Freshness Identification Of Chilled Beef Based On Machine Vision And Texture Analysis

The freshness detection of chilled meat products plays an important role in solving food safety problems,protecting consumers’ rights and interests and standardizing the chilled meat market.At present,the meat freshness detection method in the market is mainly aimed at the detection of volatile base nitrogen content in meat products.This method has the defects of long cycle and high cost,so it is hard to detect the freshness of chilled meat in circulation and sale.Therefore,to satisfied the different needs of market detection scenarios,this study takes the chilled beef stored at 4℃ as the research object,and proposes two methods for freshness detection.One is suitable for the off-line detection scenario with high accuracy requirements,and constructs the image texture information fusion model to judge the freshness of cchilled beef,and the other can realize the fast on-line detection of chilled beef freshness,The convolution network is used to pick up properties of chilled beef automatically,and the freshness of chilled beef is judged by picked-up features.Firstly,Graded the freshness of chilled beef and collected texture characteristics of chilled beef.Under the 4°C and sealed storage condition,detected the total volatile base nitrogen content of chilled beef stored 11 days.According to the relevant national standards,the chilled beef was divided into three categories : fresh,sub-fresh and spoilage.Used Texture analyzer to detect texture characteristics of cold fresh beef with different freshness and constructed texture characteristic data set.Based on the texture data set,analyzed the correlation between each texture feature and freshness of chilled beef,which provides the basis of texture data set and theoretical feasibility for subsequent freshness identification.Secondly,Carried out the image acquisition of chilled beef with different freshness.Based on the task requirements,the industrial camera,lens selection and lighting equipment are set up to complete the image acquisition platform.Based on the image acquisition platform,colledted660 groups of chilled beef images stored at 4°C through experiments.Used MATLAB-GUI toolbox to design a man-machine interactive interface integrating image processing and image feature extraction,which was used to carry out image processing,feature extraction and other work.Obtained the feature data of chilled beef images and constructed the image feature data set.On the basis of image feature data set,carried out the correlation analysis between each image feature attribute and freshness of chilled beef,which provides image data set basis and theoretical feasibility for subsequent freshness recognition.Then,an information fusion model was constructed based on BP neural network,and an efficient off-line chilled beef freshness detection method based on information fusion was proposed.Information fusion mainly includes two steps.The first step is to determine the fusion data set.The determination method is to conduct principal component analysis on the collected10 image features and 5 texture features,and select the attributes with a cumulative contribution rate of 99.84% to form the fusion data set.The second step is to use the fusion data set to train the BP neural network.In order to prevent the model from entering the local optimal state,used genetic algorithm to improve the parameters of the BP neural network.The improved network can achieve a recognition rate of 97.00%.The prediction results show that the recognition accuracy has been significantly improved compared with single image feature or single texture feature.Therefore,as an off-line detection method of chilled beef freshness,this method can replace the traditional detection method,and has high reliability while ensuring objectivity.Finally,in order to adapt to different market application scenarios,proposed another fast detection method based on machine vision technology.This detection method expands the collected image data set through rotation,flip,clipping and other operations,and introduces the concept of transfer learning into the training of neural network.The experimental results show that the expansion of image data set and transfer learning can effectively alleviate the overfitting problem of small data set on complex network structure.Finally,the recognition rate of network model for chilled beef freshness is 92.8%.While ensuring a certain accuracy,its advantage is that online detection is faster.