CLASSIFICATION OF LAND COVER USING MODEL BASED DECOMPOSITION TECHNIQUES FOR PALSAR DATA

Abstract

Microwave Remote sensing data are broadly used in detection and analysis of the land cover and land use features. The purpose of this dissertation is to examine the application of multi-polarized data of ALOS-PALSAR for land cover and land use mapping. For land cover /use classification, the application aspects of information has been analyzed, which are obtained through the fully polarimetric SAR data. There are numerous techniques available in the literature for land cover classification; however uncertainty still persists during the labeling of various clusters to their classes in absence of any other additional information. Henceforth, the theme of our present research work is centered on analyzing the usefi.zl intrinsic information - taken from SAR observables that are obtained through the various model based decomposition techniques. ALOS PALSAR L-Band fully polarimetric data was used in this study for the entire backscatter analysis. The decomposed outcomes which were retrieved from the aforesaid approach were correlated with the decomposition outcomes from the deorientation method, given by Yamaguchi et.al (2005) utilizing the coherency matrix's rotation, and also from the direct decomposition method without compensating for the shift in orientation angle. As expected beforehand, double bounce scattering from the urban scatters were seen to be enhanced, but the volume scattering component reduced. The response of the several features in the decomposition method using the cosine squared function was analyzed. The relation between the co-polarized and cross-polarized response with the four scattering components was also investigated. The Model Based Decomposition methods (Freeman & Durden 3-component decomposition, original 4-component decomposition, 4-component decomposition with rotation of coherency matrix, new general 4-component decomposition with unitary transformation) have been used to differentiate the classes based on scattering mechanisms. The precision and utility of three supervised classifiers (minimum distance, maximum likelihood, and parallelepiped) have been analyzed for Roorkee region using time-series ALOS-PALSAR data. Here we have compared the different classification methods and their performances or accuracy. The comparison among three classification methods that are maximum likelihood classification, minimum distance and parallelepiped classification have been done. The overall accuracy of these classification methods are critically examined and analyzed. The results obtained from maximum likelihood classifier method, confirm the suitability for the classification of land cover of ALOS PALSAR data very significantly. In coming years this classification method, it would be very essential for iv various environmental and socioeconomic applications which comprises the flood mapping, threst and agriculture monitoring.