Determination of robust spectral features for identification of urban surface materials in hyperspectral remote sensing data

Hyperspectral remote sensing data open up new opportunities for analyzing urban areas characterized by a large variety of spectrally distinct surface materials. Spectroscopic analysis using diagnostic spectral features yields the potential for automated identification and mapping of these materials. This study proposes a new approach for the determination and evaluation of such spectral features that are robust against spectral overlap between material classes and within-class variability. Analysis is based on comprehensive field and image spectral libraries of more than 21,000 spectra of surface materials widely-used in German cities. The robustness of the interactively defined spectral features is evaluated by a separability analysis. This method is performed based on confusion matrices for each material computed from classification results. For comparison this analysis is also performed for material-specific gray values of selected bands. The obtained commission and omission errors show superiority of the spectral features compared to gray values for most of the investigated materials. The results indicate that robust spectral features yield the potential for unsupervised detection of endmembers in hyperspectral image data.