Mapping hidden heritage: Self-supervised pre-training on high-resolution LiDAR DEM derivatives for archaeological stone wall detection

A novel self-supervised learning framework, DINO-CV, achieves high accuracy (68.6% mIoU) in mapping hidden archaeological stone walls using high-resolution LiDAR DEM derivatives, even with limited labeled data (63.8% mIoU at 10%). This approach addresses vegetation occlusion and data scarcity challenges by learning invariant geometric features from multi-directional hillshade and topographic visualization. Applied to the UNESCO-listed Budj Bim Cultural Landscape in Australia, it demonstrates potential for scalable, automated heritage mapping in complex environments, with implications for environmental monitoring and preservation.

Huang, Z., Islam, M., Armstrong, B., Bell, B., Khoshelham, K., & Tomko, M. (2026). Mapping hidden heritage: Self-supervised pre-training on high-resolution LiDAR DEM derivatives for archaeological stone wall detection. Science of Remote Sensing, 13, 100372. https://doi.org/10.1016/j.srs.2026.100372