EUVIMEDEUVIMED entdecken
Open Access · CC BY-SA 4.0

Analisis Komparatif Arsitektur CNN dan ViT dengan Variasi Augmentasi untuk Klasifikasi Karakter Mandarin

Stefanus Eko Prasetyo, Haeruddin, Wilsen Lau

Bitnet: Jurnal Pendidikan Teknologi Informasi · 30.01.2026 · 11 (1)

DOI: 10.33084/bitnet.v11i1.11930 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Bitnet: Jurnal Pendidikan Teknologi Informasi, online erschienen am 30.01.2026. Verfasst von Stefanus Eko Prasetyo und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Handwritten Chinese Character Recognition (HCCR) is a fundamental task in computer vision that poses significant challenges due to the large number of character classes and high variability of writing styles. This study aims to perform a systematic comparative evaluation between Convolutional Neural Network (ResNet-50) and Vision Transformer (ViT-B/16) architectures in handling these challenges under limited data scenarios. The research utilizes a specific subset of the CASIA-HWDB1.1 dataset, consisting of 10 fine-grained character classes selected for their visual similarity. To ensure robust evaluation a 5-Fold Cross-Validation method was employed across five experimental scenarios (Baseline, Geometric, Elastic Transform, Random Erasing, and CutMix). Experimental results demonstrate that ResNet-50 consistently outperformed ViT-B/16 in terms of accuracy and stability. The highest performance was achieved by ResNet-50 using Elastic Transform augmentation with a test accuracy of 90.52%, whereas the best performance for ViT-B/16 was achieved using Random Erasing at 88.69%. The study also reveals that ViT exhibits higher performance variability compared to CNN. These findings conclude that for HCCR tasks with limited data, CNN possess a superior inductive bias for capturing local stroke features while data augmentation must be tailored specifically to the architecture type to maximize performance.

Zitieren

APA

Prasetyo, S. E., Haeruddin, & Lau, W. (2026). Analisis Komparatif Arsitektur CNN dan ViT dengan Variasi Augmentasi untuk Klasifikasi Karakter Mandarin. Bitnet: Jurnal Pendidikan Teknologi Informasi, 11(1). https://doi.org/10.33084/bitnet.v11i1.11930

Vancouver

Prasetyo SE, Haeruddin, Lau W. Analisis Komparatif Arsitektur CNN dan ViT dengan Variasi Augmentasi untuk Klasifikasi Karakter Mandarin. Bitnet: Jurnal Pendidikan Teknologi Informasi. 2026;11(1). doi:10.33084/bitnet.v11i1.11930

BibTeX
@article{Prasetyo2026,
  author = {Prasetyo, Stefanus Eko and Haeruddin and Lau, Wilsen},
  title = {Analisis Komparatif Arsitektur CNN dan ViT dengan Variasi Augmentasi untuk Klasifikasi Karakter Mandarin},
  journal = {Bitnet: Jurnal Pendidikan Teknologi Informasi},
  year = {2026},
  volume = {11},
  number = {1},
  doi = {10.33084/bitnet.v11i1.11930}
}
Kostenlos starten →