Azimuth Sub-band and Eigenspace Decomposition for High Resolution SAR Image Analysis

With the increase of the Synthetic Aperture Radar (SAR) sensor resolution, a detailed analysis of SAR images over urban areas is needed. The high diversity of man-made structures combined with the complexity of the scattering processes makes the analysis and information extraction from high resolution SAR images over such areas non-trivial. In order to simplify interpretation and information extraction, the detection of the so-called Relevant Scatterers (RSs), is proposed in this paper. The advantage of such RSs, is that they have a stable description, allowing a better discrimination from the rest of the scene. This work addresses a RSs detection/characterization problem for high resolution SAR Automatic Target Detection (ATR), based on the covariance analysis combined with the azimuth sub-band decomposition. Indeed, the covariance matrix and its spectrum of eigenvalues are of great interest in the analysis and modeling of experimental data. The detection/characterization is obtained by performing projections of the training data in the eigenspace generated by the covariance formalism. In this article, both a description of our work and a presentation of our preliminary performance results will be provided. Keywords— SAR, relevant scatterers, azimuth decomposition, covariance matrix, eigenspace.