文献类型：期刊文献

中文题名：基于近红外光谱技术和编码器-解码器的黄芪产地鉴别

英文题名：Astragalus Origin Identification Based on Near-infrared Spectroscopy Technology and Encoder-Decoder

作者：刘明奇[1];李四海[1];宋航[1]

第一作者：刘明奇

机构：[1]甘肃中医药大学医学信息工程学院,甘肃兰州730000

第一机构：甘肃中医药大学信息工程学院（教育技术中心）

年份：2025

卷号：44

期号：10

起止页码：2063

中文期刊名：分析测试学报

外文期刊名：Journal of Instrumental Analysis

收录：;北大核心:【北大核心2023】；

基金：甘肃省自然科学基金项目(21JR1RA272,22JR5RA606);甘肃省教育厅高校教师创新基金项目(2023B-105)。

语种：中文

中文关键词：黄芪;近红外光谱;CTGAN;神经网络;编码器-解码器

外文关键词：Astragalus;near-infrared spectra;CTGAN;neural network;encoder-decoder

摘要：为进行黄芪产地溯源,提出了CTGAN+1D-CNN+Encoder-Decoder(CCEN)网络模型,首先通过条件表格生成对抗网络(CTGAN)增强黄芪近红外光谱数据,解决数据较少的问题,再通过在一维卷积神经网络(1D-CNN)上加入编码器-解码器(Encoder-Decoder)结构,使网络可以同时捕获特征之间的全局关系和局部关系。实验结果表明,CTGAN和Savitzky-Golay增强后,偏最小二乘法判别分析(PLS-DA)、随机森林(RF)、K近邻算法(KNN)和1D-CNN的准确率分别提升至0.9733、0.9533、0.9600和0.9733。加入编码器-解码器后,1D-CNN准确率提升至0.9778。最终CCEN模型在黄芪数据集上的准确率、召回率和F1值分别达到0.9867、0.9872和0.9868,均优于对比模型。结果证明CCEN模型适用于近红外光谱这类结构复杂、样本有限的一维信号数据,为黄芪中药材道地性产地识别研究提供了新方法。
To trace the origin of Astragalus,a CTGAN+1D-CNN+encoder-decoder(CCEN)network model was proposed.First,conditional generative adversarial networks(CTGAN)enhance As?tragalus near-infrared spectral data to address limited data.Then,integrating an encoder-decoder structure into a 1D-CNN enables the network to capture both global and local relationships among features simultaneously.Experimental results demonstrate that after CTGAN and Savitzky-Golay enhancement,the accuracy rates of partial least squares discriminant analysis(PLS-DA),random forest(RF),K-nearest neighbors(KNN),and 1D-CNN improved to 0.9733,0.9533,0.9600,and 0.9733,respectively.The incorporation of the encoder-decoder architecture further elevated the 1D-CNN accuracy to 0.9778.Ultimately,the CCEN model achieved accuracy,recall,and F1 scores of 0.9867,0.9872,and 0.9868,respectively,on the Astragalus dataset,outperforming all comparison models.The results demonstrate that the CCEN model is suitable for one-dimensional signal data with complex structures and limited samples,such as near-infrared spectra,providing a novel approach for identifying the origin of Astragalus.