A Data-Driven Approach for Batik Pattern Classification Using Convolutional Neural Networks (CNN)

Batik is one of Indonesia's cultural heritage with complex and diverse patterns, high artistic value and deep philosophy. Manually classifying batik patterns takes time and depends on expert knowledge, making the process inefficient. It is very difficult to distinguish batik pattern motifs because of the similarity between one and the other. This research aims to develop a batik pattern classification model using a Convolutional Neural Network (CNN) with a data-based approach, allowing pattern recognition and classification to be carried out automatically and accurately. The data set used consisted of 4,284 batik images divided into five pattern classes: kawung, slope, ceplok, machet, and nitik. In this study, the CNN model was developed by utilizing transfer learning techniques with MobileNetV3 that had been trained previously on large datasets. The training process involves adding data to increase the robustness of the model against variations in batik patterns. Evaluation is carried out by measuring the accuracy and loss of the model. The results showed that the CNN model achieved an average accuracy of 93.42% on training data and 93.88% on test data. This study shows that a data-driven approach using CNN is effective for the classification of batik patterns, providing more accurate results compared to manual methods and offering an efficient solution for the digitization of the batik industry. Data driven helps produce a general model and is able to recognize batik motifs that have never been seen before. By using data-driven, larger, more diverse data allows the model to better understand variations in patterns and textures. The developed model can serve as the basis for wider applications in cultural preservation and technological advancement based on artificial intelligence.