Mapping land covers of brussels capital region using spatially enhanced hyperspectral images

Hyperspectral data provide indispensable timely information for environmental monitoring. It has become one of the most sought after data set for many specific applications. However, for large areal coverage, spaceborne hyperspectral data are currently acquired at low resolution. Due to the proven usefulness of hyperspectral data and its potential in newer applications, many researchers have investigated novel enhancement methods for Earth Observation hyperspectral images. We have examined four different enhancement methods using a classification scheme at medium level of difficulty. Two of the examined methods are pansharpening methods and the other two are sub-space methods. The results do not show improvements in classification using spatially enhanced images except for the class of Pine trees. However, using full groundtruth of road and buildings, it is clear that spatially enhanced hyperspectral images achieve substantial improvement in classifying small sized houses. Better characterization of road networks can be visualized and also higher accuracy is observed but to a lesser extent than buildings. Among the four methods, a pansharpening method performed best.