目的 基于智能感官技术对不同发酵程度的制枳壳饮片外观性状进行分析，并构建制枳壳发酵程度的快速辨识模型以及外观性状与内在质量的逐步回归模型，以期为制枳壳饮片质量评价提供参考。方法 通过测色仪、电子鼻测定不同发酵程度制枳壳的外观性状指标（颜色、气味），结合主成分分析（principal component analysis，PCA）、正交偏最小二乘-判别分析（orthogonal partial least squares-discriminant analysis，OPLS-DA）、Fisher线性判别分析、反向传播（back propagation，BP）神经网络算法等多种化学计量学方法，建立不同发酵程度制枳壳饮片快速辨识模型。采用HPLC法对枳壳中8种黄酮类成分（柚皮苷、新橙皮苷、芸香柚皮苷、橙皮苷、枸橘苷、橙皮素-7-O-葡萄糖苷、柚皮素、橙皮素）进行定量测定，并将制枳壳饮片外观性状与黄酮类成分进行相关性分析和逐步回归分析，建立颜色、气味与内在成分之间的定量模型。结果 单源的色度值、气味特征值不能将生枳壳与不同发酵程度制枳壳完全区分，基于“色度-气味”数据融合建立的BP神经网络判别模型较Fisher线性判别模型的分类预测效果更好，能够快速、准确地辨识不同发酵程度制枳壳饮片。相关性分析结果显示，制枳壳饮片色度值与所含8种黄酮类成分含量具有显著的相关性，电子鼻气味响应值与成分之间呈现不同程度的相关性，进一步构建的逐步回归模型可通过颜色、气味的外观性状参数快速预测制枳壳饮片主要黄酮类成分的含量变化。结论 基于“色度-气味”构建的BP神经网络辨识模型可以快速准确判别制枳壳发酵程度，“外观性状-成分”回归模型的建立为制枳壳质量的快速检测提供科学依据。

Objective The appearance characteristics of processed Aurantii Fructus (PAF) with different fermentation degrees were analyzed using intelligent sensory technology. A rapid identification model of fermentation degrees of PAF and a stepwise regression model of appearance and intrinsic quality were established to provide a reference for the quality evaluation of PAF pieces. Methods The appearance characteristics (color and odor) of PAF with different fermentation degrees were determined by spectrophotometric colorimeter and electronic nose. Multivariate statistical analysis techniques including principal component analysis (PCA), orthogonal partial least squares-discriminant analysis (OPLS-DA), Fisher linear discriminant analysis, and back propagation (BP) neural network algorithm were applied to establish a rapid identification model of PAF with different fermentation degrees. The contents of eight flavonoids in PAF (naringin, neohesperidin, narirutin, hesperidin, poncirin, hesperetin-7-O-glucoside, naringenin, and hesperetin) were determined by HPLC to establish correlation between their chemical components and appearance characteristics. The quantitative model between color, odor, and intrinsic components was established by stepwise regression analysis. Results The colorimetric values and odor characteristic value of a single source could not completely differentiate between Aurantii Fructus (AF) and PAF with different fermentation degrees. The BP neural network discrimination model based on source fusion of “chroma-odor” had better classification and prediction effect than the Fisher linear discrimination model, and could quickly and accurately identify the fermentation degree of PAF. Correlation analysis found that chromaticity values of PAF were significantly correlated with the contents of eight flavonoids, and electronic nose odor response values showed different degrees of correlation with the components. The changes in the content of main flavonoids in PAF could be quickly predicted through the parameters of color and odor using the stepwise regression model. Conclusion The BP neural network discrimination model based on “chroma-odor” can identify the fermentation degree of PAF quickly and accurately, and the establishment of the “appearance characteristics-composition” regression model can provide a scientific basis for rapid quality detection of PAF.

R283.6

国家自然科学基金资助项目（81873003）;国家中医药管理局科技司项目“特色炮制技术规律发掘—蒸制”（GZY-KJS-2022-054）