基于SAGAN数据增强与双向AT-LSTM神经网络的黄芪源

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）21-0077-06

Astragalus Membranaceus Traceability Based on SAGAN Data Enhancement and Bidirectional AT-LSTM Neural Network

LYUYuexian，ZHAOFeiyan，CHENGJing，FANJilan （SchoolofPhysicsandElectronic InformationEngineering，NingxiaNormalUniversity，Guyuan756o99，China）

Abstract：Inorder to explore theapplicationofSelf-Atention Generative AdversarialNetwork （SAGAN）andBidirectional Attention-basedLongShort-TermMemoryNetwork （BidirectionalAT-LSTM）inthetraceabilityofAstragalusmembranaceus origin，thispaperuses theRamanspectraldataofAstragalusmembranaceussamples fromIer Mongolia，Schuan，hanxi， andGansu provinces for experiments.The dataof 16O original samples of Astragalus membranaceus Raman spectroscopyare enhancedbySAGAN，andthesamplesize isexpandedto5Oo.TheresultsshowthattheBidirectionalAT-LSTMmodelchieves a classification accuracy of 97% in the recognition task of Astragalus membranaceus from four producing areas.Compared with thebenchmarkmodelssuchasRandomForest，SVM，XGBoost，andKN，theclassificationaccuracyofthemodelisimpoed by no less than 2.75% ，and it performs best in the origin classification task of Astragalus membranaceus in Iner Mongolia andGansu.Theresearchshows thattheBidirectionalAT-LSTMmodelhassignificantadvantages inthe spectraltraceabilityof Astragalus membranaceus，which canopenupanew way forthe qualitycontrolandtraceabilityresearch ofTraditional Chinese Medicine.

KeyWords：SAGANdata enhancement; LSTMNeural Network; Astragalus membranaceus traceability;Raman spectroscopy

0 引言

黄芪被认为是一种强效免疫调节剂，能够增强机体免疫功能[]，提升抗肿瘤免疫效应，可用于临床免疫功能调节及肿瘤等疾病的辅助治疗。（剩余7276字）