Automated Currency Identification: A CNN Approach to Philippine Banknote Recognition

Currency banknotes are essential parts of people's daily monetary transactions. Banknotes, such as Philippine banknotes, incorporate enhanced security features, making it counterfeit-proof, and improved designs for easier recognition. Due to their prominent features for each denomination, banknotes can be distinguished by the naked eye. However, elderly and visually impaired individuals struggle to accept and identify currency, making them vulnerable to fraud and scams. This study developed three convolutional neural networks, with different architectures that can recognize Philippine banknotes using Python and Jupyter Notebook. The models were trained and evaluated using a self-generated dataset consisting of 7000 images of 20-, 50-, 100-, 200-, 500-, and 1000-peso bills and nonPH currencies captured in varied conditions, occlusions, illumination, and orientation. The training and hold out validation results showed that Model C outperformed other models with 95.93% accuracy. Models A and B exhibit slight difficulty in classifying the 50- and 100-peso bills, while all models performed the least in classifying non-PH currencies. Further, 140 images of unseen data were used for single prediction, where Model C performed best with 95.71% accuracy.