Polygonatum kingianumの九蒸九烘加工品のFT-NIRおよびATR-FTIRと深層学習を組み合わせた迅速同定

The nine-steaming and nine drying process is the traditional preparation method for Polygonatum sibiricum, involving repeated steaming and drying nine times to optimize its dual medicinal and edible value, making it the preferred technique. Though costly, unscrupulous merchants cut corners by reducing production steps to lower costs, yet the differences are nearly undetectable to the naked eye. Therefore, it is urgently necessary to establish a rapid and accurate identification method. This study employed a combination of Fourier transform near-infrared (FT-NIR) spectroscopy and attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy with deep learning to identify nine processed samples of nine steaming and nine drying of Polygonatum kingianum. The results demonstrated that the residual neural network (ResNet) model based on synchronous two-dimensional correlation spectroscopy (2DCOS) images exhibited stable performance and high accuracy. Both the test set and training set achieved an accurate rate of 100 %. Subsequently, to explore the critical factors in different treated samples, we employed principal component analysis (PCA) to interpret the spectral variables and identified that the key variables were related to carbohydrates. Therefore, we analyzed the sugar metabolites of nine treatments using gas chromatography-mass spectrometry (GC-MS) and identified d-Xylulose, d-Ribose, d-Arabinose, and l-Rhamnose as the key metabolites distinguishing the ninth steaming of Polygonatum kingianum. Ultimately, the rapid identification of Polygonatum kingianum processed by the traditional "nine steaming and nine drying" method based on FTIR and deep learning was achieved, and its identification features were analyzed using metabolomics.