A differential SAR Tomography inversion method based on Distributed Compressive Sensing

Abstract

Over the last 20 years, SAR Tomography has emerged as one of the most powerful tools to map and monitor the 3D/4D structure of the earth surface, especially for complicated urban scene. Corresponding reconstruction algorithms have been evolving toward higher accuracy and finer resolution. Conventional spectrum estimation methods, i.e., plain Fourier transform, Capon filtering, MUSIC, SVD etc., suffer from limited baseline span and irregular spatial sampling. Whereas Compressive Sensing (CS) based methods i.e., standard CS method, Distributed Compressive Sensing (DCS) method, Multi-look Compressive Sensing method (MCS), can achieve much higher accuracy and super-resolution by exploiting sparsity or multi-looking. In a complicated urban scene, pixels of different characteristics correspond to different optimal CS-based methods. To achieve the best comprehensive reconstruction result for a whole scene, an adaptive selection of methods based on analysis of pixel scattering characteristics is proposed. Real data experiment is conducted on 31 TerraSAR-X/TanDEM-X strip mode images, and result comparison with the well-known SL1MMER algorithm has validated the proposed method.