Andy Pramono, I-Cheng Chang, Betty Dewi Puspasari, Mitra Istiar Wardhana, Nuril Kusuma Wardani
The preservation of intangible cultural heritage, like Indonesian wayang kulit, is more important than ever because of the world becoming more connected and technology changing quickly. Wayang kulit is an art form that transmits moral, spiritual, and cultural values. However, the younger generation's interest in wayang kulit is declining, and weak digital records risk its life. Recent advances in artificial intelligence (AI) and seeing computers bring neat ways to wake up and save this custom. This study introduces DeepWayang, a novel hybrid deep learning architecture that cleverly merges the big-picture understanding of EfficientNetV2B3 with a specialized neural network's keen eye for intricate textures. Through a comprehensive comparative analysis, the proposed model performs better in classifying visually complex wayang kulit characters, outperforming recent research in wayang kulit in an accuracy test of 98.37%. This study also presents predictions in the Confusion Matrix and Visualization using Class Activation Mapping, further confirming the model's ability to focus on distinctive features important for character recognition. These findings highlight the potential of advanced digital technologies to preserve and renew traditional art forms and ensure their transmission to future generations. © 2025 IEEE.
State University of Malang, Department of Animation, Malang, Indonesia; National Dong Hwa University, Department of Computer Science and Information Engineering, Hualien, Taiwan; Institut Teknologi Nasional Malang, Department of Informatics Engineering, Malang, Indonesia
