Category discrimination and quality prediction of instant green tea based on the fusion of computer vision, E-nose and E-tongue

• Different machine learning algorithms were compared for classification and prediction of instant tea. • The data fusion strategies enhanced the classification and prediction accuracy for instant tea. • The effective feature selection enabled high-precision predictions of instant tea quality indicators. This study developed an integrated multi-sensor framework combining computer vision, electronic nose (E-nose), and electronic tongue (E-tongue) for category identification and quality prediction of instant green tea. The correlation between chemical components and features of different instant green teas was first established. Individual intelligent technologies combined with machine learning algorithms (SVM, KNN, and RF) were used to construct variety classification models. and prediction models for chemical components of instant green tea were developed using machine learning algorithms (PLSR, SVR, and RF). Computer vision manifested a significant advantage in the classification of instant green tea compared to the E-nose and E-tongue, especially for tea powder. Meanwhile, computer vision could effectively capture color information in instant green tea, while E-tongue could obtain taste information, better reflecting the characteristics of non-volatile substances. Feature-level fusion and decision-level fusion models were developed and different feature selection methods were introduced in category identification and quality prediction of instant green tea for establishing a better evaluation platform. The results showed that feature fusion of computer vision and E-tongue achieved superior classification accuracy using RF than single-modality approaches. Particle Swarm Optimization (PSO)-enhanced RF models exhibited robust predictive capability (R² > 0.94) for catechins, caffeine, and amino acids. This work established a rapid and non-destructive strategy for instant tea variety classification and quality prediction, highlighting the synergistic potential of multi-sensor systems combined with optimized machine learning.