摘要
应用激光拉曼光谱技术结合化学计量学方法实现了三组分食用调和油中菜籽油、花生油和芝麻油的快速定量检测。分别采用标准正态变量变换（SNV）＋去趋势（de-trending）算法和正交信号校正（OSC）算法对600~3 000cm-1波段的原始拉曼光谱进行预处理。建立了基于非线性支持向量机（SVM）和线性偏最小二乘（PLS）回归算法的定量分析模型,并采用19个预测集通过外部交叉验证法对模型进行验证。实验结果显示：对含有菜籽油、花生油和芝麻油的三组分食用调和油,以OSC预处理后建立的线性PLS模型预测效果最好,其验证集决定系数R2p分别为0.990 4,0.965 8,0.977 1,均方根误差（RMSEP）分别为0.018 8,0.037 9,0.026 2。研究结果表明,利用激光拉曼光谱结合化学计量学方法快速定量检测三组分食用调和油中菜籽油、花生油和芝麻油的含量具有可行性,并获得了较高的预测精度。
The rapeseed oil, peanut oil and sesame oil contents in ternary blended edible oil were measured quantitatively by laser Raman spectroscopy combined with the chemometric resolution method. Two kinds of pretreatment methods called Standard Normal Variate（SNV）＋de-trending and Orthogonal Signal Correction（OSC） were used to process the original Raman spectra at 600-3 000 cm^-1 , respectively. The quantification calibration models based on nonlinear Support Vector Machine（SVM） and linear Partial Least Squares （PLS） for the rapeseed oil, peanut oil and sesame oil were established, and nineteen prediction samples were predicted using the external cross-validation to verify the accuracy of these models. The results show that linear PLS model after the OSC spectral processing obtains the best result for ternary blended edible oil containing rapeseed oil, peanut oil and sesame oil. The determination coefficients（R2p）of the prediction are 0. 990 4, 0. 965 8, 0. 977 1 and Root Mean Square Error of Prediction （RMSEP） are 0. 018 8, 0. 037 9, 0. 026 2, respectively. The research results indicate that laser Raman spectroscopy combined with chemometric resolution can be used for the determination of the contents for rapeseed oil, peanut oil and sesame oil in blended edible oil, and can achieve excellent prediction precision. method ternary
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2015年第9期2490-2496,共7页
Optics and Precision Engineering
基金
国家自然科学基金项目资助(No.61178036
No.31160250)
江西省光电检测工程技术研究中心资助项目(No.赣科发财字[2012]155号)
关键词
拉曼光谱
食用调和油
支持向量机
偏最小二乘
定量检测模型
Raman spectroscopy
blended edible oil
Support Vector Machine（SVM）
Partial LeastSquare（PLS）
quantitative determination model
