法布里干涉近红外光谱仪定量测定大豆、玉米主要成分

Research on the quantitative determination of main ingredients in soybean and maize by Fabry-Perot interferometer NIR spectrometer

采用Axsun近红外光谱仪采集40个大豆样品、46个玉米样品的近红外光谱,采用PLS算法分别建立大豆粗蛋白质、脂肪含量的定量校正模型和玉米粗蛋白质、脂肪、淀粉含量的定量校正模型。结果表明,大豆粗蛋白质、脂肪定量校正模型的模型维数分别为6、5,决定系数(R^2)分别为97.94%、93.66%,校正均方根误差(RMSEC)分别为0.40、0.36.交互验证均方根误差(RMSECV)分别为0.49、0.44,相对预测性能(RPD)分别为5.18、3.15;玉米粗蛋白质、脂肪和淀粉含量的定量校正模型的模型维数分别为4、4和6,R^2分别为90.15%、95.22%和87.81%,RMSEC分别为0.25、0.12和0.53.RMSECV分别为0.37、0.15和0.72,RPD分别为2.57、3.57和2.42。F检验表明,上述校正模型的预测值与化学值具有极显著的相关关系。研究结果表明,该仪器可以用于大豆、玉米主要成分定量现场快速测定。

The NIR spectra of 40 soybean samples and 46 maize samples were collected by Axsun portable NtR spectrometer. The quantitative calibration models of crude protein and fat of soybean and the quantitative calibration models of crude protein, fat and starch of maize were developed by PLS algorithm respectively. The result indicated that for the quantitative calibration models of soybean crude protein and soybean fat. the model dimensions are 6 and 5 respectively, the determination coefficients （R2） are 97.94% and 93.66% respectively, the root mean square errors of calibration （RMSEC） are 0.40 and 0,36 respectively, the root mean square errors of cross validation （RMSECV） are 0.49 and 0.44 respectively, the relative performance deviations （RPD） are 5.18 and 3.15 respectively; for the quantitative calibration models of maize crude protein, maize fat and maize starch, the model dimensions are 4, 4 and 6 respectively, R2 are 90.15%. 95.22% and 87.81% respectively, RMSEC are 0.25, 0.12 and 0.53 respectively, RMSECV are 0.37, 0,15, 0.72 respectively, RPD are 2.57. 3,57 and 2.42 respectively. The result of F-test indicated that a very remarkable correlation exists between the estimated and specified values of each calibration model mentioned above. This research indicated that the spectrometer can be applied in the on-site rapid quantitative determination of the main ingredients of soybean and maize.