Automatic Multi-class Classification of Indonesian Traditional Food using Convolutional Neural Networks

Abstract

The culinary industry in Indonesia is growing fast in this era of globalization. With the increasing trend in fast food and other foreign culinary, Indonesia is threatening its local culinary existential. A lack of database about local culinary causing a lack of information in its people. Therefore, there is a need for a system model that can identify a local culinary. Hence, it can provide easy access to information to Indonesian and provide a database. This database can also help the government to promote Indonesian food and a chance to keep their existence. This research proposes to make a model that can identify Indonesian food with deep learning techniques. Convolutional Neural Network is chosen as a deep learning technique to recognize ten types of Indonesian food, namely kue rangi, kue putu, bika ambon, ayam taliwang, putu mayang, kerak telor, kue ape, papeda, gudeg, and sate bandeng. We used ResNet50 architecture to classify multi-class labeling. Datasets will consist of 200 images and will be duplicated into three models of separating datasets. In the first model, the dataset has a composition of 75% training dataset and a 25% testing dataset. Similarly, the second model, the dataset has 80:20 of composition, and the third model has 85:15 of composition for training and testing dataset. The experimental results show the third model has the best accuracy of 100%, with 30/30 images predicted correctly.