DEVELOPMENT OF A RUPIAH BANKNOTE IDENTIFICATION MODEL IN THE CURRENCY SENSE APPLICATION FOR VISUALLY IMPAIRED

Abstract

This study develops and evaluates an Android based assistive application, Currency Sense, designed to help visually impaired individuals recognize Indonesian Rupiah banknote denominations independently. The application employs a Convolutional Neural Network to perform image based currency classification and delivers the recognition results through audio feedback. The research methodology includes the development of a CNN based classification model, system implementation on the Android platform, and performance evaluation using a confusion matrix. The experimental results demonstrate that the proposed model achieves an accuracy of 89.28%, with the highest recognition performance observed for the Rp50,000 and Rp100,000 denominations. Several misclassifications were identified among visually similar denominations, primarily influenced by variations in lighting conditions, image orientation, and image noise. In addition to model evaluation, black box testing was conducted to assess application functionality across four main interface pages using ten test scenarios, all of which produced valid results. These findings indicate that the application functions reliably and meets functional requirements. The proposed system contributes to artificial intelligence–based accessibility solutions by offering a practical tool to enhance autonomy in financial transactions for visually impaired users.