Fusion of spectral and shape features for identification of urban surface cover types using reflective and thermal hyperspectral data

The urban environment is characterized by an intense multifunctional use of available spaces, where the preservation of open green spaces is of special importance. For this purpose, area-wide urban biotope mapping based on CIR aerial photographs has been carried out for the large cities in Germany during the last 10 years. Because of dynamic urban development and high mapping costs, the municipal authorities are interested in effective methods for mapping urban surface cover types, which can be used for evaluation of ecological conditions in urban structures and supporting updates of biotope maps. Against this background, airborne hyperspectral remote sensing data of the DAIS 7915 instrument have been analyzed for a test site in the city of Dresden (Germany) with regard to their potential for automated material-oriented identification of urban surface cover types. Previous investigations have shown that the high spectral and spatial variabilities of these data require the development of special methods, which are capable of dealing with the resulting mixed-pixel problem in its specific characteristics in urban areas. Earlier, methodological developments led to an approach based on a combination of spectral classification and pixel-oriented unmixing techniques to facilitate sensible endmember selection based on the reflective bands of the DAIS instrument. This approach is now extended by a shape-based classification technique including the thermal bands of the DAIS instrument to improve the detection of buildings during the process of identifying seedling pixels, which represent the starting points for linear spectral unmixing. This new approach increases the reliability of differentiation between buildings and open spaces, leading to more accurate results for the spatial distribution of surface cover types. Thus, the new approach significantly enhances the exploitation of the information potential of the hyperspectral DAIS 7915 data for an area-wide identification of urban surface cover types.