Reconnaissance de billets d’argent algérien par l’apprentissage profond

Abstract

This project consists of the application of artificial intelligence for the detection and recognition of Algerian currency notes. To do this, deep learning based on YOLOV5 is used. A database of Algerian currencies has been created to learn or train the YOLOV5 model. After the training stage, a test phase was carried out, the results obtained are satisfactory and the model has reached the precision of; 88.19%, 92.86%, 96.58% and 96.55% for the notes of 1000 (side 1), 1000 (side 2), 2000 (side 1) and 2000 (side 2) respectively. An additional step based on image processing is introduced to determine the dimensions of the banknote, in particular, the width and the length and therefore the ratio (width/length). This ratio is used to accept or reject the banknote.