AN OPTIMUM TOTAL NUMBER OF CLASSES OF THE UNSUPERVISED WISHART CLASSIFER FOR FULLY POLARIMETRIC SAR IMAGE DATA

摘要

An optimum total number of classes of the unsupervised classifier is derived for fully polarimetric SAR image data. The classifier is based on the ISODATA method with the Wishart distance. The optimum total number is fixed by inspecting results with which two sets of different initial clusters converged. Filling in the details, we vary the total number N of classes of each set simultaneously, and the optimum number is the iVwhich makes the concordance rate of the converged results maximum. In order to evaluate this methodology, we check whether the characteristics of the converged classes can be interpreted by the four-component scattering model. For the fully polarimetric SAR image of ALOS PALSAR, the characteristics can be clearly interpreted by the model.