流过式介质阻挡放电电离质谱法结合随机森林模型鉴别名贵木材种类  

Identification of Valuable Wood Species Using Flow-Through Dielectric Barrier Discharge Ionization Mass Spectrometry Combined with Random Forest Model

作　　者：尚宇瀚 孟宪双[1] 吕悦广 马强[1] SHANG Yu-han;MENG Xian-shuang;LYU Yue-guang;MA Qiang(Chinese Academy of Inspection and Quarantine,Beijing 100176,China)

机构地区：[1]中国检验检疫科学研究院,北京100176

出　　处：《质谱学报》2024年第4期500-509,I0003,共11页Journal of Chinese Mass Spectrometry Society

基　　金：国家市场监督管理总局科技计划项目(2021MK163);中央级公益性科研院所基本科研业务费专项资金项目(2022JK17)。

摘　　要：本研究开发了电烙铁灼烧-流过式介质阻挡放电电离质谱法快速准确鉴别名贵木材制品的种类,并根据质谱指纹图谱数据,建立了基于随机森林算法的预测模型。结果表明,电烙铁灼烧-流过式介质阻挡放电电离质谱法无需样品前处理、操作简便,单次分析用时仅为4~5 s,符合快速分析要求。优化后的随机森林模型经过袋外误判率和十折交叉验证误判率分别为4.76%和4.74%,模型分类准确率大于95%。该方法能够准确区分黄檀属、古夷苏木属和紫檀属木材样品,并成功应用于网售名贵木材制品种类的快速鉴别,可为名贵木材制品的真伪鉴别与品质评价提供科学依据与技术参考。To achieve rapid and accurate identification of valuable wood products,an analytical method was developed by combining electric soldering iron cauterization with soft ionization by chemical reaction in transfer-mass spectrometry(SICRIT-MS).SICRIT is a flow-through dielectric barrier discharge ionization technique pioneered by Zenobi et al.in 2016.The electric soldering iron cauterization-SICRIT-MS method requires no sample pretreatment,easy operation and a single analysis in less than 5 s,meeting the demands of rapid analysis.Operating parameters for the soldering iron and SICRIT ion source were optimized to achieve maximum total ion current intensity under soldering iron temperature of 450℃,ion source AC voltage amplitude of 2000 V,and sample transfer line temperature of 150℃.With the optimized parameters,the SICRIT-MS method was applied to analyze valuable wood samples,including 29 certified standard wood samples and 6 online-purchased real samples,resulting in a dataset of 210 sets of mass spectral fingerprint data.Based on the mass spectral fingerprint data acquired under positive ion mode,a predictive model was trained using the random forest algorithm.The random forest model underwent optimization for the number of decision trees,max feature algorithm,and feature selection criteria,was evaluated through out-of-bag and 10-fold cross-validation.The results showed the error rates of out-of-bag and 10-fold cross-validation are 4.76%and 4.74%,respectively.The established random forest model can accurately distinguish wood samples from the genera Dalbergia,Guibourtia,and Pterocarpus with a classification accuracy of larger than 95%.The importance of features in distinguishing the three wood genera was investigated through binary classification modeling,revealing features 269.1,270.1,255.1,159.0,182.1,102.1 and 83.1 as crucial in classification.These features may correspond to characteristic compounds in different wood species or differences in the content of the same compound across species.The predictive mod

关 键 词：流过式介质阻挡放电电离质谱 随机森林模型 名贵木材 种类鉴别 

分 类 号：O657.63[理学—分析化学]