分析测试学报2026，Vol.45Issue(6)：1212-1220,9.DOI:10.12452/j.fxcsxb.25110701
血府逐瘀浸膏质量智能评价体系构建:基于物理指纹图谱、机器视觉与大语言模型的联合应用
Construction of Intelligent Quality Evaluation System for Xuefu Zhuyu Extract:Combined Application of Physical Fingerprint,Machine Vision and Large Language Model
摘要
Abstract
In order to solve the limitation of relying on a single index or experience to judge the con-centration end point in the production of traditional Chinese medicine extract,this study took Xuefu Zhuyu extract as the object,and innovatively proposes an intelligent quality control system of"physi-cal fingerprint+machine vision+large language model"multi-modal fusion strategy.Seven core physical indexes such as density and pH value were screened and standardized to construct physical fingerprint.Then,the abnormal samples were eliminated by Mahalanobis distance,and the compre-hensive score was calculated by"CRITIC+entropy weight+TOPSIS"method.With 0.6 as the quali-fied threshold,100 batches of samples were divided into 62 batches of qualified samples and 38 batches of unqualified samples.At the same time,the extract image features were collected and ex-tracted,and six machine learning models were constructed.Among them,the XGBoost model had the best performance,with accuracy,precision,recall,F1 score,and AUC reaching 0.933 3,1.000 0,0.833 3,0.909 1,and 0.963 0,respectively.Based on this,an intelligent assessment platform integrating large language models can be further developed,which can complete the analy-sis and generate actionable process suggestions within about ten seconds.This system provides an in-tegrated quality control scheme of"objective grading+rapid prediction+intelligent"suggestion for Xuefu Zhuyu extract,and also provides a reusable technical paradigm for multi-dimensional quality evaluation of Chinese medicine extract intermediates,which effectively promotes the transformation of Chinese medicine production to"data-driven".关键词
血府逐瘀浸膏/物理指纹图谱/机器视觉/大语言模型/质量控制Key words
Xuefu Zhuyu extract/physical fingerprint/machine vision/large language model/quality control分类
化学化工引用本文复制引用
沙鑫,常淏,宋纹,余河水,李正,李文龙..血府逐瘀浸膏质量智能评价体系构建:基于物理指纹图谱、机器视觉与大语言模型的联合应用[J].分析测试学报,2026,45(6):1212-1220,9.基金项目
国家重点研发计划(2023YFC3504502) （2023YFC3504502）
现代中医药海河实验室-天津宏仁堂药业有限公司合作项目(XMH2022004) （XMH2022004）
