Skip navigation
st. Mary's University Institutional Repository St. Mary's University Institutional Repository

Please use this identifier to cite or link to this item: http://hdl.handle.net/123456789/7698
Full metadata record
DC FieldValueLanguage
dc.contributor.authorMelaku, Mikias-
dc.date.accessioned2023-08-02T11:53:09Z-
dc.date.available2023-08-02T11:53:09Z-
dc.date.issued2023-06-
dc.identifier.uri.-
dc.identifier.urihttp://hdl.handle.net/123456789/7698-
dc.description.abstractThe technology of currency identification plays a vital role in automated self-service equipment such as ATMs, vending machines, and smart card charging machines. These devices require accurate banknote recognition, counterfeit detection, serial number recognition, and fitness classification. However, existing banknote detectors are often tailored to specific countries and cannot be easily reprogrammed for currency identification. Moreover, banknote recognition algorithms based on deep learning suffer from small training datasets and lower accuracy. In this study, we address these challenges by focusing on Ethiopian banknotes. We collected a diverse dataset of Ethiopian real and counterfeit banknotes, including varying ages and conditions. The dataset size and quality significantly impact the performance of the recognition system. To extract effective features from the banknotes, we employed convolutional neural networks (CNNs) using popular architectures such as InceptionV3, VGG16, MobileNet, and ResNet50. We conducted experiments with different optimization approaches, including Adam and Stochastic Gradient Descent (SGD). These optimization approaches influence the training process and model performance. Additionally, we compared the accuracy of the models to determine the most effective solution for Ethiopian banknote identification. Our evaluation metrics included accuracy, which measures the overall correctness of the banknote recognition system. Among the models tested, MobileNet trained with SGD optimization and a batch size of 32 achieved the highest training accuracy of 99.6% and overall accuracy 97%. This outperformed the other deep learning models considered in this study. The MobileNet model with SGD optimization is implemented in both a web-based application and Android applications designed specifically for Android mobile devices. This research contributes to the development of a reliable system for Ethiopian banknote identification. By leveraging deep learning techniques and optimization approaches, we address the limitations of existing systems, such as small datasets and lower recognition accuracy. The findings demonstrate the potential of using MobileNet with SGD optimization as an effective solution for banknote recognition in Ethiopia, paving the way for improved currency identification in automated self-service equipment.en_US
dc.language.isoenen_US
dc.publisherST. MARY’S UNIVERSITYen_US
dc.subjectEthiopian Currency, Currency Recognition, Counterfeit Detection.en_US
dc.titleETHIOPIAN CURRENCY DETECTION AND COUNTERFEIT VERIFICATION USING DEEP LEARNINGen_US
dc.typeThesisen_US
Appears in Collections:Master of computer science

Files in This Item:
File Description SizeFormat 
Designing Ethiopian Paper Currency Detection-V5.pdf5.8 MBAdobe PDFView/Open
Show simple item record


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.