Site detection across Europe and the HillfortFinderApp – latest results from a Deep Learning based hillfort search

Abstract The widely increasing availability and improving resolution of LiDAR data are revolutionising landscape archaeology, allowing the identification of previously unknown archaeological features and thus aiding research as well as heritage protection strategies. This can be achieved even on a countrywide scale by using effective automation techniques. Our work aims at developing an approach to automatically detect hillfort sites cross-regionally, using Britain to train a neural network tool and Italy and Germany to test its transferability to different regions and landscapes. Hillforts are amongst the most evident and iconic archaeological sites in Europe. Usually associated with the Iron Age, they can date from the Neolithic to early Medieval times. The Atlas of Hillforts (Lock and Ralston 2017) provides a comprehensive and up-to-date record of these sites, over 4000 in total, and an invaluable database for further research (e.g., Maddison 2022). The large number of hillforts known in Britain constitutes the best training dataset available to date for developing effective automation for recognising hillforts in remotely sensed data. The variety of landscapes present in Italy and Germany, on the other hand, makes these areas an optimal case study for testing the transferability of this type of automation in different regions