Potential of deep learning segmentation for the extraction of archaeological features from historical map series

Historical maps present a unique depiction of past landscapes, providing evidence for wide range information such as settlement distribution, land use, natural resources, transport networks, toponymy and other cultural data within an explicitly spatial context. Maps produced before the expansion large-scale mechanized agriculture reflect landscape that is lost today. Of particular interest to us great quantity archaeologically relevant these recorded, both deliberately incidentally. Despite importance they contain, researchers have only recently begun automatically digitize extract from coherent information, rather than manually examine raster image. However, new approaches focused on specific types cannot be used directly archaeological or heritage purposes. This paper provides proof concept application deep learning techniques historical in automated manner. Early twentieth century colonial map series been chosen, provide enough time depth avoid many recent modifications cover very large areas (comprising several countries). The use common symbology conventions enhance applicability method. results show efficient tool recovery georeferenced, represented conventional signs, line-drawings text maps. method can excellent when adequate training dataset has gathered therefore at its best applied supply information. described here open up possibility sites features across entire much more quickly coherently available methods, opening potential reconstruct landscapes continental scales.