| Abstract [eng] |
Numismatics, the study of coins and currency, is very important for enhancing our understanding of historical economies, cultures, and societies. Coins serve as records of the past, offering insights into the art, politics, religion, and commerce of different eras. However, the classification and analysis of coins is a complex task that traditionally relies on highly skilled experts. This complexity arises from the intricate details and vast variations among coins. Additionally, many coins suffer from damage or corrosion over time, which eliminates critical features and poses significant challenges for accurate classification. This paper introduces an automated coin classification framework leveraging the Swin Transformer deep learning model. The Swin Transformer’s hierarchical architecture and shifted window mechanism enable it to effectively capture intricate features and contextual information within the coin images. This makes it particularly well-suited for detailed numismatic datasets that require high-level feature representation and spatial awareness. By processing images of both the obverse and reverse sides of coins and integrating physical metadata like weight and diameter, the system enhances classification accuracy. The image preprocessing process addresses challenges such as damaged or corroded surfaces, while data augmentation simulates real-world conditions to improve robustness and classification accuracy. The Swin Transformer’s hierarchical architecture effectively captures intricate features, making it well-suited for detailed numismatic datasets. Evaluated on a diverse collection of coins, the system demonstrates high scalability and accuracy, offering a valuable tool for numismatic research and cultural heritage preservation. Rather than replacing human expertise, this tool is designed to assist experts by enhancing the classification process through collaboration between technology and human insight. It automates the initial stages of classification, allowing experts to focus on more nuanced analyses and interpretations that require human judgment. |
