摘要
本研究旨在比较以检测值为定标参考值和以理论值为定标参考值2类近红外光谱(NIRS)定量分析模型的预测精准度差异。试验制备了225个维生素预混料样品,利用傅里叶变换近红外光谱仪采集样品在10000~4000 cm^(-1)内的NIRS,基于同一维生素预混料定标集,分别以检测值、理论值为定标参考值,借助偏最小二乘方法,筛选最优光谱预处理方式,构建维生素E的NIRS定量分析模型,并根据F-检验,采用同一独立验证集,评价NIRS定量分析模型的预测效果。结果表明:维生素E的NIRS定量分析模型A(Model-A),以检测值为定标参考值,最优光谱预处理方式为多元散射校正,定标集决定系数(R^(2))为0.992,最小交互验证标准误差(RMSECV)为4.50×10^(3) IU/kg,相对分析误差(RPD)为11.2;维生素E的NIRS定量分析模型B(Model-B),以理论值为定标参考值,最优光谱预处理方式为变量标准化,定标集R^(2)为0.991,RMSECV为4.94×10^(3) IU/kg,RPD为10.6。在置信水平α=0.05下,2个模型的F值分别为5.48和3.77,均高于F_(0.05)(2,60)=3.15。由此可见,在置信水平α=0.05下,选择以理论值为参考值构建维生素E的NIRS定量分析模型具有可行性;但是,维生素预混料存放7 d后,预测误差将增大。
The purpose of this study was to compare the prediction accuracy difference of two kinds of near-infrared reflectance spectroscopy(NIRS)quantitative analysis models which with the detected value as the calibration reference value and with the theoretical value as the calibration reference value.A total of 225 vitamin premix samples were prepared,the NIRS of all samples with a range of 10000 to 4000 cm-1 were collected by Fourier transform near-infrared spectrometer.Two NIRS quantitative analysis models of vitamin E were constructed based on the detection value and theoretical value as the calibration reference value,respectively,with same calibration set of vitamin premix,partial least square method and the optimal spectral pretreatment method.According to the F-test,the same independent validation set was used to evaluate the prediction effect of NIRS quantitative analysis model.The results showed that the NIRS quantitative analysis model A(Model-A)of vitamin E taken the detected value as the calibration reference value,the optimal spectral preprocessing method was multivariate scattering correction,the determination coefficient(R^(2))of the calibration set was 0.992,the root mean square error of cross validation(RMSECV)was 4.50×10^(3) IU/kg,and the ratio performance deviation(RPD)was 11.2.The NIRS quantitative analysis model B(Model-B)of vitamin E taken the theoretical value as the calibration reference value,the optimal spectral pretreatment method was standard normal variate,the R^(2) of the calibration set was 0.991,the RMSECV was 4.94×10^(3) IU/kg,and the RPD was 10.6.At confidence levelα=0.05,the F-values of the 2 models were 5.48 and 3.77,respectively,which are higher than F_(0.05)(2,60)=3.15.In conclusion,it is feasible to choose the theoretical value as the calibration reference value to construct the NIRS quantitative analysis model of vitamin E at the confidence level ofα=0.05,however,the prediction error will increase after the vitamin premix is stored for more than 7 days.
作者
樊霞
姜训鹏
王石
宋金彩
田淑丽
郑学聪
FAN Xia;JIANG Xunpeng;WANG Shi;SONG Jincai;TIAN Shuli;ZHENG Xuecong(Institute of Quality Standards&Testing Technology for Agro-Products,Chinese Academy of Agricultural Sciences,Beijing 100081,China;Bluestar Adisseo Nanjing Co.,Ltd.,Nanjing 210048,China;COFCO(Beijing)Feed Technology Co.,Ltd.,Beijing 100081,China)
出处
《动物营养学报》
CAS
CSCD
北大核心
2022年第3期1975-1983,共9页
CHINESE JOURNAL OF ANIMAL NUTRITION
基金
国家重点研发专项课题(2016YFF0201105)。