Abstract
At present, the mainstream restaurant automatic pricing system realizes automatic pricing by using the object detection technology based on deep learning to locate plate and identify its category. In order to make the accuracy to reach practical application, collecting and labeling lots of plate images with kinds of foods is required, it increases the labor and costs. This paper notes that detection and identification of the plate is different from that of the conventional object. A plate is a container, in which any foods and items can be placed, and the foods are unknown, and the fine-grained detection and identification is reached through the shape, edge and color of a plate. In order to improve the plate recognition rate, for the first time this paper releases a dataset named EP-20 which contains images of the empty plate without foods, and a testing dataset named EP-Test which is collected under actual using scenes. In addition, this paper proposes the data enhancement method based on the attention mechanism, which is that images of the empty plate are filled. This method guides the neural network to pay more attention to and learn more the features of the plate’s edge and color, and learn less the features of the food in the plate, which are used as the plate’s features. The method can achieve a highest accuracy of 89.63%, the highest accuracy can be improved by 58% compared to not using the method.
Highlights
With the development of science and technology and the improvement of human living standards, the food identification technology has received great attention, and there are various studies in the field of food image [1,2]
Most of existing object detection methods based on deep learning rely on large datasets to achieve a high accuracy
The main contributions of this paper: 1)We publish a small dataset named EP-20, which contains images of the empty plate and includes 20 classes, 2)We publish a testing dataset named EP-Test, which is collected under actual using scenes and 460 images, 3)We propose three data enhancement methods based on the attention mechanism, which are that images of the empty plate are filled
Summary
With the development of science and technology and the improvement of human living standards, the food identification technology has received great attention, and there are various studies in the field of food image [1,2]. In response to the problem of a high cost, we propose three methods for the plate automatic identification based on a small dataset, which only contains a few images of the empty plate. Because these methods use a small dataset, the workload of collection and labeling is much less. The main contributions of this paper: 1)We publish a small dataset named EP-20, which contains images of the empty plate and includes 20 classes, 2)We publish a testing dataset named EP-Test, which is collected under actual using scenes and 460 images, 3)We propose three data enhancement methods based on the attention mechanism, which are that images of the empty plate are filled.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
