Micro-Doppler Separation Based on U-Net and Plot-Curve Association for Ballistic Target

The high-precision separation of micro-Doppler curves is the key to micro-motion feature extraction and parameter estimation for ballistic target in midcourse. The micro-Doppler curves of each scatter overlap seriously in the time-frequency domain and are also affected by nonideal scattering phenomena such as strong noise and occlusion effects, which poses a significant challenge to the traditional curve separation methods. Aiming at this problem, a micro-Doppler curve separation algorithm under nonideal scattering conditions is proposed in this paper. First, the micro-Doppler curve and noise are separated through the U-Net model in the time-frequency domain. Then on the basis of eliminating the effect of redundant and pseudo plots by plot condensation and plot processing, the micro-Doppler plots are associated and regrouped based on interpolation and curve smoothness function. Finally, the effectiveness and robustness of the proposed algorithm have been illustrated by extensive simulation experiments.