Evaluating Texture Metrics on L-band SAR Data and their Impact on Classification Accuracy
Liesenberg, Veraldo
TU Bergakademie Freiberg, GERMANY

Dual-polarization mode data were evaluated for the characterization and classification purposes of different land use classes in an area with shifting cultivation practices located in the Eastern Amazon (Brazil). The Advanced Land-Observing Satellite (ALOS) Phased Array L-band Synthetic Aperture Radar (PALSAR) data were acquired during the dry season of 2007. Intensity backscattering intensity were combined with two different texture parameters. The first one was obtained from the Gray-Level Co-occurrence Matrix (GLCM) texture measurements (Harlick et al., 1973). Whereas the second one was obtained from proper SAR derived parameters based on Intensity and Amplitude information (Oliver and Quegan, 1998). Support vector machines (SVM) was applied as classification method. Multitemporal Landsat data was used as ground reference. The overall classification accuracy employing intensity backscattering data alone is low. Misclassifications were reduced by integrating texture. SAR derived texture based on intensity and amplitude information performed slightly better than GLCM parameters. A moderate window size performed better. SVM was not affected by SAR dimensionality and feature selection technique reveals that texture were important ranked features after backscattering intensity. SAR and their derived parameters encourage the further development of joint techniques in frequent cloud covered areas under the recent "Reducing Emissions from Deforestation and Degradation" (REDD+) protocols.