第
53
卷第
10
期
2013
年
10
月
电讯技术
Vol.
53
No.
10
Oct.
2013
Telecommunication Engineering
doi:
10.
3969/j.
issn.
1001-893x.
2013.
10.
004
基于变分贝叶斯
ICA
的遥感图像混合像元分析*
’
j
ttt
品
(上海大学叶算机工程与科学学院,上海
2001
l
l)
摘
要:混合像元已成为遥感图像处理、分类的难点和重点
υ
独立分量分析(
ICA
)能够实现图像的
去相关性以及得到相互独立的分量,但是,由于
ICA
模型的各成分独立性和数据统计分布规律的不
变假设,影响了遥感图像分类精度。针对这一问题,提出了基于变分贝叶斯
ICA(
VBICA
)的遥感图
像分析方法,并利用遥感图像进行验证,结果表明:
VBICA
方法提取的独立分量具有均方根误差小、
迭代次数少和稳定性较好的特点;基于
VBICA
方法的遥感分类精度达到了
91.
55%
,且目视效果较
好;
VBICA
方法突破了
ICA
的局限性,提高了遥感图像自动分类精度,具有很好的应用前景。
关键词:遥感图像;混合像元;独立分量分析;
FastICA
;变分贝叶斯
ICA
;自动分类
中图分类号:
TN911.
73
文献标志码:
A
文章编号:
1001-893X
(
2013)
10-1274-05
Mixed Pixel Analysis
of
Remote Sensing Image Based on
Variational Bayesian ICA Method
DONG
Jiang-shan,
LI
Cheng-fan,
ZHAO
Jun-juan,
YIN
Jing-yuan,
SHEN
Di,
XUE
Dan
(School
of
Compuler
Engineering
and
Science,
Shanghai
Universily,
Shanghai
200444, China)
Abstract:
The mixed pixels have
become
the
difficulty
and
keystone of remote
sensing
image analysis
and
classification.
Independent
component
analysis (
ICA)
not only removes the
correlation,
but
also
can
ob-
tain
mutual
independent
band
images.
However,
the
restrictions of
independence
and
the fixed statistical
distribution
rule
of ICA model itself affect the classification
accuracy
of remote sensing images.
In
order
to
overcome this
problem,
the
variational Bayesian ICA (
VBICA)
method
is proposed to analyze remote sens-
ing
images,
and
it is verified by
the
true
remote
sensing
images.
The
empirical
results show
that:
the
inde-
pendent
components
extracted
by
the
VBICA method is
characterized
by less RMSE, iteration
numbers
and
good
stability;
the classification
accuracy
of remote sensing images
reaches
91.
55
0
毛,
and
the
method
has
preferable visual effect;
the
VBICA
method
has
a big
breakthrough
on
the
limitation of traditional ICA
method
and
improves the automatic classification
accuracy
of remote
sensing
images,
and
has
a good appli-
cat10n prospect.
Key
words:
remote
sensi
吨
image;
mixed
pixel;
independent
component
analysis(
ICA)
;
Fas
ti
CA;
variation-
al
Bayesian
ICA;
automatic classification
1
引
含多种地物,即混合像元。传感器获取的绝大多数
像元基本上都是混合像元:
3
J
。混合像元的存在引
起遥感分类误差,导致分类精度下降,不能真实反映
地表覆盖状况。如何对遥感图像混合像元进行有效
随着传感器技术的进步遥感已成为研究地球
资源环境的重要手段之一,应用也越来越广泛。像
元是构成遥感影像的基本单元:
l
气每个像元中包
*
收稿日期:
2013
05
06
;修回日期:
2013
06 05
Received
date:
2013
05 06
;
Revised
date
:
2013
06
05
基金项目:国家自然科学基金资助项目(
41172303)
Foundation
ltem·The
National
刊
atural
Science
卡。
undation
of
China(
No.
41172303)
'*
通讯作者:
lrhf@
shu.
edu.
rn
Corresponding
author:
lchf@shu.edu.
en
•
1274.
