YOLOv10.
• To improve the accuracy and efficiency of ship detection in real-world maritime scenarios.

Main Methods:

• Integration of Iterative Attentional Feature Fusion (iAFF) with the C2F module (C2F_iAFF) for optimized feature fusion.
• Introduction of Mixed Local Channel Attention (MLCA) and BiFPN module to enhance multi-scale feature integration.

Main Results:

• YOLO-HPSD achieved an F1-score of 97.88% and mAP@0.5 of 98.86% on a ship dataset.
• Demonstrated a 1.22% and 0.31% improvement in F1-score and mAP@0.5 over YOLOv10n, respectively.
• Achieved a fast detection time of 20.6ms per image.

Conclusions:

• YOLO-HPSD offers a robust solution for high-accuracy and high-speed ship target detection.
• The model provides essential technical support for real-time maritime surveillance and edge computing development.