基于MCCV奇异样本筛选和CARS变量选择的蜂蜜pH值和酸度的近红外光谱检测

采用Norris平滑加一阶微分数据预处理，蒙特卡洛交互验证(MCCV)的奇异样本筛选和CARS(competitive adaptive reweighted sampling)变量选择法，用Kennard-Stone(KS)法划分训练集和预测集，偏最小二乘(PLS)回归近红外光谱建模，对蜂蜜pH值和酸度进行定量分析。pH值和酸度校正模型的交互验证决定系数(Rcv2)、交互验证均方差(RMSECV)、预测集决定系数(Rp2)、预测均方差(RMSEP)分别为0.8516和0.8723、0.1214和2.1734、0.8205和0.8250、0.1196和2.4674。结果表明，该方法适于蜂蜜pH值的测定，而不宜用于测定蜂蜜酸度。

Analysis of pH and Acidity of Honey by Near Infrared Spectroscopy Based on MCCV Outlier Detection and CARS Variable Selection

The near infrared spectra of honey samples were calculated by the method of Norris smoothing combined with first derivative. The outliers were detected by Monte Carlo cross validation (MCCV), and the variables were selected by competitive adaptive reweighted sampling (CARS). The samples were divided into calibration set and validation set by Kennard-Stone (KS) algorithm. Partial least squares (PLS) regression was applied to build a quantitative calibration model of pH and acidity. The coefficient of cross-validation of the calibration set (Rcv2) was 0.8516, and the root mean square error of cross-validation (RMSECV) was 0.1214.The coefficient of determination of the validation set (Rp2 ) was 0.8205, and the root mean square error of prediction (RMSEP) was 0.1196 for pH value. For acidity, the Rcv2, RMSECV, Rp2 and RMSEP were 0.8723, 2.1734, 0.8250 and 2.4674, respectively. The finding shows that this method is suitable for quantitative analysis of honey pH, while caution is needed for honey acidity analysis.