Real-Time Recognition of Malagasy Banknotes (Ariary)

This research focuses on real-time recognition of Malagasy banknotes to introduce an innovative solution benefiting the financial sector. The main goal is to develop an automatic system for banknote recognition, applicable in Automated Teller Machines (ATMs) and mobile devices for visually impaired individuals. The study begins with an analysis of real-time object recognition models, such as Faster Region-based Convolutional Neural Network (Faster RCNN), Single Shot MultiBox Detector (SSD), and You Only Look Once (YOLO). Yolov8 is selected due to its superior accuracy, especially for small objects, leveraging transfer learning. Model performance is enhanced through hyperparameter tuning and improved data preprocessing, including padding for image standardization and instance normalization to reduce class biases. Evaluation is conducted on Android mobile phones, simulating real-world conditions. Results confirm that preprocessing improvements significantly enhance model accuracy and stability, ensuring reliable recognition of Malagasy banknotes. This research successfully develops a real-time recognition system using Yolov8, providing an effective solution for financial transactions and inclusive access for visually impaired individuals. Future work includes extending recognition to other currencies, counterfeit detection, and integration into diverse financial devices.