Classification of glacier with supervised approaches using PolSAR data.

Glacier comprises distinct features (snow, ice, and debris cover) and their identification and classification using satellite imagery is still a challenging task. Classification of different glacier features (zones) using remote sensing data is useful for numerous environmental and societal applications. The purpose of this study is to develop the fully polarimetric SAR (PolSAR) deep neural networks classification approach for the extraction of different features of the alpine glaciers. The developed approach was tested and classification results were compared with the support vector machines-based classification over the part of two glaciers: Siachen glacier and Bara Shigri glacier. The overall accuracy (OA) of GF-DNN classification is relatively high (91.17% for Siachen and 89% for Bara Shigri) with a good kappa coefficient (0.88 for Siachen and 0.85 for Bara Shigri) as compared to SVM for both the selected glaciers. An improvement of more than 10% is achieved in the OA of GF-DNN classification as compared to SVM for both the glaciers. The obtained classified results and accuracy demonstrates the potential of deep neural networks-based glacier features classification approach for glaciated terrain features.