红外光谱结合化学计量学鉴别西红花的产地

Identification of Origin of Crocus Sativus by Infrared Spectroscopy Combined with Stoichiometry

西红花不仅是在世界范围内广泛使用的传统香料,也是一味著名的中草药,其品质与产地密切相关。迫切需要寻找一种能快速对西红花进行产地判别的方法。基于不同产地西红花样品近、中红外光谱数据特征,采用Savitzky-Golay平滑(SG平滑)、多元散射校正、标准正态变量变换、一阶导数和二阶导数等预处理算法对光谱数据进行降噪处理,可以减小试验样本、测定环境以及仪器噪音对光谱数据的影响。结合偏最小二乘判别分析(Partial least square-discriminant analysis,PLS-DA)、决策树(Decision tree,DT)和支持向量机(Support vector machine,SVM)方法分别建立近红外光谱、中红外光谱、近中红外融合光谱分类模型,并采用移动窗口偏最小二乘法(Moving window partial least square method,MWPLS)提取光谱特征区间可以提升建模速度和分类精度。基于预测结果的准确率、混淆矩阵(Confusion matrix)和ROC曲线下面积(Area under curve,AUC)以选择最优分类模型。结果显示:近红外光谱、中红外光谱与融合光谱的最佳模型组合均为基于Savitzky-Golay平滑(SG平滑)预处理的偏最小二乘判别分析法(PLS-DA),其测试集准确率分别达到94.00%、94.00%和96.00%,ROC曲线下面积分别为0.9974、0.9963和0.9990,表明融合光谱可提升分类精度。特征波段的选择增强了单一光谱的分类准确率,但降低了融合光谱的分类性能。可见,结合移动窗口偏最小二乘法(MWPLS)的单一光谱技术可以精简、优化模型。所采用的方法实现了对不同产地西红花的快速准确鉴定,为西红花的产地判别提供一种新颖、快速的解决方案,有利于维护西红花市场的秩序,有助于其产业健康发展。

Crocus is not only a traditional spice widely used in the world,but also a famous Chinese herbal medicine.Its quality is closely related to the origin.It is urgent to find a method that can be used for quick identification of the origin of crocus.The near-mid-infrared data characteristics of crocus samples from different production areas were extracted.The data were preprocessed by Savitzky-Golay smoothing,multivariate scattering correction,standard normal variable transformation,first-order derivative and second-order derivative algorithms to reduce the test samples,environmental effect and instrument noise on spectral data.Combined with partial least square-discriminant analysis(PLS-DA),decision tree(DT)and support vector machine(SVM)methods were used to establish the classification models of near-infrared,mid-infrared,and near-mid-infrared fusion spectral,respectively.The moving window partial least square method(MWPLS)used for the extraction of the spectral feature interval can improve the modeling speed and classification accuracy.The optimal classification model was selected based on the accuracy of the prediction results,the confusion matrix and the area under the ROC curve(AUC).The results showed that the best model combination of near-infrared spectrum,mid-infrared spectrum and fusion spectrum was partial least squares discriminant analysis(PLS-DA)based on Savitzky-Golay smoothing(SG smoothing)preprocessing.The accuracies of the test sets reached 94.00%,94.00%,and 96.00%,and the area under the ROC curve reached 0.9974,0.9963,and 0.9990,respectively,indicating that the fusion spectrum could improve the classification accuracy.The selection of characteristic bands enhanced the classification accuracy of single spectrum,but degraded the classification performance of fusion spectrum.It can be seen that the single spectral technique combined with moving window partial least squares(MWPLS)could simplify and optimize the model.The proposed method provides a novel and fast solution for the identification of origin of crocus,which is conducive to maintain the order of crocus market,and contributes to the healthy development of its industry.