Missing information reconstruction for multi-aspect SAR image occlusion

When synthetic aperture radar is imaging a ground target from side view, the shadows of surrounding tall buildings or trees are cast on the target, which is lead to occlusion. Occlusion causes partial information loss of the target and seriously affecting the performance of target detection, recognition and tracking. SAR images are sensitive to azimuth, and the correlation between adjacent azimuth images is strong, which is helpful for the reconstruction of missing information. This paper combines the multi-aspect SAR image sequence with tensor decomposition for the first time, and proposes a multi-aspect SAR image reconstruction of target missing information. First, a multi-way delay embedding transform is introduced to convert the multi-aspect SAR image sequence into a block Hankel tensor to obtain the latent structural relationship between adjacent azimuth images. Then, Tucker decomposition is performed in the embedded high-dimensional space to realize the reconstruction of the missing information of the block Hankel tensor. Finally, the reconstructed block Hankel tensor is subjected to multi-way delay embedding inverse transformation to realize the reconstruction of the target missing information. The experimental results of MSTAR data show that our method can realize the reconstruction of the vehicle target missing information in occlusion and better than HaLRTC.