The detection of archaeological features hidden beneath vegetation remains a major challenge in landscape archaeology, particularly in forested and densely vegetated environments. Airborne LiDAR have demonstrated strong potential to penetrate vegetation and reveal micro-topography, yet many challenges in data processing are still present.
This PhD research aims to develop new and efficient AI-based solutions for processing LiDAR data (2D/raster and 3D/point clouds) and improve the detection of sub-canopy archaeological features across diverse environments.
The main research tasks include:
– Collection and pre-processing of airborne LiDAR datasets for the generation of multi-resolution terrain representations and training data and archaeological signatures (e.g., circular mounds, linear ditches, rectangular foundations, etc.) tailored for AI applications
– Feature engineering and representation learning to enhance archaeological signatures (e.g., micro-relief, edge structures, etc.)
– Design and implementation of new deep learning architectures (both supervised and unsupervised/few-shot, 2D and 3D) for an efficient and automatic detection, segmentation and classification of sub-canopy archaeological features such as earthworks, burial mounds, walls, etc.
– Investigation of interpretable/explainable AI methods, supporting archaeologists to understand AI results
– Generalization and transferability analyses, considering domain adaptation and transfer learning strategies to ensure model robustness across different geographic regions, vegetation types and LiDAR densities
– Improve processes to automatically derive vector information from the detected under-canopy structures
– Validation and benchmarking
The ideal candidate should have proficiency in coding, knowledge of GIS environments and point cloud processing, strong interest in heritage scenarios as well as a collaborative attitude for interdisciplinary work between computer scientists and heritage people.