Why, when and how? Context and computer vision in archaeological prospection and interpretation

Dave Cowley, Karsten Lambers, Arianna Traviglia


As the session abstract identifies, the debate over applications of semi-automated or supervised feature extraction techniques in archaeology has tended to be polarised between believers and doubters. In large part the polarisation of discussion probably stems from a lack of explicitness about how archaeological feature identification is undertaken and how the processes, whether ‘automated’ or ‘human’, of identifying patterns, shapes and features interrelate with archaeological interpretation. Furthermore, the varying contexts of these processes in multi-scaled archaeological prospection have not necessarily been defined as fully as they might. Such ambiguity is unhelpful to exploration of the potential of new techniques, and the interfaces between traditional archaeological skills/knowledge and emergent tools. It also makes definition of common purpose difficult. This paper will explore these issues, looking to address the reasons why the application of computer vision to archaeological prospection is vital, but also examining the workflows and outcomes of different approaches and processes, both manual and automated. These are important steps in addressing some key questions, including: how can we create clarity about why and when automated approaches are desirable?; what are the roles of (traditional/manual) archaeological experience and skills in designing algorithms?; and how can automated/manual approaches be used iteratively to improve archaeological detection?