High Resolution SAR Tomography 3-D Imaging via Sparse Recovery Deep Learning Network

Tomographic synthetic aperture radar (TomoSAR) can achieve high-precision elevation inversion through interferometric phase, and realize three-dimensional (3-D) SAR imaging owing to the virtual elevation synthetic aperture formed by multi-pass. However, traditional high resolution imaging algorithms based on compressive sensing sparse recovery, need to set algorithm parameters and iterations artificially. Moreover, the set value has a great influence on the final imaging quality. In order to automatically adjust the parameters to the optimal state, we propose an efficient unfolded deep shrinkage-thresholding network (UDST-net) for TomoSAR 3-D imaging. The network can realize nonlinear sparse transformation and end-to-end learning through convolution layer, which improves the efficiency of imaging. The results of airborne experiments demonstrate that the UDST-net outperform some traditional CS-based algorithms.