Supervised classification of scatterers on SAR imaging based on incoherent polarimetric time-frequency signatures

This paper deals with the analysis of the non-stationary behavior of scatterers in polarimetric SAR imaging. A method based on continuous wavelet and incoherent polarimetric decompositions is proposed to extract the polarimetric time-frequency signatures of scatterers. These signatures characterize scatterers according to their polarimetric /or energetic behavior versus the emitted frequency and the observation angle. Then, signatures from reference targets are used to train a multi-layer perceptron (MLP). All in all, SAR imaging data are classified by the MLP. The efficiency of this method is demonstrated, for the deterministic targets (man-made targets). It can be explained by the fact that the man-made targets present a strong non-stationary behavior. But for the vegetation and canopy the results are not convincing. It can be interpreted by the fact that the behavior of vegetation is stationary.