College of Pharmaceutical Sciences, Zhejiang University, Hangzhou 310058, China.
SPH Liaoning Herbapex Pharmaceutical (Group) Co., Ltd., Benxi 117200, China.
Spectrochim Acta A Mol Biomol Spectrosc. 2018 Feb 15;191:233-240. doi: 10.1016/j.saa.2017.10.019. Epub 2017 Oct 10.
Near infrared (NIR) spectroscopy coupled with chemometrics was used to discriminate the geographical origin of Herba Epimedii in this work. Four different classification models, namely discriminant analysis (DA), back propagation neural network (BPNN), K-nearest neighbor (KNN), and support vector machine (SVM), were constructed, and their performances in terms of recognition accuracy were compared. The results indicated that the SVM model was superior over the other models in the geographical origin identification of Herba Epimedii. The recognition rates of the optimum SVM model were up to 100% for the calibration set and 94.44% for the prediction set, respectively. In addition, the feasibility of NIR spectroscopy with the CARS-PLSR calibration model in prediction of icariin content of Herba Epimedii was also investigated. The determination coefficient (R) and root-mean-square error (RMSEP) for prediction set were 0.9269 and 0.0480, respectively. It can be concluded that the NIR spectroscopy technique in combination with chemometrics has great potential in determination of geographical origin and icariin content of Herba Epimedii. This study can provide a valuable reference for rapid quality control of food products.
本工作采用近红外（NIR）光谱结合化学计量学方法来区分淫羊藿的地理来源。构建了四种不同的分类模型，分别为判别分析（DA）、反向传播神经网络（BPNN）、K-最近邻（KNN）和支持向量机（SVM），并比较了它们在识别准确率方面的性能。结果表明，SVM 模型在淫羊藿的地理来源识别方面优于其他模型。最优 SVM 模型对校准集的识别率高达 100%，对预测集的识别率高达 94.44%。此外，还研究了 NIR 光谱结合 CARS-PLSR 校准模型预测淫羊藿中淫羊藿苷含量的可行性。预测集的决定系数（R）和均方根误差（RMSEP）分别为 0.9269 和 0.0480。可以得出结论，NIR 光谱技术结合化学计量学在确定淫羊藿的地理来源和淫羊藿苷含量方面具有很大的潜力。本研究可为食品的快速质量控制提供有价值的参考。
