Background: Circulation of counterfeit currency presents a serious threat to the economic stability of the country. A report recently released by the Reserve Bank of India (RBI) shows that over the past few years there has been an annual increase in counterfeit currency confiscations, which indicates that there is a need for better methods of detecting fake money. 
Objective: This study proposes a deep learning framework using Convolutional Neural Networks (CNNs) with transfer learning to automate and improve the detection of counterfeit Indian currency notes. 
Methods: A complete dataset of 5,400 high definition images was created from a total of six different values (₹10, ₹50, ₹100, ₹200, ₹500 & ₹2000). The high-resolution images that make up the dataset are made available processed and prepared according to pre-existing standards (sized, normalised and augmented). Alternatively, the ResNet 50 backbone has been deconstructed and reconstructed using an array of new component parts (customised fully connected layers) specifically for classification purposes (binary). 
Results: Model accuracy was 98.4% with a precision, recall, and F1 score of 97.9%, 98.1%, and 98.0% respectively, indicating a significant increase over SVM (87.3%) and Random Forest (89.6%) baseline results. The Grad-CAM visualisation supported the conclusion that the model was targeting security features such as watermarks, thread security and microprint. 
Conclusion: The system demonstrates high accuracy, denomination-based robustness, and practical deployability at banks with ATM machines, retail POS terminals and border security checkpoints; creating a scalable and understandable solution to the counterfeit currency problem.