Indian Currency Note Classification Using Deep Learning

Abstract

Detecting and classifying currency notes is very important to protect financial security, especially in a cash-based country like India. Even though online payments are growing, people still use cash widely, and fake note detection remains a big challenge. Traditional image processing and machine learning methods are still often used, but the performance is unsatisfactory when there are changes in lighting, note orientation, or note quality. In this study, towards increasing performance propose a Convolutional Neural Network (CNN)-based deep learning model is proposed for Indian currency detection. The model classifies the currency note images in the seven denominations: ₹10, ₹20, ₹50, ₹100, ₹200, ₹500, and ₹2000. The dataset, collected from the Kaggle repository with a total of 3,166 images, was expanded using augmentation techniques to improve feature learning. The CNN architecture includes two convolutional layers with max pooling, two dense layers, and a softmax classifier. The proposed model achieved a training accuracy of 98.93% and a validation accuracy of 90.82% at 15 epochs. Furthermore, when the ₹2000 denomination (which is no longer in circulation) was excluded, the model achieved a training accuracy of 99.50% and validation accuracy of 83.95%.