Currency Note Classification Using EfficientNet

Authors:

G.Sridevi¹, Nanduri Tejaswini², Yalla Harikrishna³, Pritam Parida⁴, Simhadri Venkata Meena⁵, Cherukuri Gunalakshmi⁶

¹Assistant professor, Centurion University of technology and management, Odisha.

²Assistant professor, Vignan's institute of Engineering for Women, Vishakapatnam.

³Assistant professor, Avanthi institute of engineering and technology(A), Tagarapuvalasa.

⁴Teaching Assistant, Centurion University of technology and management, Bhubaneswar.

⁵Department of Computer Science and Engineering, Centurion University of Technology and Management, Andhra Pradesh.

⁶Assistant professor, MVGR College of Engineering and Technology(A), Vizianagaram.

Abstract - In the fields of finance and security, banknote classification is essential for processing automated teller machines (ATMs) and counterfeit detection, among other uses. This article introduces a new method of classifying banknotes using the cutting-edge EfficientNet convolutional neural network architecture. EfficientNet has demonstrated outstanding performance in a variety of computer vision applications, and its use in banknote classification could improve reliability, efficiency, and accuracy. We present a large collection of banknote photos in this work, spanning a variety of denominations, orientations, and conditions. To make sure the dataset is compatible with the EfficientNet architecture, we pre-process it by removing noise and standardising image sizes. We use transfer learning to fine-tune the banknote classification model, starting the network with pre-trained weights on a sizable dataset and adapting it to our specific task.We evaluate the performance of our banknote classification system using standard metrics such as accuracy, precision, recall, and F1-score. We also examine the model's robustness to changes in lighting, rotation, and noise, as well as its generalisability to previously unknown banknotes. Based on experimental data, we show that our EfficientNet-based technique achieves more accuracy and efficiency than previous approaches, which makes it a useful tool for practical banknote classification applications. By utilising state-of-the-art deep learning algorithms to classify banknotes, this research provides a dependable and efficient solution that advances the development of advanced financial and security systems. When EfficientNet's accuracy and efficiency are combined with a well chosen dataset, the resilience and dependability needed in actual banknote processing scenarios are ensured.

Key Words: Feature Extraction, Robust edge detection, EfficientNet, Convolutional neural networks, Currency Detection