Balinese Carving Recognition using Pre-Trained Convolutional Neural Network

The preservation of Balinese carvings in traditional buildings is needed to preserve by collecting Balinese carving data. Balinese carving data collection is an attempt to save important patterns in Balinese carvings to become a reference for repair Balinese carvings that are beginning to erode by age. Balinese carving recognition is the first step to preserve cultural heritage by collecting Balinese carving motifs on traditional sacred buildings. In this study, we compare the performance of Convolutional Neural Network pre-trained models for Balinese carving recognition. We use transfer learning using four pre-trained models, i.e., MobileNet, Inception-v3, VGG16, and VGG19, to train the recognition model. In the model training process, we fine-tuned the number of parameters trained on each pre-trained model to produce the best performing model. Based on eight experimental scenarios, the VGG19 can produce the best performance with a recognition accuracy of 87.50%.