摘要
该研究采用傅里叶变换近红外光谱(FT-NIR)、衰减全反射傅里叶变换红外光谱(ATR-FTIR)及二维相关光谱(2DCOS)技术,结合化学计量学与深度学习建立偏最小二乘判别分析(PLS-DA)和残差卷积神经网络(ResNet)判别模型,对7个主产区(221份)的草果样本进行快速、准确溯源。结果表明:ATR-FTIR光谱数据经二阶导数(2nd)+标准正态变换(SNV)预处理后建立的PLS-DA模型性能最好(95.31%),但FT-NIR光谱数据的最佳预处理为2nd。基于FT-NIR和ATR-FTIR的同步2DCOS图像建立的ResNet模型不需要筛选最佳预处理和复杂的数据转换,即可达到100%的准确率。其中,基于FT-NIR数据转化的同步2DCOS图像建立的ResNet模型的迭代次数最少、耗时最短、成本最低。该研究为鉴别不同地理来源的草果提供了一种快速、准确的新方法,为草果质量等级评价体系的进一步研究奠定了基础。
In this study,Fourier transform near-infrared spectroscopy(FT-NIR),attenuated total reflection-Fourier transform infrared spectroscopy(ATR-FTIR)and two-dimensional correlation spectroscopy(2DCOS)techniques,combined with chemometric and deep learning were adopted to establish partial least squares discriminant analysis(PLS-DA)and Residual convolution neural network(ResNet)discriminant models for rapid and accurate traceability of A.tsaoko samples from seven main production areas(221 samples).The results indicated that the PLS-DA model established after the second derivative(2nd)+standard normal variate(SNV)preprocessing of ATR-FTIR spectral data showed the best performance(95.31%),but the optimal preprocessing for FT-NIR spectral data was 2nd.The ResNet model based on FT-NIR and ATR-FTIR synchronized 2DCOS images could achieve 100%accuracy without the need for optimal preprocessing and complex data conversion.Among them,the ResNet model established for 2DCOS images converted from FT-NIR had the least number of epochs,the shortest time consumption,and the lowest cost.This study provides a fast and accurate new method for identifying A.tsaoko from different geographical origins,laying the foundation for further research on the quality rating system of A.tsaoko.
作者
苏俊宇
杨绍兵
王元忠
SU Jun-yu;YANG Shao-bing;WANG Yuan-zhong(Institute of Medicinal Plants,Yunnan Academy of Agricultural Sciences,Kunming 650200,China)
出处
《分析测试学报》
北大核心
2025年第6期1043-1054,共12页
Journal of Instrumental Analysis
基金
云南省科技人才与平台计划(202405AD350072)
云南省农业联合专项-面上项目(202301BD070001-050)。
作者简介
通讯作者:杨绍兵,硕士,副研究员,研究方向:中药资源栽培,E-mail:ysb9-116@163.com;通讯作者:王元忠,博士,副研究员,研究方向:中药资源开发与利用,E-mail:boletus@126.com。
