Data Augmentation Optimization for Indonesian Batik Motif Classification Using Transfer Learning on Convolutional Neural Network

Abstract

Batik is an Indonesian cultural heritage that has various distinctive motifs from various regions. However, manual classification of batik motifs often requires special skills and considerable time.1 To overcome this, the Convolutional Neural Network (CNN)-based classification method with the help of transfer learning is a promising solution. This research uses a dataset of 500 batik images consisting of 10 different motif classes, each class containing 50 images. Data augmentation techniques are applied to expand the variety of training data with transformations such as rotation, zoom, and flipping to reduce overfitting and improve the generalization ability of the model. The MobileNetV2 model was selected as the base model of transfer learning due to its efficiency and ability to extract features from limited data. Experiments show that the MobileNetV2 model with fine-tuning produces the highest classification accuracy of 86%, which is superior to conventional CNNs that achieve accuracies between 59% and 65%. As a result, the combination of data augmentation and transfer learning in MobileNetV2 proved to be effective in improving the accuracy and efficiency of batik motif classification on small data.