Deep Learning Model for Automated Classification and Condition Assessment of Indian Currency Notes
Payal Chhabra, Aastha Jain, Mahi, Tarang Priyadarshi · 2025
VeriTrust is a cutting-edge deep learning framework designed to improve financial safety through the automation of detecting currency denominations and assessing the condition of banknotes. It addresses the issue of damaged notes in circulation by processing images through steps like acquisition, edge detection, segmentation, and feature extraction. These features are analyzed using a custom Convolutional Neural Network (CNN) to categorize notes as damaged or fit for use, ensuring that compromised notes are efficiently filtered out. Initially developed with TensorFlow and Keras, employing the Adam optimizer alongside binary cross-entropy loss, VeriTrust achieved commendable 93% precision. Recent improvements include the integration of PyTorch-based CNN and Grad-CAM-based CNN models, which achieved accuracies of 93% and 84%, respectively. This paper outlines the VeriTrust pipeline, its CNN architecture, and its potential to strengthen transaction security, while also highlighting future plans to improve accuracy and broaden its applications.