AIDA: A SYSTEM FOR THE KNOWLEDGE BASED INTERPRET ATION OF REMOTE SENSING DA TA *

Because of the increasing amount of remotely sensed imagery there is a growing need for efficient data analysis techniques. Here to automate the image interpretation a knowledge based approach is suggested. The presented scene interpretation system AIDA uses semantic nets for the explicit repre sentation of the prior knowledge about objects expected in the scene. It exploits the knowledge base to generate a scene description assigning symbolic meanings to the image primitives. The information of a GIS database is used as partial interpretation to produce reliable hypotheses for the expected objects. This initial scene description is verified consecutively in the remote sensing imagery. Multiple sensors can be investigated simulta neously. The explicit knowledge representation eases the adaptation to dif ferent tasks. The system was tested successfully in applications like verification of GIS data, recogni tion of complex structures, the automatic search of tie points for the registration of remotely sensed images, and the object specific 3D modelling of landscapes and buildings. The recognition of land use changes for map updating and environmental and agricultural monitoring represents a major topic of remote sensing. Due to the large amount of acquired data algorithms for the automatic extraction of objects from sensor data are investigated. This contribution suggests a knowledge based approach for image interpretation using semantic nets. The presented scene interpretation system AIDA generates a symbolic scene description which can be used for recognition tasks (Koch 1997) as well as for 3D reconstruction of the detected objects (Grau 1997, Tonjes 1996) or automatic registration of multi sensor data (Growe 1997). A lot of scene interpretation systems have been developed in the past. They dif fer in a number of aspects like the control strategies, the knowledge representation or the application domain. Concerning the representation of the scene knowledge a lot of systems for aerial image interpretation like SPAM (McKeown 1985) use rules. Production rules and semantic nets are found in MESSIE (Clement 1993), while ERNEST (Niemann 19990) uses semantic nets only. SIGMA (Matsuyama 1990) is organized in three expert modules using frames and rules. In AIDA the knowledge about the scene objects is formulated in a semantic net, while the control knowledge is represented by rules.