Land Cover Classification Using NASA/JPL Polarimeric Synthetic Aperture Radar (POLSAR) Data

This study intends to investigate the use of NASA/JPL POLSAR data (multilook Cand L-bands) for classifying land cover features, such as vegetation (i.e. grass, rice paddy, rubber), natural feature (i.e. river), and man-made features (i.e. canal, highway, runway, built-up area). Covering the area of Jitra, the 10 meters resolution air-borne POLSAR data acquired on 3 December 1996 was used in this study. Prior to the classification, the complex covariance matrix based Lee and Mean polarimetric filters were separately applied for evaluating their speckle suppression performance. In unsupervised classification, the scattering behavior of each pixel in Lee filtered images was analyzed, based on a multi-pixel algorithm and the phase difference. The supervised Wishart classifier was then used to reclassify the pixels of different scattering categories into the corresponding land cover classes. The Kappa statistics computed for both Cand L-band classified images were 0.73 and 0.76, respectively.