英红九号发酵叶中茶褐素近红外定量模型的优化与验证

Optimization and Verification of a Near Infrared Quantitative Model for the Theabrownin in Yinghong No.9 Fermented Leaves

摘要:

为提高近红外光谱分析方法快速实现对测定红茶在制品中的茶褐素的定量模型精度无损、快速检测，该试验利用近红外光谱技术对以英红九号发酵叶中茶褐素进行采集、提取和分析的检测为例，对其近红外定量检测模型的构建与优化进行了研究。首先，采用规范化处理（Normalize）、基线校正（Baseline）、S-G一阶导数（Savitzky-Golay，1st S-G）、S-G二阶导数（2nd S-G）、标准正态变量变换（Standard Normal Variate Transform，SNV）五种预处理方法对原始光谱进行预处理分析。然后，采用效果最好的一阶导数预处理方法进行波长特征提取，分别使用间隔偏最小二乘算法（Interval Partial Least Square，iPLS）、竞争自适应加权算法（Competitive Adaptive Reweighted Sampling，CARS）、变量迭代空间收缩方法（the Variable Iterative Space Shrinkage Approach，VISSA）提取波长特征。最后，使用偏最小二乘回归（Partial Least Square，PLS）预测模型进行回归建模。研究结果表明：使用一阶导数进行预处理、同时使用CARS方法建立的1st-CARS-PLS模型效果特征更显著，特征值数量为53个。研究表明，该试验采用的模型方法能够快速、无损地检测英红九号发酵叶中的茶褐素含量。

Abstract:

In order to improve the near-infrared spectroscopy analysis method for realizing quickly the use of the quantitative model for non-destructive and rapid detection of the theabrownin in black tea products, this research used near-infrared spectroscopy to collect, extract and analyze the theabrownin in the fermented leaves of Yinghong No.9 as the example. The construction and optimization of the near-infrared quantitative detection model are performed. Firstly, the original spectra were preprocessed and analyzed by five preprocessing methods: Normalization, baseline correction, S-G first derivative (Savitzky-Golay, 1st S-G), S-G second derivative (2nd S-G) and standard normal variable transform (SNV). Then, the best 1st S-G preprocessing method was used to extract the wavelength features, using the interval partial least squares algorithm (iPLS), competitive adaptive weighting algorithm (CARS) and the variable iterative space shrinkage approach (VISSA), respectively. Finally, the partial least squares regression (PLS) prediction model was used for regression modeling. The results show that the 1st-CARS-PLS model established by using the first-order derivative for preprocessing and the CARS method has more significant effect characteristics, with the number of eigenvalues being 53. The research shows that the model method used in this experiment can rapidly and non-destructively detect the theabrownin content in the fermented leaves of Yinghong No.9.