This work presents an automated approach utilizing superpixel segmentation for detecting spectrally Neutral Regions (NR) in hyperspectral images. NRs are often used in planetary geology as spectral divisors to Regions of Interest (ROI), both to enhance key mineralogical signatures and correct for systematic errors such as residual atmospheric distortion. We compare automated NR selections to handpicked examples with mineralogical summary products used in analysis of data from the Compact Reconnaissance Imaging Spectrometer for Mars (CRISM). We also present a new summary product to quantify the level of atmospheric distortion in a CRISM spectrum. We find that the automated algorithm matches manual NR detection with regards to mineral spectral contrast and outperforms manual selection for reducing atmospheric distortion.
展开▼