第
53
卷第
10
期
2013
年
10
月
电讯技术
Vol.
53
No.
10
Oct.
2013
Telecommunication Engineering
doi:
10.
3969/j.
issn.
1001-893x.
2013.
10.
004
基于变分贝叶斯
ICA
的遥感图像混合像元分析*
’
j
ttt
品
(上海大学叶算机工程与科学学院,上海
2001
l
l)
摘
要:混合像元已成为遥感图像处理、分类的难点和重点
υ
独立分量分析(
ICA
)能够实现图像的
去相关性以及得到相互独立的分量,但是,由于
ICA
模型的各成分独立性和数据统计分布规律的不
变假设,影响了遥感图像分类精度。针对这一问题,提出了基于变分贝叶斯
ICA(
VBICA
)的遥感图
像分析方法,并利用遥感图像进行验证,结果表明:
VBICA
方法提取的独立分量具有均方根误差小、
迭代次数少和稳定性较好的特点;基于
VBICA
方法的遥感分类精度达到了
91.
55%
,且目视效果较
好;
VBICA
方法突破了
ICA
的局限性,提高了遥感图像自动分类精度,具有很好的应用前景。
关键词:遥感图像;混合像元;独立分量分析;
FastICA
;变分贝叶斯
ICA
;自动分类
中图分类号:
TN911.
73
文献标志码:
A
文章编号:
1001-893X
(
2013)
10-1274-05
Mixed Pixel Analysis
of
Remote Sensing Image Based on
Variational Bayesian ICA Method
DONG
Jiang-shan,
LI
Cheng-fan,
ZHAO
Jun-juan,
YIN
Jing-yuan,
SHEN
Di,
XUE
Dan
(School
of
Compuler
Engineering
and
Science,
Shanghai
Universily,
Shanghai
200444, China)
Abstract:
The mixed pixels have
become
the
difficulty
and
keystone of remote
sensing
image analysis
and
classification.
Independent
component
analysis (
ICA)
not only removes the
correlation,
but
also
can
ob-
tain
mutual
independent
band
images.
However,
the
restrictions of
independence
and
the fixed statistical
distribution
rule
of ICA model itself affect the classification
accuracy
of remote sensing images.
In
order
to
overcome this
problem,
the
variational Bayesian ICA (
VBICA)
method
is proposed to analyze remote sens-
ing
images,
and
it is verified by
the
true
remote
sensing
images.
The
empirical
results show
that:
the
inde-
pendent
components
extracted
by
the
VBICA method is
characterized
by less RMSE, iteration
numbers
and
good
stability;
the classification
accuracy
of remote sensing images
reaches
91.
55
0
毛,
and
the
method
has
preferable visual effect;
the
VBICA
method
has
a big
breakthrough
on
the
limitation of traditional ICA
method
and
improves the automatic classification
accuracy
of remote
sensing
images,
and
has
a good appli-
cat10n prospect.
Key
words:
remote
sensi
吨
image;
mixed
pixel;
independent
component
analysis(
ICA)
;
Fas
ti
CA;
variation-
al
Bayesian
ICA;
automatic classification
1
引
含多种地物,即混合像元。传感器获取的绝大多数
像元基本上都是混合像元:
3
J
。混合像元的存在引
起遥感分类误差,导致分类精度下降,不能真实反映
地表覆盖状况。如何对遥感图像混合像元进行有效
随着传感器技术的进步遥感已成为研究地球
资源环境的重要手段之一,应用也越来越广泛。像
元是构成遥感影像的基本单元:
l
气每个像元中包
*
收稿日期:
2013
05
06
;修回日期:
2013
06 05
Received
date:
2013
05 06
;
Revised
date
:
2013
06
05
基金项目:国家自然科学基金资助项目(
41172303)
Foundation
ltem·The
National
刊
atural
Science
卡。
undation
of
China(
No.
41172303)
'*
通讯作者:
lrhf@
shu.
edu.
rn
Corresponding
author:
lchf@shu.edu.
en
•
1274.