应用近红外光谱技术结合化学计量学建立花椒代表性成分的定量分析模型.采用紫外可见分光光度法测定不同批次花椒总酰胺和总黄酮含量,并测定挥发油含量.采集50批次花椒样品的近红外光谱,应用 Kennard-Stone 算法划分样本集.进一步采用偏最小二乘回归(partial least squares regression,PLSR)和支持向量机(support vector machine,SVM)建立3个指标的含量预测模型,并比较各模型的性能.不同批次花椒样品总酰胺、总黄酮和挥发油含量分别为10.40％～29.09％、10.33％～24.73％、2.72％～8.04％.近红外光谱分别经MSC、SG平滑、SG平滑+MSC预处理后,应用SVM构建的花椒总酰胺、总黄酮和挥发油定量模型准确度较PLSR高,校正集决定系数(RC2)分别为0.818,0.655,0.927,预则集决定系数(RP2)分别为0.898,0.856,0.916.文章所建立的近红外光谱结合PLSR和SVM定量测定模型可以实现花椒类调味品的品质快速评价.
Study on Rapid Evaluation of Zanthoxylum bungeanum Quality Based on Near Infrared Spectroscopy Combined with Chemometrics
A quantitative analysis model for the representative components of Z.bungeanum is established using near infrared spectroscopy combined with chemometrics.The content of total amides and total flavonoids of different batches of Z.bungeanum is determined by UV spectrophotometry,and the content of volatile oil is determined.The near infrared spectra of 50 batches of Z.bungeanum samples are collected,and Kennard-Stone algorithm is used to divide the sample set.Furthermore,partial least squares regression(PLSR)and support vector machine(SVM)are applied to establish the content prediction models of the three indexes,and the performance of each model is compared.The content of total amides,total flavonoids and volatile oils in different batches of Z.bungeanum samples is 10.40％～29.09％,10.33％～24.73％,2.72％～8.04％respectively.After the pretreatment of near infrared spectroscopy with MSC,SG smoothing,SG smoothing+MSC respectively,the accuracy of the quantitative models of total amides,total flavonoids and volatile oils of Z.bungeanum established by SVM is higher than that of PLSR.The determination coefficients of calibration set(RC2)are 0.818,0.655,0.927 respectively,and the determination coefficients of predition set(RP2)are 0.898,0.856,0.916 respectively.The quantitative determination model of near infrared spectroscopy combined with PLSR and SVM established in this paper can achieve rapid evaluation of quality of Z.bungeanum seasonings.
Zanthoxylum bungeanumnear infrared spectroscopypartial least squares regressionsupport vector machinevolatile oil
万方数据
维普
