Classification of Indonesian Batik Patterns Using Convolutional Neural Network with MobileNetV3 Architecture

Abstract

Indonesia has a diversity of ethnic groups that give rise to cultural richness, one of which is batik as a national identity. According to the Great Dictionary of the Indonesian Language, batik is a patterned cloth made by applying wax onto the fabric and processed in a certain way. Batik is an Indonesian cultural heritage with a variety of motifs that reflect regional identity and philosophical values. However, public understanding of the range of motifs is still limited, so technological support is needed for automatic identification. Although batik has become a symbol of national culture, the public's knowledge about the types and meanings of its motifs remains limited. This study develops a system for classifying Indonesian batik motifs using a Convolutional Neural Network with MobileNetV3 Small, MobileNetV3 Large, and MobileNetV2 architectures based on transfer learning. The dataset consists of 3,000 batik images with 20 motif classes from public sources, processed through image resizing to 224×224 pixels, augmentation (rotation, flip, zoom, and random brightness), and splitting into training, validation, and test sets with proportions of 80%, 10%, and 10%. Evaluation was conducted using accuracy, precision, recall, and F1 score. The results showed that all architectures achieved accuracy above 88%, with the best values exceeding 99%. MobileNetV3 Large consistently maintained accuracy up to 99% in several configurations and proved to be the most stable architecture, whereas MobileNetV2 reached a maximum accuracy of 99.33% at a learning rate of 0.0001. Therefore, MobileNetV3 Large is recommended as the primary architecture for batik motif recognition applications.