A Food Monitoring System using Image Segmentation and Machine Learning

Alatawi, Mahmud I

Abstract

In recent years, the number of people requiring dietary monitoring has risen, and dietary records are an essential part of the diagnosis and treatment of many health problems. This project designs and evaluates a food monitoring system for use in institutions such as hospitals. It includes hardware for image capture and data processing, as well as software components such as a food recognition system, a food and nutrition database, and an interface to allow users to interrogate and interact with the system. Its objectives are to specify, acquire and evaluate an image capture system suitable for an institution such as a hospital; to develop and evaluate a food recognition system compatible with the image capture system; and develop and evaluate a system to estimate the amount of food eaten by comparing images before and after eating.
A novel method is developed based on image segmentation and a machine learning algorithm. The first stage is to remove any non-food object from the image: the colour technique was tested using almost 3000 images, and yielded near 98% accuracy. The second stage used the K-means++ clustering algorithm to group parts of the image into coherent regions, each assumed to be a food type; the average accuracy for all types of food was 94%. The third stage identified the foods within each segment of the image, through machine learning; the best accuracy for food classification was 98.7%. This algorithm was able to estimate food eaten with 86% accuracy. Tests indicate that this automated system could replace the paper-and-pen approach used in Hull and East Yorkshire Hospitals, and yield similar or better nutritional metrics.

Citation

Alatawi, M. I. (2019). A Food Monitoring System using Image Segmentation and Machine Learning. (Thesis). University of Hull. https://hull-repository.worktribe.com/output/4912448

• Thesis Type: Thesis
• Deposit Date: Nov 12, 2024
• Publicly Available Date: Nov 12, 2024
• Keywords: Medical engineering
• Public URL: https://hull-repository.worktribe.com/output/4912448
• Additional Information: Department of Medical Engineering; University of Hull
• Award Date: Oct 7, 2019