Mapping Hidden Heritage: Self-supervised Pre-training for Archaeological Stone Wall Mapping in Historic Landscapes Using High-Resolution DEM Derivatives

A groundbreaking approach using self-supervised deep learning has achieved high accuracy in mapping dry-stone walls within the Budj Bim Cultural Landscape, a UNESCO World Heritage site. The method, DINO-CV, leverages cross-view pre-training on high-resolution Digital Elevation Models (DEMs) derived from airborne LiDAR data, effectively addressing challenges like vegetation occlusion and limited labeled training data. Achieving a mean Intersection over Union (mIoU) of 68.6% in test areas and maintaining 63.8% mIoU with just 10% labeled data, this framework demonstrates potential for large-scale, automated mapping of cultural heritage features. The approach not only advances archaeological documentation but also offers scalable solutions for environmental monitoring and heritage preservation in complex landscapes.

Huang, Z., Islam, M., Armstrong, B., Bell, B., Khoshelham, K., & Tomko, M. (2025). Mapping Hidden Heritage: Self-supervised Pre-training on High-Resolution LiDAR DEM Derivatives for Archaeological Stone Wall Detection (Version 3). arXiv. https://doi.org/10.48550/ARXIV.2510.17644