Applying Coherent and Incoherent Target Decomposition Techniques to Polarimetric SAR Data

Classification of polarimetric SAR images has become a very important topic after the availability of Polarimetric SAR images through different sensors like SIR-C, ALOS-PALSAR etc. In this paper we studied effect of different decomposition techniques on the classification accuracy for polarimetric SAR data. We applied different target decomposition techniques (both coherent and incoherent) on ALOS-PALSAR data over Sunderban, West Bengal district of India and later classified the data using various classification techniques. The same training areas are used for classification of decomposed images. A comparative study has indicated that Van Zyl decomposition gives better classification accuracy than other decomposition techniques. It is also observed that the H-A-Alpha decomposition along with the volume scattering Fv or Vv from the volume scattering component from Freeman and the VanZyl decomposition contributes a significant part to the improvement of classification. The accuracy for H-AAlpha classified image is improved from 76.4% to 95.6% and to 96.2% after combining the volume scattering component from Freeman decomposition and Vanzyl decomposition respectively. The accuracy is further improved to 97% when the odd bounce component is combined along with volume scattering to H-AAlpha for both the decompositions. It is observed that among the different classifiers applied, Maximum Likelihood Classifier gives highest accuracy.