藏红花和天然牛黄是广泛应用于临床实践的中药材, 由于产量较低、 药用价值和价格高, 市场需求量大等因素, 掺伪和伪品较多, 不仅严重损害患者健康, 而且妨碍市场正常运转。 传统的“一看、 二闻、 三泡”等经验鉴别方法已经越来越难以分辨高仿伪品; 而通过化学提取和色谱、 质谱等理化检测方法往往步骤繁琐、 费时, 且对检测环境、 人员及设备的要求和依赖度较高, 不能适应现场、 快速、 简便等实际需求, 亟需探索新的有效检测方法和鉴别技术。 太赫兹时域光谱(THz-TDS)不但具有单纯化合物的高度专属性和特异性, 又具有混合体系的“宏观指纹特征”, 可以鉴别混合物化学成分的多样性和复杂性。 另外, 主成分分析(PCA)作为一种常用的统计分析手段, 主要是用少数几个且能最大解释原始数据方差的综合变量来取代原始变量, 可以对不同种类的样本进行模式识别。 采用粉末研磨压片技术分别压制了藏红花和草红花样品各18个、 天然牛黄和人工牛黄各20个, 并利用太赫兹时域光谱测试技术分别测量了两种名贵中药材及其伪品在0.3~2.5 THz范围的吸收光谱, 最后利用主成分分析方法对获得的光谱数据进行分类识别。 为了提高PCA对测试数据的鉴别能力, 一方面将数据集映射到一组基(特征向量)进行简化, 选用较大的特征值代替原来的主要光谱信息; 另一方面, 为了消除无关因素对分类处理的干扰, 在进行PCA之前采取了Savitzky-Golay(S-G)平滑处理进行降噪, 去除冗余、 不相关的光谱特征; 然后通过Fisher诊断线进行判别分析。 对比未处理和经过S-G平滑处理的主成分得分图, 可以看出平滑处理后的分类效果明显优于未做处理的, 在未处理的得分图中, 两类样品点重叠比较严重, 而经过平滑后的得分图却只有相对较少的部分样品点重叠, 由此可以看出SG平滑在光谱识别中的重要性; 另外, 前两个主成分(PC1和PC2)已经基本能反映光谱之间的差异性。 分类结果显示, 藏红花和草红花具有明显的聚类趋势, 分类鉴别准确率均为100%; 而人工牛黄和天然牛黄的类内样品基本聚在一起, 但是类间略有重叠, 分类鉴别准确率分别为100%和90%。 除此之外, 样本的主成分得分图还可以反映样本的内部特征和聚类信息。 其中, 藏红花样本由于藏红花素、 藏红花酸等化合物成分含量较高, 聚合度较好, 分布范围相对集中; 反之, 天然牛黄为胆囊分泌物, 成分较为复杂, 聚类效果较差, 分布范围较广。 研究结果表明, 太赫兹光谱技术结合主成分分析可以区分藏红花和草红花以及天然牛黄和人工牛黄, 结果可靠。 该研究结果为丰富中草药的质量标准提供检测手段和理论依据。

Saffron and natural bezoar are two traditional Chinese medicines widely used in clinical practice. Due to their lower yields, high medicinal value and price, market demand and other factors, more and more adulteration and counterfeit goods not only seriously damage the health of patients but also hinder market normal operation. However, the empirical methods based on observation, smell and soak have become increasingly difficult to distinguish high imitation counterfeits. In addition, the traditional physical and chemical detection techniques through chemical extraction, chromatography and mass spectrometry are cumbersome and time-consuming, and have high requirements and reliance on testing environments, professional ability and equipment. They cannot meet the actual needs of on-site, rapid and simple identification. Thus, it is urgent to explore new and effective detection methods and identification techniques. Terahertz radiation has very low energy and terahertz time-domain spectroscopy (THz-TDS) possesses not only the high specificity of pure compounds but also the “macroscopic fingerprint characteristics” of the hybrid system to identify the diversity and complexity of the chemical composition in the mixture. In addition, as a common statistical analysis method, principal component analysis (PCA) mainly replaces the original variables with a few comprehensive variables that can explain the variance of the original data to the greatest extent and can perform pattern recognition on different kinds of samples. In this work, 18 pieces of saffron and safflower samples as well as 20 groups of natural and artificial bezoar were respectively compressed by using pellet compression. The absorption spectra of two kinds of valuable Chinese medicinal materials and their counterfeit products, saffron and safflower as well as natural and artificial bezoar, were measured using THz time-domain spectroscopy in the range of 0.3~2.5 THz. Finally, the principal component analysis (PCA) was used to identify the obtained data set. In order to improve the identification ability of PCA, on one hand, the data set was mapped to a set of bases (feature vectors) for simplification, and larger eigenvalues were selected instead of describing the original main spectral information; on the other hand, in order to eliminate the impact of noise on the classification process, we adopted Savitzky-Golay(S-G)smoothing before PCA to remove the redundant and irrelevant spectral features; the discriminant analysis was then performed using Fisher’s diagnostic line. Comparing the principal component scores with and without S-G smoothing, classification results with S-G smoothing were obviously distinguished and the first two principal components could basically reflect the differences between spectra. It could be clearly seen that in the unprocessed score plots, the overlapping of the two types of samples is severe, whereas only a relatively small number of sample points overlap in the smoothed score plots, indicating the role of SG smoothing in spectral identification. The classification results showed that the saffron and safflower had obvious clustering trends, and the accuracy of classification identification of saffron and safflower were both 100%; while there was a slight overlap of artificial bezoar and natural bezoar even though the intra-class samples basically gathered together, and the classification accuracy was 100% and 90%, respectively. Furthermore, the principal component score of the sample can also reflect the internal characteristics of the sample and the clustering information. Among them, the saffron sample contains higher compounds of crocin, crocetin and other content, so that better degree of polymerization has been obtained and the distribution is relatively concentrated; on the other hand, the compounds contained in the natural bezoar are more complex. Consequently, the clustering effect is poor and the distribution range is wide. The reliable results based on the THz-TDS and PCA not only distinguish between saffron and safflower as well as natural and artificial bezoar, but also provide the means and theoretical basis for enriching the quality standard of Chinese herbal medicine.