Differentiation of urban surfaces based on hyperspectral image data and a multi-technique approach

Airborne hyperspectral data yield a new potential for spectrally-based identification, but also raise new challenges in image analysis caused by a high spatial and spectral variability of the urban environment. The algorithms have to analyze spectrally mixed and non-mixed-pixels of various classes which often show spectrally similar characteristics. In this context we developed a multi-technique approach which combines linear spectral unmixing and spectral classification for a complete inventory of main urban surface cover types. Despite the good results, problems remained in differentiation of spectrally similar surfaces, such as buildings and sealed open surfaces. In this paper we present an improved approach including a new algorithm for shape-based detection of buildings and new rules for an optimized pixel-oriented endmember selection. The approach was developed using DAIS hyperspectral image data of the reflective and thermal wavelength ranges covering a study area in the city of Dresden (Germany). In the result a much improved identification of urban surfaces was achieved due to the incorporation of shape-based techniques.