The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLVI-2/W1-2022
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVI-2/W1-2022, 365–372, 2022
https://doi.org/10.5194/isprs-archives-XLVI-2-W1-2022-365-2022
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVI-2/W1-2022, 365–372, 2022
https://doi.org/10.5194/isprs-archives-XLVI-2-W1-2022-365-2022

  25 Feb 2022

25 Feb 2022

SEEING AMONG FOLIAGE WITH LIDAR AND MACHINE LEARNING: TOWARDS A TRANSFERABLE ARCHAEOLOGICAL PIPELINE

G. Mazzacca1,2, E. Grilli1, G. P. Cirigliano3, F. Remondino1, and S. Campana3 G. Mazzacca et al.
  • 13D Optical Metrology (3DOM) unit, Bruno Kessler Foundation (FBK), Trento, Italy
  • 2Centre of Geotechnologies, University of Siena, Italy
  • 3Department of History and Cultural Heritage, University of Siena, Italy

Keywords: point cloud classification, machine learning, remote sensing, landscape archaeology

Abstract. Airborne LiDAR technology has become an essential tool in archaeology during the last two decades since it allows archaeologists to measure and map items or structures that would otherwise be hidden under vegetation. In order to detect and characterise the archaeological evidence, it is a common practice to extract accurate digital terrain models (DTM) by filtering out the vegetation from Airborne Laser Scanning (ALS) datasets. Although previous approaches have performed well in ALS filtration, they are still subject to several variables (flight height, forest cover, type of sensors utilised, etc.) and are frequently integrated into expensive commercial software or customised for specific locations. This study presents a workflow for treating ALS archaeological datasets using machine learning algorithms for both filtering the vegetation and detecting hidden structures. The workflow is applied to two different archaeological environments (in terms of structures, vegetation, landscape, point density), and results demonstrate that the pipeline is rapid and accurate, and the prediction model is transferable.