基于电子鼻的茶油掺伪定性和定量检测

Qualitative and quantitative detection of camellia oil adulteration based on electronic nose

目的:实现对茶油中掺入菜籽油、大豆油、玉米油的掺伪检测。方法:采用电子鼻检测平台对茶油中分别掺入不同比例菜籽油、大豆油、玉米油进行掺伪检测,运用线性判别分析(LDA)和支持向量机(SVM)进行茶油掺伪定性鉴别分析,并使用多层感知器(MLP)和偏最小二乘回归(PLSR)建立茶油掺伪定量预测模型。结果:最佳输入参数下的SVM对茶油掺伪鉴别准确率高于LDA,其平均精确率、平均召回率、平均F-分数分别为94.85%,96.11%,95.34%,比LDA的提高了5.17%,4.44%,5.29%;MLP对茶油掺伪比例预测结果优于PLSR,对于掺入菜籽油、大豆油、玉米油的茶油,MLP预测的决定系数分别为0.98,0.99,0.98,均方根误差分别为4.02%,1.45%,3.74%。结论:基于电子鼻平台建立的SVM茶油掺伪鉴别模型和MLP茶油掺伪比例预测模型可有效实现茶油的鉴伪。

Objective:This study aims to realize the adulteration detection of camellia oil mixed with rapeseed oil,soybean oil and corn oil.Methods:The electronic nose detection platform was used for the adulteration detection of camellia oil mixed with different proportion of rapeseed oil,soybean oil and corn oil.Firstly,the linear discriminant analysis(LDA)and support vector machine(SVM)were used for the qualitative identification of camellia oil adulteration.Then multilayer perceptron(MLP)and partial least squares regression(PLSR)were used to establish quantitative prediction models for camellia oil adulteration.Results:The accuracy of SVM for qualitative authentication was higher than that of LDA,and the average precision rate,average recall rate and average F1-score were 94.85%,96.11%and 95.34%,respectively,which were 5.17%,4.44%and 5.29%higher than those of LDA.For quantitative prediction,MLP outperforms PLSR.In particular,the determination coefficients of MLP were 0.98,0.99 and 0.98,and the root mean square errors were 4.02%,1.45%and 3.74%,respectively,for camellia oil mixed with rapeseed oil,soybean oil and corn oil.Conclusion:The SVM-based identification model and MLP-based prediction model can effectively detect oil adulteration by using electronic nose platform.