Banknotes play a crucial role in various financial transactions, particularly in daily electronic currency exchanges. The ability to distinguish between authentic and counterfeit currency is a major challenge in the realm of computer vision. Despite the availability of different digital currency transaction platforms, physical cash continues to be the favoured choice for everyday transactions at vending machines, banks, shopping centres, railway station counters, ATMs, and foreign exchange bureaus. Moreover, there are cases where individuals unknowingly come into possession of fake money. Regrettably, this problem persists, and financial institutions are unable to outfit all their branches with the essential counterfeit currency detection mechanisms. Hence, this study introduces a banknote recognition system based on convolutional neural networks, which utilises a range of security features found on Nigerian banknotes. The experimental findings demonstrate the potential of the suggested methodology, yielding highly notable outcomes. It effectively carries out a proficient and resilient classification by utilising actual scene images obtained from the Internet and those images captured under both natural and artificial lighting conditions. Furthermore, it remains unaffected by the rotation and translation of banknotes. Notably, a commendable recognition rate was attained for Nigerian banknotes. Currently, the outcomes presented in this study represent a significant advancement compared to the existing state of the art.
Author(s): D. Olusoga Akinmosin (1) and O. S. Adewale (2)
