Automatic Segmentation and Identification of Minerals in Chang’E 5 Lunar Soil Particles

Abstract

<p>In 2020, Chang'e-5 successfully collected samples in the lunar storm ocean region and returned to Earth, completing China's first unmanned lunar sample return mission. A large number of Chang'e-5 lunar samples were prepared into smooth section samples encapsulated in resin for researchers to carry out testing and analysis. Scanning electron microscopy (SEM), as an important tool for rapid analysis of lunar regolith samples, can be used to reveal key information such as sample origin and space weathering. However, due to the large number of lunar regolith particles collected by Chang'e-5, manual identification of SEM samples is not only time-consuming and tedious, but also difficult to achieve efficient repeated verification. To address these challenges, in this study, an automatic identification algorithm for lunar regolith minerals was designed and implemented based on the gray characteristics of backscattered electron images and the element distribution characteristics of energy dispersive spectroscopy. Firstly, the K-means clustering algorithm is used to divide the gray level of the backscattered electron image into regions, and the connected region analysis is used to complete the particle segmentation. Then, the relative element density values of the particles are calculated by combining the mask technique and the energy dispersion spectrum, and these density values are clustered by AGNES, so as to construct an exponential mineral particle segmentation model containing element characteristics. Finally, the mineral particles are classified according to the classification density values of mineral element characteristics, and the automatic identification of six common minerals of lunar regolith, including quartz, plagioclase, pyroxene, olivine, ilmenite and troilite, is realized. The verification results show that the recognition accuracy of the automatic recognition algorithm is more than 97% in the number of particles and more than 95% in the area of particles, showing high performance. Although the algorithm is limited by the detection ability of scanning electron microscope, it has broad application prospects in the rapid analysis of lunar samples, sample screening and large data processing.</p>

2020年，嫦娥五号在月球风暴洋地区成功采集样品并返回地球，完成了中国首次无人月球采样返回任务。大量嫦娥五号月球样品被制备成封装于树脂中的光滑切面光片样品，以便于研究人员开展测试与分析。扫描电子显微镜(SEM)作为快速分析月壤样品的重要工具，可用于揭示样品来源及太空风化等关键信息。然而，由于嫦娥五号采集的月壤颗粒数量庞大，人工鉴定SEM样品不仅耗时繁琐，还难以实现高效的重复验证。为应对这些挑战，本研究基于光片样品背散射电子图的灰度特征和能量色散谱图的元素分布特征，设计并实现了一种月壤矿物自动识别算法。该算法首先利用K-means聚类算法对背散射电子图灰度进行区域划分，并通过连通区域分析完成颗粒分割。随后，结合掩膜技术与能量色散谱图计算颗粒的相对元素密度值，并对这些密度值进行AGNES聚类，从而构建包含元素特征的指数级矿物颗粒分割模型。最终，依据矿物元素特性的分级密度值对矿物颗粒进行分类，实现了对石英、斜长石、辉石、橄榄石、钛铁矿和陨硫铁六类月壤常见矿物的自动识别。验证结果表明，该自动化识别算法在颗粒数量的识别准确率超过97%，在颗粒面积的准确率超过95%，展现了较高的性能。尽管算法受限于扫描电子显微镜的探测能力，但其在月球样品的快速分析、样品筛选以及大数据处理等方面，具有广阔的应用前景。