Related Experiment Video
Updated: May 20, 2025
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
8.9K
Improved food recognition using a refined ResNet50 architecture with improved fully connected layers.
Pouya Bohlol1, Soleiman Hosseinpour1, Mahmoud Soltani Firouz1
1Department of Agricultural Machinery Engineering, Faculty of Agricultural Engineering and Technology, University of Tehran, Karaj, Iran.
Current Research in Food Science
|March 24, 2025
Summary
This study developed a deep learning system using ResNet50 to accurately identify food items from videos, aiding in health impact analysis in dining settings. The optimized model achieved 97.25% accuracy in food recognition.
Area of Science:
- Computer Science
- Artificial Intelligence
- Food Science
Background:
- Food consumption significantly impacts human health, influenced by factors like quality, quantity, freshness, and color.
- Accurate identification of consumed food is crucial for evaluating health impacts, especially in controlled environments like hospitals and restaurants.
Purpose of the Study:
- To develop and evaluate a machine vision system utilizing deep learning for food identification across 16 categories.
- To assess the impact of food consumption on human health by analyzing meal data from hospital and restaurant settings.
Main Methods:
- A dataset of 12,000 food images was created and augmented to 66,000 images.
- Five deep learning algorithms were employed, with ResNet50 showing superior performance.
- Hyperparameter tuning and transfer learning were used to optimize the ResNet50 model, including a customized fully connected layer.
Main Results:
- The optimized ResNet50 model achieved 97.25% accuracy and 0.2 loss in food recognition.
- The best-performing model utilized the Adam optimizer, an initial learning rate of 10⁻³, batch size of 4, and image size of 340x640.
- Training and response times were 5.30 hours and 1.2 seconds, respectively.
Conclusions:
- ResNet50, particularly with a customized fully connected layer, is highly effective for accurate and efficient food recognition.
- This technology can support health impact assessments by providing precise data on food consumption in various settings.
