Abstract:The Naira is Nigeria's official currency in daily transactions. This study presents the deployment and evaluation of Deep Learning (DL) models to classify Currency Notes (Naira) by denomination. Using a diverse dataset of 1,808 images of Naira notes captured under different conditions, trained the models employing different architectures and got the highest accuracy with MobileNetV2, the model achieved a high accuracy rate of in training of 90.75% and validation accuracy of 87.04% in classification tasks and demonstrated substantial performance across various scenarios. This model holds significant potential for practical applications, including automated cash handling systems, sorting systems, and assistive technology for the visually impaired. The results demonstrate how the model could boost the Nigerian economy's security and efficiency of financial transactions.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: **How to accurately classify different denominations of the Nigerian currency, the Naira, using deep - learning techniques**. Specifically, the paper aims to improve the accuracy of Naira banknote classification by deploying and evaluating deep - learning models (especially Convolutional Neural Networks, CNN), thus addressing the following challenges:
1. **Limitations of traditional methods**: Traditional banknote recognition methods mainly rely on manual verification. This method is time - consuming and error - prone, especially in high - volume trading environments such as banks and retail stores.
2. **The threat of counterfeit currency**: With the progress of counterfeiting techniques, the reliability of traditional currency verification methods has declined, and more efficient automated systems are urgently needed.
3. **The dominance of cash transactions**: Although Nigeria is promoting a cashless policy and digital payment methods, cash transactions still dominate. As high as 80% of transactions in 2022 were still cash transactions.
4. **The assistance needs of the visually impaired**: Existing solutions have not fully met the needs of the visually impaired for currency recognition. Therefore, it is necessary to develop an automated system that can assist the visually impaired.
To address these challenges, the paper proposes using deep - learning models (such as MobileNetV2, InceptionV3, etc.) to classify different denominations of Naira banknotes and trains and evaluates them through a dataset containing 1,808 Naira banknote images under different conditions. The research results show that MobileNetV2 performs best during training and validation, achieving accuracies of 90.75% and 87.04% respectively.
In addition, the paper also emphasizes the challenges in the data collection process, especially the shortage in obtaining images of low - denomination banknotes (such as ₦5 and ₦10), which points out the direction for improvement in future research. Finally, the application prospects of this research are extensive, including automatic cash - handling systems, currency sorters, and tools to assist the visually impaired. These applications are expected to significantly improve the efficiency, accuracy, and security of financial transactions in the Nigerian economy.
### Summary
The core problem of this paper is to develop an efficient, automated system based on deep - learning to accurately classify different denominations of Nigerian Naira banknotes, thereby improving the security and efficiency of financial transactions and meeting the needs of the visually impaired.