Food Image Classification Based On Self Supervised Preprocessing

With the rapid development of social networks,people usually upload,share and record food images,so the application value of food image classification is also increasing,which has a positive impact on nutrition collocation,food recommendation,catering,social and other aspects.Although food image classification has great application potential,it is still a challenging task to recognize food from images.The challenges mainly come from three aspects: 1)lack of large-scale data sets for food recognition;2)Different kinds of food may be very similar in appearance,and the similarity between them is very high;3)A food may have thousands of different appearances,but it is essentially the same food.In order to solve the problem of fine-grained food recognition,this paper proposes a food image classification model based on self supervised preprocessing,and designs and implements a food recognition system based on Android system.The main contents and contributions of this paper are as follows:(1)This paper summarizes the research status of food image classification technology and self supervised learning,and systematically introduces the basic theories of convolution neural network and self supervised learning.(2)In order to solve the problem that the existing food image classification technology can not obtain and distinguish the detailed features of the image well,this paper proposes a food image classification method based on self-monitoring preprocessing.The method model is built on the basis of Dense Food,a food image classification model based on dense connected network,and adopts the self-monitoring strategy of context recovery,the trained network weights are used to initialize Dense Food model,and training fine-tuning is used to complete the classification task.The self-monitoring strategy of context recovery and dense connection network are both focused on the extraction of image features.This paper combines the two methods to fully learn the food image features to achieve higher accuracy of food image classification.For performance comparison,VIREO-172 data set is used to train four food image classification models: self supervised preprocessing based food image classification model,non preprocessed food image classification model Dense Food,Dense Net based on Image Net data set training preprocessing and Res Net based on Image Net data set training preprocessing.The experimental results show that the accuracy of Top-1 and Top-5 is 84.25% and 96.97% respectively,which is better than other strategies.(3)This paper designs and implements a food recognition system based on Android system,mainly studies the specific implementation of the food image recognition system,including the introduction of the system development platform,the deployment of the food classification network model on Android system,the design of the system function module and the specific implementation sequence diagram,etc.Finally,it shows the test effect of the food recognition system function.