Is Endmember Extraction a Critical Step in the Analysis of Hyperspectral Images in Mining Environments?

Hyperspectral imaging systems (HSIs) are becoming widespread in the mining industry for mineral classification. The spectral features detectable from near infrared to long-wave infrared make HSIs a potentially efficient tool for exploration, clay mapping, and leach pad modeling. However, the redundancy of hyperspectral data makes the analysis of hyperspectral images complicated and slow. Many researchers have proposed different algorithms and strategies to speed up processing and increase accuracy. These procedures rely on endmember extraction as one of the critical steps. However, no one has tested whether endmember extraction actually improves accuracy under all circumstances. Eliminating endmember extraction, if possible, would speed up the analysis of hyperspectral data. This study tested whether endmember extraction improves the accuracy and efficiency of mapping materials at leach pads, which are among the most complicated situations in mining environments. We compared the accuracy of abundance maps produced with fully constrained least squares (FCLS) (a) with endmember extraction by N-FINDR and (b) without endmember extraction, using a spectral library instead. The results from endmember extraction showed lower accuracy than the results from using a spectral library, probably because the spectral data were noisy and the scanned materials were mixtures. The application of FCLS to hyperspectral images provides useful information for metallurgists. The abundance maps showed that kaolinite, muscovite, and precipitation (hexahydrite and pickeringite) were the dominant minerals on the leach pad. The abundance maps of pipes and precipitation can be used to monitor leaching conditions. Lixiviant ponds mapped out in the abundance map of water can indicate saturation. This technique can also detect organic leakage and agglomeration effectiveness, but it will need different wavelength ranges and more future study. This paper also suggests best practices for using hyperspectral imaging systems to map leach pads.