Bangladeshi Bank Note Detection Using Deep Convolutional Neural Network

Abstract

This report presents a Bangladeshi Banknote detection system using a Deep Convolutional Neural Network. This project is usually designed for people who do not recognize or cannot see the Bangladeshi banknotes. Visually impaired humans face trouble in figuring out and spotting the unique types of banknotes due to a few reasons. Many projects like this have been followed before this project was completed. Some of the works of others are also mentioned in this paper. The detection system is also capable of identifying the Bangladeshi Banknotes that are rumpled, decrepit, or may be worn. The detection system consists of image preprocessing, image evaluation, and image recognition. In this project, 3000 images have been used. There are 50 taka, 100 taka, 200 taka, 500 taka, 1000 taka. This project has been completed using CNN, Vgg16, Transfer learning, and transfer learning-based improved CNN model. The system can identify five banknotes used in Bangladesh with an accuracy of 91% by Deep CNN, 95% by Improved Deep CNN, and 81% by VGG16. Visually impaired human beings might be capable of using it effortlessly in day-by-day transaction.