Ghanem Ahmed M, Youness Hassan A, Wahba Mohamed, Abdelaal Hammam M
Department of Computers & Systems Engineering, Faculty of Engineering, Minia University, Minya, 61519, Egypt.
Technology Center, Kafrelsheikh University, Kafr el-Sheikh, 33516, Egypt.
Sci Rep. 2025 Oct 1;15(1):34288. doi: 10.1038/s41598-025-20646-x.
This study presents a novel real-time Egyptian currency recognition system designed to assist visually impaired individuals in performing financial transactions independently and securely. The system leverages advanced deep learning models-YOLOv8, YOLOv9, and YOLOv10 to achieve high accuracy and low latency in identifying Egyptian banknotes. Evaluated on a comprehensive dataset of 2,000 annotated images, the models incorporate innovations such as context aggregation, GELAN, and NMS-free training to enhance performance. A review of prior systems highlights their limitations, especially concerning regional currencies. YOLOv10 achieved the best performance, with a precision of 0.9678, F1 score of 0.9715, and mAP@0.5 of 0.9934, surpassing both YOLOv8 and YOLOv9. Compared to traditional techniques, this approach offers significant improvements in accuracy and processing speed, providing a scalable and practical solution for accessible AI applications. These contributions promote financial independence and inclusion for visually impaired users, supporting ongoing advances in assistive technology.
本研究提出了一种新颖的实时埃及货币识别系统，旨在帮助视障人士独立、安全地进行金融交易。该系统利用先进的深度学习模型——YOLOv8、YOLOv9和YOLOv10，在识别埃及纸币方面实现高精度和低延迟。在一个包含2000张带注释图像的综合数据集上进行评估时，这些模型融入了上下文聚合、GELAN和无NMS训练等创新技术以提升性能。对先前系统的回顾凸显了它们的局限性，尤其是在处理区域货币方面。YOLOv10表现最佳，精度为0.9678，F1分数为0.9715，mAP@0.5为0.9934，超过了YOLOv8和YOLOv9。与传统技术相比，这种方法在准确性和处理速度方面有显著提升，为无障碍人工智能应用提供了一种可扩展且实用的解决方案。这些成果促进了视障用户的金融独立和包容，支持了辅助技术的持续进步。
