Feature-to-Image RGB Encoding of Physical Tri-Features for Maritime Radar Small Target Detection

Detection of slow-moving small maritime targets under strong sea clutter remains a challenging problem, particularly in low signal-to-clutter ratio conditions where target energy leakage produces ambiguous secondary cells. Conventional tri-feature detectors, such as those based on relative average amplitude (RAA), relative Doppler peak height (RDPH), and relative vector entropy (RVE), improve target-clutter separability but typically rely on point-wise decision strategies and often exclude secondary cells during evaluation, which limits precise localization capability in realistic scenarios. This article proposes a physically informed feature-to-image encoding framework that enables joint spatial interpretation of multiple radar-derived features without modifying standard deep learning (DL) detection architectures. Instead of processing features independently or employing multistream fusion networks, the proposed method maps RAA, RDPH, and RVE onto orthogonal channels of a unified three-channel representation, preserving spatial continuity along the range dimension and facilitating structured context learning. Based on this encoding, lightweight You Only Look Once (YOLO) detectors are adapted for radar target detection with a scale-consistent head design and a Neyman-Pearson-based statistical thresholding strategy to ensure controllable false alarm performance. Experiments conducted on the measured IPIX maritime radar dataset demonstrate that the proposed approach achieves superior detection probability under low signal-clutter ratio (SCR) conditions while maintaining strict false alarm control, particularly in scenarios where secondary cells are retained. The results indicate that appropriate multifeature representation and spatially aware decision mechanisms can substantially enhance maritime radar small target detection without increasing system complexity.