论文部分内容阅读
对乳粉中蛋白质和脂肪近红外定量模型的优化进行了研究。结果表明:利用全波段光谱建立的模型,蛋白质和脂肪的模型预测评价参数分别为:校正集预测偏差(RMSECV)为2.837、2.984,内部交互验证标准偏差(RMSEP)为3.406、3.751,交互检验相对偏差(RPD)为2.6、2.5。经过波长优选后,蛋白质和脂肪优选的波数范围分别为(9403.5~7498.1)cm-1、(9403.5~6098)cm-1,蛋白质和脂肪的模型预测评价参数有一定的提高,分别为RMSECV 1.963、2.317,RMSEP 2.396、3.035,RPD 2.9、2.6。通过正交试验对定标方法和光谱预处理方法进行了优化:蛋白质的最佳参数组合为:定标方法为改进偏最小二乘法,多元离散校正,一阶导数处理,导数处理间隔点为8,平滑处理间隔点数为4,二次平滑处理间隔点数为1,目标函数达到98.25%;脂肪的最佳参数组合为:定标方法为偏最小二乘法,散射校正为多元离散校正,导数处理为二阶,导数处理间隔点为1,平滑处理间隔点数为4,二次平滑处理间隔点数为4,目标函数达到95.26%。
The optimization of quantitative models of protein and fat near infrared in milk powder was studied. The results showed that RMSECV and RMSEP were 2.837,2.984 and 3.406,3.751, respectively, for the model predictive model of protein and fat. The results of cross-validation Deviations (RPD) of 2.6, 2.5. The optimized wavenumber ranges of protein and fat were (9403.5 ~ 7498.1) cm-1 and (9403.5 ~ 6098) cm-1 after wavelength optimization, respectively. The prediction parameters of protein and fat model were improved to some extent, RMSECV 1.963, 2.317, RMSEP 2.396, 3.035, RPD 2.9, 2.6. The orthogonal test was used to optimize the calibration method and the spectral pretreatment method. The optimal combination of protein parameters was as follows: calibration method was improved partial least square method, multivariate discrete correction and first derivative treatment, the interval of derivative treatment was 8 , The number of smoothing interval points is 4, the number of secondary smoothing interval points is 1, and the objective function reaches 98.25%. The optimal combination of fat parameters is: the calibration method is partial least squares, the scatter correction is multiple discrete correction, the derivative processing is In the second order, the derivative processing interval is 1, the number of smoothing processing interval is 4, the number of second smoothing processing interval is 4, and the objective function is 95.26%.