<![CDATA[Indonesia boasts a rich cultural heritage in the form of batik, which has gained international recognition. However, the wide variety of batik motifs makes visual identification difficult for both locals and tourists. The limitations of manual observation and a lack of understanding regarding the significance behind each design pose a significant barrier to cultural preservation in the digital age. This study aims to conduct a comparative analysis of Deep Learning models to identify the most effective architecture for automatically classifying batik motifs. The method employed involved comparing three Convolutional Neural Network architectures: MobileNetV2, Xception, and EfficientNet. This study was conducted using a dataset containing 3,700 batik images that had been processed through a careful data distribution process. The primary objective of this evaluation is to find the optimal balance between high classification accuracy and efficient use of computational resources, enabling implementation on platforms with limited specifications. The results of this study indicate that these models can recognize complex batik patterns with outstanding validation accuracy rates ranging from 98% to 100%. These findings provide a strong technical foundation for selecting the most appropriate model architecture for developing intelligent systems aimed at preserving traditional batik. This study also shows that MobileNetv2 is the most optimal model architecture because it achieves a perfect balance between 100% accuracy and the fastest total inference time of 36.74 seconds, with an average inference time per sample of 0.0525 seconds. It is hoped that this research will make the batik identification process faster, more accurate, and accessible to the general public, thereby supporting the sustainability of Indonesia’s cultural heritage.]]>
Copyrights © 2026
