Deep Learning-Based Detection of Fake Multinational Banknotes in a Cross-Dataset Environment Utilizing Smartphone Cameras for Assisting Visually Impaired Individuals

Abstract

The automatic handling of banknotes can be conducted not only by specialized facilities, such as vending machines, teller machines, and banknote counters, but also by handheld devices, such as smartphones, with the utilization of built-in cameras and detection algorithms. As smartphones are becoming increasingly popular, they can be used to assist visually impaired individuals in daily tasks, including banknote handling. Although previous studies regarding banknote detection by smartphone cameras for visually impaired individuals have been conducted, these studies are limited, even when conducted in a cross-dataset environment. Therefore, we propose a deep learning-based method for detecting fake multinational banknotes using smartphone cameras in a cross-dataset environment. Experimental results of the self-collected genuine and fake multinational datasets for US dollar, Euro, Korean won, and Jordanian dinar banknotes confirm that our method demonstrates a higher detection accuracy than conventional "you only look once, version 3" (YOLOv3) methods and the combined method of YOLOv3 and the state-of-the-art convolutional neural network (CNN).