Recognition of Balinese letters with convolutional neural network

Abstract

Indonesia is one of the countries which is famous for having various kinds of distinctive cultures. Almost every region has a unique culture, one of which is the province of Bali which is known as "the island of the gods". Bali Province has very strong customs and culture. This study aims to train computers to recognize Balinese letters. Prior to this study, there had been several studies with similar aims to other recognition methods. This study uses the Convolutional Neural Network (CNN) method. The process of recognizing Balinese script has its own difficulty level in terms of collecting data that will be used for the training process for the Balinese script recognition system. Broadly speaking, the process of introducing Balinese script can be divided into 3 stages, namely; the Pre-Processing stage, the Training stage (Training), and the Testing stage. The introduction of Balinese script was carried out using the Tensorflow and Keras frameworks with a dataset of 1440 images. The dataset is divided into 80% training data and 20% testing data. This Balinese script recognition system is made using the Python programming language. The research was conducted using variations of the epoch in training, namely 50, 100, 150, 200, 250, 400, and 500. The highest testing accuracy was obtained at the number of epoch 500 with an accuracy rate of 89.93%.