YOLOv5 Implementation for Image Classification in Indonesian Cuisine Calorie Estimation System

Abstract

In this era of continuously evolving technology, calorie counting applications have become crucial for individuals who are concerned about their eating habits and health. However, most of these applications have not fully accommodated the variety of dishes commonly consumed in Indonesia, especially the popular dishes in Java Island, which has the largest population in Indonesia. To address this limitation, this research introduces an innovative solution in the form of an Indonesian Cuisine Classification and Calorie Content Estimation System using YOLOv5 technology. In this approach, the YOLOv5 object classification technology is used to identify various types of Indonesian dishes, including eight classes such as satay, meatball soup, traditional soup, fried rice, mixed vegetables salad, fried chicken, beef soup, and beef stew. This system is not only capable of accurately classifying dishes but also provides calorie content estimation based on the composition of the classified food ingredients. The implementation of this research combines YOLOv5 to apply the Indonesian cuisine classification model using the nutrition API from API Ninjas to obtain the required nutrition data. This research uses datasets obtained from Kaggle website, Mendeley Data, and Roboflow, with a total of 303 images for each class of dishes. As a result, the model achieved an accuracy score of 94.2%, precision of 94.3%, recall of 93.8%, and an F1 Score of 93.8%.