Extraction of linear structures from LIDAR images using a machine learning approach

For extraction and characterization of archeological structures from LIDar data, most studies focus on manual spotting (vectorization) or automatic image processing (IP), while few studies have examined semi-automated methods based on machine learning (ML). In the context of the Solidar project, after trying to use classical image processing techniques, we propose to reflect on elements to be integrated in ML approaches for a better and a more flexible extraction and characterization of archeological structures discovered in the LiDAR datasets. Indeed, the LiDAR data reveal many varied remains over large geographic areas. Manual digitizing of these remains is a time-consuming activity and does not guarantee an exhaustive recognition of features. This article proposes to present: (1) the archaeological context of this work, (2) the searched objects in this study, (3) the first tests and (4) how the data will be processed in the near future.