基于近红外光谱(NIR)和潜在语义分析(LSA)方法, 对５种典型壮阳中药材进行分类鉴别研究。利用潜在语义分析对光谱预处理后的5种壮阳中药材光谱数据进行特征提取和鉴别分类后, 将经光谱预处理和主成分分析(PCA)提取特征后的光谱特征数据分别带入K近邻(KNN)、BP神经网络(BP-ANN)和偏最小二乘支持向量机(LSSVM)三种典型的分类模型进行分类, 并将结果与潜在语义分析模型结果进行对比。 在4119.20~9881.46 cm-1波数范围内, NIR光谱数据经多元散射校正(MSC)预处理后, 代入潜在语言空间维数为3时所建立的LSA分类模型, 训练集和测试集准确率均达到了100%。 结果表明, 在壮阳类中药材的近红外光谱分析鉴别中, 潜在语义分析可以作为一种全新的提取光谱信息并分类的方法, 具有较好的运用前景和实际意义。

Five kinds of typical Yang-boosting Chinese herbal medicine are identified and classified based on near infrared spectroscopy (NIR) and latent semantic analysis (LSA) methods. Latent semantic analysis is used for characteristic extraction and classification of preprocessed spectral data of 5 kinds of Yang-boosting Chinese herbal medicine. The spectral characteristic data, after spectral pretreating and characteristic extraction by principal component analysis (PCA), are respectively subjected into the K-nearest neighbor (KNN), BP-artifical neural networks (BP-ANN) and least squares support vector machine (LS-SVM) classification models whose results then are compared with the result of latent semantic analysis model. In the characteristic wavenumber range of 4119.20~9881.46 cm-1, spectral data pretreated by multiplicative scatter correction (MSC) are substituted to LSA classification model when spacing dimension of underlying language is 3, and accuracy rates of both training set and test set are 100%. The results show that latent semantic analysis, which has a good application prospect and practical significance, can be used as a new method for spectral information extraction and classification in the near-infrared spectroscopy identification of Yang-boosting Chinese herbal medicine.