CORE-Net: A Collaborative Optimization Framework for Rotated Ship Detection in Complex SAR Scenes.

Area of Science:

• Maritime remote sensing
• Computer vision
• Synthetic Aperture Radar (SAR) image analysis

Background:

• Rotated ship detection in complex SAR scenes is crucial for maritime surveillance but faces challenges.
• Existing methods suffer from inconsistent multi-scale features, unstable angle regression, and non-uniform training supervision.
• These issues lead to high false alarms, missed detections, and poor localization, especially in challenging inshore environments.

Purpose of the Study:

• To propose CORE-Net, a novel framework for robust and accurate rotated ship detection in complex SAR scenes.
• To address the core bottlenecks of existing rotated ship detection methods.
• To improve both the robustness and fine-grained localization capabilities for maritime remote sensing.

Main Methods:

• CORE-Net integrates Rotation-Consistent Feature Pyramid (RCFP), Progressive Cascade Rotation Head (PCR Head), and Orientation-Aware Regression Enhancement Unit (OAREU).
• An Uncertainty-Aware Sample Reliability Steering (UARS) module optimizes training by downweighting unreliable positive samples.
• The framework focuses on collaborative optimization of feature representation, rotated regression, and sample reliability.

Main Results:

• CORE-Net consistently improves AP50:95 while maintaining high Recall and Precision on public SAR datasets (RSDD-SAR, SSDD+, RSAR).
• The method demonstrates enhanced robustness and fine-grained localization in complex SAR scenes.
• Large-scene inference experiments show successful extension to high-resolution SAR scene interpretation using a sliding-window strategy.

Conclusions:

• Joint optimization of feature representation, rotated regression, and sample reliability is effective for rotated ship detection.
• CORE-Net offers a significant advancement in detecting rotated ships in challenging SAR environments.
• The proposed method provides a scalable solution for maritime remote sensing applications.