Infrared-spectrum-effect combined with deep learning to predict the origin of Gentiana rigescens Franch.

Abstract

Gentiana rigescens Franch. (GR) is a high-value medicinal plant and is widely used as food additive and beverage. Due to the influence of the environment, the accumulation of active ingredients of GR from different origins varies and produces different brand values, which is of great significance for the certification of the GR origin. This study employs the infrared-spectrum-effect to reflect the differences among different origins. The partial least squares-discriminant analysis (PLS-DA) and data-driven version of SIMCA (DD-SIMCA) models were used to determine origin. The Residual Neural Network (ResNet) model was constructed using two-dimensional correlation spectra (2DCOS) and three-dimensional correlation spectra (3DCOS) to discriminate between different origins. Maximum Entropy (MaxEnt) was used to screen out environmental variables that have a significant effect on the accumulation of active ingredients. The conclusion is that the ResNet model based on synchronous 2DCOS and 3DCOS has better performance, the accuracy of training and test sets were 100 %.