Insights into Lunar Mineralogy: An Unsupervised Approach for Clustering of the Moon Mineral Mapper (M3) spectral data (2411.03186v1)

Abstract: This paper presents a novel method for mapping spectral features of the Moon using machine learning-based clustering of hyperspectral data from the Moon Mineral Mapper (M3) imaging spectrometer. The method uses a convolutional variational autoencoder to reduce the dimensionality of the spectral data and extract features of the spectra. Then, a k-means algorithm is applied to cluster the latent variables into five distinct groups, corresponding to dominant spectral features, which are related to the mineral composition of the Moon's surface. The resulting global spectral cluster map shows the distribution of the five clusters on the Moon, which consist of a mixture of, among others, plagioclase, pyroxene, olivine, and Fe-bearing minerals across the Moon's surface. The clusters are compared to the mineral maps from the Kaguya mission, which showed that the locations of the clusters overlap with the locations of high wt% of minerals such as plagioclase, clinopyroxene, and olivine. The paper demonstrates the usefulness of unbiased unsupervised learning for lunar mineral exploration and provides a comprehensive analysis of lunar mineralogy.

• The paper presents an unsupervised machine learning approach using CVAE and k-means clustering on Moon Mineral Mapper data to analyze and map lunar surface mineralogy.
• The analysis identified five key spectral clusters that map to distinct lunar surface regions and mineralogical features across the Moon.
• This unsupervised clustering method provides an effective, unbiased approach for mapping mineralogical heterogeneity on extraterrestrial surfaces for planetary exploration and resource assessment.

The paper "Insights into Lunar Mineralogy: An Unsupervised Approach for Clustering of the Moon Mineral Mapper (M3) Spectral Data" investigates the lunar surface's mineral compositions using unsupervised machine learning techniques applied to hyperspectral data. The data, obtained from the Moon Mineral Mapper (M3) imaging spectrometer on Chandrayaan-1, is analyzed to establish spectral clustering and mapping of lunar surface materials.

Methodology

1. Data Acquisition and Preprocessing:
   • The paper utilizes M3 spectrometer data ranging from 0.43 μm to 2.5 μm, excluding thermal correction required bands, to ensure high-quality hyperspectral information.
   • Data is selected based on angular and reflectance criteria to improve signal quality and mitigate observational distortions.
2. Dimensionality Reduction:
   • A convolutional variational autoencoder (CVAE) is employed to compress the high-dimensional spectral data into low-dimensional latent variables. This structure facilitates the extraction of meaningful spectral features while preserving essential compositional information.
3. Clustering:
   • The latent variable space is subjected to k-means clustering to delineate five distinct spectral groups, corresponding to various mineral compositions on the Moon's surface. Silhouette and Davies-Bouldin metrics are used to optimize the number of clusters.

Results

• Global Moon Spectral Cluster Map:
  • The analysis identifies five principal spectral clusters on the lunar surface. These clusters are spatially mapped to highlight regions with dominant mineralogical features. Notably, specific clusters are predominant in lunar maria, polar regions, and distinctive terranes like Procellarum KREEP Terrane (PKT) and South Pole-Aitken Terrane (SPAT).
• Spectral Analysis:
  • The reflectance spectra for clusters reveal variations in absolute absorption depths at characteristic wavelengths, such as Band I (~1 μm) and Band II (~2 μm).
  • Observations show discrepancies among clusters, such as stronger Band I absorption typical of pyroxenes (orthopyroxene and clinopyroxene) and olivines, and weaker absorptions related to plagioclase's influence.
• Comparative Study:
  • The spectral clusters are compared with established mineral maps from the Kaguya mission, evidencing overlaps with mineral-rich regions, such as plagioclase and pyroxene distributions, affirming the clustering method's validity.

Inferences and Implications

• The paper demonstrates that unsupervised clustering, without prior mineralogical assumptions, effectively discerns structural patterns in multispectral data.
• Spectral clusters reflect lunar mineral assemblages and geologic heterogeneity, corroborating known distributions of anorthosites, basalts, and mixed igneous compositions.
• This approach offers a methodology for unbiased mineralogical surveys on extraterrestrial surfaces, emphasizing its utility in planetary exploration and potential resource assessment for in-situ utilization.

Conclusion

The paper underscores the efficacy of convolutional variational autoencoders in conjunction with k-means clustering for lunar mineral analysis, providing a nuanced understanding of lunar surface mineralogy. This methodological framework facilitates effective data-driven mineralogical mapping, enhancing our exploration and geological understanding of the Moon.

The paper concludes with the prospective integration of multi-instrumental datasets and enhanced spatial information in future work to refine the clustering outcomes and mineralogical interpretations.