ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002
SEMANTIC INTERPRETATION OF REMOTE SENSING DATA
J. Bückner, S. Müller, M. Pahl, O. Stahlhut
Institute of Communication Theory
and Signal Processing
Hannover, Germany
geoaida@tnt.uni-hannover.de
Working Group VII/7
KEY WORDS: Photogrammetry, Remote Sensing, Semantic Network, Knowledge Based System.
ABSTRACT
The interpretation of remote sensing data is a complex task because of the high variability of the image material under
investigation. At the Institute of Communication Theory and Signal Processing procedures for maximizing the degree of
automation of the interpretation of remote sensing data are developed. In this paper the system GEOAIDA is presented,
which allows an intelligent, concise and flexible control of a scene interpretation by utilizing a semantic scene description.
The system produces a hierarchic, pictorial description of the results as well as the structural context of the identified
objects including the associated attributes. The output of GEOAIDA can be used for update of geographic information
systems and for map generation.
1 INTRODUCTION
Based on good experience with the system AIDA (Tónjes
et al., 1999) (Tónjes, 1999) (Growe, 2001) developed at
the Institute of Communication Theory and Signal Pro-
cessing, the system GEOAIDA was developed to put the
experimental oriented system AIDA into practice. Parts of
this contribution were submitted for a special issue of the
ISPRS Journal of Photogrammetry and Remote Sensing,
planned to appear in April 2003. The basic approach to
integrate additional knowledge in terms of a semantic net-
work into the analysis is adopted.
Originated in all examined problems the conception of
GEOAIDA (Biickner et al., 2000) is focused on the inter-
pretation of remote sensing data. Hereby an exclusive hier-
archical description of the problem in a semantic network
arose. Furthermore the possibility to add holistic opera-
tors is integrated. In AIDA only the leaf nodes of the used
semantic network contain image processing operators to
extract objects of the image data. The following grouping
of the objects implied the problem, that the combinational
diversity was often very high, because all objects extracted
from the image had to be taken into account at the same
time.
The so-called holistic operators can reduce the problem
of the combinational diversity, by interpreting one image
holistically. Holistic image processing operators can be
connected to all nodes of the semantic network. The task
of holistic operators is to divide a region into sub-regions
and to reduce the possible alternative interpretations if ap-
plicable. The structural interpretation of the sub-regions
follows and can verify or disprove the holistic results.
The aims of the system GEOAIDA are:
e Generation of a structural description of a given
scene.
e Generation of hierarchical thematic maps.
e Multi-sensor analysis.
e Flexible integration of any image processing
operator to the system.
e Treatment of alternative hypothesis.
e Inclusion of object relations to the analysis.
e Introduction of previous knowledge, e.g. knowledge
of geographic information systems.
e The system offers a clear sequential control.
e Continuous geo-references of all objects.
e Clear and flexible structuring of the problem
description.
e Parallelizing of all processes to accelerate the
analysis.
e Easy to learn and expand by use of
XML -interfaces.
The basis for interpretation of remote sensing data are re-
sults generated with image processing operators. In this
context, image processing operators are all operators, that
generate a labeled result image of a given image under use
of a function. Each label has a special meaning. Such im-
age processing operators are denoted classifying operators.
They can fulfill simple threshold operations, texture based
or model based methods and build the basis for an inter-
pretation.
The different results of such operators are structured in
GEOAIDA for optimal use. The problem, that different
image processing operators can generate different informa-
tion of the same region in the image, or that image process-
ing operators generate wrong part interpretations is solved
by use of addition knowledge, e.g. neighborhood relations.
This additional knowledge is formulated in the nodes of the
semantic network. Here, part-of relations or neighborhood
relations of objects are specified.
Examples for typical input data of the system are shown in
figure 1. The figure shows a region, recorded with three
different sensors. The upper part is an example taken from
an image recorded with a sensor in visual range, the middle