SerendibCoins: Exploring The Sri Lankan Coins Dataset
NH Wanigasingha, ES Sithpahan, MKA Ariyaratne, PRS De Silva
TL;DR
This work tackles the lack of public datasets for Sri Lankan coin recognition by introducing the Sri Lankan CoinVision dataset and benchmarking multiple classifiers. It combines a reproducible data-collection methodology with three feature/extraction strategies: OpenCV ORB with traditional ML, MobileNetV2 as a feature extractor, and a from-scratch custom CNN. Key findings show that SVM outperforms KNN and RF among traditional methods, the MobileNetV2-based approach yields perfect accuracy on the tested splits, and the custom CNN achieves near-perfect performance with only a few misclassifications, establishing a strong upper bound for this task. The dataset, with 24,867 images across eight coin classes (old and new variants), is publicly available on Kaggle and provides a solid baseline for automated coin recognition and future regional currency research.
Abstract
The recognition and classification of coins are essential in numerous financial and automated systems. This study introduces a comprehensive Sri Lankan coin image dataset and evaluates its impact on machine learning model accuracy for coin classification. We experiment with traditional machine learning classifiers K-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Random Forest as well as a custom Convolutional Neural Network (CNN) to benchmark performance at different levels of classification. Our results show that SVM outperforms KNN and Random Forest in traditional classification approaches, while the CNN model achieves near-perfect classification accuracy with minimal misclassifications. The dataset demonstrates significant potential in enhancing automated coin recognition systems, offering a robust foundation for future research in regional currency classification and deep learning applications.