Research On Risk Ealy Warning Of Typical Food Safety Based On Machine Learning

In recent years,China has continuously promoted the development of domestic food quality and safety monitoring and early warning,greatly reduced the occurrence of food safety incidents,and ensured the food safety and health of the people as much as possible.However,with the rapid development of food production technology,there are more unprecedented safety events.Food safety is still a major issue related to people’s livelihood,which is highly concerned and valued by the government and the people.In the two typical food of Chinese liquor and infant powder,the traditional Chinese liquor quality identification methods include taste,electronic nose,and the chromatograph.However,these methods are time-consuming and laborious,and have some limitations.The quality and safety early warning methods of infant powder are often highly subjective.Only relying on some detection data of relevant institutions,the safety early warning performance needs to be improved.In view of the two typical food safety problems,such as fake Chinese liquor that are represented by industrial alcohol blending liquor and infant powder,this paper studied the classification of quality of Chinese liquor and the early warning of infant powder safety risk based on machine learning.The main work is as follows:1.Aiming at the classification of quality of Chinese liquor,a deep learning Li Net network combining the Generative Adversarial Network WGAN-GP and CNN-based SAE was proposed.The network is mainly composed of two parts: data enhancement part and the classifier network.First,the WGAN-GP learned data distribution of the ion migration spectrum of Chinese liquor,and the data with the same distribution were generated to achieve the purpose of data enhancement.Secondly,based on the traditional Stacked Auto Encoder(SAE),a CNN-based SAE network suitable for the data of ion migration spectrum of Chinese liquor was designed by one dimension CNN network.Experimental results showed that compared with the existing temporal classification network,the Li Net network proposed in this paper had better classification performance on the Chinese liquor data set.2.A infant powder risk level early warning method of STL-XGBoost based on temporal series decomposition(Seasonal-Trend Decomposition Procedure Based on Loess,STL)was proposed.Firstly,the sampling data of pollutants in infant powder were decomposed into trend component,seasonal component and residual component by STL method,and the risk level of pollutants was determined by the softmax method in combination with the national detection standard.At the same time,this paper proposed to use entropy weight to determine the weight of different pollutants on the risk level of infant powder.Then,XGBoost was used to train the infant powder data set.And we utilized Bayesian optimization to get the best parameters of XGBoost.Finally,combined with the risk grade and pollutant weight predicted by the model,the pollution risk grade of infant powder was given.The experimental results showed that compared with the current machine learning methods,the STL-XGBoost method had better early warning performance on the infant powder data set.