Statistical Assessment of Eigenvector-Based Target Decomposition Theorems in Radar Polarimetry

The performance of quantitative remote sensing based on multidimensional synthetic aperture radars (SARs), and polarimetric SAR systems in particular, depends strongly on a correct statistical characterization of the data, i.e., on a complete knowledge of the effects of the speckle noise. In this framework, the eigendecompostion of the covariance or coherency matrices and the associated decomposition have demonstrated the potential for quantitative estimation of physical parameters. In this paper, we present a detailed study of the statistics associated with this decomposition. This analysis requires the introduction of mathematical tools that are not well known in the remote sensing community. For this reason, we include a review section to present them. Using this work, we then present an expression for the probability density function of the sample eigenvalues of the covariance or coherency matrix. The availability of this expression allows a complete study of the separated sample eigenvalues, as well as, the entropy H and the anisotropy A. As demonstrated, all these parameters must be considered as asymptotically nonbiased with respect to the number of looks. In order to reduce the biases for a small number of averaged samples, a novel estimator for the eigenvalues is proposed. The results of this work are analyzed by means of simulated and real airborne SAR data. This analysis permits us to determine in detail the effects of the number of averaged samples in the estimation of physical information in radar polarimetry.