Entropy based landcover classification using polarimetric SAR images and GMM method

An unsupervised classification scheme based on the use of polarimetric entropy, alpha angle and complex Wishart classifier is widely used for landcover classification. The segmented zones in the entropy/alpha plane is fed as the initial input to Wishart based classifiers. The Wishart based classifiers highly depend on this initial input. However as the entropy/alpha boundaries are fixed, this scheme does not perform satisfactorily in some cases. We propose a modified version of this scheme in which the entropy/alpha boundaries are set based on the nature of the dataset. The popular Gaussian mixture model clustering method is used in deciding the boundaries. The proposed procedure which is reported in this paper is found to enhance the landcover classification capability and computational efficiency of the classic entropy based Wishart classifiers.