Fake Currency Detection using Modified Faster Region-Based Convolutional Neural Network

Abstract

Significant technological advancements in the printing and scanning industries exacerbated the counterfeiting problem. As a consequence, counterfeit currency has an impact on the economy and diminishes the value of genuine currency. Therefore, it is essential to detect the counterfeit currency. The majority of previous methods rely on hardware and image processing techniques. Using these methods to detect counterfeit currency is inefficient and time-consuming. We have proposed a system for the detection of counterfeit currency using a modified faster region-based convolution neural network (Faster R-CNN) to circumvent the aforementioned issue. This study identifies counterfeit currency by analyzing images of currency. One thousand images of currency note are used as dataset to train a Faster-RCNN model on inception V2 architecture to learn the feature map of currencies. Upon successful training and validation of the model, 500 images of counterfeit currencies were used to test the model. The proposed method efficiently identifies 96% of counterfeit currency images tested. Other evaluation means such as mean average precision and detection accuracy show that the developed system has an accuracy of 97%.