Evaluation of spectral unmixing algorithms for mineral identification on Mars.

摘要

Spectral unmixing by linear least-squares is a technique that is widely used to identify and estimate mineral abundance in hyperspectral images of Earth and Mars. It is used to obtain the best fit between laboratory spectra measured of known minerals and the observed field and airborne data sets.; The drawback of this established method is that mathematically it can return a negative abundance of a material, even though that is physically impossible especially when the minerals in the target are unknown. The accuracy of the abundance is also affected by the spectral band depth of the mineral and the viewing angle. Various constrained and unconstrained least-squares algorithms were employed to analyze using IDL software along with the ENVI 4.0 package on synthetically created hyperspectral images and on terrestrial Mars analog sites imaged using airborne (SEBASS) and ground-based (Tonka) hyperspectral imaging systems. The results show that the constrained methods do not result in negative abundance values but also show no improvement in the accuracy of the identification of minerals when compared to unconstrained methods.