A DEEP LEARNING COREGISTRATION APPROACH FOR DISTRIBUTED GEOSYNCHRONOUS SAR THREE-DIMENSIONAL DEFORMATION RETRIEVAL

Geosynchronous Synthetic Aperture Radar(GEO SAR) has become a hot spot because of short revisit time and wide coverage. Compared with single satellite, distributed GEO SAR provides rich observation angles which makes high-accuracy three-dimensional(3D) deformation retrieval possible. However, there are significant differences in the resolution and texture of Interferometric Synthetic Aperture Radar(InSAR) image at different observation angles, which will lead to reduced accuracy of 3D deformation retrieval. In terms of problems above, Pseudo-CycleGAN is proposed in this paper based on phase unwrapping Deep Neural Network(DNN) and CycleGan. It can improve the accuracy of 3D deformation retrieval through texture assimilation of interferogram with high phase accuracy.