Non-destructive determination of taste-related substances in fresh tea using NIR spectra

Non-destructive determination of taste-related substances using near-infrared spectroscopy (NIRS) is of great significance for effectively evaluating tea quality. In the present research work, NIRS (400–2400 nm) were correlated with tea polyphenol (TP), free amino acid (FAA), caffeine (CAFF), and total sugar (TS) content in 187 tea samples by processing of spectra using continuous wavelet transform (CWT). Some effective variable selection algorithms (variable combination population analysis and iterative retained information variable algorithm (VCPA-IRIV), competitive adaptive reweighted sampling (CARS), variable iterative space shrinkage approach (VISSA), bootstrapping soft shrinkage (BOSS), and genetic algorithm (GA) were used for the quantification of taste-related substances of tea samples. The developed models were established and evaluated for the content of taste substances in several varieties of tea, including PLS and the potential of some machine learning models such as Gaussian process regression, support vector regression, and random forest regression. The efficiency of the developed model was significantly enhanced with the use of CWT-BOSS-PLS for monitoring the state of each component. More than 99.34% of variables was reduced. The predicted R values for TP, FAA, CAFF, and TS were 0.6891, 0.8385, 0.6810, and 0.8638 with accuracy improved by 6%, 3%, 45%, and 8%. Overall, this study provides an important support method for the practical application of content analysis for tea taste components.