Development of a CNN-Based Knowledge System for Rupiah Currency Authenticity Detection and Nominal Classification

The circulation of counterfeit money in Indonesia inflicts substantial losses on the public and financial institutions. Manual verification of money is inefficient and error-prone, especially during high transaction volumes, because counterfeit bills exhibit physical characteristics nearly identical to genuine currency. To uncover counterfeit notes, an ultraviolet lamp exposes invisible ink. This research employs the Convolutional Neural Network (CNN) to detect authenticity and classify Indonesian rupiah banknotes. The CNN is trained using images of authentic banknotes captured with a camera and ultraviolet light across various denominations. The system stores the images and trains the model to identify authenticity and denomination features. Experimental results demonstrate that the proposed approach achieves high classification accuracy in distinguishing genuine and counterfeit Rupiah banknotes, as well as in recognising their respective denominations. The testing phase introduces real notes exposed to ultraviolet light, producing images that reveal invisible ink patterns. The authenticity detection achieved a 100% success rate, while the denomination recognition rates were 70% for Rp. 5,000 notes, 80% for Rp. 10,000 and Rp. 20,000 notes, and 90% for Rp. 50,000 and Rp. 100,000 notes. The system’s overall success rate is 82%.