Moving from pixels to parcels: The use of possibility theory to explore the uncertainty associated object oriented remote sensing

This paper explores the issues relating to uncertainty in the application of object oriented classifications of remote sensing data. Object oriented remote sensing software such as eCognition (now known as Definiens Developer) provides the user with flexibility in the way that data is classified through segmentation routines and user-specified fuzzy rules. However the aggregation of fuzzy data objects such as pixels to higher level parcels for the purpose of policy reporting is not straightforward. This paper explores the uncertainty issues relating to the aggregation from fine detailed (uncertain) objects of one classification system to coarser grain (uncertain) objects of another classification scheme. We show Possibility Theory to be an appropriate formalism for managing the non-additive uncertainty commonly associated with classified remote sensing data. Results are presented for a small area of upland Wales to illustrate the value of the approac