Classification of Currency Denominations in Folded Condition using Convolutional Neural Network

Identifying the denomination of money is a complex issue for blind people. Typically, denomination identification relies on the sense of touch by feeling certain parts of the money. This method also presents significant difficulties if the money being held is crumpled. The need for money denomination identification for the visually impaired becomes crucial to help them recognize the denominations. This research proposes a Convolutional Neural Network (CNN) method for detecting and classifying money denominations from digital images taken by a camera. The CNN model will learn from various types of currency data and then classify the denominations of Indonesian rupiah from thousands to hundreds of thousands with high accuracy. Besides general money denomination identification, this research also focuses on identifying worn, folded, and torn money. This will facilitate blind people in classifying money without having to place the money in an ideal position. The ideal position refers to new money that is well-captured by the camera. Based on testing results by performing simulations of the number of epochs and stopping callbacks, an accuracy above 90% was achieved with a loss of 0.3.1.