Abstract
Object-based remote sensing image classification is known for its ability to elicit objects that correspond one-on-one with real-world objects. However, it is still subject to a two-stage linear segmentation and classification process and a limited ability to use geometry, class identity and neighbourhood information in that process. This paper explores the scope of intelligent vector agents (VAs), potentially unifying segmentation and classification, and, as implemented through the Geographic Automata System framework, explicitly modelling a set of vector objects with (1) elastic geometry, (2) states, (3) neighbourhoods and embedded rules connecting all three. A brief illustration involving geometry, geometry rules and states is presented.