Image Classification and Segmentation for Estimating Calories of Food
54
Issued Date
2025-01-01
Resource Type
Scopus ID
2-s2.0-105004560576
Journal Title
10th International Conference on Digital Arts, Media and Technology, DAMT 2025 and 8th ECTI Northern Section Conference on Electrical, Electronics, Computer and Telecommunications Engineering, NCON 2025
Start Page
525
End Page
530
Rights Holder(s)
SCOPUS
Bibliographic Citation
10th International Conference on Digital Arts, Media and Technology, DAMT 2025 and 8th ECTI Northern Section Conference on Electrical, Electronics, Computer and Telecommunications Engineering, NCON 2025 (2025) , 525-530
Suggested Citation
Chanthanuwatkun U., Jitpattanakul A., Hnoohom N. Image Classification and Segmentation for Estimating Calories of Food. 10th International Conference on Digital Arts, Media and Technology, DAMT 2025 and 8th ECTI Northern Section Conference on Electrical, Electronics, Computer and Telecommunications Engineering, NCON 2025 (2025) , 525-530. 530. doi:10.1109/ECTIDAMTNCON64748.2025.10962099 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/110122
Title
Image Classification and Segmentation for Estimating Calories of Food
Author(s)
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Accurate estimation of calorie content in meals is essential for promoting healthy dietary practices and supporting healthcare efforts. However, the lack of nutritional knowledge poses challenges for individuals in estimating the calorie content of foods. This study aims to develop a deep learning-based model for estimating the calorie content of real foods through image classification and segmentation techniques. Image classification was employed to identify various one-dish Thai menus, while image segmentation categorized food components into fats, carbohydrates, and proteins. A new dataset, consisting of 15 one-dish Thai menus, was created using a combination of publicly available datasets (THFOOD-50 and FoodyDudy) and self-collected images. Nutritional analysis was based on the Thai nutritional table from the Bureau of Nutrition, Department of Health, Ministry of Public Health. We compared the performance of YOLOv5x, YOLOv8x, and YOLOv9e on the ThaiFood dataset using identical hyperparameters. The results showed that YOLOv8x achieved the highest mAP@0.5 of 0.883, with an accuracy of 0.829 and a recall of 0.902. YOLOv5x closely followed with a mAP@0.5 of 0.881, an accuracy of 0.822, and a recall of 0.945. YOLOv9e demonstrated competitive performance, achieving a mAP@0.5 of 0.846, an accuracy of 0.775, and a recall of 0.91. These findings highlight YOLOv8x as the most suitable model for Thai food classification and detection tasks under the evaluated conditions.