Automatic food recognition based on efficientnet and ResNet

Abstract

Automatic Food recognitionbased image is an emerging topic, which aims to extract the feature of given food image and then predict its category. Compared with the ordinary image recognition task, the food recognition is more challenging due to the high similarity between subcategories and the high inter-class variability of specific category. Early food image recognition methods are mainly based on manual features such as color, shape, texture and other information, which cannot meet actual application requirements limited by insufficient feature expression capabilities. Thanks to the rapid development of convolutional neural network, the methods based on deep learning boost the recognition accuracy and speed significantly. In this paper, we construct the food recognition models based on the EfficientNet and ResNet, respectively. Extensive experiments have been conducted on Food-101 dataset. All the results show the effectiveness of our models, which can recognize various foods accurately.