BengaliTaka: A Comparative Analysis of Transformer and CNNs on Bangladeshi Currency Recognition
Jasmin Jahan Puspo, M. Shahidur Rahman · 2025
Automated banknote detector shows the pioneering of technology in terms of financial assistance and visually impaired people. Manual verification is often time-consuming and suffers from limitations in accuracy and efficiency. This research introduces a transformer model for the Bangla currency classifier compared to other traditional CNN architectures. The proposed model utilizes a customized dataset, which is publicly available. To ensure higher efficiency, the proposed pipeline used augmentation, pre-processing, and hyperparameter tuning. The performance of the model was evaluated by various metrics and achieved 98.00 % accuracy. Results demonstrate that the ViT model shows efficiency with real-world currency images.