The data explosion: tackling the taboo of automatic feature recognition in the use of airborne survey data for historic environment applications (original) (raw)

The data explosion: tackling the taboo of automatic feature recognition in the use of airborne survey data for historic environment applications

The increasing availability of multi-dimensional remote-sensing data for large geographical areas is generating a new wave of landscape-scale research that promises to be every bit as revolutionary as the application of aerial photographic survey during the 20th century. Data are becoming available to historic environment professionals at higher resolution, greater frequency of acquisition and lower cost than at any previous time. However, to take advantage of this explosion of data a paradigm change is needed in the methods used to routinely evaluate aerial imagery and interpret archaeological evidence. Central to this shift is a fuller engagement with the concept of computer aided methods of feature detection as a viable way to analyse airborne and satellite data. This requires a reassessment of workflows and an understanding of the different types of information that may be generated, developing a multi-stage assessment based on explicit predicates vital to providing confidence in outputs, while improving our ability to work with vast datasets. Automated and semi-automated image analysis is routine in fields such as environmental remote sensing, where they underpin the analysis of extensive datasets (see Lasaponara and Mansini (2012) for a review of these techniques which is not replicated here). While aspects of these developments have made their way into archaeological applications, they remain far from routinely used and are often viewed with suspicion. This paper outlines the status quo and key issues for readers unfamiliar with debates that have mainly played out at conferences and steering groups but are seldom committed to print. We summarise the difficulties surrounding the creation of geographically extensive, systematic datasets for heritage management and research from high-volume data derived from airborne laser scanning (ALS), satellite imagery and airborne digital spectral data. At the heart of this issue is the potential to incorporate methods that exponentially increase the rapidity with which initial historic environment datasets can be created, moving beyond the ‘human-timescales’ within which most archaeological survey is undertaken (i.e. ‘manual’ analysis), to fuller exploitation of computer assisted techniques. In order to do this we must recognise the fundamentally different but complimentary types of information that the two approaches produce, and assess their value and contribution as part of broader research objectives. Throughout we use the term computer vision to define all methods by which imagery can be processed, analysed and understood using computer generated algorithms (including classifications) and in particular with reference to the electronic replication of the abilities of human ocular perception (Sonka et al. 2008).