In recent years, the automation of currency classification has gained significant attention due to its potential applications in banking, retail, and financial services. This study presents a deep learning-based approach for automating the classification of Indian currency notes, specifically targeting denominations of ₹10, ₹20, ₹50, ₹100, ₹500, ₹1000 and ₹2000. Utilizing a dataset consisting of images from these currency classes, we explore the effectiveness of various convolutional neural network (CNN) architectures, including MobileNet, ResNet, and DenseNet. Among these, we propose a hybrid model combining the strengths of MobileNet, ResNet and DenseNet, leveraging their capability to capture intricate features while maintaining computational efficiency. Our experiments demonstrate that the proposed model outperforms traditional methods, achieving high accuracy in currency classification. This advancement not only enhances transaction efficiency but also contributes to the ongoing development of automated financial solutions in India.

Keywords: Indian currency classification, deep learning, CNN, MobileNet, ResNet, DenseNet, automation, image recognition.