摘要
该研究基于高光谱技术结合化学计量学方法,建立丹参主要酚酸成分的快速无损定量检测模型。采集来自山东、河北、山西、四川和安徽的420份丹参样品高光谱数据,采用一阶导数(D1)、二阶导数(D2)、savitzky-golay平滑(SG)、多元散射校正(MSC)和标准正态变换(SNV)进行预处理。通过偏最小二乘回归(PLSR)、反向传播神经网络(BPNN)与随机森林回归(RFR)构建丹酚酸B、紫草酸、丹参素、丹酚酸Y及迷迭香酸的含量预测模型,并采用经迭代保留信息变量(IRIV)、连续投影(SPA)、非信息变量剔除(UVE)和变量迭代空间收缩(VISSA)方法筛选特征波长。研究表明,PLSR模型预测表现最佳;特征提取方法的波段数量差异显著,直接影响模型性能;预处理方法应与特征提取适应性组合,以优化模型稳健性。各成分最佳模型分别为:丹酚酸B、紫草酸和丹酚酸Y采用VISSA-D1-PLSR模型,预测集决定系数(R_(p)^(2))分别为0.942、0.818、0.797,残差预测偏差(RPD)为4.158、2.308、2.186;丹参素和迷迭香酸以全波段SG-PLSR模型较优,R_(p)^(2)为0.803、0.702,RPD为2.162、1.782。该研究建立了高光谱技术结合化学计量学方法对丹参药材中多种酚酸成分的预测输出模型,可为现场场景下丹参药材的快速检测需求提供技术支持。
This study aims to establish a model for rapid non-destructive quantitation of main phenolic acids in Salviae Miltiorrhizae Radix et Rhizoma by hyperspectral imaging combined with chemometric methods.Hyperspectral imaging was performed for 420 Salviae Miltiorrhizae Radix et Rhizoma samples collected from Shandong,Hebei,Shanxi,Sichuan,and Anhui.Original spectral data(ORI)were preprocessed by first derivative(D1),second derivative(D2),savitzky-golay(SG)smoothing,multiplicative scattering correction(MSC),and standard normal variate(SNV)algorithms.Models for predicting the content of salvianolic acid B,lithospermic acid,danshensu,salvianolic acid Y,and rosmarinic acid were constructed via partial least squares regression(PLSR),backpropagation neural network(BPNN),and random forest regression(RFR).Additionally,iterative retained information variable(IRIV),successive projections algorithm(SPA),uninformative variables elimination(UVE),and variable iterative space shrinkage approach(VISSA)were employed for feature extraction to refine model efficiency.The results demonstrated that the PLSR model exhibited the best predictive performance.Significant differences in the number of wavelengths selected by different feature extraction methods directly impacted model performance.Therefore,preprocessing methods should be combined adaptively with appropriate feature extraction techniques to enhance model robustness.The optimal models for each component were as follows.VISSA-D1-PLSR models were selected for predicting salvianolic acid B,lithospermic acid,and salvianolic acid Y,with prediction set coefficient of determination(R_(p)^(2))values of 0.942,0.818,and 0.797 and residual predictive deviation(RPD)values of 4.158,2.308,and 2.186,respectively.For danshensu and rosmarinic acid,the full-spectrum SG-PLSR models demonstrated the best performance,with R_(p)^(2) values of 0.803 and 0.702 and RPD values of 2.162 and 1.782,respectively.This study demonstrates that HSI technology combined with chemometric methods can provide reliable technical support for predicting multiple phenolic acids in Salviae Miltiorrhizae Radix et Rhizoma and the real-time rapid evaluation of this medicinal material.
作者
栾美琪
熊丰
戴瑶瑶
詹志来
洪家顺
宁知贵
白瑞斌
杨健
LUAN Mei-qi;XIONG Feng;DAI Yao-yao;ZHAN Zhi-lai;HONG Jia-shun;NING Zhi-gui;BAI Rui-bin;YANG Jian(State Key Laboratory for Quality Ensurance and Sustainable Use of Dao-di Herbs,National Resource Center for Chinese Materia Medica,China Academy of Chinese Medical Sciences,Beijing 100700,China;Evaluation and Research Center of Dao-di Herbs of Jiangxi Province,Ganjiang New District 330000,China)
出处
《中国中药杂志》
北大核心
2025年第22期6319-6327,共9页
China Journal of Chinese Materia Medica
基金
中国中医科学院科技创新工程项目(CI2023E002)
国家重点研发计划项目(2024YFC3506800)
中国中医科学院中药研究所中药全产业链质量技术服务平台项目(2022-230-221)
国家中医药管理局高水平中医药重点学科建设项目(ZYYZDXK-2023244)
财政部和农业农村部国家现代农业产业技术体系项目(CARS-21)。
关键词
丹参
高光谱技术
酚酸类成分
化学计量学
含量预测
特征筛选
Salviae Miltiorrhizae Radix et Rhizoma
hyperspectral imaging
phenolic acids
chemometrics
content prediction
fea-ture selection
