Multi-source data fusion enhanced online monitoring of catechins during ultrasound-assisted enzymatic hydrolysis of summer-autumn tea

Fusion of Vis-NIR, NIR and CIELAB color parameters enhances data complementarity • The mid-level fusion strategy improves the performance in multi-source data mining • Demonstrates potential for online catechin monitoring during ultrasound-assisted enzymatic hydrolysis Catechins are important contributors to the bitter and astringent taste of summer–autumn tea and play a key role in determining the sensory quality of tea extracts. Conventional analytical techniques for monitoring ultrasound-assisted enzymatic hydrolysis of catechins are often time-consuming and unsuitable for rapid analysis. Although single-spectroscopy techniques enable online catechin monitoring, their analytical performance remains constrained by noise and external interferences. To address these limitations, this study integrates visible-near-infrared spectroscopy (Vis-NIR), near-infrared spectroscopy (NIR) and CIELAB color parameters to develop three multi-source data fusion strategies, including dual-spectral fusion and low- and mid-level dual-spectral-color fusion, with the aim of enriching dataset information and enhancing feature representation capability. The mid-level fusion model combined with variable combination population analysis-partial least squares (MID-MF-VCPA-PLS) achieved coefficient of determination for prediction (R 2 p ) values of 0.9695, 0.9533, and 0.9666 for gallic acid, non-galloylated catechins, and galloylated catechins, respectively, with corresponding residual predictive deviation (RPD) values of 5.78, 4.67, and 5.52. Compared with the low-level fusion model, which yielded R 2 p values of 0.9569, 0.9250, and 0.9604 and RPD values of 4.87, 3.69, and 5.08, the proposed model demonstrated improved overall predictive capability. The proposed method enables rapid monitoring of catechin dynamics during laboratory-scale ultrasound-assisted enzymatic hydrolysis and demonstrates its potential as a monitoring approach for tea enzymatic hydrolysis. Download: Download high-res image (263KB) Download: Download full-size image