节点文献

基于激光诱导击穿光谱的类火星岩石化学成分反演研究

LIBS Quantitative Analysis of Martian Analogues Library (MAL)

  • 推荐 CAJ下载
  • PDF下载
  • 不支持迅雷等下载工具,请取消加速工具后下载。

【作者】 刘长卿凌宗成

【Author】 LIU Chang-qing;LING Zong-cheng;Shandong Key Laboratory of Optical Astronomy and Solar-Terrestrial Environment, School of Space Science and Technology, Institute of Space Sciences, Shandong University;

【通讯作者】 凌宗成;

【机构】 山东大学空间科学与技术学院,空间科学研究院,山东省光学天文与日地空间环境重点实验室

【摘要】 激光诱导击穿光谱(LIBS)是一种元素发射光谱,可快速获取样品的化学成分信息。天问一号是我国首次火星探测任务,祝融号火星车搭载的火星表面成分探测仪载荷(MarSCoDe),采用LIBS光谱技术获取火星表面化学成分信息。然而,火星表面大气环境变化复杂,岩石和矿物种类多样,将影响LIBS等离子体的形成和演化过程,进而影响基于MarSCoDe-LIBS光谱数据反演物质成分的准确性和可靠性。针对MarSCoDe-LIBS实测数据的反演需求,该研究采用自主构建的LIBS光谱数据库探索不同算法对LIBS光谱数据反演的性能,为构建适用于MarSCoDe的模型提供支撑。选用351种类火星地质标样,在模拟火星环境下构建LIBS光谱数据库,基于数据库和标样的化学成分信息,采用机器学习、集成学习和深度学习等9种不同的算法,构建了SiO2、 TiO2、 Al2O3、 Fe2O3T、 MgO、 CaO、 Na2O、 K2O等主量元素反演模型。建模采用交叉验证的方式进行模型调参,并基于测试集误差评估模型性能。研究发现,除普通最小二乘法外,其他8种模型训练集和测试集的RMSE接近,说明没有明显的过拟合现象;在所有模型中,多层感知机回归和梯度提升回归模型对主量元素反演可取得最优性能,模型RMSEP与ChemCam、 SuperCam等国际LIBS数据库公布的反演模型误差相当,证明本研究构建的模型效果较好,可基于未知样品的LIBS光谱准确反演其化学成分信息。该研究对不同模型的调参过程及反演结果评价,为构建适用于天问一号MarSCoDe-LIBS实测数据解译的模型提供了重要的算法参考。

【Abstract】 Laser-induced breakdown spectroscopy(LIBS) is a valuable technique for elemental analysis from a laser-induced plasma. The Zhurong rover in the Tianwen-1 Mars exploration mission carries a payload named Mars Surface Composition Detector(MarSCoDe), which can obtain geochemical compositions on Mars. However, the interpretation of MarSCoDe-LIBS spectra will be affected by the complex environment and rock types. With an intent to acquire accurate chemical compositions on Mars using MarSCoDe-LIBS spectra, this work evaluates the performance of several algorithms using the independent third-party LIBS spectral library. This work uses 351 Martian Analogues Library(MAL) to build the LIBS spectral library in a simulated Martian environment. Several models are built based on the LIBS spectra and chemical compositions using nine different algorithms, including machine learning, integrated learning, and deep learning, to derive the major elements(SiO2, TiO2, Al2O3, Fe2O3T, MgO, CaO, Na2O, and K2O). The parameters of these models are confirmed using the cross-validation method, and the performance of these models is evaluated using the RMSE values of the test set. The training set and test set for most models have similar RMSE values except for the ordinary least square method, suggesting no obvious over fitting for these models. In addition, the MLP and GBR models perform better for major elements. Moreover, the RMSE values of the models are similar to those of the published models for ChemCam and SuperCam, suggesting these models have a good performance and can acquire accurate chemical compositions of unknown targets based on their LIBS spectra. This work is valuable for building models suitable for interpreting MarSCoDe-LIBS spectra acquired on Mars.

【基金】 国家自然科学基金项目(U1931211,12303067);科工局民用航天预研项目(D020102);山东省自然科学基金项目(ZR2023QD106);中国博士后科学基金项目(2023M732044)资助
  • 【文献出处】 光谱学与光谱分析 ,Spectroscopy and Spectral Analysis , 编辑部邮箱 ,2025年03期
  • 【分类号】TP18;P185.3
  • 【下载频次】132
节点文献中: