Prediction of lipid oxidation and polycyclic aromatic hydrocarbons in grilled sausages based on optimized back propagation neural network and computer vision

Abstract

This study developed a model to predict multiple quality parameters of grilled sausages using a computer vision system (CVS) integrated with a back propagation neural network (BP-NN) optimized by particle swarm optimization (PSO). The non-contact characteristic of CVS allowed for the quick and non-destructive acquisition of color information, which was used as input variables. Peroxide value (POV), 2-thiobarbituric acid reactive substances (TBARS), and polycyclic aromatic hydrocarbons (PAHs) were used as output layer parameters. The TBARS, POV, and PAH4 models outperformed other prediction models in terms of accuracy, with prediction error under 10%, a maximum mean squared error (MSE) of 0.037, and correlation coefficients (R) above 0.9. In contrast, the PAH15 prediction model had a prediction error under 16%, an MSE of 0.083, and correlation coefficients ranging from 0.76 to 0.83. Furthermore, sensitivity analysis revealed the a-value as the most influential parameter for predicting lipid oxidation and PAH4 levels in grilled sausages. Overall, these results suggest that the color parameters extracted by the CVS combined with the PSO-BP-NN model have great potential for predicting lipid oxidation and PAHs formation during grilled sausage processing in a rapid and non-destructive manner.