US 12,632,754 B2 · Remote sensing image feature discretization method based on rough-fuzzy model

Abstract

Provided is a remote sensing image feature discretization method based on rough-fuzzy model, comprising the following steps: listing the digital number in each band and category of a selected sample in remote sensing images, and building an image information decision table based on the digital number and category; initializing the class center of each category and the membership degree of a sample example relative to the class center; updating the class center of each category and the membership degree of the sample example relative to the class center iteratively, and obtaining the final value of the class center of each category and the final value of the membership degree; building a rough-fuzzy set, computing the mean approximation accuracy of the rough-fuzzy set, discretizing the image information decision table, evaluating the discretization results based on the mean approximation accuracy and a genetic algorithm, and selecting an optimal discretization solution.

A remote sensing image feature discretization method based on a rough-fuzzy model, comprising the following steps:

listing a digital number in each band and category of a selected sample in the remote sensing image, and building an image information decision table based on the digital number and category, wherein building the image information decision table based on the digital number and category, comprising:
selecting a plurality of samples from remote sensing images;
obtaining the digital number in each band of each selected sample by image sampling, and labeling a corresponding land cover type in each selected sample, wherein each selected sample constitute a sample set;
and
building a decision table matrix with each band as a condition attribute and with a corresponding land cover type as a decision attribute according to the sample set with digital number and land cover type label;
initializing a class center of each category and a membership degree of a sample example relative to the class center, wherein initializing the class center of each category and the membership degree of the sample example relative to the class center, comprising:
initializing the class center
c
j
0
of each category as:
P
⁢
M
0
=
[
f
1
(
x
1
)
f
1
(
x
2
)
…
f
1
(
x
N
)
f
2
(
x
1
)
⁢
·
·
·
f
2
(
x
2
)
⁢
·
·
·
…
⁢
⋯
⋯
⋯
f
2
(
x
N
)
⁢
·
·
·
f
C
(
x
1
)
f
C
(
x
2
)
…
f
C
(
x
N
)
]
f
j
(
x
i
)
=
{
1
,
x
i
⁢
belongs
⁢
to
⁢
class
⁢
j
0
,
otherwise
where
1
≤
i
≤
N
⁢
and
⁢
1
≤
j
≤
C
wherein N is the number of samples, and C is the number of categories;
and
initializing the membership degree of the sample example relative to the class center as:
u
i
⁢
j
0
=
1
/
∑
k
=
1
C
(
∑
h
=
1
M
(
x
ih
-
c
jh
)
2
∑
h
=
1
M
=
(
x
ih
-
c
k
⁢
h
)
2
)
wherein M is the number of attributes, xih is a value of the sample xi on the h the attribute, and cm is a value of the current class center cj on the h th attribute;
updating the class center of each category and the membership degree of the sample example relative to the class center iteratively, and obtaining a final value of the class center of each category and a final value of the membership degree, wherein updating the class center of each category and the membership degree of the sample example relative to the class center iteratively, comprising:
updating the class center
c
j
0
iteratively by the following formula:
c
j
1
=
∑
j
=
1
N
⁢
(
(
u
ij
0
)
2
×
x
i
)
/
∑
i
=
1
N
⁢
(
u
ij
0
)
2
wherein
c
j
1
is an updated class center;
and
computing the new membership degree
u
ij
1
iteratively according to the updated class center
c
j
1
;
building a rough-fuzzy set, computing a mean approximation accuracy of the rough-fuzzy set, discretizing the image information decision table, evaluating discretization results based on the mean approximation accuracy and a genetic algorithm, and selecting an optimal discretization solution, wherein building the rough-fuzzy set, and computing the mean approximation accuracy of the rough-fuzzy set, comprising:
creating the rough-fuzzy set based on each category:
Aj(xi)=uij,1≤i≤N,1≤j≤C;
computing a lower approximation of the rough-fuzzy set:
R
-
⁢
A
j
(
x
i
)
=
inf
y
∈
U
⁢
{
A
j
(
y
)
|
(
x
i
,
y
)
∈
R
}
;
computing an upper approximation of the rough-fuzzy set:
R
-
⁢
A
j
(
x
i
)
=
sup
y
∈
U
⁢
{
A
j
(
y
)
|
(
x
i
,
y
)
∈
R
}
;
computing the mean approximation accuracy of the rough-fuzzy set:
η
_
=
1
C
⁢
∑
j
=
1
C
❘
"\[LeftBracketingBar]"
R
-
⁢
A
j
❘
"\[RightBracketingBar]"
❘
"\[LeftBracketingBar]"
R
-
⁢
A
j
❘
"\[RightBracketingBar]"
;
and
stopping iterative computations of uij and cj, and obtaining the final value of the class center of each category and the final value of the membership degree when the following conditions are met:
max
ij
{
❘
"\[LeftBracketingBar]"
u
ij
t
+
1
-
u
ij
t
❘
"\[RightBracketingBar]"
}
<
ε
wherein tis the number of iterations, & is an error threshold.

350 more elements

Remote sensing image feature discretization method based on rough-fuzzy model

Life of this patent

Remote sensing image feature discretization method based on rough-fuzzy model

Life of this patent

Abstract

Claims3