LE 
LE 
  
  
  
&& 
&& 
  
  
&& 
&& 
  
  
&& 
&& 
&& 
&& 
&& 
  
  
  
  
  
  
  
  
multispectral data 
  
maximum 
likelihood 
     
  
; digital landscape 
panchromatic model 
  
classification classification 
n s à settlement 
2 probability 
  
  
  
  
  
or^ & 
|" probability 
ie — 
situation forest 
forest 
laver layer 
  
 _-- BB 
  
  
  
  
  
  
     
  
Rule-Base 
with conditions 
   
  
Fig.l: Flow chart diagram 
Landuse 
  
Fig. 1 shows the flow chart diagram of the complete 
classification procedure. As a result we obtain the 
landuse-layer at a spatial resolution of 15 m x15 m with a 
class and a reliability code assigned to cach pixel of the 
landuse image. 
8 EXAMPLES 
5.1 Thematic Mapper + IRS-1C-pan 
In a practical test example we combined a Landsat TM 
image with an IRS-1C-Pan image. The acquisition dates 
were 1991 and 1996. respectively. The TM image was 
Fig.2: IRS-1C-pan original and sigma filtered 
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
classified as described in chapter 2. As a result we obtain 
the two classifications with the highest and the second 
highest probability. By comparing the two results it is 
obvious, that "Water", "Forest" and "Grassland" are quite 
certain classes, while the difference between the two 
classifications for "Fields" and "Settlements" is not 
significant shwoing that these two classes are relatively 
uncertain. We shall pay attention to that fact when these 
data are combined in the rule based classifier. 
The panchromatic image IRS-1C-Pan turned out to be 
rather noisy. Filtering the image in advance seemed to be 
advantageous. In order to leave the edges and the textural 
image properties unaffected as much as possible (they 
bear valuable information necessary for texture analysis) 
  
Fig. 3 Detected points and edges and the settlement class