Land cover mapping method for polarimetric SAR data

We consider a land cover mapping method for polarimetric SAR data analysis. The method is based on a neural network whose input data are elements formed by the Stokes matrix. In this case, we must select a suitable combination of complex elements as a feature vector. After forming the probability density for each element and comparing the characteristics between JM distances, we determine a specific feature vector as the input for the network. As a result of experiments using SIR-C data, average accuracy for classification results is 86.40 percent, where (i) the 8D feature vector with backscattering coefficients and pseudo-phase differences between HH and VV from L and C bands and (ii) the competitive neural network with 8 input and 40 output neurons are simultaneously employed. In comparison, the proposed method outperforms other methods in average accuracy.