<![CDATA[The application of deep learning has rapidly advanced across various sectors due to its ability to enhance effectiveness and efficiency in solving complex tasks. One promising implementation is in object detection, particularly in recognizing and identifying banknotes. This innovation is highly beneficial for individuals with visual impairments who often struggle to distinguish between currency denominations. Currently, people with vision loss still rely on manual techniques such as sorting by size or folding bills in specific patterns—methods that are inefficient and prone to error. This study developed a real-time banknote detection system using moving image input through a combination of Convolutional Neural Network (CNN) and the YOLOv4-tiny algorithm. The research employed a quantitative approach with software engineering methods and a Research and Development (R&D) design. The research subjects consisted of annotated digital images of the 2016 Indonesian Rupiah banknotes, comprising seven denominations and two image classes (front and back). A total of 1,260 images were used for training over 7,000 iterations, yielding a mean Average Precision (mAP) of 97.65%. Validation was conducted on 140 test images using cross-validation, resulting in an mAP of 97.5%. Additional evaluation was performed on banknotes in imperfect physical conditions (e.g., crumpled or folded), achieving an mAP of 88%. These findings indicate that the model demonstrates strong robustness under real-world conditions. Overall, the integration of CNN and YOLO in this system provides a practical solution for individuals with low vision to independently recognize Indonesian Rupiah banknotes, thereby enhancing financial accessibility and inclusion.]]>
Copyrights © 2025
