Land cover classification of urban and sub-urban areas via fuzzy nearest-mean reclustering of SAR features

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. Pixel vectors constituted by features calculated from the backscattering coefficient(s) in one or more bands and/or polarizations are clustered. 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. To yield the weighted distances from a pixel vector, its features are weighted by means of progressively refined coefficients, whose calculation still relies on the membership function through a least squares algorithm. Refinements of the feature-dependent weights are introduced to optimize individual classes. Possible "a priori" knowledge coming from ground truth data may be used to initialize the procedure, but is not required. Experimental results carried out on SIR-C 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 bunt-up areas in particular, from SAR observations analogous to those which will be routinely available from EnviSat. A training set, even of very small size, may be utilized. However, its knowledge affects initialization only and is unnecessary for the iterative refinement procedure. Pixel-based classification attains almost 70% accuracy without any postprocessing.