Establishment of a Predictive Model for the Quality Assessment of Chilled Meat Using a Moth-Flame Optimization BP Neural Network

To quickly and accurately evaluate and predict the quality and safety of food products, in this study, storage temperatures (0, 4 and 25 ℃) were used to investigate the effects on the total number of colonies, TVB-N, pH, moisture content, color, and biogenic amine content of chilled meat. It aimed to accurately predict the pattern of change in the quality of chilled meat in storage, as well as its quality and safety, and determined the characteristic quality indexes of chilled meat. Based on Backpropagation (BP) neural network and Moth-Flame Optimization (MFO) BP neural network, the quality prediction model of chilled meat under different storage temperatures was constructed using characteristic indexes as the training data to quickly and accurately evaluate and predict the quality and safety of food. The results showed that the total number of colonies, pH, TVB-N, color and biogenic amine content of chilled meat under different storage temperatures showed an increasing trend with the extension of storage time (P<0.05), and the pattern of change of each index under different storage temperatures was inconsistent, and the higher the temperature, the faster the rate of corruption and deterioration. Based on the results of this part of the experiment, the total number of colonies and TVB-N as feature indicators were used as training data by the Backpropagation (BP) neural network and Moth-Flame Optimization (MFO) BP neural network to construct a quality prediction model for chilled meat under different storage temperatures. The findings demonstrated that the MFO-optimized BP neural network outperforms the standalone BP neural network model. Specifically, the R-values for the colony count and TVB-N indicators achieved through training with the BP neural network model were 0.95018 and 0.94283, respectively. In contrast, the R-values obtained through the optimized MFO algorithm reached 0.97538 and 0.98001, respectively, which closely approached 1. Furthermore, the optimized values of RMSE, MSE, and MAE indicated smaller discrepancies, indicating a better fit of the model. Consequently, the MFO-optimized BP neural network exhibited superior prediction performance throughout the entire storage duration, showcasing heightened accuracy. Hence, this model can effectively forecast the patterns of change in chilled meat quality during storage.