经19种光谱预处理筛选建立3种连翘叶茶近红外定性定量模型

Establishment of the near infrared qualitative and quantitative models of 3 kinds of Forsythia suspense leaves tea by 19 kinds of spectral pretreatment and screening

中文关键词:  连翘叶茶  近红外光谱技术  总黄酮  总多酚  模型

英文关键词:Forsythia suspensa leaves tea  near infrared spectroscopy  total flavonoids  total polyphenols  models

中文摘要:

      目的 建立一种能快速判定连翘叶绿茶、红茶及黑茶类别并预测其总黄酮、总多酚含量的分析模型。方法 以3种连翘叶茶的近红外(near infrared, NIR)光谱和总黄酮、总多酚含量为研究对象, 对19种NIR光谱预处理方法进行比较, 在最佳预处理方法下, 用判别偏最小二乘法(discriminant partial least-squares regression, DPLS)、马氏距离法(Mahalanobis distance, MD)对连翘叶茶建立定性模型, 用偏最小二乘法(partial least squares, PLS)对连翘叶茶建立定量模型, 通过内部交互验证和外部验证, 筛选出最佳的定性定量模型。结果 DPLS定性模型可100%识别3种连翘叶茶类别; MD定性模型可100%实现3种连翘叶茶两两间的类别判定; PLS建立的总黄酮定量模型中绿茶、红茶及黑茶的预处理方法分别为矢量归一化、矢量归一化、标准正态变量变换(standard normal variate, SNV); 预测含量与实测含量间回归方程的相关系数分别为0.9653、0.9764、0.9952; 平均相对误差分别为0.26%、7.70%、0.63%; 总多酚定量模型中绿茶、红茶及黑茶的预处理方法分别为卷积(savitzky-golay, S-G)平滑+矢量归一化+多元散射校正、SNV、S-G平滑+矢量归一化+SNV; 预测含量与实测含量间回归方程的相关系数分别为0.7902、0.9614、0.7568; 平均相对误差分别为6.29%、2.55%、5.67%。结论 通过对19种光谱预处理方法进行筛选, 所建立的3种连翘叶茶定性定量模型稳定可靠, 预测精度更高, 可用于连翘叶茶样品质量品质的快速判别与检测。

英文摘要:

      Objective To establish an analytical model which can quickly distinguish the types of Forsythia suspense leaves green tea, black tea and dark tea and predict the content of total flavonoids and total polyphenols. Methods The near infrared (NIR) spectra of 3 kinds of F. suspense leaves tea and the content of total flavonoids and total polyphenols were taken as the research objects. The pretreatment methods of 19 kinds of NIR spectra were compared. Under the optimal pretreatment method, the discriminant partial least-squares regression (DPLS) method and Mahalanobis distance (MD) method were used to establish the qualitative model of F. suspense leaves tea. Partial least squares (PLS) method was used to establish a quantitative model of F. suspense leaves tea. The best qualitative and quantitative models were screened by internal interactive verification and external verification. Results The accuracy of DPLS qualitative model in identifying 3 kinds of F. suspense leaves tea was 100%. MD qualitative model could achieve 100% pairwise classification of 3 kinds of F. suspense leaves tea. In the quantitative model of total flavonoids established by PLS method, the pretreatment methods of green tea, black tea and dark tea were vector normalization, vector normalization and standard normal variate (SNV) respectively. The correlation coefficients of regression equations between predicted content and measured content were 0.9653, 0.9764 and 0.9952, respectively. The average relative errors were 0.26%, 7.70% and 0.63%, respectively. In the quantitative model of total polyphenols, the pretreatment methods of green tea, black tea and dark tea were savitzky-golay (S-G) smoothing+vector normalization+multiplicative scatter correction, SNV, S-G smoothing+vector normalization+SNV, respectively. The correlation coefficients of regression equations between predicted content and measured content were 0.7902, 0.9614 and 0.7568, respectively. The average relative errors were 6.29%, 2.55% and 5.67%, respectively. Conclusion Through screening 19 kinds of spectral pretreatment methods, the established qualitative and quantitative models of 3 kinds of F. suspense leaves tea are stable and reliable, and the prediction accuracy is higher, which can be used for the rapid identification and detection of the quality of F. suspense leaves tea samples.