Deep Learning for Meal Recognition and Calorie Estimation

Abstract

Accurate calorie estimates from foods are prerequisite for diet following and health monitoring. Manual calorie estimations according to age-old methods mostly tend to be inaccurate. This paper proposes the use of convolutional neural networks (CNNs) for precise identification from food images and prediction of meal calories to solve the concern. Therefore, the objective is to create a model capable of recognizing foodstuff besides estimating their caloric content. Developing a model that could correctly identify food ingredients and calculate their energy value through training and testing was important in this project. Our aim here was to verify the accuracy of the model using systematic reviewing means as well as an interface where it can be tested. A dataset of 1,337 high-quality images divided into 12 culinary classes cake, hamburger, noodles, spaghetti, pizza, chicken curry, croissant, French fries, fried chicken, roast chicken, lobster nasi goreng, and waffle was obtained from Roboflow Universe and used for this project. The selection of technique which is YOLO (You Only Look Once) model architecture and flow design because it proved to be highly efficient for real-time object recognition.