Quality assessment of Gastrodia elata Blume (Tianma) based on Vis-NIR spectroscopy: Discrimination of harvest times and prediction of quality indicator contents

Abstract

Gastrodia elata Blume (Tianma) is widely cultivated and consumed as a food and herbal material in China. With its increasing use as a functional food, Tianma is gaining popularity in the market. However, during the cultivation process, the harvest time affects its maturity and storage time, while the total content of gastrodin (GAS) and p-hydroxybenzyl alcohol (HA) affects its nutritional value, making it difficult to control the quality of Tianma. In this study, a combination of visible-near-infrared (Vis-NIR) spectroscopy and chemometrics was used to identify the harvest time and determine the quality indicator content. A Vis-NIRRamanNet modeling method based on the architecture of RamanNet was proposed and compared with harvest time discrimination models constructed using partial least squares discriminant analysis (PLS-DA), k-nearest neighbor (KNN), support vector machine (SVM), and one-dimensional convolutional neural network (1D-CNN). The results demonstrated that Vis-NIRRamanNet exhibited superior performance, achieving accuracy rates of 0.995 and 0.973 on the training and testing sets, respectively. Subsequently, Vis-NIR spectroscopy combined with support vector regression (SVR) was employed for quantitative analysis of quality indicator content. Wavelength selection methods such as uninformative variable elimination (UVE), genetic algorithm (GA), and competitive adaptive reweighted sampling (CARS) were utilized to optimize the model. The results indicated that the CARS-SVR model provided the best prediction of quality indicator content, with root mean square error of prediction (RMSEP) and coefficient of determination (Rp2) values of 0.257 and 0.901, respectively. Therefore, the combination of Vis-NIR spectroscopy and chemometrics shows significant potential in assisting the selection of high-quality Tianma.