International Archives of the Photogrammetry, Remote Sensing 
and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
5. ANALYSIS 
Statistical indicators of automatic classification (dendogram, 
error ellipsis, co-occurrence matrices) make us aware of the 
fact that the combination of input data containing TIR 
component yield better results than those not containing it. It 
is obvious that the basic topological objects can be well 
differentiated in the field through automatic classification 
itself. Hence, the resulting presentation of the field with 3 
classes illustrate very well the relationships in the field. Three 
  
    
File Optica te ly 
Highest Cooccurrence: | ij I 1 Lowest Separabilituf 11 EI 
  
    
p.i 
623,302 
246.579 - 155.761 2d. o6 
BEL 3 555 
  
presented. classes are (given in the chapter 3): vegetation, zum ir 
À : x file Options Welp | 
rocks and stones, and bare soil. The best result is obtained Mishest Cooccurrences PETITS Lowest Sersesbilitef T= LI 10 
through automatic classification of the following input data 
combinations: TIR, 3 VNIR and 2 Volter/Unsharp filtered 
presentations (VNIR-G, VNIR-R). 
  
    
  
  
  
  
  
   
  
    
  
  
       
  
  
  
  
  
      
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
s D -— Tl 2448 Han ~§74,216 
File View Options Help Ur [ wen wl eu EC 
Shuster eli 0.00 49,66 Miche 148,99 198,66 : EN inel, 
ER: (045.460 ^ : : , = 
[) 6 (6)<14,26%) 
a ce Fgure 12. Coocurenc matrix with minimum and maximum Fist 
5 ox SLR coocurenci and separabiliti of best auto-corelation with 3 IR, R, 
[12 (X 9.740 classes (GREEN: vegetation, BLUE: rocks and stones, chan 
Bl 4 (4)(28.79%) BLACK: bear land) 
6. CONCLUSION 
Fi— Iu | 
Out of the above given data it can be seen that no | 
Figure 10. Dendogram of the best auto-classification combination includes DOF. It has been done on purpose | 
= is because the observed part of the field (due to large height 
tuesg C differences) is largely covered with shadows that make a 
Fils Uptime pete good insight into the field impossible and introduce 
wR anomalies into the radiometric processing. The contribution 
wo es ere f Saez E 7 "s "Ere ion of TR ea EN SOT sendy in he V ME | 
sir ree | crum | mae | mn se ci C ass] ication ane t vous h simple connection 9 indivi na | Ë 
RE | | m m | Ar an | aires 2 classes after the classification. While the result of the | 
| Ae [Hes rn n [i 094 automatic classification of VNIR components only indicates | 
+ UE > | cm ae 1e Sm the connection between the series of images making the 
rdi s Fuse WAGE imas ZZ ; mosaic, this line is not visible in the automatic classification 
of TIR and VNIR components. The influence ol the terrain 
parts covered with shadows is also less expressed in the | 
re results of automatic classification with TIR components. The | 
- inaccuracy of the automatic classification can be noticed in 
2154 s the class snow that is attached to the rocks. 
“631 206.579 | 497,506 06.004 7 
SS EUG | 3013.753 A = ne Figui 
UU IM E RH um E redu 
we m * EN 874,218 22st , 3U4 G R E E N 
Fgure 11. Coocurenc matrix with minimum and maximum 
coocurenci and separabiliti of best auto-corelation with all 
classes Krauss, 
Bonn 
Castlelm 
1258 
    
5 
125.536 
54.771 
125. 761 
À rames EATER totu orsus 
  
  
  
Figure 13. Result of auto-classification with all mentioned 
inputs except TIR component 
Internc 
Hall Inte 
Kasser, | 
& Franci 
Olujié, ^ 
sateliti,s