Q型因子光谱转移法用于烟草化学成分近红外预测模型的传递

Using Q-factor spectral transfer method to improve transferability of near-infrared prediction models for tobacco chemical components

为了解决不同近红外光谱仪器采集相同样品的光谱不一致的问题,并提高烟草化学成分近红外预测模型的传递效果,建立了一种Q型因子光谱转移法(QFST)。该方法根据类间可分性原则,将众多原始变量归结为几个综合因子,利用综合因子重组主机和子机光谱矩阵,利用广义逆矩阵得出主机和子机光谱转换关系矩阵,从而实现主机和子机间的光谱转移。并将建立的QFST模型转移法与光谱空间转换(SST)和分段直接校正法(PDS)两种常用的模型转移方法进行了比较。结果表明:(1)对于烟草70种化学指标,QFST法和SST法的模型预测效果相差不大,整体预测效果优于PDS法。(2)应用QFST法、SST法和PDS法3种模型转移方法后进行预测,总植物碱、还原糖、总糖、总氮和氯等常规化学成分指标的R2均达到0.9以上。对于氨基酸和Amadori化合物等烟草中质量分数较低的化学成分来说,采用QFST法进行模型转移后预测的效果优于SST法。

In order to solve the problem of spectral inconsistency of the same leaf sample collected by different NIR spectroscopies and to improve the transfer effects of NIR prediction models for chemical component analyses in tobacco,a Q-factor spectral transformation(QFST)method was established.According to the principle of inter-class separability,QFST classified many original variables into several comprehensive factors,which were used to reorganize the spectral matrixes.The QFST method derived the transformation relationship matrix by a generalized inverse matrix to realize the spectral transfer between the master and slave instruments.The QFST method was compared with the spectral space transformation(SST)and piecewise direct standard(PDS)methods,which were the common model transfer methods.The results showed that:1)For 70 different chemical indexes of tobacco,the prediction accuracies of the QFST and SST methods showed no major difference,and the overall prediction effects of both methods were better than the PDS method.2)The R2 of conventional chemical components such as total plant alkaloids,reducing sugars,total sugars,total nitrogen and chlorine were all above 0.9 after using the three model transfer methods.For amino acids and Amadori compounds,which are present at lower mass fractions in tobacco,the QFST method was better than the SST method in model transfer prediction.