A Novel High-Resolution Fully Polarization Nonparametric Spectrum Estimation Method for Forest SAR Tomography

The polarimetric synthetic aperture radar tomography (TomoSAR) technique has proven to be a highly promising cutting-edge microwave remote sensing technique for obtaining forest vertical structure parameters because of its ability in three-dimensional imaging. For distributed scatterers like forests, tomograms are generally estimated from the polarimetric covariance matrix. And the estimation accuracy of the polarimetric covariance matrix directly determines the accuracy of parameter inversion. However, most of the current covariance estimation methods are relatively simple, and it is difficult to ensure their estimation accuracy. Moreover, most current tomographic focusing estimators suffer from low elevation resolution and are vulnerable to sidelobe interference, leading to inadequate spectral estimation and parameter extraction performance. Based on this, this article first proposes a high-resolution fully polarimetric nonparametric spectrum estimation algorithm with low sidelobes. In this method, the polarimetric covariance matrix of the target pixel is estimated by searching its homogeneous pixels in the neighborhood based on the nonlocal means method, and the interference caused by the neighborhood pixel with great difference in statistical characteristics is eliminated as far as possible. Then, the obtained covariance matrix is reconstructed by eigenvalue decomposition to enhance the real reflected signal, and reduce sidelobe. To validate the proposed method in this article, we perform experimental verification and accuracy evaluation using TropiSAR 2009 fully polarized airborne data collected in French Guiana. The results demonstrate that the new method has more advantages than other methods in forest tomographic focusing and parameters estimation.