COMPUTER IMPLEMENTED METHOD OF SPECTRAL UN-MIXING

摘要

A computer implemented method (70) of un-mixing mixed spectrum data of a physical location received from a spectroscopic measurement device (71). A spatial image of the physical location is received and segmented into a plurality of segments (72). A plurality of sets of endmember spectra are generated with each set comprising an endmember spectrum assigned to each of the segments of the spatial image (73). Abundance optimisation operations are performed (74) to generate abundance values for each endmember spectrum that optimise a cost function (75), thereby generating modelled spectrums. The modelled spectrum that optimises the cost function is output as the unmixed spectrum (78). The cost function (75) comprises a spectral similarity term (76) quantifying a difference between the mixed spectrum data and the respective modelled spectrum, and a structural similarity term (77) quantifying a difference between the spatial image and respective modelled spectrum. Using both terms ensures the un-mixed spectrum more accurately reflects the physical attributes of the physical location from which the mixed spectrum was observed. Particularly suited to the field of hyperspectral imagery.