Spektrale Inventur städtischer Oberflächen
(Den Text gibt es leider nur auf Englisch)
Successful identification of surface materials in hyperspectral image data requires detailed knowledge about the spectral characteristics of urban surfaces. For this purpose spectroscopic laboratory studies and field investigations have been carried out that have revealed the great spectral variability of urban surface materials as well as the necessity of relating this knowledge to the analysis of hyperspectral image data as an important prerequisite for an automated identification of urban surface materials. The representation of urban surface materials in hyperspectral image data is also characterized by high within-class variability.
This fact is caused by several factors, such as color, coating, degradation and illumination of the materials as well as preprocessing of the image data. The highest number of spectrally distinct materials and the greatest spectral variability can be found for roof materials. This is caused by the wide range of available roofing materials and varying orientations of the roofs towards the sun and the sensor.
Field measurements (Figure 1) are performed using a field spectrometer (Analytical Spectral Device (ASD) Field Spec Pro FR) with a fore optic lens of 8� and a spectralon panel (Halon) as white reference. The ASD records data in 2151 bands in the wavelength range between 0.35 �m and 2.50 �m using three integrated spectrometers. The sampling interval ranges from 1.4 nm in the VIS/NIR region and 2 nm in the SWIR region resulting in a spectral resolution of 3 nm (VIS/ NIR) and 10 nm (SWIR).
The selection of the image spectra is performed in an interactive way including additional information, such as the results of field investigations, the field spectral library and color infrared aerial imagery. The goal is to identify target areas that represent the materials of interest as homogeneously as possible. The resulting pixels are regarded as spectrally pure pixels. However, due to the spatial resolution of the image data and the natural variability of the surfaces the purity that can be achieved is limited in comparison to laboratory investigations. Thus, a high number of image spectra needs to be determined with the goal of covering the existing variation for each material.