Application Of Neural Networks In Food Image Classification

Catering industry is one of the pillar industries of national economy,and food materials are the key link in the process of food quality control and food safety monitoring.How to fast classify and accurately detect food is one of the core and key issues.Traditional food procurement stay on the ways by phone,SMS,WeChat way,which are communication inefficient and error-prone.The classification of food based on text classification can not meet the growing demand of business.Thus,how to quickly and efficiently classify food becomes one of the urgent problems to be solved.As a new kind of image classification method,image classification based on convolutional neural network uses different levels of the underlying portfolio to produce high-level image features,whose characteristics are well with human visual principle.It uses repetitive iterations on model to better extract image features,which can get more accurate classification and more efficient recognition speed,compared with traditional image classification,and it has special important significance for image classification.First of all,it bases on the analysis of the traditional image classification method,and selects restaurant food image classification as the research background and then studies the classification of food image based on convolution neural network.Compared with traditional image classification,image classification based on convolutional neural network uses different levels of the underlying features combination to produce high-level image features,and then constantly doing iteration over the model to extract more effective features of images.At the same time,it introduces a dropout strategy and ReLU acti,vation function to reduce over-fitting phenomenon and generation of gradient problems,then applies the AlexNet,VGG-16,and CaffeNet convolution network on food image classification,and finally tunes hyper-parameters of the network structure,making the image recognition reaches up to 90%of classification accuracy with 140ms of recognition time.For the problem that traditional manual cleaning need to spend a lot of manpower and material resources,what's worse,more time cost in the process of pre-processing,additionally,at the same time,the cleaning process itself is very easy to produce wrong results due to human factor.Therefore,we proposed an multitasking Auto-Clean convolution neural network model,this model considers data cleaning model and image classification model as two tasks,and does training on the model to realize automatic cleaning and classification,effectively solving the problem of data auto-cleaning of pre-processing phase.Finally,through comparison of the Auto-Clean multi-task convolution neural network model with single-task convolution neural network,it proves that the methods can automatically complete the work of image cleaning and get higher precision at the same time.