小波变换结合连续投影算法优化茶叶中咖啡碱的近红外分析模型

Wavelet Transform Combined with SPA to Optimize the Near-infrared Analysis Model of Caffeine in Tea

本文着重探讨小波变换及其它光谱预处理方法对连续投影算法(SPA)波长筛选优化及建模效果的影响。以158个不同茶叶样本作为研究对象,将各种预处理方法单独或组合后与SPA结合使用,并通过偏最小二乘法(PLS)建立咖啡碱定量模型。其中一阶微分(WT-1stD)-SPA组合建立的模型最佳,预测相关系数达到0.9481,均方根误差达到0.3053,验证集相对分析误差达到3.1959,建模变量由1038减小为10,其挑选出的波长数10和通过交叉验证确定的最佳PLS成分数7也比较接近,并包含在茶叶咖啡碱主要吸收谱带范围内。结果表明,小波变换结合WT-1stD方法在消除光谱部分散射误差和高频噪声的同时,能有效提高茶叶光谱的分辨率,有助于SPA算法筛选出更少、代表性和独立性更优的特征波长组合,并极大地改善了模型的精度,为茶叶中咖啡碱的近红外分析建模提供了一种快速、简便的方法。

In this paper,the influence of wavelet transform and other spectral preprocessing methods on wavelength selection and model optimization of successive projection algorithm(SPA)is focused on for the investigation.Before building NIR analytical model for caffeine by partial least squares(PLS),158 tea samples were collected independently and their spectra were pretreated by various methods individually or in combination with SPA for wavelength selection.Among the established models,the combined method of WT-1stD-SPA-PLSR model gave the best results with the correlation coefficient(R)of 0.9481 and the predicted root mean square error of 0.3053.Only 10 wavelengths from 1038 were introduced in the model which covered the main absorption band of the tea caffeine as well as the number of which were close to the PLS components 7 by cross-validation.The results showed that the WT-1stD method could improve the resolution of the spectrum effectively while eliminating part of the spectral scattering error and high-frequency noise.The advantage is helpful to the SPA algorithm’s screening out fewer optimal characteristic wavelengths combinations with more representativeness and relative independence as well as great improvement of the prediction accuracy.The proposed method provides a rapid and simple modeling strategy for the near infrared analysis of caffeine in tea.