A high resolution SAR image spectral analysis framework for multiple class image mining in urban areas

Traditional approaches for Synthetic Aperture Radar image mining aim at finding accurate representations and models for specific categories of targets. Usually the task of achieving classifications with large number of classes is left for high resolution multispectral images or polarimetric SAR images, although for the latter the number of discoverable classes is significantly lower. This paper discusses the opportunity to use high-resolution SAR image spectra to obtain an increase in information content that can be extracted from the data, in order to identify a large number of classes directly from SLC images. The discussion features three spectrum processing algorithms, incorporating an information theory method, an approach for spectrum description and a spectrum estimation method. The results show that high-resolution SAR images can be used to obtain a statistical description of high accuracy for urban areas.