<![CDATA[Indonesia has many variations of traditional food and interesting tourist destinations. The large number of tourist destinations make people like traveling and try to enjoy their traditional food. However, when trying traditional foods, especially foods that are new to them, they must be more careful, because the various ingredients contained in them have an impact on health. This research will try to make an application that can recognize Indonesian traditional food. The hope is that it can provide complete information, so that it can be used to develop calorie counter applications in the future. This study aims to design a system that can classify Indonesian traditional food images to help recognize food names with a certain level of accuracy using the EfficientNetV2 architecture. EfficientNetV2 is a new family of deep learning that excels in training as well as parameter efficiency. Deep Learning is a method often used to classify complex images. The EfficientNetV2 used in this study consists of four different architectures namely EfficientNetV2_S_21k, EfficientNetV2_M_21k, EfficientNetV2_L_21k, and EfficientNetV2_XL_21k. The dataset used comes from three types of data source categories, namely from Google Images, direct image capture using a Smartphone camera, and a combination of both. Each dataset category consists of 18 classes with a total of 1,800 images from Google Images, 1,800 images from Smartphone cameras, and 3,600 images from a combination of Google Images and Smartphone cameras. The dataset is taken from three categories to compare the level of accuracy and get the best accuracy value. The results of this study indicate that EfficientNetV2 can classify images of Indonesian traditional food with the highest test accuracy value of 99.4% from the EfficientNetV2-L(21k) model and the results obtained do not occur overfitting.]]>
Copyrights © 2023
