基于不同颜色模型定量预测永川秀芽在制品含水率

为定量预测永川秀芽在制品的含水率，基于不同颜色模型探究在制品的色泽变化，并结合偏最小二乘（partial least square，PLS）法建立含水率的定量预测模型。结果表明：在永川秀芽初制过程中，在制品的红绿度、蓝色通道均值增高，含水率和亮度、黄蓝度、红色通道均值、绿色通道均值、色调均值等15 个颜色模型分量降低，即色泽表现为变暗、变黄；通过热图与聚类分析，可将在制品分为2 个大类、4 个亚类，且理条工序对在制品含水率、色泽的影响最为显著；利用17 个颜色模型分量和PLS方法建立了含水率的定量预测模型，以校正集相关系数（Rc）、交互验证均方根误差（root-mean-square error of cross-validation，RMSECV）、预测集相关系数（Rp）、预测均方根误差（root-mean-square error of prediction，RMSEP）、相对分析误差（relative percent deviation，RPD）为评价指标。模型的Rc、Rp、RMSECV、RMSEP分别为0.979、0.980、0.044 7、0.044 3。RMSECV、RMSEP的差值仅为0.000 4，且RPD达到5.04，表明模型具有极好的预测能力和泛化能力，为实现永川秀芽在制品含水率的在线监测提供了一种新方法。

Quantitative Prediction of the Moisture Content in Work-In-Process Yongchuan Xiuya Tea Based on Different Color Models

A quantitative prediction model for the moisture content of work-in-process Yongchuan Xiuya tea was established using partial least squares regression (PLSR) based on its color changes as evaluated using different color models. The results showed that during the initial production process, the red-green and mean blue channel value increased, while the moisture content and 15 other color model components such as lightness, yellow-blue, mean red channel value, mean green channel value and mean hue value decreased, suggesting that the color became darker and yellower. Through heatmap and cluster analysis, the samples were divided into two categories and four sub-categories, and the carding process had the most significant impact on the moisture and color of the products. Based on the 17 color model components, the predictive model was established, and its performance was evaluated by considering correlation coefficient of calibration set (Rc), root-mean-square error of cross-validation (RMSECV), correlation coefficient of predication set (Rp), root-mean-square error of prediction (RMSEP) and relative percent deviation (RPD). The values of Rc, Rp, RMSECV and RMSEP were 0.979, 0.980, 0.044 7, and 0.044 3, respectively. The difference between RMSECV and RMSEP was merely 0.000 4, and the RPD value was 5.04, indicating that the model had excellent prediction capacity and generalization capacity and could provide a new method for the online monitoring of the moisture content in work-in-process Yongchuan Xiuya tea.