Supervised classification by neural networks using polarimetric time-frequency signatures

In radar imaging, the assumption is made that scatterers are white in the emitted frequency band and isotropic for all direction of observation. Nevertheless, new capacities in radar imaging, using a wideband and a large angular excursion, make these hypotheses not valid. Time-frequency analysis highlight this point of view and show some scatterers are anisotropic and/or dispersive. This information source can be completed by radar polarimetry. This paper suggests a supervised classification of scatterers using neural networks based on polarimetric time-frequency signatures. This method is applied here on anechoic chamber data, however can be generalized to SAR or circular SAR imaging.