Ethiopian Fake Note Currency Detection Using Deep Learning Approach

Abstract

Ethiopian Government has introduced new currency notes to fight organized crimes and counterfeiting on September 2020. The currency notes have got a complete new feature including color changes. It is also believed that the new features will make counterfeiting impossible, but within just a few weeks after new currency is introduced fake notes of the new currencies being circulating in the market becoming bottleneck for the country economy, increase inflection and reduces the value of original money. And to address the problem macro and giant organizations like Bank using a special machine that uses UV light to detect the fake notes, but this is not applicable for every one especially for the individual since the machine is expensive relatively. Thereby, we consider the CNN model with better architecture and investigating and developing AI model that can detect Ethiopian Note Currency either it is fake or genuine from images. In the entire research work we have followed the design science methodology that focusing on the development and validation of prescriptive knowledge and model development. In the compression of others work we have used two art of science pre trained model to implement the transfer learning while the other model is based on CNN that we have designed from scratch. And the three models (Our CNN, VGG16 and EfeceinetB0) performance is 98.6%, 99.8% and 98.9% respectively on test data. After all performance of VGG16 have been reported as highest accuracy. Based on that we have recommend using bigger data set with larger Network like Resnet50, Resenet150, and EffeceinetB7 with high optimized layers.