History gets a high-tech ally
Machine learning accelerates discovery of ancient stone monuments in Arabia
August 19, 20:16
Archaeologists are turning to artificial intelligence to locate ancient stone structures in the Arabian Peninsula. In a recent project, researchers used AI to survey five areas totaling around 2,500 square kilometers in the southern Nefud Desert of Saudi Arabia, concentrating on terrain near hills and mountains—especially around Jubbah, a site known for prehistoric stone architecture and rock carvings.
Three deep-learning models—MA-Net, SegFormer, and U-Net—were trained for the task. Of these, MA-Net performed best, slightly outperforming SegFormer in accuracy. The team also found that doubling satellite image resolution markedly improved detection precision, although it raised training time from 10 to 22 minutes.
The AI did particularly well at recognizing mustatils, large rectangular ritual structures that are relatively easy to identify. Smaller features, such as stone cairns, pendant tombs, and triangular formations, proved harder to spot because of their complicated shapes or resemblance to the natural landscape. According to the researchers, machine learning won't replace on-the-ground excavation, but it can serve as a powerful tool for the initial mapping of ancient sites.
This work underscores how modern computational tools can enhance archaeology. By combining field methods with AI-driven analysis, scientists can map archaeological features faster and more efficiently, especially in remote or underexplored regions—opening up new avenues for understanding humanity's past.