Automated Detection and Recognition of Sinhalese Inscriptions Using YOLOv5

Abstract

This study proposes to recognize words in Sinhalese inscriptions using computational techniques as inscriptions form a foundational element of Sri Lanka’s historical record and cultural heritage. Computational Archaeology plays a vital role in fields of Archaeology, Linguistics, and Anthropology. There is a significant advancement in using computational techniques to recognize text on inscriptions since manual interpretation and transcription are highly effort intensive, resource demanding and prone to inaccuracies. The model trained and evaluated on a curated dataset of 300 tokenized Sinhalese inscription images. The performance of You Only Look Once (YOLO); a deep learning model, was analyzed based on standard evaluation metrics—accuracy, precision, recall, and F1 score. Results indicate promising average accuracy of 90% and demonstrates YOLOv5’s superiority in handling the unique challenges of ancient Sinhalese epigraphy such as irregular layouts, paleographic variations, and surface degradation compared to Optical Character Recognition systems. This research not only seeks to preserve and enhance access to Sri Lanka’s rich cultural heritage but also provides a wider scope for linguistic and scholarly inquiry by facilitating more efficient analysis of ancient texts through automated recognition of Sinhalese inscriptions.