By STEPS 5,6,7 and8, each image pointis assigned 
its (single or multi) category land cover which is 
based on "Principle of Convergence" and “Possibilistic 
Function of AND operator". The Possibilstic Func- 
tion has been developed on the basis as suggested 
by Dubois and Prade (1988). The details of the same 
are given below: 
Let X={x,,x,,...,x,} be the set of values of N linguistic 
variables for a given pixel X. The possibilistic func- 
tion of AND operator representing the possibility of 
X to belong to class C is 
T(C,,X)=min {n, 04} 
where k=1,2.,..... ‚N and 
p, (X) represents the possibility of belonging of the 
pixel, having attribute value x of the linguistic 
variable k, to the class C, . 
In STEP 9, the possibilitic function of OR operator 
representing the hard class, C of the mixed pixel X 
is 
TT(C,X)= max (t(C,,X)) when varying C,. 
By performing STEP 10, the desired output i.e., the 
classified land cover image, is obtained through the 
understanding system. 
5. RESULTS 
As a result of domain knowledge, heuristics and 
thresholding through training samples and conver- 
gence of evidences, a rule base is prepared to infer 
pure and mixed pixels. Some of the rule base used 
in determining the component proportion of the 
mixed pixel are 
defind as follows: 
5.1 Rule Base 
RULE # 
IF NDVI is High 
AND HUE is (in overlapping of) Low and Medium 
AND TONE is High 
THEN cover type is (mixture of) WATER and SOIL. 
RULE # 
IF NDVI is (in overlapping of) Medium and High 
AND TONE is (in overlapping of) Low and Medium 
THEN cover type is (mixture of) SOIL and VEGETA 
TION. 
RULE# 
IF NDVI is High AND HUE is (in overlapping of) Low 
and Medium 
AND TONE is (in overlapping of) Low , Medium and 
High 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
THEN cover type is (mixture of) WATER , SOIL and 
VEGETATION. 
RULE# 
IF . NDVI is Medium 
AND HUE is (in overlapping of) Medium and High 
AND TONE is (in overlapping of) Low , Medium and 
High 
THEN cover type is (mixture of) WATER , SOIL and 
VEGETATION. 
As the pure pixels of each class have distinct char- 
acteristics as defined inthe domain knowledge, only 
NDVI and HUE is sufficient to satisfy the conver- 
gence in determining the cover type, whereas, mixed 
pixel gets identified and estimated using the over- 
lapping of labels. The variables addressed by the 
rule-base are considered for the possibilistic calcu- 
lation. 
5.2 lllustrative Exapmples 
EXAMPLE 1. 
A sample, X1(Refer fig.1) which has NDVI and HUE 
values of 0.26 and 9.83 respectively represents 
fuzzy labels as Medium and Low. As both the labels 
represents the same land cover type i.e. soil, so by 
"Convergence of Evidence" principle, it can be 
inferred that the land cover type is SOIL. 
In order to find the membership category, the mem- 
bership values of fuzzy labels are found to be 
respectively 1.0 and 1.0 ( refer Fig.1). By applying 
the Possibilstic function of AND operator, the mem- 
bership value of the pixel as soil cover is found to be 
1.0 i.e.,it is a pure pixel of SOIL. 
EXAMPLE 2. 
Sample, X2 (Refer fig.1) has NDVI and HUE values 
of -0.02 and 229.1 respectively. itis found that NDVI 
gives a fuzzy label of low and medium i.e., the 
sample is a mixed pixel of water and soil, whereas 
HUE gives a fuzzy label of Medium to High i.e., the 
sample contains both water and vegetation. As the 
labels represent different land cover types, so in 
order to resolve the ambiguity, the fuzzy variable 
TONE is further examined and the value is 22. The 
corresponding fuzzy label is Medium and High, thus 
indicating that the pixel as a mixture of soil & water. 
By applying "Convergence of Evidence" principle, it 
can be now inferred that the sample is a mixture of 
WATER and SOIL only, and that vegetation is ex- 
cluded as it appears only in one case. 
For determining the category membership of the 
pixel, different membership values for water are 
0.12, 0.996 and 1.0 and for soil these are 0.88 and 
1.0 from different linguistic variables. Using 
Possibilistic Function of AND operator, the mem- 
bership value of water and soil are found to be 0.12 
and 0.88. The hard class, using possibilistic OR 
operator, of the pixel is classified as SOIL. 
242 
   
  
     
  
  
  
   
  
  
   
  
   
   
    
  
  
   
   
  
  
   
   
  
   
   
  
   
  
    
  
   
  
   
   
  
   
   
   
   
   
  
  
   
    
   
   
   
  
   
    
  
   
   
  
   
  
   
  
    
    
EXAN 
A sai 
value 
NDVI 
samp 
label 
both | 
repre 
resoh 
exam 
& Me 
water 
Evide 
isan 
For d 
pixel, 
0.55, 
and fi 
Funct 
watel 
and 
possi 
WAT! 
The : 
plain: 
Ther 
com; 
tion 
meth 
ods. 
autor 
so, it 
andr 
requi 
to re 
mum 
types 
catec 
the r 
defin 
may 
of im 
whic] 
to th 
Dubc 
Plen 
Este 
tal o!