Classification of Monkey Characters Using CNN-VGG16 and ResNet50 Architectures

Abstract

Wayang kulit is one of Indonesia’s cultural heritages that possesses significant artistic and philosophical value. In the Balinese Ramayana puppet tradition, monkey characters exhibit highly similar visual characteristics, making manual identification difficult and requiring specialized expertise. To address this challenge, this study proposes an image classification approach based on Convolutional Neural Networks (CNN) using transfer learning with the VGG16 and ResNet50 architectures. We utilize a dataset consisting of 270 images divided into 15 classes, including 14 monkey puppet character classes and 1 non-monkey puppet class. This study conducts multiple experimental scenarios involving different dataset partitioning strategies and hyperparameter configurations to analyze model performance comprehensively. Furthermore, we apply data augmentation techniques to the training dataset to improve model generalization and reduce overfitting. Based on the obtained results, VGG16 achieves the best performance with a testing accuracy of 98.67%, while ResNet50 achieves 98.33%. We also obtain stable training and validation performance from both architectures, indicating strong capability in learning visual patterns from limited datasets. However, several classification errors still occur in classes with high visual similarity, demonstrating the challenges of fine-grained image classification. This study demonstrates that transfer learning-based CNN architectures can effectively classify Balinese Ramayana monkey puppet characters. It also contributes to the development of intelligent systems for cultural heritage preservation under limited dataset conditions.