Remote Sensing, Formation of Objects and Determination of Quality

Object-oriented procedures are gaining increasing importance in the theory and practice of the analysis of remote sensing data. Unlike pixel-based procedures, object-oriented procedures are based on a priori defined objects or segmentation of regions. A priori defined objects can be derived e.g. from the ATKIS. Using computer-assisted segmentation, the picture is divided into more or less homogeneous regions. These regions can be characterized by features of different origin, e.g. spectral features, texture features, form characteristics and context relations. The representation and combination of the knowledge gained is implemented by modelling task-oriented concepts in semantic networks. At present the image interpretation system AIDA and the commercial software eCognition provide specially adapted environments for the objectoriented analysis of remotely sensed images. The operation requires the construction of proper rule networks which support the task-oriented extraction of objects. Beyond this procedures are to be developed which support the selection of relevant features, specify the force of an expression of a feature in relation to a target object, permit reliable classification of the object and allow quantitative predictions about the quality or classification accuracy of an object. 1. Why use Objectand Knowledge-Based Techniques? The last decade, especially the past few years in remote sensing work, have seen rapid progress in research and development work objectand knowledge-based techniques aimed at automatic image interpretation. This trend has been driven in part by the awareness that traditional approaches (e.g. pixel-based Maximum Likelihood classifications) fail to deliver the kind of information demanded by the consumer. Most remote sensing activities to date have been projects performed in specific locations and over a specified time, with classification techniques adapted to the specific domain. To this extent it may be claimed that, except for some applications, remote sensing cannot be called operational! Even if remote sensing is 1 DELPHI-IMM, Winkelmannstr. 24, D-39108 Magdeburg, Email: [email protected], Internet: http://www.delphi-imm.de