Pengenalan Aksara Karo dengan Menggunakan ResNet50 Secara Real Time

Abstract

Most of the Karo community understands the Karo language verbally rather than in its written form, known as Aksara Karo. This is due to a lack of awareness and interest in learning the script. To address this challenge, this study employs the Convolutional Neural Network (CNN) method using the ResNet-50 architecture to recognize Aksara Karo through images uploaded by users. The dataset consists of 9,500 images grouped into 19 words in Aksara Karo. The developed model achieved an accuracy of 93%. Real-time application testing results demonstrate that the ResNet-50 method is effective in classifying the script with a high level of precision. However, challenges such as a lack of data variation, low image quality, and visual similarity among similar characters still impact the model's performance. This research aims to contribute to the preservation of Aksara Karo and to increase the interest of younger generations in learning it.