Abstract
To achieve rapid, non-destructive, and accurate digital evaluation of core quality components in Chongqing white tea, samples manufactured by multiple tea enterprises were investigated. Wet chemical analytical methods were conducted to determine primary biochemical components, followed by principal component analysis (PCA) and correlation analysis to identify the core quality-characterizing indicators. Near-infrared spectroscopy (NIRS) coupled with back-propagation artificial neural network (BP-ANN) models incorporating three transfer functions (linear [−1, 1], logistic, and tanh) were subsequently developed, and independent external samples were used to validate model generalization. The results indicated that the cumulative variance contribution rate of the first five principal components of biochemical components reached 90.957%. Ranked by their contribution to white tea sensory quality, free amino acids (AAs) served as the pivotal quality-determining indicator (r = 0.53, p < 0.05). Multiplicative scatter correction combined with first derivative (MSC + FD) was identified as the optimal spectral preprocessing method. Synergy interval partial least squares (si-PLS) screened four characteristic spectral regions: 4377.6–4751.7 cm−1, 4755.6–5129.7 cm−1, 6262.7–6633.9 cm−1, and 7386.0–7756.3 cm−1. The cumulative contribution rate of the first three principal components derived from the selected characteristic spectra reached 97.16%. The BP-ANN model utilizing the tanh transfer function achieved optimal predictive performance (Rp2 = 0.961, RMSEP = 0.253%, RPD = 4.336) and showed reliable practical generalization capability on external validation samples (Rp2 = 0.978, RMSEP = 0.231%). This study realizes rapid, non-destructive, and digitized quantitative determination of free amino acid content in Chongqing white tea, providing robust technical support for intelligent quality control in the white tea industry.