POLARIMETRY BASED LAND COVER CHANGE DETECTION

Abstract

The main aim of this dissertation is to implement the change detection techniques using polarimetric SAR data. Many SAR change detection techniques have been explored but polarimetric SAR interferometry and full polarimetric have not been studied much. In this change detection and monitoring of different classes like urban, water, vegetation, bare soil areas etc. using SAR datasets is an important topic for research. The main classification techniques (supervised: maximum likelihood, minimum distance, parallelepiped and unsupervised: wishart) have been applied to see differences in SAR observables in terms of information and their usefulness in classifying land cover types. In this project, land cover classification technique is developed for labeling clusters into their own classes. A decision tree classification (DTC) is developed that is knowledge based approach which is implemented by data obtained by experimental validation and their empirical evidence. In this standard polarimetric parameters such as polarised backscatter coefficients (linear, circular and linear 45), cross and co-pol ratios for both linear and circular polarisation are used to bearing features for making decision boundries. Classification approaches have been evaluated for fully polarimetric ALOS PALSAR-2 DATA. The five classes of land cover are categorized as: water, urban, bare soil, short and tall vegetation surface which use the dtc to classify individual pixel. DTC based land cover classification is performed by two methods such as OTSU thresholding and histogram based with same training and testing ROI points and evaluation of their accuracy assessment.