A brave new world for archaeological survey: Automated machine learning-based potsherd detection using high-resolution drone imagery

Abstract

Archaeological pedestrian survey is one of the most popular techniques available for primary detection of archaeological sites and description of past landscape use. As such it is an essential tool not just for the understanding of past human distribution, economy, demography and so on but also for cultural heritage management and protection. The most common type of pedestrian surface survey consists of fieldwalking relatively large tracts of land, recording the dispersion of items of material culture, predominantly pottery fragments, by teams of archaeologists and students. This paper presents the first proof of concept for the automated recording of material culture dispersion across large areas using high resolution drone imagery, photogrammetry and a combination of machine learning and geospatial analysis that can be run using the Google Earth Engine geospatial cloud computing platform. The results show the potential of this technique, under appropriate field circumstances, to produce accurate distribution maps of individual potsherds opening a new horizon for the application of archaeological survey. The paper also discusses current limitations and future developments of this method.