Texture Analysis in Polarimetric Sar Using the Covariance Matrix

Abstract

© IEEE 1992. In the last decade much interest has been devoted to the use of polarisation information in Synthetic Aperture Radar (SAR). Several studies have shown that improved information about the electrical and geometrical properties of surfaces can be inferred using multipolarisation and multifrequency SAR. Typically for rough surfaces, due to the presence of speckle, second order descriptors such as Covariance or Mueller matrices are necessary to fully characterise the backscattered signal. These matrices contain information related to the imaged scene and are used to characterise its backscattered electromagnetic properties. In this paper the Covariance matrix approach is utilised. In the backscatter direction for practically any natural surface (even if it is inhomogeneous and anisotropic), the Covariance matrix contains three real and three complex independent parameters. Since it is a positive semi-definite matrix its three eigenvalues are never negative and the corresponding eigenvector are orthogonal. The eigenvalues are analytically demonstrated to be associated with the intrinsic spatial variability of the span image ('texture'). Experimental data taken from the multifrequency and multipolarisation P L airborne SAR campaign are analysed. The theoretical results show promising agreement with the experimental data, demonstrating the concept of using the Covariance matrix properties for classification or calibration purposes.