| Today,with the development of society and the improvement of people’s concept of healthy life,healthy eating has been paid more and more attention in our lives.In addition,obesity and overweight have become a global public health problem and have led to the complication of chronic diseases such as diabetes and cardiovascular disease.The follow-up chronic disease control and treatment of these patients with chronic diseases often require intervention through standardized diets.However,the current hospitals’ diet follow-up methods for patients with chronic diseases are generally questionnaire surveys and log reports.This intervention method is not only cumbersome,but also inaccurate,and has significant drawbacks.With the development of mobile Internet and computer vision technology,the demand for dietary intervention for patients with chronic diseases and other diet-related mobile applications continues to grow.In response to the above-mentioned existing phenomena,at the same time in order to cooperate with the practical application of the Big Health management system and Fatty liver intervention module of this research group.After researching and comparing the current mature classification algorithms and analyzing the application and construction technology of neural networks in the mobile Internet,this thesis completed the training of the food recognition model,and realized the deployment of its Web applications and the trial development of mobile applications.The research in this thesis can be highly packaged and transplanted,providing a direct food identification interface for other modules or related mobile application development.The main research contents and results of this thesis are as follows:(1)Based on the data of Food-101 Dataset and food-172,this thesis first amplifies and cleans the data and recalibrates the label,and establishes 262 types of food data including Chinese food,fast food,Western food,Japanese food,etc.set.Then this thesis uses Resnet-34 network to build the migration model,uses the adaptive learning rate optimization algorithm to optimize the model parameters,fine-tune the output layer neurons,and then train the model.The accuracy of the food recognition model is79.04%.Afterwards,network reasoning and performance evaluation of the obtained model are carried out,and comparative experiments are carried out with Alex Net and VGG network to compare the classification effect of the model obtained in this thesis.(2)The model obtained by the classifier is used as the input of the Web application,and the Flask service framework is used to build the application and realize the function.The part of the built local warehouse deployment is packaged into a Docker image,and then the system is made into a container according to the image,and then Deploy the container to the Heroku server to complete the deployment of the Web application.Meanwhile each module designed by its application is tested for function and performance,so as to provide corresponding interface services for mobile applications.(3)According to the research of food recognition model and Web application in the early stage,the requirements of the application of this work under the mobile Internet are analyzed.Later,the Web application was encapsulated into API interface calls,and an application demo based on the Android operating system was designed to test the feasibility of the previous work of this thesis.At the same time,the natural data set was used to test the Web applications and mobile applications.The feasibility and practicality of the identification model and Web application interface in the above research content of this thesis,and initially realize the mobile interface Applications. |
