Sun Chen, Xu Weijie, Tan Yongqi, Zhang Yuqing, Yue Zengqi, Zou Long, Shabbir Sahar, Wu Mengting, Chen Fengye, Yu Jin
School of Physics and Astronomy, Shanghai Jiao Tong University, Shanghai, 200240, People's Republic of China.
Sci Rep. 2021 Nov 1;11(1):21379. doi: 10.1038/s41598-021-00647-2.
With the ChemCam instrument, laser-induced breakdown spectroscopy (LIBS) has successively contributed to Mars exploration by determining the elemental compositions of soils, crusts, and rocks. The American Perseverance rover and the Chinese Zhurong rover respectively landed on Mars on February 18 and May 15, 2021, further increase the number of LIBS instruments on Mars. Such an unprecedented situation requires a reinforced research effort on the methods of LIBS spectral data analysis. Although the matrix effects correspond to a general issue in LIBS, they become accentuated in the case of rock analysis for Mars exploration, because of the large variation of rock compositions leading to the chemical matrix effect, and the difference in surface physical properties between laboratory standards (in pressed powder pellet, glass or ceramic) used to establish calibration models and natural rocks encountered on Mars, leading to the physical matrix effect. The chemical matrix effect has been tackled in the ChemCam project with large sets of laboratory standards offering a good representation of various compositions of Mars rocks. The present work more specifically deals with the physical matrix effect which is still lacking a satisfactory solution. The approach consists in introducing transfer learning in LIBS data treatment. For the specific application of total alkali-silica (TAS) classification of rocks (either with a polished surface or in the raw state), the results show a significant improvement in the ability to predict of pellet-based models when trained together with suitable information from rocks in a procedure of transfer learning. The correct TAS classification rate increases from 25% for polished rocks and 33.3% for raw rocks with a machine learning model, to 83.3% with a transfer learning model for both types of rock samples.
通过ChemCam仪器，激光诱导击穿光谱技术（LIBS）通过确定土壤、地壳和岩石的元素组成，相继为火星探测做出了贡献。美国的“毅力号”火星车和中国的“祝融号”火星车分别于2021年2月18日和5月15日登陆火星，进一步增加了火星上LIBS仪器的数量。这种前所未有的情况需要加强对LIBS光谱数据分析方法的研究力度。尽管基体效应是LIBS中的一个普遍问题，但在火星探测的岩石分析中，由于岩石成分变化大导致化学基体效应，以及用于建立校准模型的实验室标准样品（压制粉末颗粒、玻璃或陶瓷）与火星上天然岩石表面物理性质的差异导致物理基体效应，基体效应变得更加突出。在ChemCam项目中，通过大量能很好代表火星岩石各种成分的实验室标准样品解决了化学基体效应问题。目前的工作更具体地处理仍然缺乏令人满意解决方案的物理基体效应。该方法包括在LIBS数据处理中引入迁移学习。对于岩石的全碱-二氧化硅（TAS）分类这一具体应用（无论是抛光表面还是原始状态），结果表明，在迁移学习过程中，当与来自岩石的合适信息一起训练时，基于颗粒的模型的预测能力有显著提高。对于两种类型的岩石样品，正确的TAS分类率从使用机器学习模型时抛光岩石的25%和原始岩石的33.3%，提高到使用迁移学习模型时的83.3%。
