Definition and identification of objects on a high spatial resolution image. Application to the differentiation of chestnut orchard types

Abstract

High-resolution spatial data are bound to be more and more reachable for all users. Facing so many data, we propose to define pertinent object, both on thematic and methodological levels. The aim is to automate the extraction of objects that are suitable for research and management. This process is used to distinguish classes of chestnut tree orchards and copses with infrared aerial photos of IFN. A chestnut tree classification is elaborated so as to associate shapes with agricultural practices. Not only do objects differ in colour but also in component (chestnut tree crowns) spatial distribution. CLAPAS is selected: the classification integrates points proximity which allows to distinguish classes with the same colour but differing in texture. Reference classes are chosen on the image thanks to vegetation observation (assisted classification). The analysis shows that, in spite of confusions, output fit with sites checking.