近红外光谱模型的适用性考察

Applicability of Near Infrared Spectrum Model

为了更好地维护近红外光谱模型,选择了15个密封的灭活烟叶样品和分装在3个密封样品瓶中的微晶纤维素,在3台不同型号的近红外仪上对样品分别进行检测。采用方差分析、差谱分析考察样品瓶、仪器光源更换、样品测试时间对同一样品光谱的影响,并采用基于主成分分析(PCA)与马氏距离的判别分析模型进行样品的定性分析。结果表明:材质相同且均匀的样品杯及仪器光源的更换对样品光谱及模型预测准确率几乎无影响;同一仪器测得同一样本的光谱随测试时间的推移有一定变化,因此,所建模型的预测准确性也会逐渐降低;数学模型具有一定的适用期,当模型对检验集样品的识别准确率低于某一限定阈值时,则需要重新建模或对新测样本光谱进行校正。

In order to well maintain near infrared spectrum model, 15 sealed inactivated tobacco samples and microcrystalline cellulose samples sealed in 3 bottles were separately detected by 3 NIR instruments of different types. Variance analysis and differential spectrum analysis were used to investigate the effects of sample bottle, alteration of light source in instrument and testing time on spectra of the same sample. Furthermore, the samples were qualitatively analyzed by the discriminant model based on Principal Component Analysis （PCA） and Mahalanobis distance. The results showed that sample bottles of identical and homogeneous mate- rial and light source alteration hardly affected the spectra of sample and the prediction accuracy of model, while the spectra of the same sample detected by the same instrument changed to a certain extent with the lapse of time and the prediction accuracy of the model also decreased gradually. It means that the mathematical model has a limited service life, once the prediction accuracy is lower than the specified threshold, it has to be replaced by a new model or the spectra of newly detected sample has to be corrected.