基于矿物元素与同位素联合分析的吉林大米产地判别的研究

Study on rice origin discrimination in Jilin province based on combined analysis of mineral elements and isotopes

为实现更加准确、可靠的吉林省大米产地判别分析,利用电感耦合等离子质谱仪(ICP-MS)和同位素质谱仪(IRMS)测定样品中的12种矿物元素和δ^(13)C、δ^(15)N同位素含量,结合线性判别分析(LDA)和支持向量机(SVM)模型探究同位素数据的加入对产地确证结果的影响;利用Xgboost模型和交叉验证,筛选判别因子最优合集。结果表明,在12种矿物元素数据联合δ^(13)C、δ^(15)N数据分别代入模型后,线性判别分析结果准确率和回代正确率从79%、73.3%提升到80.1%、74.0%;支持向量机模型结果的准确度和精确度从83.019%、82.589%提高到90.245%、87.549%。得到吉林省大米最优判别因子合集为{Si、Mn、Ca、Mg、Al、Na、Cr、P、δ^(13)C},为快速准确实现吉林省大米产地溯源提供理论参考。

In order to achieve more accurate and reliable discriminant analysis of rice producing area in Jilin province,the contents of 12 mineral elements and δ^(13)C and δ^(15)N isotopes in samples were determined by inductively coupled plasma mass spectrometry(ICP-MS)and isotope mass spectrometry(IRMS).Linear discriminant analysis(LDA)and support vector machine(SVM)models were used to explore the influence of isotope data on the results of origin confirmation.Xgboost model and cross validation were used to screen the optimal set of discriminant factors.The results showed that the accuracy and iteration accuracy oflinear discriminant analysis were increased from 79%and 73.3%to 80.1%and 74.0%after the data of 12 mineral elements were combined with δ^(13)C and δ^(15)N,respectively.The accuracy and precision of support vector machine model were improved from 83.019%and 82.589%to 90.245%and 87.549%.The optimal discriminant factor set of rice in Jilin province was{Si,Mn,Ca,Mg,Al,Na,Cr,P,δ^(13)C},which provides theoretical reference for fast and accurate rice origin tracing in Jilin province.