A Novel Decoupled Feature Pyramid Networks for Multi-Target Ship Detection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Maritime Technology

Background:

• Current ship detection methods struggle with complex scenarios, diverse scales, and slow training.
• Accurate ship detection is crucial for maritime safety, harbor management, and ocean surveillance.

Purpose of the Study:

• To propose a novel multi-target ship detection method addressing limitations of existing approaches.
• To enhance accuracy and efficiency in detecting ships across various scales and complex environments.

Main Methods:

• Introduced a feature decoupling module to separate contour and position features.
• Constructed a feature pyramid with a gating attention module to improve small ship detection.
• Developed a multi-feature fusion algorithm for adaptability to scale variations.

Main Results:

• The proposed Decoupled Feature Pyramid Network (DFPN) achieved significant improvements over YOLOv4, Faster R-CNN, Mask R-CNN, and DetectoRS.
• DFPN demonstrated a 6.3% mAP increase compared to YOLOv4, with 20 FPS higher speed.
• Achieved superior performance in multi-target ship detection datasets, showing high accuracy and fast detection.

Conclusions:

• The DFPN method effectively overcomes challenges in multi-target ship detection.
• The algorithm offers high accuracy and speed, making it suitable for real-world maritime applications.
• DFPN enhances feature resolution and adaptability to scale changes for robust ship detection.