Related Experiment Video
Updated: Aug 5, 2025
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
9.3K
Capturing children food exposure using wearable cameras and deep learning
Shady Elbassuoni1, Hala Ghattas2,3, Jalila El Ati4
1Computer Science Department, American University of Beirut, Beirut, Lebanon.
PLOS Digital Health
|March 27, 2023
Summary
This study introduces a novel machine learning system using wearable cameras to objectively track children's food exposure, overcoming limitations of self-reported dietary data. The system accurately identifies and categorizes food-related visuals, offering a more reliable method for understanding environmental influences on children's eating habits.
Area of Science:
- Nutrition Science
- Computer Science
- Public Health
Background:
- Children's dietary habits are shaped by multifaceted environmental factors.
- Traditional data collection methods rely on self-reports, which are susceptible to recall bias.
- Objective methods are needed to accurately assess children's food exposure in their daily environments.
Purpose of the Study:
- To develop and evaluate a machine learning-based system for objectively capturing children's food exposure.
- To assess the cultural acceptability of using wearable cameras for food exposure data collection in urban Arab settings.
- To train machine learning models for identifying and classifying food-related images from wearable camera footage.
Main Methods:
- A user-centered design study was conducted to assess the acceptability of wearable cameras among school children in Beirut and Tunis.
- A machine learning pipeline was developed, including models for food image detection, classification of food-related content (items, ads, outlets), and consumption identification.
- Models were trained using web data, deep learning techniques, public datasets, and crowdsourced information.
Main Results:
- The study reports on the user-centered design findings regarding the acceptability of the wearable camera system.
- The development and training process for the machine learning models designed to detect and classify food exposure are detailed.
- The system's successful integration and performance in a real-world case study are presented.
Conclusions:
- The developed machine learning system offers an objective and culturally acceptable method for collecting data on children's food exposure.
- This technology can provide valuable insights into environmental determinants of children's dietary habits, moving beyond self-reported data.
- The system's components and their performance in a real-world setting demonstrate its potential for future research and public health interventions.
