Classification of Batak Toba Ulos Motifs Based on Transfer Learning with MobileNetV2

Abstract

Indonesia possesses a rich cultural heritage, including the traditional Batak Toba Ulos textile, which is known for its diverse motifs and deep philosophical meanings. However, the preservation and visual recognition of Ulos remain challenging, particularly in terms of systematic documentation and automated classification. This study presents a visual recognition system for Batak Toba Ulos motifs using a transfer learning approach based on the MobileNetV2 architecture. The methodology involves the construction of a curated dataset of Ulos images, the application of data augmentation and preprocessing techniques, and model training utilizing ImageNet pre-trained weights. The system’s performance was evaluated using accuracy, precision, recall, and F1-score metrics. Results show that the model is capable of accurately classifying all 12 Ulos classes, achieving F1-scores ranging from 0.93 to 0.97. These findings demonstrate that transfer learning is effective in overcoming the limitations of culturally specific, small-scale datasets. This research contributes to the development of artificial intelligence tools for cultural preservation and supports the digital documentation and promotion of Batak Toba Ulos to younger generations and broader audiences in an efficient and scalable manner.