红外光谱结合化学计量学鉴别獐牙菜属植物

目的:采用傅里叶变换红外光谱(FTIR)结合化学计量学实现獐牙菜属植物快速、准确鉴别。方法:采集川东獐牙菜、青叶胆、紫红獐牙菜、狭叶獐牙菜和西南獐牙菜不同部位(根、茎、叶) 543份样品红外光谱信息,原始数据经标准正态变量(SNV),多元散射校正(MSC),平滑(SG),一阶导数(1D),二阶导数(2D),三阶导数(3D)等处理后删减4 000～3 700,2 799～1 800 cm~(-1)和682～653 cm~(-1)波段,建立偏最小二乘判别分析(PLS-DA)和支持向量机(SVM)模型。结果:5种獐牙菜属植物相同部位平均光谱较为相似,无法区分,同一物种不同部位光谱特征峰有差异,复杂程度为叶茎根;根、茎和叶3个部位PLS-DA和SVM模型均能准确鉴别5种獐牙菜属植物,且MSC+SG+2D预处理效果最佳。PLS-DA模型R~2Y 0. 8,RMSEP RMSECV,所建模型稳定,效果好,Q~2超过0. 6,预测集正确率达到100%,所有预测集样品分类正确,模型预测能力强。根、茎和叶SVM模型最优惩罚参数c分别为22. 627 4,2和1. 414 2,均在正常范围内,预测集正确率均为100%,分类准确率高。结论:FTIR结合PLS-DA和SVM模型能准确区分不同獐牙菜属植物,模型预测效果好,为其他植物鉴别提供一定的参考依据。

Discrimination of Different Species in Swertia Using FTIR Combined with Chemometrics

Objective:To realize the rapid and accurate discrimination of Swertia plants by Fourier transform infrared spectroscopy (FTIR) and chemometrics. Method:The original infrared spectra data from different parts (roots,stems,leaves) of all of the 543 samples of S. davidii,S. mileensis,S. punicea,S. angustifolia and S. cincta were collected and preprocessed by multiplicative scatter correction (MSC),standard normal variate (SNV),Savitzky-Golay filter (SG),first derivative (1D),second derivative (2D),third derivative (3D). Then,the spectral ranges of 4 000-3 700,2 799-1 800 cm^-1 and 682-653 cm^-1 were deleted before PLS-DA and SVM analysis. Result:The samples of the five species could not be distinguished with similar averaged infrared spectra in the same part. The characteristic peaks of different parts in the same species were different, and the sequence of complexity was leaves > stems > roots. The five species of Swertia could accurately be identified by PLS-DA and SVM models established by spectra data in roots, stems and leaves. MSC+SG+2D showed the best preprocessing effect,and the prediction accuracies of all models were 100%. The values of R²Y in PLS-DA of all of the parts were more than 0.8, and the RMSEP was less than RMSECV,indicating that the model was stable and more effective. Furthermore,the value of Q² exceeded 0.6, and the accuracy of prediction set reached 100%, indicating a high classification accuracy. It showed that PLS-DA models had a strong prediction ability. The c values in SVM model of roots,stems and leaves were 22.627 4,2 and 1.414 2,respectively,which were all within the normal ranges. The accuracy of prediction set was 100%, suggesting a high accuracy. Conclusion:FTIR combined with PLS-DA and SVM could accurately distinguish different species from Swertia, and the model has a good prediction effect and provides certain reference for the identification of other plants.