Convolutional Neural Network Based Food Calorie Estimation System for Dietary Tracking

This study provides a novel approach that enhances the efficiency and accuracy of dietary tracking through automated image recognition technology. This is aimed at promoting the consumption of a healthy diet. This is important due to the fact that if the correct calories are not taken, it could lead to overweight or obesity which causes certain cardiovascular diseases like hypertension, stroke, and coronary artery disease. A dataset named Food 101 containing 101,000 food images was sourced from Kaggle.com and was split into 70% for training, 15% for testing, and 15% for validation. The data went through normalization, resizing, and augmentation for preprocessing. Feature extraction was then carried out using edge detection method. Two trained convolutional neural network models- ResNet50 and google gemini pro vision were used in this work. ResNet50 was used for food type recognition while google gemini pro vision was used to calculate the calories in the food(images). The overall model was deployed as a web application using python programming language. When put to test, the system was able to validate 85% accuracy, ultimately providing a valuable tool for individuals to manage their dietary intake effectively.