Predicting Thamudic inscriptions pre and post-sequence using deep learning

This research examines the translation of ancient Thamudic inscriptions into contemporary Arabic script through the application of advanced deep learning methodologies. Convolutional neural networks (CNNs) alongside advanced models like ResNet-50, ResNet-18, Inception-v3, and GlyphNet are utilized to develop an automated system for translating Thamudic inscriptions. A dataset comprising 13,450 images of Thamudic inscriptions was developed. The models underwent training, validation, and testing with this dataset, resulting in high accuracy rates and strong performance in the identification and classification of Thamudic inscriptions. The ResNet-18 model demonstrated exceptional performance on all critical metrics, achieving an accuracy of 99.59%, precision of 99.50%, recall of 99.55%, and an F1-score of 99.52% indicating its effectiveness in identifying and translating Thamudic inscriptions with high precision and minimal error. This contribute to the solution of assisting historians and researchers in interpreting ancient texts while supporting Saudi Arabia’s preservation and promotion of its historical and cultural heritage.