Klasifikasi Citra Pada Wayang Kulit Menggunakan Convolutional Neural Network

Abstract

This research aims to develop a Convolutional Neural Network (CNN)-based shadow puppet image classification system by utilizing the ResNet-18 architecture, which is known to be efficient in handling image data and has a high level of accuracy. The system is designed to classify the Punakawan characters in shadow puppets, namely Bagong, Gareng, Petruk, and Semar, which are part of Indonesia's cultural heritage. In addition, this study also compares the performance of ResNet-18 with two other architectures, namely MobileNetV2 and DenseNet121. The dataset used consists of 2,148 images, which were obtained through live shooting and online searches. The images were processed using augmentation techniques and divided in a ratio of 70:15:15 for training, validation, and testing. The model was trained using optimal hyperparameters, such as learning rate 0.001 and batch size 32, to evaluate the performance of the three architectures. The evaluation results showed that the ResNet-18 architecture, as the main focus of the research, achieved an overall accuracy of 93.90%, with precision, recall, and F1-score of 94% each. In comparison, MobileNetV2 produced the highest validation accuracy of 96%, with better performance in generalization, while DenseNet121 produced a validation accuracy of 95%. This result confirms that although MobileNetV2 has the best performance in shadow puppet image classification, ResNet-18 still shows excellent results with simpler complexity, so it can be an efficient solution for the implementation of Punakawan shadow puppet classification system.