Effect Of Speckle On The Hh-vv Statistics

Abstract

Speckle phenomenon is intrinsic to SAR imagery and the singlechannel aspect of it has been extensively studied in the litterature. This paper concentrates on the multi-channel aspect of the problem, more specifically it addresses the effect of speckle on the statistics of the HH VV phase difference. This phase difference depends on the type of scattering taking place in the resolution element and can be used to discriminate between single bounce and double bounce scatterings. Furthermore, the standard deviation of the phase depends on the homogeneity of the scattering mechanism in the test area. However, the variability of the scattering area is not solely responsible for the standard deviation: speckle also widens the distribution. This study aims at understanding and quantifying the effect of speckle in the standard deviation of the phase difference. The approach is two-fold: 1) a two-channel speckle model which accounts for the HH VV correlation is developed; 2) The statistical behavior of the HH VV phase difference obtained from simulated data is compared with the one observed using the JPL AIRSAR multifrequency, multipolarization data. The simulation predicts that for a perfectly homogeneous area, .the standard deviation of the phase difference decreases with increasing number of looks; in this case the phase standard deviation is only due to speckle. In the case of natural targets, the width of the phase distribution is a result of both speckle and the intrinsic variability in the target backscattering properties. One of the motivations of the study is to separate the part of the standard deviation due to the target variablility from the part due to speckle. Several target types are analysed and a measure of their intrinsic variability is presented. This measure is then compared to the coefficient of variation and the correlation coefficient as these two coefficients can also characterize the degree of polarization. A speckle sensitivity analysis on these two coefficient is also performed.