A Deep Learning Approach to Recognizing New Thai Currency Using Transfer Learning for Visually Impaired Individuals
Irfan Ahmad, Abubakar Sharif, Aneeka Azmat, Muhammad Usman Jamil · 2024
Assistive mobile applications play a pivotal role for visually impaired individuals worldwide. These applications often face challenges in currency recognition due to varying perspectives, inconsistent illumination, and background clutter. This issue is especially pressing in developing countries like Thailand, where there is a notable gap in robust currency recognition systems, particularly for the new Thai currency notes. This study employs a deep learning approach using a convolutional neural network (CNN) to automate the recognition of the new Thai currency notes. Using transfer learning, we fine-tuned the CNN using the Xception model, renowned for its depth-wise separable convolution. The network trained on a meticulously curated dataset comprising 3600 images (without data augmentation) of five different denominations of the new Thai currency (20, 50, 100, 500, and 1000 baht) notes, captured under various conditions. The resulting model achieved an average training accuracy of 99.5% and a validation accuracy of 99.8%. Given its robustness and high accuracy, the model can be integrated into an Android application. Such an application would offer a user-friendly and reliable tool for visually impaired individuals to effortlessly identify the new Thai currency notes in their everyday transactions.