Classifying Javanese Letters with Convolutional Neural Network (CNN) Method

HARJOSEPUTRO, YULIUS (2018) Classifying Javanese Letters with Convolutional Neural Network (CNN) Method. In: The First International Conference and Exhibiton on Sciences and Technology (ICEST), 25 27 Oktober 2018, Labuan Bajo.

Preview	Text ICEST UNDANA -Classifiying Java.pdf Download (2MB) | Preview
Preview	Text Peer Review ICEST UNDANA -Classifiying Java.pdf Download (679kB) | Preview

Abstract

One crucial issue related to comput er vision is image classification which is commonly used to detect an object in an image. The i mage classification process done by a computer is considered complicated therefore res earchers prefer to implement some methods such as the most classic one called artificial neural network. However, there are still limitations in using this method including the number of neurons used that the results used for image classification are still not optimal. Meanwhile, one of the most popular met hods which is lately used to deal with the limitations of the previous method is one technique from Deep Learning using the Convolutional Neural Network (CNN) method. The research result s of implementing CNN method for classifying Javanese lett er show that the accuracy level for training is 90% with training time using GPU of 409.25 seconds. The level of accuracy for the Java letters classification test is quite good, which is 85%. It is considered highly potential to be developed into a system that can recognize the Javanese letters systems.

Actions (login required)

View Item