Automated Mineral Identification and Rock‐Type Classification of Lunar Mare Basalts Using SEM Images

Journal article

Plain Language Summary: Scientists who study rocks often spend a great deal of time identifying minerals in thin sections using high‐resolution microscopy, such as scanning electron microscopy (SEM). Here, we developed a computer vision–based deep learning model that automatically identifies minerals and classifies rock types in lunar volcanic rock samples using SEM images. Trained on a variety of returned Apollo samples, our model achieved good overall accuracies >80% in mineral identification and >78% in rock type classification. At the same time, our method struggles to reliably identify rare minerals and is influenced by how humans labeled the training images. As a result, expert interpretation is still necessary for detailed mineralogical and petrological analyses. Overall, our system provides a practical and scalable starting point for automated analysis of lunar rocks, enabling much faster mineral identification of existing returned samples, lunar meteorites, and future Moon mission materials.

Authors: Jung, J; Tikoo, SM; Chung, J; Nichols, CIO

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: Wiley
• Journal: Journal of Geophysical Research: Machine Learning and Computation
• Volume: 3
• Issue: 4
• Article number: e2025JH001124
• Publication date: 2026-06-27
• Acceptance date: 2026-05-21
• DOI: 10.1029/2025jh001124
• EISSN: 2993-5210
• ISSN: 2993-5210
• Language: English
• Keywords: SEM images; rock‐type classification; image segmentation; lunar petrology; computer vision; lunar mare basalts