Automated differentiation of urban surfaces based on airborne hyperspectral imagery

The urban environment is characterized by an intense use of the available space, where the preservation of open green spaces is of special ecological importance. Because of dynamic urban development and high mapping costs, municipal authorities are interested in effective methods for mapping urban surface cover types that can be used for evaluating ecological conditions in urban structures and supporting updates of biotope mapping. Against this background, airborne hyperspectral remote sensing data of the DAIS 7915 instrument have been analyzed for their potential in automated area-wide differentiation of ecologically meaningful urban surface cover types for a study area in the city of Dresden, Germany. The small urban structures and the high spectral information content of the hyperspectral image data require the development of special methods capable of dealing with the resulting large number of mixed pixels. In this paper, a new approach is presented that combines advantages of classification with linear spectral unmixing. Since standard unmixing techniques are not suitable for an area-wide analysis of urban surfaces representing a large number of spectrally similar endmembers (EMs), the mathematical model, were extended and a new method for pixel-oriented EM selection was developed. This method reduces the number of possible EM combination for each pixel by introducing spectrally pure seedlings and a list of possible EM combinations into a neighborhood-oriented iterative unmixing procedure. The results and their comparison with standard spectral classification methods show that the new pixel-and context-based approach enables reasonable material-oriented differentiation of urban surfaces.