Study On Hyperspectral Rapid Detection Of Preservatives Commonly Used In Pure Milk

At present,the development of China's dairy industry is in a critical period from quantity expansion to quality efficiency transformation.Pure milk is a kind of common milk drink.By ensuring the quality and safety of pure milk,consumers' health interests are helped to be safeguarded,and the dairy industry is conducive to being pushed towards higher quality.In China,preservatives are banned from pure milk.Ultra-high temperature instantaneous sterilization is usually used to extend the shelf life.High cost and strict process are required.So preservatives are illegally added to the process of producing milk by some illegal vendors to replace the strict sterilization process and reduce the cost.The health of consumers is seriously endangered.However,long and complicated detection cycle is needed by traditional preservative detection method.It is not suitable for the detection of a large number of samples.In view of this phenomenon and problem,the commonly used food preservatives sodium benzoate and potassium sorbate were took as the research object,and hyperspectral imaging technology was used to explore the rapid detection method of the commonly used preservatives in pure milk in this paper.The main results of this paper are as follows:(1)The factors affecting the acquisition of hyperspectral images of samples were studied.And suitable experimental conditions were determined.The spectral reflectance of different objective table color,container height and sample loading degree were compared through experiments.The appropriate experimental conditions were determined by combining with the modeling method.Samples should be filled with containers.When the color of the objective table was white,the appropriate height of the container was 3cm.When the color of the objective table was black,the appropriate height of the container was 2cm.(2)The dichotomous detection model for sodium benzoate and potassium sorbate in pure milk were established respectively.Firstly,Spectral pre-processing methods,namely standard normalized variable(SNV),savitzky-golay smoothing(SG),first derivative(FD),second derivative(SD),mean centering(MC),SNV-FD,SNV-SD,SG-FD and SG-SD,were used to reduce the noise of the original spectrum.Later,competitive adaptive reweighted sampling(CARS),sparse autoencoder(SAE),principal component analysis(PCA)and CARS-SAE were used to extract characteristic wavelengths.Finally,the detection effect of support vector machine(SVM)and kernel extreme learning machine(KELM)were compared.In order to solve the influence of super parameter setting on KELM model prediction,whale optimization algorithm(WOA)was used to optimize KELM in this study.The results showed that the optimal combination method detection model of sodium benzoate in pure milk was SNV-SAE-WOA-KELM,the accuracies of model training set and testing set were 90.5% and 90.12%,and the kappa coefficient was 0.86.The optimal combination method detection model of potassium sorbate in pure milk was SNV-CARS-SAE-SVM,the accuracies of model training set and testing set were 99.14% and 98.76%,and the kappa coefficient was 0.98.(3)The influence of concentration on the detection effect of the model was studied.And the detection effect of multi-classification model was explored.Five intervals of 0.05?1g/L,1.05?2g/L,2.05?3g/L,3.05?4g/L and 4.05?5g/L were divided according to the addition concentrations of samples with preservatives.For each interval,samples mixed with preservatives and pure milk samples were used to establish a dichotomy model.The accuracies of model training set and testing set of sodium benzoate detection model were greater than 85%,and the kappa coefficient was greater than 0.61.The accuracies of model training set and testing set of the potassium sorbate detection model in the five intervals were greater than 90%,and the kappa coefficient was greater than 0.81.The results show that among the two kinds of preservatives,detection effects of the hyperspectral qualitative analysis models for sodium benzoate and potassium sorbate were all better at different concentrations.The models had good stability and reliability.In this paper,the effects of two kinds of preservatives multi-classification detection models were explored on the basis of inter-partition research.The results showed that the detection effect of the multi-classification models was not good,and the addition degree of sodium benzoate and potassium sorbate in pure milk could not be determined.