Abstract:
Abstract: Silage is a type of storage fodder from green foliage crops to reduce the cost of feed and environmental pollution. The silage can be preserved by fermentation to the point of acidification. Among them, microbial growth can dominate in the silage quality. Especially, the proliferation of harmful microorganisms has also posed a great threat to crop resources, and ruminantia production, such as clostridium, acetic acid bacteria, and yeast. However, the commonly-used plate counting and turbidimetry for microbial growth in the laboratory cannot accurately characterize the growth state of silage microorganisms in time, due to tedious steps, time-consuming, and slow response rate. This study aims to effectively monitor the growth of silage microorganisms (lactic acid bacteria, acetic acid bacteria, and clostridium butyricum) separating from the silage as the indicator strains. A systematic investigation was made for the three-dimensional fluorescence spectra, the number of microbial colonies, and the absorption of 105 samples at the seven growth time points (0, 2, 4, 8, 12, 24 and 48 h). The chemometrics analysis and spectroscopic techniques were combined for the rapid screening of microbial growth. Parallel factor analysis was applied to resolve the three-dimensional fluorescence data. Back Propagation (BP) neural network was also used in the material quantitative analysis in the field of machine learning, due to its powerful nonlinear ability. The three-dimensional Synchronous Fluorescence Spectra (SFS) showed that there were two strong fluorescence peaks at about 225 and 275 nm, respectively. The main fluorescence peaks were the microbial endogenous tyrosine and tryptophan. The fluorescence intensity increased gradually with the increasing culture time, where the position of the fluorescence peak shifted the red. Meanwhile, the width of the fluorescence peak increased significantly. The parallel factor analysis showed that there was a significant difference in fluorescence information, where the characteristic wavelength Δλ was 50 nm with six components. In addition to the two characteristic peaks, there were two weak fluorescence peaks at 310-360 and 370-390 nm. The two wave peaks at 340 and 380 nm were the microbial metabolism products or acids. There was a positive correlation between the intensity of natural fluorescence peak at 380 nm during culture time. Outstandingly, there was more information on fluorescence components in the two-dimensional fluorescence spectra from the parallel factor analysis. In terms of two-dimensional spectral data, the number of microbial colonies, and the absorbance were taken as the input or the output values of the BP neural network model, respectively. The modeling was constructed for the microbial growth of different detection. The experimental results showed that the correlation coefficients of the two models were close to 1.0, and the Mean Square Error (MSE) was all very small. A very reliable model was achieved in the neural network with a high fitting ability. Therefore, the three-dimensional fluorescence spectroscopy combined with the chemometrics was feasible to monitor the microbial growth in the silage. The finding can also provide a new technical approach for the rapid determination of the fermentation silage stage.