Food Calorie Estimation System Based on Semantic Segmentation Network

• Published in: Sensors and materials Vol. 35; no. 6; p. 2013
• Main Authors: Kong, Xiang-Yong, Sun, Xiao-Han, Wang, Yu-Ze, Peng, Rui-Yang, Li, Xin-Yue, Yang, Yi-Heng, Lv, Ying-Rui, Tseng, Shih-Pang
• Format: Journal Article
• Language: English
• Published: Tokyo MYU Scientific Publishing Division 01.01.2023
• Subjects: Artificial neural networks; Food cans; Food intake; Image contrast; Image segmentation; Machine learning; Pattern recognition; Semantic segmentation
• ISSN: 0914-4935; 2435-0869
• DOI: 10.18494/SAM4061

The food calorie estimation system (FCES) is designed to record dietary information for diabetic patients to monitor their dietary intake to estimate the number of calories they are consuming. Deep learning technologies have recently been used for FCESs. In this work, we use the neural network for the pattern recognition of food images to calculate the number of calories. In contrast to the traditional convolutional neural network, we build a semantic segmentation network model based on SegNet + MobileNet to segment the food images and extract the area feature of food images. By determining the corresponding relationship between the area feature of the food image and the food calorie value, the number of calories in the food can be estimated and realized. The experimental results show that the accuracy of food recognition reached 97.82% and that of calorie estimation was above 84.95%.