Search

This study aims to test the relevance of synthetic aperture radar images of medium and high resolution on the characterization of the types of land use in urban areas. To this end, we relied on textural approaches based on second-order statistics. Specifically, we look for texture parameters most relevant for discriminating urban objects. In this regard, we used Radarsat-1 in fine polarization mode and Radarsat-2 HH fine mode in dual and quad polarization and ultrafine mode HH polarization. The land uses sought were dense buildings, medium density buildings, low density buildings, industrial and institutional buildings, low density vegetation, dense vegetation, and water. We identified nine texture parameters for analysis, grouped into families according to their mathematical definitions in a first step. The parameters of similarity and (or) dissimilarity include homogeneity, contrast, the differential inverse moment, and dissimilarity. The parameters of disorder are entropy and the second angular momentum. Standard deviation and correlation are the dispersion parameters, and the average is a separate group. It is clear from experience that certain combinations of texture parameters from different groups used in classifications yield good results, while others produce kappa values of very little interest. Furthermore, we realize that if the use of several texture parameters improves classifications, its performance reaches a maximum from three parameters. The calculation of correlations between the textures and their principal axes confirms the results. [ABSTRACT FROM AUTHOR]