Food Image Recognition Using CNN, Faster R-CNN and YOLO

Abstract

Keeping a track on daily dietary intake is the first step to promote good health. Counting calories helps to keep a check on dietary assessment. The objective of this research work is to help the people to move towards the automated dietary assessment so as to ensure healthy lives and encourage well-being at all ages. It has been carried out to promote a healthy lifestyle by keeping a check on one's calorie intake. Computer- aided system helps the users to accurately calculate the food type along with the portion size. Computer—aided system has many advantages over the traditional approaches. Machine learning has unfolded its potential in the various spheres and this research work has inscribed the proficiency of machine learning in food computing. Automatic dietary assessment commences with the classification of food. This study initiates with using Convolutional Neural Network (CNN), Faster R-CNN and YOLO as a classifier for food image recognition and to provide a comparative analysis between them. CNN as a classifier is used for classification of food images. Faster R-CNN has also been used in this work for food image recognition. In addition to this YOLO as a classifier has also been implemented on the food database for improving the classification accuracy. This work has been performed on UEC FOOD-100, a dataset containing images of 100 food categories. Results show that YOLO has outperformed all other combinations for classification accuracy.