Speckle filtering influence on unsupervised terrain classification based on polarimetric scattering mechanisms

Two scattering model-based decompositions (the first one, developed by Freeman and Burden (1998), where pixels are divided into three scattering categories: surface, volume, and double bounce, and the second, developed by Cloude and Pottier (1997), where scattering mechanisms, characterized by entropy H, and α-angle, are used for classification) were investigated in terms of speckle filtering influence on classification parameters. Freeman and Burden decomposition gives us less problems with relation to speckle filtering. Whereas in Cloude and Pottier decomposition, speckle filtering and averaging are very critical moments. Polarimetric radar images of Moscow region acquired in 1994/10/09 by SIR-C/X SLC L-band data are used for illustration. Results from unfiltered and Lee-filtered (Lee et al., 1999) Burden and Freeman decomposition images and entropy images are compared. A great influence of speckle filtering on Cloude and Pottier classification parameters was demonstrated.