MCEM: Multi-Cue Fusion with Clutter Invariant Learning for Real-Time SAR Ship Detection

Area of Science:

• Maritime surveillance
• Remote sensing
• Artificial intelligence in defense

Background:

• Small-vessel detection in Synthetic Aperture Radar (SAR) imagery is crucial for maritime security.
• Existing methods struggle with weak target signatures, sea clutter, and computational demands, limiting performance.
• High-resolution representation learning is hindered by trade-offs between noise suppression and feature preservation.

Purpose of the Study:

• To develop an advanced detection framework for small vessels in SAR imagery.
• To overcome limitations of current sea-clutter statistical models and deep learning detectors.
• To enhance maritime surveillance capabilities in challenging oceanic environments.

Main Methods:

• Proposed the Multi-cue Efficient Maritime detector (MCEM), an anchor-free framework.
• Integrated a Feature Extraction Module (FEM) with scale-adaptive convolutions.
• Incorporated a Feature Fusion Module (F2M) for target-background decoupling and a Detection Head Module (DHM) for accuracy-efficiency balance.

Main Results:

• MCEM achieved state-of-the-art performance on challenging SAR datasets (HRSID and SSDD).
• Demonstrated superior performance compared to baseline models like YOLOv8, with significant improvements in Average Precision (APS) and Average Recall (AR).
• Achieved 45.1% APS on HRSID and 77.7% APL on SSDD, showcasing high accuracy in high-clutter conditions.

Conclusions:

• The MCEM framework offers robust maritime surveillance, particularly for small target detection in high sea clutter.
• The synergistic integration of FEM, F2M, and DHM effectively addresses limitations of previous methods.
• MCEM provides a computationally efficient and accurate solution for critical maritime security applications.