AI-Powered Mobile Application for Image-Based Food Ingredient Detection and Recipe Generation

This research explores the critical impact of dietary habits on health, addressing the lack of awareness regarding nutritional content that leads to issues such as malnutrition. The proposed mobile application utilizes the YOLOv8 algorithm with a deep convolutional neural network (CNN) architecture, deployed on the Roboflow server, a local MySQL database server, and ChatGPT. The architectural framework of this application is integrated through the NodeJs server via an API system embedded within the Android application. Despite this approach, the study highlights the need for further refinement in the accuracy of the ingredients’ object detection model. The refined YOLOv8 model, trained on an extensive dataset of 24,583 images across 13 food ingredient classes, outperforms the initial YOLOv8 model trained on 9,867 images. Improvements are notable, particularly in addressing misclassifications of ingredients such as beef, pork, onion, and egg in the initial model. However, the YOLOv8 model can still be improved by introducing more ingredient variations and expanding the training dataset. This application can detect multiple food ingredients from an image and evaluate their nutritional value, specifically in calorie count. It also provides tailored recipes to help users meet their daily nutritional requirements, offering a practical solution for users to achieve their dietary goals. This research connects computer vision and nutrition, providing insights for developing practical solutions in this domain.