Classification of Javanese Script Characters Using ResNet-50 Architecture Based on Transfer Learning

Abstract

Javanese script is one of Indonesia’s cultural heritages that has complex visual characteristics and high similarity among characters, making automatic character recognition a challenging task. Recent advances in deep learning, particularly Convolutional Neural Networks (CNNs), have shown promising performance in image classification problems. This study aims to develop and evaluate a Javanese character classification model using the ResNet-50 architecture based on transfer learning. The dataset consists of digital images of basic Javanese characters that were preprocessed through image resizing, normalization, and data augmentation. ResNet-50 was employed as the main feature extractor, with modifications applied to the final classification layer according to the number of character classes. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics, supported by confusion matrix analysis. The experimental results indicate that the ResNet-50 model achieves stable and reliable performance in classifying Javanese characters. The use of transfer learning and data augmentation contributes to improved generalization capability. These findings demonstrate that deep CNN architectures such as ResNet-50 are effective for Javanese character recognition and have potential applications in digital learning and cultural preservation systems.