菊花为菊科植物菊的头状花序, 滁菊、 贡菊、 杭菊和亳菊是常见的几类药用品种菊花。 不同品种菊花在外观上具有极大的相似性, 非专业人员仅凭肉眼难以对其进行准确鉴别分析。 常规仪器分析法检测成本较高, 分析时间较长, 且需要对样品进行破坏性处理, 影响了产品的二次销售。 近红外光谱技术作为近年来快速发展起来的一种绿色、 简单、 快速的新型检测技术, 在中药鉴别领域取得了很大的进展。 基于便携式近红外光谱仪结合化学计量学方法建立了一种菊花品种无损鉴别方法。 利用便携式近红外光谱仪采集了滁菊、 贡菊、 杭菊和亳菊完整以及粉末状两种物理形态样品的光谱, 采用单一以及组合光谱预处理方法消除光谱中存在的干扰, 结合不同模式识别方法(主成分分析法、 软独立模式分类法和Fisher线性判别分析法)分别构建了不同品种菊花的鉴别模型。 结果表明: 由于仪器的限制及样品物理性状的原因, 光谱中存在较为明显的背景、 基线漂移以及噪声的干扰, 完整样品由于物理性状的原因, 基线漂移干扰尤为严重; 采用主成分分析法结合光谱预处理方法无法实现不同品种菊花的准确鉴别, 完整样品最佳鉴别正确率仅为8.33%, 粉末样品最佳鉴别正确率为52.38%; 通过软独立模式分类法结合预处理方法可以得到较为准确的鉴别结果, 完整样品光谱数据经一阶导数+多元散射校正优化后鉴别正确率为95%, 粉末状样品数据采用原始数据的鉴别正确率为92.5%; Fisher线性判别分析方法结果最佳, 完整样品数据经连续小波变换优化后可以得到97.5%的鉴别正确率, 粉末状样品采用原始光谱便可得到100%鉴别正确率。 以上结果表明, 当采用合适的预处理和建模方法, 完整样品和粉末状样品鉴别结果较为一致, 基于便携式近红外光谱仪结合化学计量学可实现对不同品种菊花的准确无损鉴别分析, 为食药同源产品的无损鉴别分析提供了新途径。

Chrysanthemum is derived from the capitulum of Chrysanthemum. Chuju, Gongju, Hangju and Boju are common medicinal chrysanthemums. Different chrysanthemum varieties have great similarities in appearance, and it is difficult for laypeople to identify them accurately only by naked eyes. The conventional instrumental analysis method has the disadvantages of high detection cost, long analysis time, and destructive treatment of samples, which affects the secondary sales of the products. As a green, simple and rapid detection technology, near-infrared spectroscopy has made great progress in traditional Chinese medicine identification. This study established a nondestructive identification method of different Chrysanthemum varieties based on portable near-infrared spectrometer and chemometric methods. The spectra of complete and powder samples of Chuju, Gongju, Hangju and Boju were collected by grating portable near-infrared spectrometer. The single and combined spectral pretreatment methods were used to eliminate the interferences in the spectra. The identification models of different Chrysanthemum varieties were constructed by combining principal component analysis, soft independent modeling of class analogy and Fisher linear discriminant analysis methods. The results show that: due to the restrictions of the current measure instruments and the difference of sample particle size and distribution, there are obvious interferences of background, baseline drift and noise in the spectra. The baseline drift interference is particularly serious for the analysis of the complete samples. The principal component analysis combined with spectral pretreatment methods could not identify different varieties of chrysanthemum. The best identification accuracy of complete samples was only 8.33%, and that of powder samples was 52.38%. The soft independent modeling of class analogy can obtain more accurate identification results with preprocessing methods. The identification accuracy of complete sample data is 95% with first derivative+multiple scattering correction, while the identification accuracy of powder sample data is 92.5% with the original data. The results of Fisher linear discriminant analysis are the best. When the complete sample spectra were optimized by continuous wavelet transform, the identification accuracy was 97.5%. When the original spectra of powder samples were used, the identification accuracy could reach 100%. The above results show that the complete and powder samples’ identification results are consistent when the appropriate pretreatment and modeling methods are used. Based on the grating portable near-infrared spectrometer combined with chemometrics methods, the accurate identification of different Chrysanthemum varieties can be realized, which provides a new way for the nondestructive identification of food and drug homologous products.