Urban area classification with polarimetric statistical features of simulated data in PolSAR images

A new scheme for pixel-based polarimetric synthetic aperture radar (PolSAR) classification of the urban area was proposed. First, the characteristic of urban backscattering was analysed and it was found that the backscattering of buildings is very sensitive to the orientation of buildings. Second, by utilising Euler rotation to the polarimetric coherency matrix, a sequence of data with different rotation angles was simulated. Then a polarimetric statistical feature vector would be extracted from the simulated data. At last, the feature vector together with four components decomposition result would be put into a multiple layer perceptron neural network to get the classification result. The proposed scheme can improve the accuracy of urban area classification in a PolSAR image and be verified by using AIRSAR image data of San Francisco.