Preliminary results from the model-based classification of airborne polarimetric SAR images

In recent years, the application of radar polarimetry for remote sensing of land cover types has attracted extensive interest. Numerous microwave scattering models have been developed and used to interpret the polarimetric SAR data. In this paper, existing L-band backscatter models were used to model land-cover types, such as smooth and slightly rough surfaces (single scattering), urban area and tree trunks (double-bounce scattering) and forested area (diffuse scattering). Using these models, it is possible to construct the amplitude scattering matrix, Mueller matrix, Stokes parameters, etc. for each target. However, a state-vector was created using the Stokes parameters, degree of unpolarization and the phase difference between the HH and VV polarizations. The angle between two state-vectors (the theoretical state-vector derived from the calculation using the existing models and the state-vector derived from the observation or image data) was calculated for each land cover-type. We found that there is a strong correlation between the model predicted and the observed state-vectors for the same land cover types. The angle between the calculated and observed state-vectors is very useful for contrast enhancement and classifying the polarimetric radar data. For this purpose, polarimetric L-band airborne SAR data acquired over a variety of geographic targets are analyzed with the support of field investigations of forest, bare land and smooth surface (or ground and water), urban and rough surfaces. The classification results were presented.