Physics-Guided Physical-Coordinate Center Detection for Measured Radar Point Targets Under Ground Clutter

Le résumé fourni par la source

Measured radar point-target detection under ground clutter is difficult because weak target echoes often compete with clutter peaks, and image-space detections do not directly represent physical range-Doppler coordinates. This paper proposes a waveform- and physics-guided center detector (WPG-Center) for measured radar maps. The method reformulates detection as continuous physical-coordinate center localization, constructs a four-channel clutter-aware range-Doppler (RD) representation, uses an anisotropic backbone with range-clutter response modulation, and supervises the center heatmap with waveform-resolution-aware Gaussian targets. Conditional physical-prior decoding is used as a targeted candidate reranking step for low-bandwidth, near-zero-Doppler cases. Experiments are conducted on the measured LSS-Ku-1.0 dataset using a strict blocked 5-fold linear frequency-modulated (LFM) protocol with physical localization tolerances and decoded-center detection metrics. WPG-Center achieves a probability of detection (Pd) of 0.850±0.101 and a range root-mean-square error (RMSE) of 18.16±12.19 m, giving the best average decoded-center detection probability and range accuracy among the compared learned and constant false-alarm rate (CFAR)-style baselines, including an independent radar feature pyramid network (Radar-FPN) heatmap detector. The fixed-sigma ablation is substantially weaker, supporting the need for resolution-aware supervision. Fold-wise, qualitative, and stepped-frequency waveform (SFW) analyses are reported as auxiliary evidence to define the scope of the measured LFM conclusion.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.