Enabling Automatic Coin Grading Through Late-Fusion of Vision–Language and Hierarchical Transformers
IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 4504811-4504811
DOI: 10.1109/tim.2026.3699676 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Melaku N. Getahun und 1 weiteren Autor:innen. Die Arbeit nennt 32 Literaturangaben.
Abstract (Originalsprache)
Measuring coin condition is a procedure performed by specialists in the area of numismatics and museum technologies. Although the Sheldon scale formalizes the coin conditions within a 70-point system, the grading remains a highly subjective process as it is based on the human assessment of the coin strike, luster, wear, and overall preservation. Existing approaches addressing the problem of automatic coin condition measurement are limited in scope, focusing on a small number of coin types or simplifying the grading task into a few broad categories, which restricts their real-world applicability. In this work, we propose automatic mint-state (MS) coin grading using computer vision methods. We introduce SwinCLIP, a two-stream transformer framework that integrates the Swin Transformer and contrastive language–image pretraining (CLIP) models through a late-fusion mechanism to leverage both fine-grained spatial features and high-level semantic information available during the coin image processing. Prior to image processing, we collected a diverse, fully balanced MS coin dataset across 11 distinct MS grades. Experimental results show that SwinCLIP achieves a Top-1 accuracy of 0.591, outperforming strong CNN and Transformer baselines, as well as existing coin grading studies. The model also achieves a Top-5 accuracy of 0.960, indicating reliable ranking of plausible grades. In addition, ±1 and ±2 grade tolerance accuracies of 0.785 and 0.922, respectively, further demonstrate the model’s consistency in making predictions close to the true grade. An ablation study confirms that the fusion of Swin and CLIP provides complementary strengths in texture sensitivity and semantic discrimination.
Zitieren
APA
Getahun, M. N., & Somov, A. (2026). Enabling Automatic Coin Grading Through Late-Fusion of Vision–Language and Hierarchical Transformers. IEEE Transactions on Instrumentation and Measurement, 75, 4504811-4504811. https://doi.org/10.1109/tim.2026.3699676
Vancouver
Getahun MN, Somov A. Enabling Automatic Coin Grading Through Late-Fusion of Vision–Language and Hierarchical Transformers. IEEE Transactions on Instrumentation and Measurement. 2026;75:4504811-4504811. doi:10.1109/tim.2026.3699676
BibTeX
@article{Getahun2026,
author = {Getahun, Melaku N. and Somov, Andrey},
title = {Enabling Automatic Coin Grading Through Late-Fusion of Vision–Language and Hierarchical Transformers},
journal = {IEEE Transactions on Instrumentation and Measurement},
year = {2026},
volume = {75},
pages = {4504811--4504811},
doi = {10.1109/tim.2026.3699676}
}
EUVIMED