Time-varying Segmentation for Mapping of Land Cover Changes

We propose a new set of algorithms to analyze in an automated fashion multi-temporal (vector-valued) SAR sequences, taking advantage of information redundancies and complementarities. Our goal is twofold: to automatically extract coherent regions and to analyze backscattering coef cients across these consistent regions. The proposed approach allows to discriminate between natural (special weather conditions or to the growth of crops inside elds) and human (early harvest of cultures) changes which is not possible with existing techniques. To this end, several issues are tackled: ltering, edge detection and image segmentation methods capable of exploiting this multi-temporal information. Moreover the system is designed so as to be completely sensor independent and to merge information from different spaceborne systems without any a-priori knowledge about the type of sensor used.