A Two-Stage Maritime Radar Target Detection Method Based on Multi-Domain Feature Fusion

Maritime radar is a crucial component of the Vessel Traffic Service (VTS) system, and sea clutter interference is a significant factor affecting the accuracy of radar target detection and tracking. To enhance the accuracy of maritime radar target detection, the spatiotemporal and time-frequency information in radar echoes are utilized, and a two-stage target detection method was developed. In the coarse stage, a detection process for rapidly extracting target candidate regions is designed by combining the phase and time-domain amplitude features. In the fine stage, four key features extracted from time-frequency domain are adopted, including Relative Average Amplitude (RAA), Temporal Hurst Exponent (THE), Sample Entropy (SE), and Doppler Peak Height (DPH). We use these features construct a fused feature space, subsequently, the feature pyramid is utilized to enhance feature representation capabilities across multiple scales. Finally, a simple and feasible decision tree algorithm is used to classify the fused feature space, distinguishing between sea clutter and targets. The empirical results demonstrate that compared with YOLOv8 and Constant False Alarm Rate (CFAR) detection methods, the proposed method increases accuracy to 93.2% while maintaining a recall rate of 89.2%.