A PolSAR Clustering Scheme Using the Model-free Scattering Power Components

Abstract

Using full-polarimetric Synthetic Aperture Radar (SAR) data, the model-free four-component scattering power decomposition technique overcomes several limitations of other model-based target decomposition approaches. In particular, all the power components are roll-invariant, non-negative. Alongside this, an unsupervised clustering technique was proposed with 24 clusters. However, such a large number of clusters may not be necessary for classes with scanty pixels. Hence, this study modifies the existing clustering scheme for land cover characterization based on specific criteria on the number of pixels pertaining to some particular dominant mechanism. This modification adequately captures the clusters' variations according to their physical and geometrical properties.