Combined Analysis of Hyperspectral and High Resolution Image Data in an Object Oriented Classification Approach

Many applications of remote sensing – like, for example, urban monitoring – require high resolution data for a correct determination of object geometry. These spatial high resolution image data contain often limited spectral information (e.g. three band RGB orthophotos). This poor spectral information lead often to classification errors between visible similar classes like water, dark pavements or dark rooftops. Additional information about the material of an urban object’s surface is needed to separate these classes. Hyperspectral data with the typical high number of bands could be used to provide this information and allow a differentiation of material due to their typical spectra. In the context of remotely sensed data, fusion is often performed by combining high spatial with high spectral resolution imagery on different levels. In contrast to pixel-based approaches like the IHS-transformation or PC spectral sharpening, the emphasis of this paper is fusion of data at feature level. Hyperspectral data recorded by the HyMap sensor are fused with high spatial resolution imagery (digital orthophotos) for a combined endmember selection and classification. After a segmentation of the high spatial resolution orthophotos, the resulting segments will be used to detect those pixels in the hyperspectral data set , which represent candidates for the definition of reference spectra (so called endmember). Afterwards, the segments of the high spatial resolution data will be classified based upon the classification of the hyperspectral dataset and the application of overlay rules. * Ansgar Greiwe, FZG University of Osnabrueck, PO Box 1553, 49364 Vechta, Germany