CurrencyNet: A Vision Transformer-Based Approach for Indian Currency Note Classification with Optimizer Exploration

This study proposes CurrencyNet, a new way to classify Indian rupee notes using Vision Transformer (ViT) deep architecture. CurrencyNet takes advantage of ViT's feature to manage spatial connections in images. Different optimizers, induding Adadelta, Gradient Descent, AdaGrad, RMS Prop, Adamax, Momentum, Adaptive Moment Estimation (Adam), and Nesterov Momentum, are explored to see how they affect CurrencyNet's performance. Testing results show that CurrencyNet gets 97.76 % accuracy, using both the Vision Transformer and the Adam optimizer together. Proposed CurrencyNet is compared with well-known deep architectures, such as VGG16, VGG18, Inception, Xception, ResNet, and MobileNet. The outcomes show that CurrencyNet is the best way to sort Indian rupee notes, making it the only choice.