Prediction of multi-task physicochemical indices based on hyperspectral imaging and analysis of the relationship between physicochemical composition and sensory quality of tea

Abstract

Tea is highly valued by consumers worldwide for its distinctive flavor and rich nutritional profile. Efficient and accurate assessment of tea quality is essential for both producers and consumers. This study focuses on Yongchuan Xiuya green tea and utilizes hyperspectral imaging (HSI) technology integrated with a multi-task regression (MTR) model to simultaneously predict 12 physicochemical indices (WE, SSC, FAA, TP, CAF, EGCG, EGC, EC, ECG, GA, C, GC). To develop this model, the relationship between sensory attributes and physicochemical components was first analyzed, identifying key quality indicators. The original spectral data were preprocessed using the SNV-SG method to enhance data quality. The predictive performance of various models, including partial least squares regression (PLSR), random forest (RF), and extreme gradient boosting (XGBoost), was evaluated, with XGBoost identified as the most effective. Subsequently, the Newton-Raphson-Based Optimization (NRBO) algorithm was employed to optimize the parameters of XGBoost, forming the foundation of the MTR model. By incorporating feature enhancement and correlation analysis, the MTR model effectively predicted multiple quality indices. The model exhibited high predictive accuracy, as indicated by an average R_P² of 0.9774 and an average RMSEP of 0.1097, demonstrating its robustness and reliability in assessing tea quality.