Dual-Perception Detector for Ship Detection in SAR Images

Ship detection in synthetic aperture radar (SAR) images is a critical task for maritime surveillance, port management, navigation safety, and coastal security. SAR images often contain speckle noise, cluttered backgrounds, and varying ship orientations, which make automated detection challenging. This project introduces a dual-perception detector that combines rotation-aware detection networks and YOLO-based real-time detection to improve the accuracy and efficiency of ship identification in SAR imagery. The system utilizes three models: OBB-SAR-HyperNet, Rotated-Ship-PrecisionNet, and YOLOv8. OBB-SAR-HyperNet focuses on orientation-sensitive detection, accurately capturing rotated ships and refining bounding boxes. Rotated-Ship-PrecisionNet enhances precision through rotation-aware regression, while YOLOv8 provides fast, real-time detection with robust feature extraction. The combination of these models allows the system to handle small, overlapping, and occluded ships effectively. Performance is evaluated using metrics such as precision, recall, F1-score, mean average precision (mAP), and inference speed. Comparative analysis demonstrates that the dual-perception approach outperforms conventional object detectors in both accuracy and reliability. This framework is scalable and adaptable, offering practical utility for automated maritime monitoring and security applications. The project highlights the benefits of integrating orientation-aware deep learning models with high-speed detectors to address complex detection challenges in SAR imagery.

Keywords: SAR Images, Ship Detection, Deep Learning, Rotation-Aware Detection, OBB-SAR-HyperNet, Rotated-Ship-PrecisionNet, YOLOv8, Object Detection, Maritime Surveillance, mAP.