Land Cover Classification by Decision Based Classifier using Dual Polarimetric SAR Observables

In this research work the classification of land cover of Prayagraj region is carried out using Sentinel-1 dual polarimetric C-band SAR (Synthetic Aperture Radar) data. Four classes of land covers namely water, urban, vegetation and bare soil have been considered for classification using 8 polarimetric features which are backscattering coefficients and their ratio (
$\sigma_{vv},\sigma_{vh}, \sigma_{vv}/\sigma_{vh}$
), Radar vegetation index (RVI), Normalized difference polarization index (NDPI) and Eigen value decomposition parameters (Entropy, Anisotropy and Alpha angle). The separability index criterion is applied to determine the best features capable of separating each class. For each of these selected features the threshold value is obtained from our experimental analysis for segregating the classes which are used to develop the algorithm for decision based classifier. Further, a comparison of proposed classifier has been carried out with the three supervised classifiers namely Maximum likelihood classifier, Minimum distance classifier and Random forest classifier along with the accuracy evaluation of each. It has been found that the accuracy of the proposed decision tree classifier is better as compared to the other classifiers which are considered in this work.