<![CDATA[Batik is a cultural heritage owned by Indonesia and has been inaugurated by UNESCO on October 2, 2009. In this digital era, the variety of batik motifs must be preserved, especially in Giriloyo Batik Village located in Karang Kulon, Wukirsari, Imogiri Sub-district, Bantul. The complexity and diversity of batik motifs in the area require a modern technological approach to assist the accurate classification process. This study aims to compare the performance of the two current models, EfficientNetB0 and Vision Transformer, in classifying five classic batik motifs in Kampung Batik Giriloyo. This research method combines deep learning approach based on Convolutional Neural Network (CNN) and Transformer with training process from zero without transfer learning. The research stages used include dataset collection, prepocessing, augmentation, model building and training, evaluation and visualization of result comparison. Evaluation is done using accuracy, precision, recall, F1-score and inference time efficiency metrics. The final dataset amounted to 13,128 sliced batik images. The dataset is then divided into 3 main parts, namely training data by 80%, validation data by 10% and testing data by 10% of the total dataset. The final results showed that Vision Transformer achieved the best performance with testing accuracy reaching 99.85 and the EfficientNetB0 model gave an accuracy of 98.78% with stable efficiency. This research confirms that the Vision Transformer model is superior in extracting global patterns in complex batik motifs. This research also makes a real contribution to the utilization of artificial intelligence in cultural preservation through the classification of digital batik motifs and the development of a classic batik motif classification system in Giriloyo Batik Village.]]>
Copyrights © 2025
