AN EXAMPLE OF A CONVOLUTIONAL NEURAL NETWORK FOR RECOGNIZING THE DENOMINATION OF BANKNOTES

Abstract

A common means of solving the problems of classification, recognition, segmentation of images is the use of Convolutional Neural Networks (CNN). In this paper, we reviewed the popular CNN architectures used for object recognition, including Region Based Convolutional Neural Networks (R-CNN), Fast R-CNN, Faster R-CNN, You Only Look Once (YOLO), Single Shot Detector (SSD), Feature Pyramid Networks (FPN) and RetinaNet. It is shown that the YOLO convolutional neural network is optimal in terms of speed and accuracy of recognition.

The effectiveness of convolutional neural networks for the recognition of objects in the images is shown on the example of the development of a prototype system for recognizing the denominations of Ukrainian hryvnia banknotes and finding their sum. The work of the developed prototype of such system is demonstrated for which YOLOv5 Small architecture was used and it was tested on images of Ukrainian hryvnias. For summarizing the amount of money on the photo was created a separate program on python. The characteristics of the used software and hardware are specified. The structure of datasets that were used for training and testing the network is described, quality indicators of the developed prototype and comparison with existing banknote recognition systems are given.

The average recognition accuracy for this model is 0.985. To achieve more accurate and objective results, it was proposed to expand the dataset and re-train the network.