Micro-CT characterization of lunar regolith using machine learning-based segmentation

On 17 December 2020, China's Chang'e-5 mission returned about 1.73 kg of lunar regolith from one of the youngest basalt units in northern Oceanus Procellarum. The mineralogy of the lunar surface regolith provides a wealth of information on its geological history, and the characterization of lunar regolith at high spatial resolution has been a significant goal of lunar exploration. In this study, we combine high-resolution micro-CT imaging and a state-of-the-art machine learning-based image processing approach to assess morphological and physical properties of the lunar regolith sample returned by Chang'e-5 mission. The lunar regolith sample was scanned by an X-ray micro-computed tomography (micro-CT) with a spatial resolution of 2.48 μm. A pixel-wise random forest classifier was employed to segment the volume data into regolith particles considering multiple features including intensity, edge and texture. On the basis of segmented images, the particle size distribution of the lunar regolith sample was extracted. The average density of the sample is estimated to be around 1582 kg/m³ based on a calibration of the relationship between the image intensity and material density. Particles with extremely high-density mineral phases (around 4500 kg/m³) in the sample are considered rich in metal elements such as iron and titanium. In addition, we were able to extract particles with distinguished features such as isolated pores, which implies the possible melting and solidification process related to past meteorite impacts. This study provides a workflow for micro-CT imaging-based analysis of lunar regolith.