Wiring Effects Mitigation for Through-Wall Human Motion Micro-Doppler Signatures Using a Generative Adversarial Network

Abstract

Through-wall detection and recognition of human motions via radar is of great benefit to public security and emergency service applications. The micro-Doppler signatures extracted from the targets of interest in motion typically contain distinct inner-individual motion features, which is the key to human identification and motion classification. However, no research so far considered a very common application scenario, where the conductive wires buried in the wall are in a powering on mode, let alone study its potential effect on the collected signatures of motion behind wall. As it should be anticipated, strong interference components would be brought in the obtained micro-Doppler signatures, and the subsequent motion recognition would be severely affected. In this paper, we, for the first time, report the effect of the buried live wire on the micro-Doppler signatures. Specifically, a micro-Doppler signature enhancement method, named range-max time-frequency representation (R-max TFR) is utilized to obtain feature enhanced micro-Doppler signatures of behind wall human motions. And to mitigate the clutter components introduced by the buried live wire, the effect is first modeled as an impulse response with its center located at a fixed frequency instance in the R-max TFR map. Then, a novel technique based on conditional Generative Adversarial Network (cGAN), is proposed to fulfill the goal. Both numerical and experimental results, as well as comparisons with other classical de-clutter methods, demonstrate the effectiveness and superiority of the proposed de-wiring cGAN framework in suppressing the wiring effect in behind wall micro-Doppler signatures.