Three-dimensional Initial Imaging Result of Chinese Gaofen-3 Satellite Based on CS-TomoSAR

Synthetic aperture radar tomography (TomoSAR) enables three-dimensional (3-D) reconstruction of urban buildings with a high level of details. However, traditional spectrum estimation algorithms for TomoSAR inversion are usually based on large data stacks and high-resolution synthetic aperture radar (SAR) images. For the Gaofen-3 (GF-3) dataset with few available images, due to the low image resolution and large baseline intervals, traditional methods fail to achieve accurate 3-D reconstruction of the interested area. Compressed sensing (CS) method has super-resolution imaging capability in TomoSAR, which can significantly reduce the number of samples required for 3-D imaging. With the help of multi-signal compressed sensing (MCS) theory, this paper introduces a novel processing workflow to achieve 3-D reconstruction of Chinese GF-3 Satellite dataset. This workflow firstly uses two-dimensional (2-D) building footprint geographic information system (GIS) data to extract features of target building. Then, these features are introduced into the estimation as prior knowledge to improve the accuracy of TomoSAR inversion. Finally, to ensure that scatterers on the same contour line of a building are regularly arranged, we exploit total variation (TV) to constrain the distribution of these scatterers. This paper uses the GF-3 dataset to generate high-resolution 3-D point cloud of Beijing, demonstrating the potential of GF-3 satellite for 3-D imaging.