INTEGRATION OF NEAR-INFRARED, THERMAL, RAMAN SPECTROSCOPY AND HYPERSPECTRAL IMAGING FOR MINERAL CHARACTERIZATION: IMPLICATIONS FOR FUTURE LUNAR MISSIONS

Abstract

Hyperspectral remote sensing is a valuable tool for mineral exploration and lithological mapping due to its extensive information content, including numerous bands and high spectral resolution. In recent years, numerous airborne and spaceborne hyperspectral sensors have been deployed to study minerals. In addition to these imaging sensors, non-imaging spectroscopic processing techniques such as field and lab spectra covering various wavelength ranges have been widely employed. Here, the focus was on characterizing minerals using datasets from new and recent missions such as PRISMA, DESIS, and AVIRIS-NG, as well lab spectra from Banswara in Rajasthan and field spectra from impact crater sites in India. Additionally, the study explored the potential of imaging hyperspectral data from the imaging and non-imaging VNIS data for the Moon, along with RELAB spectra from Apollo samples, to further enhance mineral characterization and identification on the lunar surface. In the series of new hyperspectral sensor, the focus is on the spectral analysis techniques, including dimensionality reduction and narrowband indices generation, were applied to selected geological regions in India to assess the capability of the PRISMA datasets. Comparative analysis of dimensionality reduction techniques, specifically Singular Value Decomposition (SVD) and Eigen Value Decomposition (EVD) based Principal Component Analysis (PCA), was conducted for PRISMA hyperspectral imageries. Both methods demonstrated advantages, with PCA using EVD facilitating better mineral mapping through band ratios. PCA with Eigen value decomposition and lower correlation between different bands, can be used for deriving various band ratios, which facilitates a better mineral mapping. Along with PRISMA the use of DESIS hyperspectral data for mineral classification in a complex geological setting is also explored. Unsupervised machine learning methods, such as K-means and ISODATA, were employed for mineral classification, while feature extraction techniques were utilized for dimensionality reduction. For DESIS, the rock type identification from the results of PCA and Kernel-based PCA was better than of MNF (Minimum Noise Fraction) and ICA (Independent Component Analysis). To reduce the vegetation effect from a spectrum an "Extended Vegetation Response Removal” (EVRR) thresholding model is proposed. Here, PCA plays an essential role in segregating the various endmembers based on spectral absorption features as noise gets segregated in a few cases.