JIITA, vol.7 no.1, p671-674, 2023, DOI: 10.22664/ISITA.2021.7.1.671
Jun-Woo Park 1) , Dong-Hoon Kwak 2) , Young-Duk Kim 3*)
1,2,3) ICT Resarch Institute of DGIST, Dageu, KOREA
Abstract: Many foods are packaged in various wrapper and distributed to people. In order to live a healthy life by eating food in a balanced manner and improving eating habits, it is necessary to identify the nutrients and ingredients of the food and intake it in a planned manner. Although ingredients are legally marked on the back of food packaging, it is sometimes difficult to grasp information in imported foods, and even the marked information is not delivered to people due to their unique package design. In this paper, we intend to improve these problems by proposing a technique that uses deep learning algorithm to recognize packaged food and deliver food information to people via smartphone applicaton. Because the performance of smartphones used by people is various, implementing deep learning internally causes a problem of performance difference between users, which does not guarantee good detection results. To this end, a separate system was established based on server and client model. First, a server system is build up for clasifying packaged food using deep learning. Sencond, an andorid appplication is implemented to capture food image data by the user. When the application(client) transmits the captured image data of food to the remotely located server, the server matches the recognized result with a predefined food information from database. Then it delivers detailed food information to users.
In addition, a gateway server is also designed for multiple user accesses with their data queries. The performance result has been demonstrated that accurate food information is delivered to the client. Future research will increase the number of foods to improve performance with deep learning network design, and further study is planned for design of enhanced detection methods.
Keywords: deep learning, packaged food, classification, server processing.
Fullpaper: