Including overall amplitude in the Geometrical Perturbation Filters classifier

Abstract:
Classification is an important topic in SAR remote sensing and polarimetry plays an important rule in this context due to the capability to discriminate among different polarimetric scattering mechanisms. The classifier proposed in this paper is a modification of the distance based on Geometrical Perturbation Filters (GPF), to include the overall amplitude (i.e. the Trace of the Covariance matrix). This is to cover situations where the Trace brings irreplaceable information to separate some typologies of targets. A test of the algorithm is provided on L-band fully polarimetric ALOS-PALSAR data. Additionally, a comparison with the ordinary GPF and the supervised Wishart classifier is presented. The results show that the classification is generally improved neglecting the overall amplitude, nevertheless there are categories of targets where including the Trace seems to avoid misclassification.