目的 探讨傅里叶变换近红外（fourier transform near infrared，FT-NIR）光谱技术在滇黄精Polygonatum kingianum质量评价中的应用，基于不同干燥处理方法、不同产地滇黄精的一维光谱与二维光谱构建鉴别模型，为滇黄精质量评价提供新方法。方法 研究采集了198份不同干燥方式和产地的滇黄精样品，将其进行不同干燥处理，分析产地、干燥处理方法等因素对滇黄精质量的影响；采集并分析滇黄精FT-NIR光谱和二维相关光谱（two-dimensional correlation spectroscopy，2DCOS）特征，采用主成分分析（principal component analysis，PCA）和层次聚类分析（hierarchical clustering analysis，HCA）可视化样品的分类趋势，并基于FT-NIR光谱建立偏最小二乘判别分析（partial least squares discriminant analysis，PLS-DA）和支持向量机（support vector machine，SVM）2种传统模型；基于同步、异步和综合2DCOS图像建立残差神经网络（residual neural network，ResNet）模型，比较3种模型的整体鉴别能力和模型性能。结果 同步2DCOS图像结合ResNet可以成功鉴别滇黄精不同产地和干燥处理方法，准确率达到100%，整体鉴别能力和模型性能均显著优于PLS-DA和SVM传统模型。结论 基于FT-NIR光谱构建的ResNet鉴别模型可以快速准确鉴别滇黄精的产地和干燥处理方法，为滇黄精质量的快速检测提供科学依据。

Objective To explore the application of Fourier transform near infrared (FT-NIR) spectroscopy technology in the quality evaluation of Dianhuangjing (Polygonatum kingianum). Based on different drying treatment methods and one-dimensional and two-dimensional spectra of P. kingianum from different origins, a discrimination model was further constructed to provide a new method for the quality evaluation of P. kingianum. Method In this study, a total of 198 samples of P. kingianum from five regions of Yunnan Province were collected and subjected to different drying treatments in order to analyze the effects of factors such as different origins and drying treatments on the quality of P. kingianum. Collect and analyze FT-NIR spectra and two-dimensional correlation spectroscopy (2DCOS) features of P. kingianum, and then use principal component analysis (PCA) and hierarchical clustering analysis (HCA) to visualize the classification trend of the samples. Based on FT-NIR spectra, establish two traditional models: partial least squares discriminant analysis (PLS-DA) and support vector machine (SVM). Establish residual neural network (ResNet) models based on synchronous, asynchronous, and integrative 2DCOS images to compare the overall discrimination ability and model performance of the three models. Results Synchronized 2DCOS images combined with ResNet can successfully identify the different origins and drying treatments of P. kingianum, with an accuracy of 100%, and the overall identification ability and model performance are significantly better than that of PLS-DA and SVM traditional models. Conclusion The ResNet identification model based on FT-NIR spectroscopy can rapidly and accurately identify the origin and drying method of P. kingianum, and provide a scientific basis for the rapid detection of the quality of P. kingianum.

R286.2

云南省重大科技专项计划（202502AS100009）；云南省农业基础研究联合专项重点项目（202401BD070001-018）