Efficient identification of Li-rich claystones based on imaging hyperspectral remote sensing

• Clay ores of different grades exhibit obvious differences in spectral characteristics. • Feature bands extracted via the CARS method significantly improve the clay lithium content estimation model. • A new technical method for clay-type lithium-rich ore exploration optimization is proposed, using imaging hyperspectral images. Lithium is a strategic rare metal and a crtical component of the rapidly growing new-energy industry. In recent years, clay-type lithium resources have attracted increasing attention because of their large resource potential, wide distribution, and relatively low development and utilization costs. However, efficient utilization of these resources remains challenging, as the introduction of gangue significantly increases beneficiation and metallurgical costs. Rapid and cost-effective identificationli-rich claystones is therefore essential for advancing exploration and development. In this study, the Xiaoshiqiao are in Yuxi, Yunnan Province, China, was selected as the study area. A total of 50 representative claystone samples were collected to evaluate the potential of imaging hyperspectral technology combined with characteristic band selection methods for identifying Li-rich claystones. The results show that claystone samples with different lithium contents exhibit distinct characteristics, with pronounced absorption features near 1.8 μm, 2.2 μm, and 2.45 μm. Logarithmically transformed spectra can effectively suppresses noise , enhances diagnostic spectral features, and strengthens the correlation between spectral signatures and lithium content. Compared with models based on original spectra, the coefficient of determination (R 2 ) and relative percent deviation (RPD) increased by 0.13 and 0.45, respectively, while the relative root mean square error (RRMSE) decreased by 0.10. Furthermore, estimation model constructed based on the competitive adaptive reweighted sampling (CARS) characteristic band selection method significantly outperform full-band models, achieving an R 2 of 0.81, an RRMSE of 0.34, and an RPD of 2.28. Applicaction of the optimized model to hyperspectral ore images enables rapid and effective discrimination of ore grades. These results demonstrate that imaging hyperspectral analysis provides an efficient tool for the preliminary screening of clay-type lithium-rich ores and the identification of lithium anomaly zones at the aerospace scale.