Classification Method Validation for Rice Mapping Using ENVISAT APS Data Erxue CHEN(1), Zengyuan LI(1), BingxiangTan(1),Wei He(1), Bingbai LI(2) (1)Institute of Forest Resources Information Techniques, Chinese Academy.

Download Report

Transcript Classification Method Validation for Rice Mapping Using ENVISAT APS Data Erxue CHEN(1), Zengyuan LI(1), BingxiangTan(1),Wei He(1), Bingbai LI(2) (1)Institute of Forest Resources Information Techniques, Chinese Academy.

Classification Method
Validation for Rice Mapping
Using ENVISAT APS Data
Erxue CHEN(1), Zengyuan LI(1), BingxiangTan(1),Wei He(1), Bingbai
LI(2)
(1)Institute of Forest Resources Information Techniques, Chinese
Academy of Forestry,
100091, Beijing, PRC. China
(2) Institute of Agriculture Modernization, Jiangsu Academy of Agriculture
Sciences,
210014, Nanjing, PRC. China
[email protected], [email protected],
[email protected], [email protected]
Outlines
Introduction;
Test site and EO data;
ASAR APS data pre-processing;
Classification evaluation methods;
Results and analysis;
Conclusions.
Introduction
• Regular monitoring of the crop is
necessary in the major rice producing
countries for political, economical and
food security reasons;
• We validated two kinds of APS
preprocessing schemes and compared
their classification performance.
Test site and EO data
• Xinghua district of Jiangsu Province, PRC.
China
Test site location on China map
•Ground true data
Landsat TM image. R: b 5; G: b4; B: b3 Ground true map generated from Landsat TM
image interpretation and aided by field work
•ENVISAT ASAR APS data used
ASAR APS data pre-processing
• Backscattering coefficient based method
(BSCBM );
• Polarimetric SAR data processing method
(PSDPM ).
•BSCBM
complex
T1
T2
T3
Radiometric calibration
Intensity
BSC1
BSC2
 i0, j 
DN i2, j sin  j
K
1
R 4j
sin  ref G  j 2 Rref
BSC3
Multi-look (6az*2rg)
Spatial speckle filtering
Lee speckle filter
Geocoded ellipsoid correction
Range-doppler Model
SAR to TM image co-registration
db
BSC1
BSC2
MLC
BSC3
INPUTS
•PSDPM
Scattering
matrix
 S11
S
 S 21
Scattering
vector
VV+VH
S12 
S 22 
0
S
 S 21
VV+VH
S12 
S 22 
Covariance matrix (multi-looked)
kp
 S12 


S
 22 
[C ]  k p k
*
p
 S12 2
C11 C12 
 
   *
C
C
22 
 21
S12  S 22
* 
S12  S 22
2 
S 22 
Lee Polarimetric Filter
H-Alpha decomposition
Entropy or H
Alpha
CMF
 S12 2
  *
S12  S 22
Intensity-VH
* 
S12  S 22
2 
S 22 
Multi-looked
and filtered
[C]
MLC
Intensity-VV
INPUTS
POLSARPro©ESA used for all these processing routes.
Classification evaluation methods
•Dataset inputs to the MLC
1.BSCBM
• 6BSC-(3VV+3VH): All 6 BSC images from three dates;
• 3BSC-VV: the 3 VV polarization BSC images from three dates;
• 3BSC-VH: the 3 VH polarization BSC images from three dates.
2.PSDPM
•
6I-(3VV+3VH): All the 6 intensity images from three dates;
•
3I-VV: the 3 VV polarization intensity images from three dates;
•
3I-VH: the 3 VH polarization intensity images from three dates;
•
12IAE-(6I+3A+3E): 6 intensity and 3 alpha and 3 entropy images.
•
6AE-(3A+3E): All the 3 alpha and entropy images.
1.BSCBM
Multi-temporal VV BSC image
Multi-temporal VH BSC image
2.PSDPM
Multi-temporal VV intensity Image
Multi-temporal VH Intensity image
2.PSDPM
Polarimetric alpha image
Polarimetric entropy image
•Training and test samples collected from ground true data
Results and Analysis
•PSDPM compared with BSCBM
classification performance of PSDPM is better than BSCBM
Results and Analysis
•VV compared with VH
• The classification accuracy of VV is better than VH.
•The contribution of Entropy and Alpha
• Intensity combined with E+Alpha leads to lower
classification accuracy,but highest rice acc.;
• E+Alpha only achieved the worst performance
Conclusions
(1)For rice mapping using ENVISAT ASAR APS
level-1 data, it has been observed that polarimetric
data processing method is better than
backscattering coefficient based processing method;
(2)Multi-temporal VV co-polarization SAR data has
higher capability for land cover classification in
agricultural region than multi-temporal VH crosspolarization SAR data in C band;
(3) Applying PSDPM to APS data and combining
all the information from it as inputs to a certain
classifier was suggested for operational rice
mapping.
ACKNOWLEDGES
Thanks for European Space Agency to
provide all the ENVISAT ASAR datasets
used in this study. Many thanks towards Dr.
Mike Wooding of RSAC of UK. for helps to
order the ASAR APS data from ESRIN.
Thanks!