Land cover classification of built-up areas through enhanced fuzzy nearest-mean reclustering of textural features from X- and C-band polarimetric SAR data

This paper describes a nonparametric algorithm based on fuzzy-reasoning concepts and suitable for land use classification, either supervised or unsupervised, starting from pixel features derived from SAR observations. To this purpose, two novel features describing scene inhomogeneity are utilized. The former relies on the joint density of estimated local standard deviation to local mean. The latter is a multiresolution coefficient of variation calculated in the. domain defined by the "a trous" wavelet transform. Pixel vectors constituted by features calculated from the backscattering coefficient(s) in one or more bands and/or polarizations are clustered. Possible "a priori" knowledge coming from ground truth data may be used to initialize the procedure, but is not required. At each iteration step, pixels in the scene are classified based on the minimum attained by a weighted Euclidean distance from the centroid representative of each cluster. Upgrade of centroids is iteratively obtained both from the previously obtained classification map, and by thresholding a membership function of pixel vectors to each cluster. Such a function has been derived based on entropy maximization of the resulting clusters configuration and has the favorable property of preserving minor clusters. Experimental results carried out on SIR-C polarimetric and X-SAR data of the city of Pavia and its surroundings demonstrate the usefulness of a nonparametric classification to discriminate land use in general, and urban and built-up areas with different degrees of building density, in particular, from SAR observations analogous to those which are routinely available from ERS-2 and EnviSat, and will be provided by the COSMO-SkyMed upcoming mission. Pixel-based classification attains over 70% accuracy without any postprocessing.