Spectra description and clustering based on morphological pattern spectum

Morphological pattern spectrum was used to shape features description and analysis of visible/near infrared spectra. Based on the firstly constructed multi-scale Gaussian structure elements, which have similar shape to local feature of spectra, we calculated pattern spectrums of 10 mineral spectra from USGS mineral spectral library, and then we compared both the similarity and k-means clustering results of the original spectrum with that of the corresponding pattern spectrum. The results show that pattern spectrum increases differences between different categories, but retains similarity within a category, and pattern spectrum is more separable than the original spectra. Mineral spectra classification has higher accuracy based on pattern spectrum rather than the original spectra.