Deepfood: Food Image Analysis And Dietary Assessment Via Deep Model

Abstract

With the growing importance of dietary awareness
and health monitoring, automated food recognition
and nutritional assessment have become critical
areas of research. This project, titled DeepFood,
introduces a deep learning-based system for food
image analysis and dietary evaluation using a
Region-Based Convolutional Neural Network
(Faster R-CNN) with a VGG16 backbone. The
system is trained on the UECFOOD 100 dataset,
which includes annotated food images with single
bounding boxes, allowing the model to learn
region-specific features for accurate food
classification. The application comprises several
modules, including dataset upload, data
preprocessing, model training, performance
visualization, and real-time food classification from
images. Upon classifying the food item, the system
retrieves and displays its dietary details to the user.
Achieving a classification accuracy of 92%, the
system demonstrates the effectiveness of Faster RCNN
in food detection tasks. Although limited to
identifying one food item per image due to dataset
constraints, the model can be extended to handle
multiple food items if multi-bounding-box datasets
become available. This system has significant
potential for health monitoring, diet tracking, and
automated food logging in real-world applications