ion C of 
objects. 
Yould be 
s should 
is in the 
f objects 
icultural 
= 4 
regation 
at are of 
se class. 
shows a 
> newly 
original 
S can be 
e shows 
latabase 
per step 
only the 
and 12. 
ed to the 
/ to jump 
step. In 
> original 
le same 
ve been 
the step 
| objects 
culture". 
re. They 
> values. 
omposite 
riculture” 
cts. The 
then be 
ariability 
fter each 
assigned 
r alarger 
n spatial 
ough the 
it also 
fe TN IT TEA CE Te 1 
| natveg | agriculture | | nat veg | 
    
i 
  
    
  
Ë 
  
[ 
   
agriculture | [natgrsind] | forest 
"agriculture | natgrsind| forest || agriculture 
[erue dm ed ed. 
  
  
   
(9,7,(8,9,14,10 (1,6 5,8125. (27) (36,582) ago. 
| S cer eu A a. un ~ E eT Sa f 
E : i à © V = D. 
   
  
(nat grsind | | dec frst ||conif frst | ar Ind | pasture| 
euo med - 
D Ed. b 
(27 (8,94; 4,10) (1,6) (88 12^ 
Si Bd 
Vv 
  
  
   
    
  
object aggregation 
class generalization 
fig. 5: A diagram representing the generalization and aggregation steps of the object generalization process of figure 
4. 
expressed by the spatial variability of the attribute values. 
When the attribute values are of the ratio scale type then 
the aggregated value can often be obtained by summation 
or by taken the average value over the objects that 
compose the new object, e.g.: 
AIO, ] = 2p e compion ALO; 1 
Examples are attributes like wood volume and crop yield 
and population. For other attributes like vegetation cover 
or population density it might be that (weighted) averages 
should be computed. 
  
attr. values n 
[ attr. values 2 
= attr. values 1 
fig. 6: The aggregation of attribute values. 
  
  
3.2. Functional Object Generalization 
It is certainly not always so that object aggregation can 
be done within the framework on one class hierarchy. In 
many cases object aggregation will imply a completely 
different thematic description of the objects, so that new 
classes should be defined. This is illustrated in figure 7 
where farm yards and fields have been aggregated into 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
  
  
  
  
  
  
  
AGGREGATION 
LEVEL 1 LEVEL 2 LEVEL 3 LEVEL 4 
farmyard 1 
feld 1 ., At! NE red P. 
field2 + lot2 T una i 
fied3 ^" 43 3 am fam district 
level 1 level 2 farm n d 
Speo ut o level 3 level 4 
fid mA, 
> DM i ds 000 
aa == [lhe 
figa | LL SES 
  
  
  
  
  
  
  
  
  
  
  
  
fig. 7: An example of a functional object generaliza- 
tion process. 
farms and these in their turn into farm districts. The aggre- 
gation hierarchy has a bottom up character in the sense 
that starting from the elementary objects composite objects 
ofincreasing complexity are constructed in an upward dire- 
ction (in figure 7 from left to right). The farm districts should 
only consist of farms and the farms should be mutually 
adjacent so that the adjacency graph (see section 4) of 
the farms that belong to one district is connected. 
The aggregation steps in figure 7 show how the fields are 
considered as elementary objects at level 1. They are 
defined per growing season as spatial units under one crop. 
For the farmer they are management units, because his 
management operations are planned and performed per 
field. They are aggregated to lots which are elementary 
objects at level 2, i.e. these objects belong to the extensions 
of classes such as "arable-lot" and "grass land". These 
are management units at a higher level; the farmer will 
549