PATTERN RECOGNI EILON 
Dipl.-Ing. Bernhard Bargel 
Introduction 
This paper gives a review of those activities of the Forschungs- 
institut für Informationsverarbeitung und Mustererkennung" (FIM, 
Karlsruhe), which are connected to the topic of automatic or 
interactive picture processing of remote sensed data. Out of 
this, the survey is further restricted to the thème of classi- 
fication of objects which could be of interest to scientists 
in the field of agricultural research, forestry or urban planninge 
Hereby the interests of FIM are concentrated mainly on the de- 
velopment of procedures to evaluate pictorial data and less on 
the problems the individual researchers or users have to inter- 
pret the special phenomena they find in the pictures. In coope- 
ration with remote sensing scientists, it is therefore, the 
task of the institute to prepare the tools for computer aided 
picture analysis and to demonstrate which patterns can be 
extracted out of mono or multispectral data. 
Multispectral Patterns 
The most common method in the classification of remote sensed 
data is the usage of multispectral patterns. Here the intensity 
of radiation and reflectance in separated parts or channels of 
the visible and infrared area of the spectrum is used to dis- 
criminate between single classes of objects. Due to the lack 
of knowledge of the signature of objects, the impact of the 
environmental conditions, the illumination of the scene and the 
sensor characteristics a completely automatic unsupervised classi- 
fication is rarely possible. One has to recourse to the super- 
vised classification with an interactive selection of training 
areas, which specify the classes of objects and the representa-