The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
211
and e could be produced to represent the corresponding
linguistic atom, which is showing in figure 3.
1
y/2
^MEC A2 (u)du
u
(13)
drop(* r y ( )
Figure 3. 1-D backward cloud generator
The combination of the two kinds of generators can be used
interchangeably in images segmentation.
3.2 Synthesized Cloud
A synthesized cloud is used to synthesize linguistic terms into a
generalized one and it can be described as a concept tree where
each leaf node is a cloud model.
Suppose there are two neighbour linguistic atoms
A l( E xl’ E nV H el) and A 2( E x2’ E n2’ H el) ? over the
A(E E ,H )
same universe of discourse U.A virtual atom v *’ n ’ e
may be created by synthesizing the two atoms using the
following definition:
ß _ E x\ E n\ JrE x2 E n2
e„;+e„;
E n ~ E n\ + E n2
_ H e \ E n\ JrE ^e2 E n2
Eni + Eni
(8)
(9)
(10)
E E
Where nl and n2 are calculated as follows. Suppose
M EE 'a\ ( w ) and MEC A2 (U) are t ^ e mathematical expected
A Al,
curves of 1 and 2 respectively. Let
MEC M (u) = {
0 , otherwise
MEC Al (u), when MEC AX {u)^MEC A2 {u)
MEC a2 (u) = {
j MEC a2 (u), whenMEC A2 (u)>MEC M (u)
, otherwise
(11)
(12)
then.
It can be deduced that the parent node generated by
synthesizing two clouds is still a cloud model and it serves as a
leaf node for higher level cloud synthesizing, which continues
until a root node of the universal concept tree is generated.
4. EXPERIMENTS AND ALGORITHMS ANALYSIS
ASAR-VV data from Zhuhai, Guangdong, China on April 6,
2004 is used to implement the comprehensive method for fish
ponds recognition in the paper. Figure 4 shows the original
image. From the original image we can see that the brightness
of the fish ponds is close to the river but the textures are
different a lot. According to the characteristic of the original
image, a comprehensive mechanism, considering both the
brightness and texture characteristic of the target to be
recognition as well as the uncertainty of the target edge, is
implemented in this paper. In order to optimize the recognition
result ,a series of mathematical morphological operators are
used .Meanwhile some comparative experiments with classical
segmentation methods are implemented.
E„i =-j^\ME c ' M (u)d u ,
Figure 4 .ASAR-VV data from Zhuhai, Guangdong, China
