多角度SAR图像非目标遮挡缺失信息重构

Missing Information Reconstruction for Multi-aspect SAR Image Occlusion

合成孔径雷达对地面目标侧视成像时,周围高大建筑或树木的阴影投射到目标上导致非目标遮挡,造成目标部分信息缺失,严重影响目标检测、识别以及跟踪的性能。SAR图像对角度敏感,相邻方位角图像之间相关性强,有助于缺失信息重构。本文首次将多角度SAR图像序列与张量分解结合,提出多角度SAR图像非目标遮挡缺失信息重构。首先,引入多路延时嵌入变换,将多角度SAR图像序列转换为块Hankel张量,以获取相邻方位角图像之间的潜在结构关系。然后,在嵌入的高维空间内进行Tucker分解,实现块Hankel张量缺失信息重构。最后对重构后的块Hankel张量进行多路延时嵌入逆变换实现目标缺失信息重构。MSTAR数据实验结果表明,该方法能实现非目标遮挡情况下车辆目标缺失信息重构,且优于经典算法HaLRTC。

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.