Scattering Feature-Guided Open-Set Recognition Model for UWB Radar Vehicles Target

This paper proposes a scattering feature guided open-set recognition (OSR) framework for ultra-wideband radar vehicle targets in open-set scenarios. Within this framework, we introduce a supervised contrastive reciprocal learning (SCRL) based open-set recognition model that leverages scattering features to address the challenges of OSR. First, a scattering feature extraction module is designed to extract target orientation, equivalent scattering center positions, and amplitude distributions from UWB radar data. Second, a scattering feature guided residual network branch is constructed based on the scattering feature extraction module. Finally, by integrating scattering features with SCRL, we propose the scattering feature guided SCRL open-set recognition (SFRL-OSR) network model. Experiments are conducted on both simulated and measured UWB open-set vehicle target datasets. The results demonstrate that the proposed method achieves an open-set classification rate exceeding 0.963 with SNR = 5dB, outperforming the residual convolutional network and the classification-reconstruction learning OSR networks. Further experiments under low signal-to-noise ratio conditions reveal that the proposed method exhibits superior generalization and robustness, making it highly effective for open-set target recognition in UWB radar applications.