A Ship Detection Model Based on Dynamic Convolution and an Adaptive Fusion Network for Complex Maritime Conditions

Area of Science:

• Computer Vision
• Maritime Surveillance

Background:

• Ship detection is crucial for maritime safety but faces challenges like false detections, complex backgrounds, and varying scales.
• Existing models struggle with performance in adverse weather conditions.

Purpose of the Study:

• To develop an advanced ship detection model, YOLO-Vessel, to overcome limitations of current methods.
• To enhance maritime safety and vessel monitoring through improved detection accuracy and speed.

Main Methods:

• Developed YOLO-Vessel based on YOLOv7, incorporating Efficient Layer Aggregation Networks and Omni-Dimensional Dynamic Convolution (ELAN-ODConv) for feature extraction.
• Introduced space-to-depth structure in the head network to improve detection of small ship targets.
• Implemented ASFFPredict for robust multiscale ship target detection.

Main Results:

• YOLO-Vessel achieved a mean average precision (mAP) of 78.3%, outperforming YOLOv7 by 2.3% and Faster R-CNN by 11.6%.
• The model operates in real-time at 8.0 ms/frame.
• Demonstrated superior performance in adverse weather conditions.

Conclusions:

• YOLO-Vessel offers a robust and effective solution for real-time ship detection.
• The model significantly enhances maritime safety and vessel monitoring capabilities.
• The proposed architectural innovations address key challenges in maritime surveillance.