Crop Classification Using Fully Polarimetric SAR Imagery

An important prerequisite for improving the classification accuracy is to fully extract the characteristics that reflect physical properties of the objects. The objective of this study is to investigate the capability of quad polarized Synthetic Aperture Radar (SAR) images for crop classification in Ontario, Canada. Multi-temporal RADARSAT-2 fine beam quad-polarized SAR data were acquired. A support vector machine (SVM) classifier was selected for the classification using combinations of the polarization characteristics and texture features. The polarimetric features, including odd scattering, double scattering and volume scattering, were extracted from classic Pauli decomposition. Eight texture features were extracted from grey level co-occurrence matrix (GLCM). Principal Component Analysis (PCA) method was applied to reduce the redundancy of texture features. The results indicated that multi-temporal SAR data achieved satisfactory classification accuracy. Texture features of SAR data were useful for improving classification accuracy. SAR data have considerable potential for agricultural monitoring.