As the next figures and next paragraphs show the mixed 
pixels are classified wrongly using different classification 
rules. 
The criteria of the first classification are the following 
ones: 
- supervised classification; 
- bands: all the recorded channels 
- maximum likelihood method; 
- . threshold: 596 
- classes: snow, green, vegetation, water, cultivated 
field, rice-field, cloud. 
The next figure 4 shows the entire classified image: in 
white the blanket of snow. 
  
    
  
   
   
Me Wa m = 20 Ww Cr in à T eel “iles 
= a M A, Fre hia N von IB dme d . 
x, Fl, a } At t > es x * Eee 2 
zy $e SE 9^ . - 
enr - ate : , = rw us 
: be uL. > m 1 > ete Yo 
‘ ur 2 ut iy P 
die s ; > ied SNS TE TRE à 
un J x net ae RE I X wm rt “2 
RE tye aa GSP SAT Vut e cam LAT 
à NES -— am cmt "NA unir 
A ROC S Pr E 
m ME p. Bag nod -— : 
: FARBE Ar eS tem ti 
- ; a N us "o 2 
= << . » P set 4 un. 
ate al BEY ee Ti 
al pt í ad. 
ie oos 
    
  
AE AE s ma ci 
Figure 4: the result of first classification. 
Figure 5 shows that the borderlines around snow layer, 
where many pixels are likely to be mixed, are classified as 
clouds. That day the weather was sunny and so it is easy 
to understand that the pixels classified as clouds along 
the borderlines are not reaily clouds but they are snow 
added to green and we can really say that this 
classification assigns part of the mixed pixels to an 
extraneous class (error Ill). 
M t ry» 
i ow EA, ; A 
m I x ; A 
-— & - i a » Et ee 
33 wry WW - t e 
^i . * A, f. 
Yor : ma; m Gite 
; # E * Ss now” és ; Le M 
- $ E 1 i N s je 3 
t ^ i a * 4 1 
; hm € pes pu TS à 
Son : z P. Fuer e a MR 
x ^ 3 - 4 is » Ber 217. P9", 
Sols pty wm is pu 4 au. ; 
(x AA .Aa 58 (l5 
4&á nonsnow 1:53  *., 
5 vu Fat e enm. 
e opu T. i eh. 
wiper - p^ E 3 £C e "ol > 
Aum Fa Fra, gn ce ms 
Meet wt uit a S ei w^ x 
ky aE, ^ a = 
WG cm d (E 
Sai ein Y sett 
y bw 
Figure 5: a zoom into the image 4. The borderline around 
the snow areas is classified as cloud (dark grey) 
In order to give different weight to thermal information 
contained in the channels 4 and 5, the same classification 
criteria are performed only with tree channels (1, 4 and 5). 
The criteria of the second classification are the following 
ones: 
-  Supervised classification; 
- bands: channels 1, 4 and 5; 
- | maximum likelihood method; 
- . threshold: 596; 
- classes: snow, green, vegetation, water, cultivated- 
field, rice-field, cloud. 
Figure 6 shows the results of the second classification. 
  
Figure 6: a zoom into the second classified image. The 
borderline around the snow areas is still classified as 
clouds (dark grey). 
Comparing figure 5 and figure 6 we can observe that they 
both assign snow-green mixed pixels to an extraneous 
class. 
A third new classification is done whit only one thermal-IR 
channel: 
- supervised classification; 
- . bands: channels 1 and 5; 
- maximum likelihood method; 
- . threshold: 596 
- classes: snow, green, vegetation, water, cultivated 
field, rice-field, cloud. 
In Figure 7 the results of the third classification are shown. 
AN LA Te "e le. duh aa 
5 STE e Uum 
S M BO i - = 
+ Br 
* 
A^ bm. 
re ae 
           
   
  
    
Figure 7: a zoom into the third classified image. The 
borderline around the snow areas is still classified as 
cloud (dark grey) 
Even using only one thermal band (channel 5) the snow- 
green mixed pixels are classified as cloud. 
5. THE SPECTRAL SIGNATURES 
In order to understand the behaviour of snow-green mixed 
pixels their spectral signatures are compared together and 
with the clouds one. 
Figures 8, 9 and 10 show the characteristic spectral 
signatures . 
330 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
  
Figure 
Figure 
Figure 
Figures 
*green 
by ban 
last col 
values 
approxi 
5096 c 
(snow+ 
mean v 
the ex 
classifie 
Even t 
classifie 
good ai