基于高光谱技术的红茶茶多酚可视化研究

Study on visualization of black tea polyphenols based on hyperspectral technology

中文关键词:  红茶  茶多酚  蒙特卡罗-高斯分布  最小二乘支持向量机  粒子群  可视化

英文关键词:black tea  tea polyphenols  Monte Carlo Gaussian distribution  least squares support vector machine  particle swarm  visualization

中文摘要:

      目的 利用高光谱技术实现对英红九号红茶茶多酚含量的快速无损、可视化检测。方法 采集128个红茶光谱数据并进行光谱预处理后, 引入蒙特卡罗-高斯分布方法寻找异常样本。经两次异常样本剔除, 各模型预测集决定系数r2均有0.2~0.4的大幅提升。为解决大样本模型训练时间长、数据冗余问题, 采用连续投影算法进行波长筛选, 共得到14个能反映红茶茶多酚含量的特征波长, 并比较了最小二乘回归、支持向量机回归、BP神经网路、粒子群优化最小二乘支持向量机回归(particle swarm optimization least squares support vector regression, PSO-LSSVR) 4种模型预测红茶茶多酚含量的精度。最后以最优模型建立茶多酚可视化模型。结果 合理剔除样本并以光谱特征为输入, 结合PSO-LSSVR方法建立的模型效果最佳, 其校正集决定系数为0.921, 预测集决定系数为0.903, 预测精度达到了90%以上, 基本实现了茶多酚含量可视化检测。结论 可视化算法有效地反映了红茶茶多酚分布情况, 适用于茶叶快速无损检测。

英文摘要:

      Objective To realize the rapid, nondestructive and visual detection of the polyphenol content in Yinghong No.9 black tea by hyperspectral technology. Methods After the spectral data of 128 black tea samples were collected and pretreated, the Monte Carlo Gaussian distribution method was introduced to search for abnormal samples. After 2 times of abnormal sample elimination, the determination coefficient r2 of each model prediction set had been greatly improved by 0.2?0.4. In order to solve the problems of long training time and data redundancy of large sample models, continuous projection algorithm was used to screen wavelengths, and 14 characteristic wavelengths that could reflect the polyphenol content of black tea were obtained. The accuracy of 4 kinds of models to predict the polyphenol content of black tea was compared, including least square regression, support vector machine regression, BP neural network, particle swarm optimization least squares support vector machine regression (PSO-LSSVR). Finally, the visual model of tea polyphenols was established with the optimal model. Results The model established by combining PSO-LSSVR method with reasonable sample removal and spectral characteristics as input had the best effect. The determination coefficient of the correction set was 0.921, the determination coefficient of the prediction set was 0.903, and the prediction accuracy reached more than 90%. The visual detection of tea polyphenol content was basically realized. Conclusion The visualization algorithm can effectively reflect the distribution of black tea polyphenols, and is suitable for rapid non-destructive testing of tea.