Hyperspectral imaging-ensemble learning algorithm-based rapid quantitative detection of pH and amino acid nitrogen in...

Fermented grains (Zaopei) form the core matrix for Baijiu fermentation, with pH and amino acid nitrogen (AAN) serving as key physicochemical indicators of fermentation quality. This study aimed to achieve rapid, accurate quantitative detection of pH and AAN levels during fermentation employing hyperspectral imaging (HSI) and ensemble learning algorithms. After evaluating various preprocessing methods and identifying standard normal variate as the optimal spectral preprocessing technique, the effectiveness of competitive adaptive reweighted sampling, principal component analysis (PCA), interval variable selection optimization (IVSO), and a combined PCA-IVSO strategy was assessed in mitigating redundancy in high-dimensional data. Subsequently, Decision Tree (DT), XGBoost, and DT-Adaptive Boosting (AdaBoost) prediction models were constructed. Notably, results showed that the PCA-IVSO strategy effectively removed irrelevant variables; when paired with DT-AdaBoost, it exhibited superior performance for pH prediction (R_P² = 0.9998; root mean square error of prediction [RMSEP] = 0.0029). For AAN prediction, the DT-AdaBoost model utilizing IVSO-extracted features performed best (R_P² = 0.9999; RMSEP = 0.0003). Altogether, these findings demonstrate that HSI coupled with ensemble learning algorithms enables quantitative detection and visual mapping of pH and AAN in fermented grains, highlighting its substantial potential for real-time quality monitoring in Baijiu brewing.