← Back to Publications List

A Deep Learning-Based Classification of Counterfeit and Genuine Bangladeshi Banknotes Using EfficientNet-B0

Students & Supervisors

Student Authors
Tanvir Ahmed
Bachelor of Science in Computer Science & Engineering, FST
Brishty Das
Bachelor of Science in Computer Science & Engineering, FST
Supervisors
Shakila Rahman
Lecturer, Faculty, FST

Abstract

The counterfeiting of Bangladesh bank notes, particularly high denominations of Taka 500 and Taka 1000 bills, is a persistent economic issue since it impacts on financial confidence and safety of cash dealings. Common methods such as ultraviolet scanning, watermark verification and manual feature checking are commonly labor intensive, prone to error and are currently hard to implement in large scale or real time applications. Despite some studies available on the topic of Bangladeshi currency recognition and counterfeit detection, most of them are constrained by a small dataset, lack of model comparison, or lack of discussion regarding computational efficiency and interpretability. Hence, a solid and effective image-based system is yet to be developed to classify counterfeit Bangladeshi banknotes. The proposed work presents an efficient deep learning model of EfficientNet-B0 that classifies counterfeit and authentic Bangladeshi banknotes. In a large-scale dataset consisting of 8,340 images of Taka 500 and Taka 1000 notes, a structured organization, duplicate verification, note-wise division, image resizing, normalization, and moderate augmentation were made. Five CNN models were trained on the same settings to achieve an equal evaluation: Baseline CNN, MobileNetV2, ResNet50, EfficientNet-B0, and EfficientNet-B2. EfficientNetB0 was used with the highest overall performance, accuracy and macro-averaged precision, recall, F1-score (98.46, 98.57, 98.00, 98.24 respectively) with only 4.01 million parameters and 0.41 GFLOPs.In general, the suggested framework offers a practical trade-off between accuracy, efficiency, and interpretability of scalable counterfeit Bangladeshi banknote classification.

Keywords

Counterfeit banknote classification Bangladeshi currency EfficientNet-B0 deep learning image classification transfer learning

Publication Details

  • Type of Publication:
  • Conference Name: IEEE International Conference on Signal Processing, Information, Communication and Systems 2026
  • Date of Conference: 13/08/2026 - 13/08/2026
  • Venue: Qadirabad, Dayarampur, Natore-6431, Bangladesh
  • Organizer: Bangladesh Army University of Engineering & Technology (BAUET)
View on Publisher