Lezing | Digital Archaeology Group
Ancient Storage and AI
- Datum
- donderdag 22 januari 2026
- Tijd
- Locatie
-
Van Steenis
Einsteinweg 2
2333 CC Leiden - Zaal
- E002a / live stream
Automated object detection of prehistoric granaries on archaeological GIS maps: A Deep Learning approach
With the growing prominence of Artificial Intelligence, new opportunities are emerging for bringing these techniques into archaeological research, though its practical use is still in its early stages. This talk presents research on the application of Deep Learning for automated feature detection in GIS maps. More specifically, the study investigates the use of the YOLOv8s object-detection algorithm to automatically identify prehistoric granaries on Dutch archaeological excavation maps, an approach that addresses the limitations of traditional manual analysis, which currently is often time-consuming, labour-intensive, and challenging when dealing with complex spatial datasets.
The results show that Deep Learning models can (to a degree) reliably detect prehistoric granaries with promising accuracy and precision, suggesting that automated approaches can meaningfully support archaeological interpretation. Subsequently, the study also identifies limitations, including the challenge of interpreting model decisions, sensitivity to background noise, and the need for comprehensive and uniform datasets. Addressing these factors is critical for ensuring that automated detection methods are reliable, responsible, and informative. While further work is certainly required to refine detection models and improve their generalisability across different archaeological sites, time periods, and structure types, this research illustrates the potential for AI-driven methods to enhance the study of spatial data. By combining computational tools with domain expertise, researchers can potentially develop more efficient workflows and explore new avenues for recognising archaeological structures.