SkyEye, a machine learning software to detect archaeological structures in LiDAR Dataset
0Nathanaël Le Voguer, Clément Laplaige, Xavier Rodier
The goal of this paper is to present the first results produced by the SkyEye software, developed by the Laboratoire d’Informatique Fondamentale et Appliquée (University of Tours), in collaboration with the Laboratoire Archéologie et Territoire (CITERES, CNRS/University of Tours) and the SOLiDAR research program. This software, by using images extracted from a DTM (Digital Terrain Model), aims to automatically detect archaeological structures. The first step is to do a manual research and vectorise some of the structures. Then, we use them to create binary images which are classified by the software and compared to the pixels on the 8 bit DTM image using an SVM (Support Vector Machine) classifier. Afterwards, the software is able to recognize these types of pixels and their neighbours based on their grayscale values (representing altitude differences). It will recognize similar structures on another DTM image by analysing the value of the pixels. For this presentation, we will focus on two examples, firstly with linear structures (embankments) and secondly with point structures (charcoal burning platforms); and on two different wooded areas in the Centre region of France. The software was first developed for linear structures, therefore it has a good recognition rate for these remains. The use for discrete structures, more recent, needs to be improved but the results are encouraging. One of the main issues is that the software is extremely dependent on the quality of the data and on the learning datasets in particular. Manual exploration is necessary to build the learning datasets as well as for post-control to confirm results and improve determination rates.