Urban features retrieved by hyperspectral multi-angle CHRIS Proba images

The Frascati territory and the contiguous Tor Vergata Campus, a large area in the outskirts of Rome, Italy, form an interesting land cover study site, given their heterogeneity and dynamics. Cereals and vegetables fields are encountered together with extended vineyards and olive groves, mixed with woodland, recent and historic residential areas, isolated buildings of various dimensions and ages, industrial and commercial complexes and miscellaneous artificial surface. This paper presents the potentialities of the multi-angle hyperspectral CHRIS Proba images to perform new land use analysis. Indeed the instrument provides different information with respect to other remotely sensed data, such as ERS and ENVISAT SAR products or high and very high multi-spectral images, also available over this area. After analyzing the multi-spectral and multi-angle signatures over different types of surfaces, a preliminary land use map obtained from CHRIS images has been produced. For the classification task, a neural network approach has been considered. Indeed, neural networks are characterized by a considerable ease in performing non linear mapping of a multidimensional input space into the output one. The obtained results show a satisfactory level of accuracy.