| Milk and dairy products have become an important part of the daily diet.However,the lacking of rapid detection of the total number of the dairy products during the industrial procession and the means for identifying the quality of the colostrum led to the frequent occurrence of milk safety incidents resulting from the total number of bacteria.Some of the most common microbial detection methods used are time-consuming,complicated operation and expensive equipment.Therefore,UV-Vis spectroscopy is applied to the detection of the total number of milk microflora in this paper,which provides a new idea for the detection of dairy microorganisms.In this study,the characteristics of milk bacteria were studied.UV-Vis spectroscopy combined with chemometrics was applied in the field of quantitative analysis of the total bacterial population in milk.The plate count method was used for calibration.The original spectrum was de-noised and smoothed by Savitzky-Golay smoothing.For the internal particles in milk,light scattering phenomenon was generated,which was corrected by multivariate scatter correction and standard normal variable transformation.In order to solve the problem of small sample size and the non-linearity caused by the complex composition of milk,a quantitative model of the total number of milk microflora was established by using partial least squares and error reverse artificial neural networks respectively.The specific research contents are as follows:The feasibility of UV-Vis spectrum detection of milk bacteria.The total number of bacteria in milk incubated at constant temperature(0,2,4,…,20h)at 32 is used ℃as the research object.The UV-Vis spectrum was used to reflect the growth trend of the milk and the plate count method as standard.And the spectral bands and other methods were extracted to verify the feasibility of detecting the total number of milk microflora by ultraviolet-visible spectroscopy.Explore the mathematical relationship between spectral information and milk bacterial content.Partial least squares was used to establish the quantitative relationship between the original spectrum and the pretreated spectral data and the total population of bacteria.The results show that the parameters of partial least squares model established by smoothing and de-noising spectral data are better than the original spectral data and other preconditioned models were established,but the root mean square error was too large and the prediction was not effective.Secondly,the quantitative model between spectral data and the total number of colonies based on the original spectra and three pretreatment methods is established by using the error reverse artificial neural network.The results shows that the model prediction effect of multivariate scatter correction is the best.The comprehensive analysis shows that the error inverse artificial neural network model with multivariate scatter correction is the best. |
