LP t 
statistics bare soil grass winterwheat the le 
VERSION 1 | non-direction dependant 35 33 30 direct 
VERSION 2 direction dependant 35 33 30 edges 
VERSION 3 direction dependant 560 73 345 decuk 
Table 1: numbers of used training areas. forest 
In VERSION 1 the means of about 30 training areas per class defined a non-direction differ: 
dependant statistic. In the two other cases, direction dependant values of the mean vectors They « 
were applied, determined by second order (bare soil, grass, winterwheat) or first order Wo pe 
polynomials (forest and water) using few (VERSION 2) or many (VERSION 3) training fields. 
Classification results with VERSION 1 and VERSION 2 show figure 4. Non-direction ne 
dependant statistics (Figure 4 a) result in misclassifications, e.g. near the edges and in The re 
4L d PAN R vw mulis 
P ur u. depen 
The su 
For ho 
distrib 
simila 
trainir 
to non 
classif 
Theref 
(a) VERSION 1 foture 
5. LIT 
[1] 
[2] 
  
(b) VERSION 2 
Figure 4: maximum likelihood classification without (a) and with (b) direction 
dependant statistics 
white bare soil dark-grey winter wheat 
white-grey deciduous forest - black-grey grass 
middle-grey coniferous forest black water (Rhine, 
hannel, 
ake) 
192 - 
E Oo qm 
IEEE Emu moo OU ee it NE ré 
" _ A MÀ idu o 7