Robust nonlinear blind SAR tomography in urban areas

Abstract

Synthetic aperture radar has been widely exploited for the reconstruction of 3-D urban models. Resolving the layovered scatterers in SAR images is typically tackled by an explicit inversion of the SAR imaging model, which is otherwise known as SAR tomography (TomoSAR). TomoSAR is essentially a spectral estimation problem. Existing algorithms usually do not have a closed-form solution, rendering them computationally expensive. This paper demonstrates a robust nonlinear blind SAR tomographic method via kernel principle component analysis (KPCA) to unmix the layovered scatterers, avoiding the computationally expensive multidimensional tomographic inversion. We demonstrate that the state-of-the-art linear PCA-based methods are limited by its strict assumption of orthogonal signals, as well as by its assumption of ergodic Gaussian samples that are often violated in urban area. Experiments on real data show that the proposed method outperforms the state-of-the-art by a factor of three in terms of the accuracy of the phase estimates of individual scatterers.