论文部分内容阅读
为快速无损检测掺伪羊乳粉中牛乳清粉的含量,采用近红外光谱法(NIR)结合ν-支持向量回归(ν-SVR)检测197个掺伪牛乳清粉的羊乳粉,并与偏最小二乘法(PLS)对比,光谱经平滑、标准变量变换及导数预处理。随机选取132个样本作为校正集建立模型,其余65个样本作为测试集,评估模型性能。结果显示:校正集最优模型为标准变量变换、Savitzky-Golay平滑和二阶导预处理ν-SVR模型,其交叉验证误差均方根(RMSECV)为0.586,交叉验证相关系数(RCV)达到0.9947。预测集验证结果ν-SVR法比PLS法更优,ν-SVR模型预测值与真实值R达到0.9958,RMSEP为0.526。试验结果表明NIR结合ν-SVR模型可用于掺伪羊乳粉中牛乳清粉的快速无损检测,而且操作简便,可为实际应用提供参考。
In order to rapidly and non-destructively detect the content of bovine whey powder in adulterated sheep’s milk powder, 197 sheep’s milk powder with adulterated bovine whey powder was detected by near infrared spectroscopy (NIR) and ν-support vector regression (ν-SVR) Partial Least Squares (PLS) comparison, spectral smoothing, standard variable transformation and derivative preprocessing. Randomly selected 132 samples as a calibration set to establish the model, the remaining 65 samples as a test set to assess the performance of the model. The results show that the RMSEV of cross-validation error is 0.586, and the cross-validation correlation coefficient (RCV) is 0.9947 for the calibration set optimal model as standard variable transformation, Savitzky-Golay smoothing and second- . The results of the predictive set ν-SVR method are better than the PLS method. The predicted and true values of ν-SVR model are 0.9958 and RMSEP is 0.526. The experimental results show that the NIR combined with ν-SVR model can be used for rapid and non-destructive detection of bovine whey powder in adulterated sheep’s milk powder, and its operation is simple and convenient, which can provide reference for practical application.