YOLO26 for automated batik pattern classification: Preserving cultural heritage through advanced computer vision

Abstract

Batik is an important cultural heritage of Indonesia, characterized by diverse motifs reflecting regional identity, philosophy, and historical background. Manual identification requires expert knowledge and is time-consuming, making automated classification a valuable research challenge. This study proposes an automated batik motif classification system using YOLO26, a modern deep learning architecture optimized for end-to-end inference without Non-Maximum Suppression. The removal of post-processing stages enables a simpler and more efficient classification pipeline, suitable for lightweight and scalable deployment. A dataset of 20 batik motif classes, including Batik Bali, Batik Parang, Batik Mega Mendung, and Batik Kawung sourced from Kaggle, was constructed and preprocessed using standardized image resizing and normalization techniques. Data augmentation strategies such as geometric and photometric transformations improved model robustness. The system was trained using GPU acceleration to ensure efficient experimentation and reproducibility. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. Experimental results show the proposed system achieved 86.44% overall classification accuracy with balanced macro and weighted F1-scores, indicating consistent performance across all batik categories. Results demonstrate that YOLO26 effectively captures fine-grained texture details and high-level motif structures, enabling discrimination between visually similar patterns. This approach contributes to automated batik recognition systems and supports digital preservation, cultural education, and practical applications in batik authentication and classification.