Pakzad, Kian 
KNOWLEDGE BASED INTERPRETATION OF MOORLAND IN AERIAL IMAGES 
Kian PAKZAD, Christian HEIPKE 
University of Hanover, Germany 
Institute for Photogrammetry and Engineering Surveys 
pakzad @ipi.uni-hannover.de 
KEY WORDS: Change detection, Knowledge representation, Model-based processing, Monitoring, Multi-temporal 
ABSTRACT 
For the interpretation of remote sensing data the traditional methods such as multispectral classification are in many 
cases not sufficient. This applies especially to more complex scenes. In order to interpret such scenes it is necessary to 
include and use more prior knowledge about the depicted objects, e.g. knowledge about the possible object structure or, 
in a multitemporal interpretation, knowledge about the possible temporal changes. 
In this paper we present an approach for the automatic interpretation of moorland from aerial images. The first step is a 
monotemporal interpretation. We use a knowledge based system with an explicit knowledge representation through 
semantic nets. This system is suitable to formulate explicitly (i.e. in a standard language) prior knowledge and to use it 
for the interpretation. In our case we divided moorland into different relevant land use classes and described them in a 
semantic net. For every class we described the obligatory parts. Obligatory parts are features and structures, which have 
to be detected in the particular areas in order to assign them the corresponding class. 
Because in moorland areas monitoring of changes is very important we extended the monotemporal system to a 
multitemporal one. The multitemporal interpretation also exploits explicitly represented prior knowledge about the 
possible temporal changes. 
The results show that the presented approach is suitable for the interpretation of moorland. The exploited additional 
prior knowledge led to an improvement of the interpretation, especially for the multitemporal one. 
1 INTRODUCTION 
Monitoring of moorland is necessary, because this area is both an environmentally sensitive region and also of interest 
for industrial work. One way to accomplish this task is to use remote sensing techniques. In multispectral, moorland is 
sometimes detected, but often appears as just one single class. At a closer look, however, a moor area consists of 
regions with quite heterogeneous use. 
In this work we present an approach for an automatic knowledge based and rule based multitemporal interpretation of 
moorland. Up to now the usual methods for an interpretation of such areas were data driven multispectral classifications 
(e.g. NMU (1997)). But standard multispectral classifications don’t use prior knowledge about the area to be interpreted. 
For moorland such prior knowledge is for example the fact that peat extraction is performed mostly by using harvester 
machines. Theses machines leave visible tracks on the ground which can be recognized as lines in the related images. 
The use of prior knowledge has the potential to improve the interpretation because it reduces the search space by 
additional constraints. 
Our interpretation is divided into two parts: The first part is the monotemporal interpretation and uses only information 
regarding geometry, radiometry and texture of the land use classes. The second part is the extension to a multitemporal 
interpretation. The necessary temporal information for this part is formulated in a diagram, which describes the most 
probable state transitions. In the monitoring process the state transition diagram is used to predict the possible land use 
changes. This leads to a reduction of the search space and improves the monitoring. 
In the next section we briefly describe the knowledge based system we used for the interpretation. In section 3 the prior 
knowledge we used for interpretation of moorland and in section 4 an overview about the recognizability of moorland 
classes is given. Section 5 describes the interpretation process: After the description of the used input data (5.1) and a 
concept for an initial segmentation of the image (5.2) the monotemporal interpretation process together with the 
obtained results will be shown in section 5.3 and the extension to the multitemporal interpretation is demonstrated in 
section 5.4. Section 6 concludes the paper and gives an outlook for further research. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1103