[2U6-IS-1c] Knowledge engineering

[2U6-IS-1c-01] Direct calorie classification of West African dishes with a small dataset

〇Michel Ezoa Avotchi DJANGORAN¹, KIKUCHI MASATO¹, TADACHIKA OZONO¹ (1. Nagoya Institute of Technology)

[[Online, Working-in-progress]]

Keywords:West Africa dishes, calorie classification, calorie estimation, food recognition

Diabetes, obesity, and high blood pressure are becoming common diseases that the African healthcare system is fighting. These diseases have been linked to a diet that is often either too rich or too poor in certain nutrients.
In addition, many Africans want to track their diet to lose weight or stay healthy, but they struggle to find an application or a good dataset with information on the daily dishes and calories that are specific to their cuisine.
This paper proposes a new, direct, real-time calorie rating system for daily dishes in West Africa. Our first study was conducted on 11 types of Ivorian daily dishes, using a dataset of 636 images. Our first baseline system groups dishes into calorie classes (high, medium, and low) and use YOLOv5 (You Only Look Once) for dish detection and calorie classification, with a small amount of data. With 88.8 % of accuracy, our model can be used in mobile applications to support the African healthcare system, which suffers from a lack of dieticians.