A robust system for classification of remote sensing images

A novel system for the robust classification of multitemporal remote-sensing images is presented. The proposed system is aimed to perform efficiently on images acquired in a specific area of interest at different times also in the cases when the corresponding training set is not available. It relies on three main modules: two modules are devoted to the extraction and selection of features that exhibit a substantially invariant behavior versus the image acquisition date. The last module is an incremental learning classifier able to learn from different training sets as they become available