Improving Real-Time Currency Recognition for Visual Impairments: Using MobileNetV2 and Extreme Learning Machines

For the majority of people living in modern cities, distinguishing denominations of banknotes is a daily routine involving no special effort. However, for people with visual impairments or blindness, this seemingly undemanding task is characterized by unimaginably tough challenges. Given that the capacity to detect and identify money denominations in a real-time environment is critical to independence, including fast transactions to meet the needs, the present study suggests a profoundly innovative solution using the convolutional neural network model for mobile devices. Given the critical role of money requirements in the daily routine and refusal to live in society for people with disabilities, the novel approach aims to enhance the system's performance efficiencies and reliabilities for both Indian currency identification and denomination detection. The novel approach to using ELMs in computer vision models and replacing the fully connected layers in the traditional MobileNetV2 model with them significantly contributes to improving the system's adaptation. The experimental results exhibit high values, with the mobile net model demonstrating an accuracy of parcel detection of 97.80%. Hence, the novel system independently and operates in real-time environments to identify currencies and help people with visual disabilities participate in economic activities.