Detecting Iranian Paper Money to Help People With Visual Disabilities

This project, undertaken just before I began my university studies, was my first application of deep learning to solve a real-world problem. The inspiration came from Microsoft’s Seeing AI app, which offers a range of tools for visually impaired individuals. One of its standout features is the ability to identify currency bills, a functionality I aimed to replicate for Iranian currency, which was not supported by the original app. A significant challenge was the lack of an existing dataset. I dedicated a month to collecting and curating a suitable dataset of Iranian banknotes. During this time, I researched various approaches and decided to use transfer learning with the MobileNet architecture, as the goal was to deploy the model on an Android device. TensorFlow Lite provided the ideal framework for on-device inference.