Land Cover Classification In L-Band SAR Images

Abstract

The intent of this research is to explore the application of Earth Observation data obtained from fully polarimetric SAR Images. A list of various classification techniques are available to distinguish the different types of objects on earth surface being observed by the remote sensor. ALOS PALSAR is the modern remote sensor SAR Satellite used to observe earth Surface by transmitting and receiving fully polarized electromagnetic wave. There is lack of labeled data available in large number. Unsupervised learning technique K-Means clustering is applied in order to see possible differences among SAR observable back-scattering modes named Single Bounce, Double Bounce, and Volume Scattering. The Overall accuracy achieved by using K-Means clustering is 89.67% signifying that the proposed approach performs to acceptable accuracy for classification of fully polarized SAR images