Research on qualitative discrimination of black tea rolling degree and quantitative analysis of tea pigment content based on near-infrared spectroscopy technology combined with multi-layer perceptron

Rolling is the key step linking withering and fermentation in black tea processing, directly affecting the formation and transformation of tea pigments. This study aims to realize the rapid detection of tea pigment content and objective evaluation of rolling degree suitability during rolling based on near-infrared spectra. Firstly, the spectral data of different rolling stages are collected. After preprocessing and feature wavelength extraction, the key spectral information related to tea pigment content is screened out. Secondly, the linear, nonlinear, and neural network models are constructed and compared, and the non-destructive detection of theaflavins (TFs), thearubigins (TRs), and theabrownins (TBs) is successfully realized. Meanwhile, the water bath method combined with machine vision is used to obtain the results of cell breakage rate as the classification standard of rolling degree. The feature wavelengths related to rolling degree are screened and the classification model of rolling degree is constructed based on multi-layer perceptron (MLP). The results show that MLP performs well in both quantitative prediction and qualitative classification tasks. The Rp values of testing set of prediction models are higher than 0.990, and the accuracy of classification model is 0.983. This study provides a potential technical solution for objective evaluation of rolling quality.