Transfer Learning on Balinese Character Recognition of Lontar Manuscript Using MobileNet

The Balinese lontar manuscripts are cultural heritage written using Balinese characters on dried palm leaves. The conservation of the lontar manuscript is carried out by understanding the meaning contained in it. This research is the first step in the conservation of the lontar manuscript by recognizing Balinese characters. In this study, we recognized Balinese characters on lontar using a transfer learning approach. Transfer learning is done by fine-tuning the number of parameters of the pre-trained model to speed up the model convergence by modifying the number of trainable parameters on the pre-trained model. We modified the number of MobileNet architecture parameters with varying the number of trainable parameters and three optimizers to produce the best performance model. Based on the experimental result, the best recognition model yields 86.23% accuracy with a combination of SGD optimizer and 60% trainable parameters.