MC-ASFF-ShipYOLO: Improved Algorithm for Small-Target and Multi-Scale Ship Detection for Synthetic Aperture Radar...

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• Synthetic Aperture Radar (SAR) ship detection is crucial for maritime safety and management.
• Current deep learning methods struggle with ship size variations, small target features, and environmental interference in SAR images.
• Existing research often addresses small targets or multi-scale detection separately, lacking integrated solutions.

Purpose of the Study:

• To present MC-ASFF-ShipYOLO, a unified framework for SAR ship detection.
• To enhance the detection of small ships and improve multi-scale detection capabilities.
• To address the limitations of existing SAR ship detection methods.

Main Methods:

• Integration of a Monte Carlo Attention (MCAttn) module in the backbone for enhanced small target focus.
• Incorporation of Adaptively Spatial Feature Fusion (ASFF) modules in the detection head for dynamic multi-scale feature fusion.
• Experimental validation on a combined HRSID and SSDD dataset.

Main Results:

• MC-ASFF-ShipYOLO demonstrated improvements: 1.39% in precision, 2.63% in recall, 2.28% in AP50, and 3.04% in AP compared to the baseline.
• The proposed method outperformed mainstream SAR ship detection models.
• The framework consistently produced high-quality detection boxes, even at high confidence thresholds.

Conclusions:

• MC-ASFF-ShipYOLO effectively addresses the challenges of small target recognition and multi-scale detection in SAR imagery.
• The novel MCAttn and ASFF modules significantly enhance detection performance and robustness.
• This framework offers a valuable advancement for SAR ship detection technology in maritime applications.