论文部分内容阅读
茶叶中茶多酚作为茶叶品质检测中常规检测成分之一,目前的常规检测方法的缺点是费时、费力,成本较高,因此本研究利用近红外光谱分析技术对茶叶中茶多酚含量快速无损检测具有很高的实用价值。为了提高近红外光谱茶多酚预测模型的精度,利用小波消噪预处理茶叶近红外光谱,滤去其中的噪声信息。再用区间偏最小二乘法(iPLS)与遗传算法(GA)相结合的PLS波长筛选法iPLS-GA建立茶多酚的预测模型:用iPLS预测前,先将整个光谱划分为40个子区间,选择交互验证均方根误差RMSECV值低于全光谱区间的第25和34子区间的组合为信息区间,共166个波数点,然后用GA全局优化组合这166个波数点,最终共有18个波数点用于建立茶多酚模型。结果表明,用小波消噪和iPLS-GA建立的茶多酚模型的预测相关系数Rc和校正均方根误差RMSEC分别为0.964 8和2.14;预测相关系数Rp和预测均方根误差RMSEP;分别为0.958 7和2.22。均比其它模型好。建模数据量从3 320个减少到18个,使模型得以简化。
Tea polyphenols tea as one of the routine testing of quality testing components, the current conventional detection methods have the drawbacks of time-consuming, laborious and costly, so the use of near infrared spectroscopy analysis of tea polyphenols quickly and without damage Detection has a high practical value. In order to improve the accuracy of prediction model of tea polyphenol by near infrared spectroscopy, the near-infrared spectrum of tea was pretreated by wavelet denoising and the noise information was filtered out. The prediction model of tea polyphenols was established by using PLS wavelength screening method iPLS-GA combined with interval partial least squares (iPLS) and genetic algorithm (GA): Before iPLS prediction, the whole spectrum was divided into 40 sub-intervals, The root mean square error RMSECV value of the cross-validation is lower than the combination of the 25th and 34th sub-intervals of the full-spectrum interval as the information interval with a total of 166 wavenumber points. Then, these 166 wavenumber points are combined by GA global optimization. Finally, there are 18 wavenumbers Used to establish a polyphenols model. The results showed that the predicted correlation coefficient (Rc) and root mean square error (RMSEC) of the tea polyphenols model established by wavelet denoising and iPLS-GA were 0.964 8 and 2.14, respectively; the prediction correlation coefficient Rp and the prediction root mean square error RMSEP were 0.958 7 and 2.22. All better than other models. The number of modeling data was reduced from 3 320 to 18, making the model easier.