Browsing by Author "Sheoran, Arjun"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
Item Land Cover/Use Classification Using Optical and Quad Polarization Radar Imagery(2009-03-26T18:41:29Z) Sheoran, Arjun; Sheoran, ArjunWith the recent increase in the availability of quad polarization (Horizontal-Horizontal, Horizontal-Vertical, Vertical-Horizontal and Vertical-Vertical) radar data, the need to assess the utility of these datasets for land cover/use classification is crucial. Historically, most spaceborne radars were single wavelength and single polarization. For this study, the Japanese ALOS (Advanced Land Observing Satellite) PALSAR (Phased type L-band Synthetic Aperture Radar) quad polarization radar data were obtained at 12.5 meter spatial resolution. The second dataset to be used in this study was acquired by Landsat TM (Thematic Mapper) at a 28.5 meter spatial resolution. The purpose of this study is to evaluate the classification of various land covers/uses using spaceborne quad polarization radar and optical TM data. Secondly, the study analyzes the utility and improvements that can be made to the radar and TM data with the help of using radar texture and multi sensor fusion techniques, e.g., layer stacking and Principal Component Analysis (PCA). Three study sites Bangladesh, California and Kenya were chosen for analysis in this study. The primary methodology was spectral signature extraction and Transformed Divergence (TD) separability measures to evaluate the relative utility of the various data types. In addition four texture measures, kurtosis, mean euclidean distance, skewness, and variance and four window sizes were analyzed. Supervised signature extraction and classification (maximum likelihood) was used to classify different land covers/uses followed by an accuracy assessment. The combination of radar and Landsat consistently provided excellent classification accuracies, well over 90%. Comparing the two datasets Landsat provided higher classification accuracies as compared to radar and radar texture analyzed individually. Variance texture was consistently the best among all four texture measures, as it showed the most improvement in the TD values. The use of texture on radar was helpful when evaluating the separability among the different land covers/uses. However, texture was not able to provide higher classification accuracies for the different land covers/uses as compared to the original radar and Landsat datasets.