Identification of Rupiah Banknote Authenticity for the Visually Impaired Using Convolutional Neural Network Architecture
Muhammad Ayyub Ramli, Faqih Hamami, Riska Yanu Fa’rifah · 2024
This study investigates the efficacy of employing advanced Computer Vision (CV) techniques, particularly Convolutional Neural Networks (CNNs) such as EfficientNetV2 and VGG-19, to facilitate the detection of counterfeit money for visually impaired individuals. A mobile application leveraging these architectures was developed, enabling authentication through smartphone cameras without the need for specialized equipment. Rigorous evaluation, including dataset parameter variations and Grad-CAM analysis, was conducted to assess model performance. Results demonstrate the superiority of the EfficientNetV2-B2 architecture, achieving an accuracy of 94% and an F1-Macro score of 93%. VGG-19 also exhibited competitive performance with 85% accuracy and a 75% F1-Macro score. Furthermore, the implementation on Android devices demonstrated the application's ability, equipped with EfficientNetV2-B2, to accurately distinguish between genuine and counterfeit currency, achieving a confidence level of 99.93% for genuine banknotes and 97.67% for counterfeit ones. These findings highlight the potential of leveraging technology to enhance financial security and independence for visually impaired individuals without using LED and UV light assistance.