Hybrid Deep Learning Algorithm-Based Food Recognition and Calorie Estimation

Every individual requires some sort of system that informs them about portions and calories of food, as well as providing them with directions on how to consume it. In our study, we propose a hybrid architecture that makes use of deep learning algorithms to forecast the number of calories in various food items on a bowl. This consists of three major components: segmentation, classification, and calculating the volume and calories of food items. When we use a Mask RCNN, the images are first segmented. Using the YOLO V5 framework, features are collected from the segmented images and the food item is categorized. In order to determine the dimensions of each food item, we identify the items first. In order to calculate the quantity of the food item, the estimated dimension must be used. The calories are then computed using the food item's volume. The aforementioned approaches, which were trained on the dataset's food images, that correctly identified and forecasted a food item's calories had an accuracy of 97.12%. To Provide directions on how to consume food is expected by individual and will be completed after knowing intake of volume of food.

Automated summary

In this study, a hybrid architecture that utilizes deep learning algorithms is proposed to forecast the number of calories in food items on a bowl. The approach consists of segmentation, classification, and calculating the volume and calories of the food items. The accuracy of identifying and forecasting the calories of food items using this method reaches 97.12%.