Mineral mapping and vegetation removal via data-calibrated pixel unmixing, using multispectral images

Publisher Website
Google Scholar
Abstract
In this paper an image unmixing process is described which is based on the assumption that pixel reflectance is a linear mix of component reflectances. Three stages are used initially to calibrate the multispectral data to shaded reflectance: (1) minimum value subtraction, (2) band-mean standardization and (3) reflectance-mean equalization. Using laboratory or field spectra, linear mixing equations are then solved for substance proportions. For a geological test site in North Queensland, Australia, two vegetation types (green and dry) and four mineral types (clay, haematite, goethite and quartz) were unmixed using NS001 aircraft scanner data. The six images created show the spatial abundance of individual vegetation and mineral types. The obscuring effect of both green and dry vegetation was removed from mineral maps by separately scaling up mineral proportions in each pixel. Unmixed mineral maps were used to locate potentially economic hydrothermal alteration zones. In addition, scaled mineral proportions were used to recalculate brightness values for the original wavelength bands producing de-vegetated geological reflectance images which are useful for identification of rock-types.