基于二维光谱结合深度学习对不同产地杜仲的鉴别  

Identification of Duzhong from different habitats based on two-dimensional spectrum combined with deep learning

作　　者：丁于刚 李鹂[3] 左智天[2] 李佑稷[3] DING Yu-gang;LI Li;ZUO Zhi-tian;Li You-ji(College of Traditional Chinese Medicine,Yunnan University of Traditional Chinese Medicine,Kunming 650500,Yunnan;Institute of Medicinal Plants,Yunnan Academy of Agricultural Sciences,Kunming 650200,Yunnan;College of Biological Resources and Environmental Sciences,Jishou University,Jishou 416000,Hunan)

机构地区：[1]云南中医药大学中药学院,云南昆明650500 [2]云南省农业科学院药用植物研究所,云南昆明650200 [3]吉首大学生物资源与环境科学学院,湖南吉首416000

出　　处：《中药与临床》2023年第1期16-19,28,共5页Pharmacy and Clinics of Chinese Materia Medica

基　　金：《基于多源信息融合技术的杜仲质量评价体系研究》,国家自然科学基金项目(No.31960323)。

摘　　要：目的:对来自全国13个省市共171个杜仲茎和叶样本进行光谱分析,结合深度学习建立不同产地判别模型,为其资源合理开发利用提供依据。方法:以13个产地的杜仲茎和叶为实验材料,分别检测其近红外光谱,结合二维光谱算法和残差卷积神经网络建立模式识别模型。结果:基于同步二维相关光谱的模型在卷积层数为26层及以上时均取得了100%的分类正确率,而基于异步二维相关光谱的模型预测集正确率低于30%。结论:表明该模型能够应用于不同产地杜仲药材的鉴别。Objective:To conduct spectral analysis on the 171 stems and leaves of Duzhong from 13 provinces in our country,and establish discrimination models for different habitats combined with deep learning,so as to provide a basis for the rational development and utilization of its resources.Method:The stems and leaves of Duzhong from 13 places of production were used as experimental materials,and their near-infrared spectra were detected respectively.A pattern recognition model was established by combining two-dimensional spectrum algorithm and residual convolutional neural network.Result:The model based on synchronous two-dimensional correlation spectrum achieved 100%classification accuracy when the number of convolution layers was above 26,while the model based on asynchronous two-dimensional correlation spectrum achieved less than 30%classification accuracy.Conclusion:It shows that the model can be applied to the identification of Duzhong from different habitats.

关 键 词：杜仲 二维相关光谱 深度学习 判别模型 

分 类 号：R282[医药卫生—中药学]