目的 探究百合制桔梗“辨色论质”的实质，形成百合制桔梗科学的质量评价方法，并为其在线质量控制提供理论依据。方法 采用精密色差仪对桔梗和百合制桔梗的色度值进行测定，运用UPLC-ELSD建立桔梗和百合制桔梗指纹图谱，通过SPSS 21.0和Origin Pro 2021软件分析其色度值与共有峰之间的相关性，根据相关程度筛选色度相关峰并进行指认。运用软件MATLAB R2016a筛选隐含节点数，建立最优的“成分-色度”人工神经网络模型。结果 百合制桔梗共有12个共有峰，指认出6个色度相关成分，分别为王百合苷A(峰1)、王百合苷F(峰2)、王百合苷B(峰3)、桔梗皂苷E(峰5)、党参炔苷(峰6)、桔梗皂苷D(峰10)。桔梗皂苷E与色度值a^*、b^*呈正相关，王百合苷A、王百合苷F、王百合苷B等化学成分均与色度值呈负相关。通过隐含节点数筛选，发现隐含节点数为8时模型整体拟合程度最高，人工神经网络模型的均方根误差(root mean square error，RMSE)较小，R²最高(RMSE＝0.251 3，R²_all＝0.926 2)，因此选择4-8-7作为模型最佳拓扑结构，经验证表明该模型预测结果良好。结论 证实百合制桔梗“辨色论质”的科学合理性，可将色度指标纳入百合制桔梗饮片质量评价体系中，为百合制桔梗的在线质量控制提供思路。

Objective To explore the essence of “color discrimination grading” of Lilii Bulbus-processed Platycodonis Radix (LB-pPR), form a scientific quality evaluation method of LB-pPR, and provide a theoretical basis for its online quality control. Methods The chromaticity values of Platycodonis Radix (Jiegeng, PR) and LB-pPR were measured by precision color difference instrument, and the fingerprints of PR and LB-pPR were established by UPLC-ELSD. The correlation between the chromaticity values and common peaks was analyzed by SPSS 21.0 and Origin Pro 2021 software, and the chromaticity correlation peaks were screened and identified according to the correlation degree. The software MATLAB R2016a was used to screen the number of hidden nodes, and the optimal “component-chromaticity” artificial neural network model was established. Results There were 12 common peaks in LB-pPR. Six color related components were identified, which were regaloside A (peak 1), regaloside F (peak 2), regaloside B (peak 3), platycodin E (peak 5), lobetyolin (peak 6) and platycodin D (peak 10). Platycodin E was positively correlated with chromaticity values a^*, b^*, and chemical components such as regaloside A, regaloside F, regaloside B were negatively correlated with chromaticity values. Through the screening of the number of hidden nodes, it was found that when the number of hidden nodes was 8, the overall fitting degree of the model was the highest, the root mean square error (RMSE) of the artificial neural network model was small, and the R² was the highest (RMSE = 0.251 3, R²_all = 0.926 2). Therefore, 4-8-7 was selected as the best topology of the model, and experience showed that the prediction result of the model was good. Conclusion This study confirmed the scientific rationality of “color discrimination grading” of LB-pPR. The chromaticity index can be included in the quality evaluation system of LB-pPR, which provides ideas for the online quality control of LB-pPR.

国家重点研发计划专项课题(2018YFC1707206)；国家自然科学基金项目(82060724)；省重点研发计划项目(20192BBG70073)；国家级大学生创新创业训练计划项目(202110412015)；校级创新团队项目(CXTD22003)