Development of an Automatic Sushi Plate Counting Application Based on Deep Learning

After finishing a meal at a conveyor belt sushi restaurant, the staff cannot immediately process the payment for customers, resulting in a poor rate of table turnover. Therefore, this paper utilizes the deep learning and image recognition technology YOLO v8, and combines it with Android Studio using the Kotlin language to develop an app. The ScreenMatch plugin is employed to ensure that the layout of components remains consistent regardless of variations in screen resolution. Additionally, a web server and database are set up using Xampp and Ngrok is used to connect to the internet. The results obtained by YOLO v8 are displayed on a web page in JSON format, and OkHttp is employed to read the JSON data into the app, displaying the number, types, and total cost of the plates. The developed app not only reduces the errors in the plate counting, but also accelerates the processing time of checkout for staff. Moreover, it eliminates the need for a substantial cost in improving the restaurant's equipment. In order to show the efficiency of this work, we used 100 images as test data for experimental verification. The experimental results showed that the model performance achieved a Precision of 0.996, Recall of 0.998, mean average precision (mAP@50) of 0.995, mAP@50-95 of 0.767, and its calculation speed at 0.21 seconds per image.