International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004 
  
temporal model. The investigated scene corresponded to a 
rural area, and the analysis of the images required specific 
agricultural knowledge, modeled into a formalism, so-called 
timed automata, as a priori knowledge about the scene. 
Suzuki et al. (2001), integrated structural knowledge to the 
image classification process. Basically, a fuzzy classifier 
generate a preliminary partition of the image and, then, the 
system tries to improve the initial classification. 
In order to evaluate the potentiality of knowledge-based 
approaches for the interpretation of low-resolution satellite 
images, the following papers were produced as part of the 
ECOWATCH project: Müller (2003) and Mota (2003) 
investigated the usage of the GEOAIDA system to perform 
knowledge-based classification of a SPOT 3 XS and 
LANDSAT 7 TM, respectively. In such essays, only spectral 
and contextual knowledge were employed. The obtained 
results demonstrated the potential of knowledge-based 
approaches to the interpretation of low-resolution satellite 
images. Pakzad et al. (2003) presented a procedure for the 
multitemporal interpretation of LANDSAT 7 TM images of 
a region covered by different categories of vegetation. In 
such article, an object oriented classification was performed 
employing the commercial software eCognition simulating a 
multitemporal reasoning. Given the previous classification of 
an object, the procedure modeled possible temporal state 
transitions, taking into account either ecological, agronomic, 
or legal restrictions. 
3. METHODOLOGY 
3.1 General model description 
The proposed framework considers data corresponding to 
different time instances. Hereafter, £ represents the time in 
which the image to be interpreted was acquired, f-/ a 
previous time instance and Af the time interval between 1-7 
and f. By analogy, the time interval between #-2 and # is equal 
to 24t. 
In figure 1 the automatic low-resolution satellite image 
interpretation framework is presented. In order to perform the 
  
  
Input images o 
(1-1) (2) 
   
  
  
  
Interpretation 
(t-1) 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
es 4 
> 4X yp. 
3 v 
Segmentation 
y Y 
Y I Interpretation 
Yeh. (t) 
.'Segments Pi Segments classifier 
ma 
| {++ A 
1 Automatic ou am 3 Py 
1 extraction of Automatic change ı 
ı spectral detection Automatic 
' knowledge ; interpretation 
| i process 
! 
Stable Segments | 
I 
Y 1 
| 
Spectral signatures 
estimation 
I 1 
Legend: 
spectral O Procedure 
[A] Input data 
contextual Ne [J Intermediate result 
multitemporal GIS Data [I Result 
attributes vector 
class description 
rules 
Specific knowledge 
  
  
  
Maps of the 
region 
Figure 1. Framework to the automatic low-resolution satellite image interpretation 
   
  
  
   
  
  
  
   
  
  
  
  
    
     
  
    
  
  
   
   
  
  
   
    
  
  
  
  
  
  
   
   
  
  
   
      
    
    
inte 
pro] 
the 
and 
cha 
clas 
the 
inte 
abo 
The 
clas 
con 
bet 
rep 
pix 
20€ 
the 
em 
cor 
int 
In 
are 
I) 
(L! 
2) 
cla 
3) 
inf 
cle 
4) 
cle 
the 
int 
all 
pri 
ini 
sp 
TI