目的 建立NIRS技术快速无损鉴别当归药材及其伪品的方法。方法 采集当归及伪品断面的近红外光谱，结合模式识别法分析药材，用主成分分析（Principal component analysis，PCA）进行定性分析；对比梯度提升决策树（Gradient Boosting Decision Tree，GBDT）、支持向量机（Support Vector Machine，SVM）、人工神经网络（Artificial Neural Network；ANN）3种当归真伪判别模型的分类效果；利用RF筛选特征波长优化所建模型。结果 PCA无法有效区别当归及其伪品；与ANN、SVM相比，GBDT具有更高的准确性，训练集与预测集的总体准确率分别为94.39%和90.38%；而后以RF选取出20个特征波长，建立的近红外特征光谱判别模型训练集和预测集的总体准确率也达到了91.59%和86.54%。结论 近红外光谱技术结合GBDT鉴别当归药材真伪鉴别是可行的，为当归药材真伪快速无损鉴别提供了一种新方法。
Objective To identify Angelica Sinensis and its adulterants by NIRS. Methods The near-infrared spectra of Angelica Sinensis were collected and further analyzed by PCA which is one kind of pattern recognition. Then three different classifiers, namely GBDT, SVM and ANN, were employed to establish discriminative models. Afterwards, an optimized model was screened out by using RF filter characteristic wavelength optimization on the basis of GBDT mode.Results PCA could not distinguish Angelica Sinensis and its adulterants effectively. Compared to SVM and ANN, GBDT creates better identification model. It showed higher accuracy, and the overall accuracy rate of training set and prediction set was 94.39% and 90.38%, respectively. Furthermore, 20 characteristic wavelengths were extracted by RF and re-establish the Angelica Sinensis authenticity characteristic identification model, the overall accuracy rate of the training set and prediction set was 91.59% and 86.54%. Conclusion An identification model of Angelica Sinensis and its adulterants is built based on NIR and GBDT, which provides a new method for traditional Chinese medicine non-invasive identification.
＊ 首都中医药研究专项重点课题 17ZY05;北京中医药大学在读研究生项目 2016-JYB-XS057＊ 首都中医药研究专项重点课题（17ZY05）：不同产区当归功效组分特异性及其形成机制研究，负责人：王大仟；北京中医药大学在读研究生项目（2016-JYB-XS057）：基于中药变质多因素筛选及相关性评估构建中药加速模型，负责人：拱健婷。