Fake Congolese Banknote Detection Based On Deep Learning

With the rapid integration of world economies,the establishment of the Eurozone,and the growth of Africa's economy in recent years,border trade and personal interactions between countries have become more common.Travellers always bring a large amount of paper currency from other nations.As a result,the chances of paper currencies from several countries becoming intertwined are becoming increasingly likely.One of the most severe issues influencing cash transactions is a forgery;the advancements in digital multi-colour printing,scanning and image processing have made it easier to create high-resolution counterfeit banknotes known as super notes.In the Democratic Republic of Congo,counterfeit banknotes are becoming a severe danger to smooth transactions.As a result,such counterfeit notes on the market necessitates the automation of money transaction systems.The banking industry is unable to make full use of self-service equipment,such as ATMs.On the other hand,banks have yet to implement a viable detection system for detecting fake banknotes.On the other hand,banks have yet to implement a viable detection system for detecting fake banknotes,which necessitates creating a more effective authenticity checking system.As a result,the ability to detect counterfeit currency is critical.The majority of the previous strategies rely on hardware and image processing techniques.These procedures are inefficient and time-consuming when it comes to detecting counterfeit currency.That necessitates developing a system and method for detecting counterfeit banknotes in less time,more efficiently,and with greater accessibility.To design the Congolese system,we looked at converting classic RGB encoding image to HSV Color encoding by keeping only one layer for the convolutional neural network as a feature extraction technique to perform the classification under pattern recognition methods.This work identifies counterfeit banknotes by examining note images using the convolutional neural network method to learn the feature map of a specific note using a trained set of one thousand Congolese notes;once the feature map is learned,the network is ready to identify the fake currency in real-time under some predefined environment.To verify banknotes,we used a deep learning approach.In this case,we do the authentication by using several image processing concepts and methodologies to create the information used in the model's training process,namely,to categorise the note based on the fundamental or trained features.The proposed system focus mainly on the banknote 1000 CDF note that we construct data-set blended with open-source authentication information published by the Central Bank of Congo to achieve a high accuracy result.We have analysed the properties of the HSV(hue,saturation and value)colour space,emphasising the visual perception of the variation in hue,saturation and intensity values of an image pixel.We extract pixel features by either choosing the hue or the intensity as the dominant property based on the saturation value.The feature extraction method has been applied for image segmentation and histogram generation applications-two distinct approaches to content-based image retrieval.Segmentation using this method shows better identification of objects in an image.The histogram retains an uniform colour transition that enables us to do a window-based smoothing during retrieval.The results have been compared with those generated using the RGB colour space.Extensive picture preprocessing tasks such as image histogram equalisation and adaptive median filter-based image de-noising reduce the effect of noisy data.According to the experimental data,we achieved 99.4 per cent recognition accuracy in categorising Congolese banknote denominations using the CNN model as a feature extractor.CNN feature surpassed the other feature extraction techniques with a 96 per cent accuracy level for counterfeit cash recognition.The system relies on a specific feature of the Congolese Banknotes and the feature is not possible to replicate for the counterfeit makers or producers.And there is no chance that they would be capable to imitate this feature even within a pretty long time.The result suggests that the potential of the method and system suggested for society and law enforcement services will improve their capacity to identify the counterfeit note in real-time that circulates in the market.The same system can also be applied in the device such as a smartphone,vending machine to an Automated teller machine by allowing the automating task that needed human supervision,such as identifying many notes.We,therefore,recommend that a further investigation on the CNN model using advanced architecture like Goog Le Net and Res Net with the larger dataset to study the banknote classification and verification system.