Evaluation of Nonparametric SAR Tomography Methods for Urban Building Reconstruction

Recently, the synthetic aperture radar tomography (TomoSAR) technique has attracted significant attention owing to its 3-D reconstruction capability of complex urban environments. The availability of a high number of images is usually a requirement for nonparametric spectral estimation methods. This letter evaluates the potential of four nonparametric spectral estimation algorithms, that is: 1) linear prediction (LP); 2) minimum norm (MN); 3) singular value decomposition (SVD); and 4) Capon for improved tomographic reconstruction of the third dimension of built-up areas with a small number of observations. The performance analysis is carried out for both simulated and real SAR datasets. The returns from the employed techniques indicate the efficient and low-computational estimator of LP by minimizing the average output signal power at the array of antenna elements and make it possible to separate multiple scatters at a distance below the Rayleigh resolution and clean sidelobes' phenomena in the elevation profiles. The experimental results of a dataset acquired by the TerraSAR-X sensor verify the effectiveness of the LP spectral estimator algorithm in the reconstruction of urban buildings. The estimated height of scatterers with the LP method is considerably similar to the ground-observed data.