A rapid method for detecting l-Theanine during kombucha fermentation using SERS combined with machine/deep learning

Abstract

The rapid detection of key indicators during the fermentation is crucial for improving product quality. In light of this, a strategy was developed for the rapid detection of l-Theanine during kombucha fermentation using surface-enhanced Raman Scattering (SERS) coupled with machine/deep learning methodologies. After analyzing the attribution of l-Theanine SERS characteristic peaks, effective variable features, converted features, and depth features were extracted, and calibration models were constructed employing Partial Least Squares (PLS), Support Vector Machine (SVM), and a one-dimensional Convolutional Neural Network (1DCNN). Based on the first 10 principal components derived via principal component analysis (PCA), the established model, PCA10-1DCNN, displayed superior performance with a determination coefficient (Rp2) of 0.9750. PCA10-1DCNN achieved commendable verification efficacy, with a maximum error of 4.4460 µg/mL, indicating the efficacy of this strategy for the l-Theanine rapid detection. This study offers a viable strategy for the l-Theanine rapid detection in the production of tea-based beverages.