Light-Weight Food Image Classification For Egyptian Cuisine

Abstract

Food is an integral aspect of daily life in all cultures. It highly affects people's diets, eating behaviors, and overall health. People with poor eating habits are usually overweight or obese, which leads to chronic diseases such as diabetes and cardiovascular disease. Today, the classification of food images has several uses in managing medical conditions and dieting. Deep convolutional neural network (DCNN) architectures provide the foundation for the most recent food recognition models. However, DCNNs are computationally expensive due to high computation time and memory requirements. In addition, food recognition of Egyptian cuisine is less well-studied than other broad food categories or particular cuisines. This study focuses on incorporating dishes from Egyptian cuisines into a lightweight food recognition model to save computational power. Accordingly, the proposed lightweight model (Enhanced MobileNet) achieved top-1 accuracy of 75%, which exceeded other related work in this area.