Integration of generalized additive models and hyperspectral imaging for quality prediction in Chaoshan beef...

Maintaining consistent quality of Chaoshan beef meatballs (CBMs) is challenging due to the intricate production processes and stringent raw material requirements, with no effective method currently available for real-time monitoring during production. In this study, a comprehensive "raw materials-processing-quality-sensory" network model to predict CBMs quality attributes by integrating generalized additive model (GAM) with (HSI) was developed. GAM showed strong predictive performance for key attributes such as the pH (R² = 0.837), salt-soluble protein content (R² = 0.909) of the beef batter, and the proportion of free water in CBMs (R² = 0.850). However, the model was less accurate for predicting for the hardness of CBMs (R² = 0.781) and total sensory score (R² = 0.772). To address this, HSI technology was introduced to enhance the generalization capability of GAM. The optimal models using HSI for predicting pH and salt-soluble protein content of beef batter were SNV-CARS-GLGCM-BP-ANN (R_p2 = 0.959, RMSE_p = 0.035) and SNV-CARS-GLGCM-PLSR (R_p² = 0.983, RMSE_p = 0.101), respectively. Incorporating these predictions into the GAM improved predictive accuracy for CBMs hardness and total sensory score, increasing from 78.1 % and 77.2 % to 83.0 % each. This work provides a theoretical basis for optimizing CBMs processing and real-time food quality monitoring.