AI Based Fake Currency Detection System Using Deep Learning and CBAM

Abstract

Counterfeit currency is a significant financial and security concern because fake banknotes can cause direct economic losses and reduce public confidence in financial transactions. Manual verification of banknotes generally depends on human expertise and inspection of security features, which may be difficult, time-consuming, and inconsistent under practical conditions. This paper presents an artificial intelligence based fake currency detection system using deep learning and the Convolutional Block Attention Module (CBAM). The proposed system uses EfficientNetB3 as the principal feature extraction backbone and incorporates an attention mechanism to improve the representation of important spatial and channel-wise features present in currency-note images. A multi-task learning framework is employed to perform two related tasks simultaneously: authenticity classification and denomination classification. The authenticity branch determines whether an input note is real or fake, while the denomination branch identifies one of seven Indian currency denominations, namely Rs. 10, Rs. 20, Rs. 50, Rs. 100, Rs. 200, Rs. 500, and Rs. 2000. A dataset containing 7,442 currency-note images is used for model development and evaluation. Image preprocessing, feature extraction, attention refinement, and classification are integrated into an end-to-end pipeline. The proposed approach is intended to provide a more informative currency verification system than a simple binary classifier by jointly considering authenticity and denomination. The experimental evaluation considers accuracy, precision, recall, F1-score, confusion matrix, and confidence-based prediction analysis. The proposed framework demonstrates the applicability of transfer learning and attention mechanisms for automated Indian currency verification and provides a foundation for future realtime and mobile-based counterfeit detection systems.