A Deep Learning-Powered Web Service for Optimal Restaurant Recommendations Based on Customers Food Preferences

The surge in food images on social media demands the advancement of effective classification algorithms for applications like restaurants recommendations, personalized health management, nutrition analysis, and dietary monitoring. Recently, food classification has witnessed substantial progress through Deep Learning driven by the availability of large-scale food datasets and enhancements in Deep Learning models. In this paper, a web service designed to assist users in identifying the optimal restaurant for their desired dish based on quality, price, and location parameters utilizing a Deep Learning (DL) engine for food image classification. The research focuses on evaluating the efficiency of several DL algorithms, namely You Only Look Once (YOLO) V8, YOLO V5, ResNet 50, ResNet 18, Inception V3, VGG 16 and MobileNet, utilizing a Jetson Nano board for training purposes (100 epochs). Food 101 dataset used in the training process while LabelImg tool is employed for annotation. The annotated version of the dataset used for training of YOLO V8 and YOLO V5. The findings reveal YOLO V8 attains a notable accuracy of 96.3%, surpassing YOLO V5, ResNet 50, ResNet 18, Inception V3, VGG 16 and MobileNet, which achieve 89.7%, 89.35%, 67.23%, 76.01%, 78%, and 57.90% accuracy, respectively. Consequently, the research advocates for the adoption of YOLO V8 as the optimal DL algorithm for the proposed web service. Future enhancements will include the integration of textual data as a feature to enhance the efficiency of the detection process. The web service will be deployed on Amazon Web Services (AWS) which opening up possibilities for expansion through the development of a mobile application accessible on platforms such as the App Store or Google Play.