Direct West African Dishes Recognition and Calorie Classification with Small Dataset

Abstract

One of the sustainable development goals (SDGs) is to ensure healthy lives and promote well-being for all, at all ages, on all continents. In recent years, diabetes, obesity, and high blood pressure have become the most common diseases that the African healthcare system is fighting, in addition to other illnesses. Furthermore, many Africans and foreign visitors who want to track their diet to lose weight and stay healthy by continuing to eat local dishes struggle to find an application or good dataset with information on the daily dishes and calories that are specific to their cuisine. This paper proposes a new system for direct, real-time calorie rating and West African daily dishes recognition. Our first study was conducted on 11 types of Ivorian daily dishes using a dataset of 718 images. Our system combines two classifiers: one for dish recognition and another for calorie estimation. The calorie classifier groups dishes into calorie classes (high, middle, low), and both dish detection and calorie classification use YOLOv5 (You Only Look Once) on small dataset. After training, our different models dish recognition and calorie classification achieved accuracies of 94.8% and 90.0%, respectively.