Food Classification Using Deep Learning: Presenting a New Food Segmentation Dataset

Abstract

Calorie estimation is of significant importance in promoting a healthy lifestyle, as it enables individuals to effectively manage their weight. Applications that calculate caloric intake by analyzing food images have the potential to save users time and effort. Consequently, the primary objective of this study is the development of a model capable of identifying food classes from images. This classification model is crucial for the first step of calorie estimation applications. While numerous food classification datasets are available online, there is a paucity of food segmentation datasets. In response to this challenge, a novel dataset for food segmentation is presented, designed to facilitate the estimation of food quantities—a critical component of the second step in calorie estimation. The performance of the MobileNetV2 model was evaluated for food classification, yielding an optimal accuracy of 93.06% and a loss of 0.31. These promising experimental results demonstrate the potential of the approach in real-time environments.