Gini index-based land-cover classification using polarimetric synthetic aperture radar

Classification of the Earth's surface types is one of the important remote-sensing applications of radar polarimetry. An unsupervised classification scheme based on the use of entropy and alpha angle is widely used for land-cover classification using multi-polarization radar images. The polarimetric entropy and the alpha angle are used to characterize a target's randomness and scattering mechanism, respectively. Here, we replace the entropy by the Gini index. Evaluation of the Gini index is computationally efficient. It also overcomes the drawback encountered in entropy evaluation, namely, the use of logarithmic operation. We develop and validate an unsupervised classification technique based on the use of the Gini index and alpha angle and show that it performs better than the classic entropy/alpha classification technique. We have also used the Gini/alpha method with anisotropy and complex Wishart distribution to design a complete land-cover classification scheme. The proposed classification scheme performs better than the entropy/alpha land-cover classification scheme.