] in 
rich 
ion. 
vere 
des 
set 
sed 
me 
ed. 
the 
ing 
the 
ere 
ed. 
on. 
ey 
um 
ing 
dy. 
ich 
the 
ich 
le- 
the 
  
  
Satellite image 
  
  
  
Y 
Y 
y 
  
  
y wv 
  
  
Segmented image 
  
Training statistics 
  
  
  
  
  
Y 
| 
"y 
Cresson 
  
Y 
y 
  
Topographic map 
  
  
  
Preclassification result 
Field measurements 
  
  
  
L 
  
  
  
  
  
Y 
Attributes 
  
  
  
  
t 
Rule base 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Figure 2. The interpretation process. 
3.3 Postclassification 
The advantage of a rule-based method is that different kinds of 
digital data can be used in the interpretation if the data is in 
image format and relevant decisions can be made from it. A 
satellite image itself is a valuable data source because of its 
multitemporal and multispectral properties, but the low spectral, 
spatial and temporal resolution limits its ability to identify 
objects with almost similar spectral signatures or objects which 
are smaller than the sensor's IFOV on the ground. In these cases 
more information is needed to correctly interpret the objects. 
The data from GIS, e.g. a topographic map, existing land use 
map. soil map and road network etc., usually provide such 
information. 
Attributes. The attributes used in the rule-based 
postclassification are presented in Table 1. They were derived 
from the preclassification result and the old topographic map. 
  
Data Attribute 
Topographic map Land use class 
Height 
  
Preclassification result Land use class 
  
  
  
  
Table 1. Attributes used in the postclassification. 
Rules. Knowledge used in the interpretation could be taken 
from the experience and knowledge of objects, history and 
development of the objects, natural law and results of 
interviews with experts and users. Considering the nature of 
such kinds of knowledge, it may be partial or unreliable. This 
leads to a degree of uncertainty regarding the conclusions that 
could be drawn from the knowledge. 
In this study, information about the landscape and its changes 
was collected by field measurements. Before a rule-base for the 
postclassification was created, investigations on the attributes 
and their relationships with ground classes were made. For 
instance, the old land use map was compared to the reference 
points and probabilities for changes in land use were calculated. 
The rules and belief values given by the rules were mainly 
based on these investigations. The rules were collected in a 
separate rule base. 
A rule can confirm or disconfirm any class in the hierarchy 
(Figure 1) by assigning a confirming or disconfirming belief to 
this class. Because of the different information contents of the 
different data sources, rules derived from the preclassification 
result give belief to the terminal classes in the hierarchy, while 
rules derived from the topographic map give belief to the major 
land use classes. Most of the rules used in this study were based 
997 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996