Based on vis-NIR combined with ANN for on-line detection of bacterial concentration during kombucha fermentation

On-line detection of key indicators is helpful to manage the food production process. This study proposes an on-line detection method and system based on visible/near-infrared spectroscopy (vis-NIR) combined with artificial neural network (ANN) for detection of bacterial concentration during kombucha fermentation. Backpropagation ANN (BP-ANN), Extreme Learning Machine ANN (ELM-ANN), and Radial Basis Function ANN (RBF-ANN) models were constructed, and their random parameters were optimized using meta-heuristic algorithms. ANN models had been significantly improved after optimization, and RBF-ANN achieved the best results with the ratio of the standard deviation of verification set to prediction set (RPD) of 6.7878. The external verification effect of ANN models performed favorably, especially BP-ANN with the maximum error of only 0.0169 Au. The results show that the developed on-line detection method and system can satisfy the detection of kombucha bacterial concentration, and provide a feasible strategy for the on-line detection of key indicators during food fermentation.