Target detection in hyperspectral mineral data using wavelet analysis

A method for the automatic supervised detection of multiple mineral targets in hyperspectral mineral data is presented in this paper. The method makes use of wavelet analysis, wavelet-based denoising using thresholding of wavelet detail coefficients, and feature reduction based on sequential forward selection, which utilises an extension of receiver operating characteristic curves to fuzzy set membership in order to measure discriminating capability. The method is shown to run in time linear to the number of hyperspectral bands, per pixel. Furthermore, an extension of this method to linear unmixing is presented, based on minimising the least-squares error between abundance estimates and actual spectra by varying a thresholding parameter to eliminate outliers and imposing a sum-to-one constraint on the abundances.