TU4.L09 - FOUR-COMPONENT SCATTERING POWER DECOMPOSITION WITH ROTATION OF COHERENCY MATRIX
Transcript of TU4.L09 - FOUR-COMPONENT SCATTERING POWER DECOMPOSITION WITH ROTATION OF COHERENCY MATRIX
Four-component Scattering Power Decomposition with Rotation of Coherency Matrix
Yoshio Yamaguchi, Niigata University, JapanAkinobu SatoRyoichi Sato
Hiroyoshi YamadaWolfgang -M. Boerner, UIC, USA
Measured
decompositionrotation
Surface scattering
Double bounce
Volume scattering
Helix scattering
Scattering Power
expansion matrix
20
* 1 0
0 0 0
1*
02
0
0 0 0
130
15 5 05 7 00 0 8
14
2 0 00 1 00 0 1
130
15 - 5 0- 5 7 00 0 8
12
0 0 0
0 1 + j
0 + j 1
Four-component Scattering Power Decomposition
Y. Yajima, Y. Yamaguchi, R. Sato, H. Yamada, and W. -M. Boerner, “POLSAR image analysis of wetlands using a modified four-component scattering power decomposition,” IEEE Trans. Geoscience Remote Sensing, vol. 46, no. 6 , pp. 1667-1673, June 2008.
Measured =
N
Google earth optical image
Ps Pv
Pd
Scattering power Decomposition
© METI, JAXA
Decomposed image (Ps, Pd, Pv)
ALPSRP172780690-P1.1__A
2009/4/22
-Kyoto-Japan
34.800N135.781E
Radar line of sight
Deorientation
� TT
Concept for new decomposition
Minimization of the HV component by rotation
Ps Pv
Pd
Too much green in urban area
Azimuth slope and Oblique wallHV component
creation
Rp ( ) =1 0 00 cos 2 sin 20 – sin 2 cos 2
Rotation of Coherency Matrix
Ensemble average in window
T33 = T33 cos2 2 Re T23 sin 4 + T22 sin22
Minimization of T33 component
= 14tan- 1
2 Re T23
T22 T33
Rotation angle
Same as azimuth slope angle
Coherency matrix elements after T33 rotation
Major terms for 4-comp. decomposition
Minor terms
Unchanged
Pure imaginary : Best fit to Helix scattering
2 = 12tan- 1
2 Re T23T22 T33
measured decomposition
rotation
4-component decomposition
New decomposition scheme
Rotation of data matrix
Four-component decomposition
- 2 dB 2 dB
yes
yesno
no
Decomposed power
Four comp. Three comp. Two comp.Three comp.
if Pv < 0 , then Pc = 0 (remove helix scattering) 3 comp. (Ps, Pd, Pv) decomposition
if
Helix scattering power
Volume scattering power
Double bounce scattering
Surface scattering
Pv + Pc > TP
C = T12 ( )
10 logT11 ( ) + T22 ( ) – 2 Re T12( )
T11 ( ) + T22 ( ) + 2 Re T12( )
C0 = T11 ( ) – T22 ( ) – T33 ( ) + Pc
Ps = Pd = 0
S = T11 ( ) - 12
PvS = T11 ( ) - 12
PvS = T11 ( ) - 12
Pv
D = T22 ( ) - 730
Pv - 12
Pc D = T22 ( ) - 730
Pv - 12
Pc
C = T12 ( ) - 16
Pv
Pc = 2 Im T23 ( )
Pv = 154
T33 ( ) – 158
Pc
Pv = 154
T33 ( ) – 158
Pc
C0 > 0
Ps = S +C
2
S
Pd = D –C
2
S
Pd = D +C
2
D
Ps = S –C
2
D
TP = T11 ( ) + T22 ( ) + T33 ( )
Ps > 0 , Pd > 0
Ps > 0 , Pd < 0
Ps < 0 , Pd > 0
Ps , Pd , Pv , Pc
TP = Ps + Pd + Pv + Pc
Pv , Pc
Pd = 0
Ps = TP – Pv – Pc
Pv , Pc
Ps = 0
Pd = TP – Pv – Pc
Pv = TP – Pc
Ps = Pd = 0
Pc
D = T22 ( ) - T33 ( )
Pv = 4 T33 ( ) – 2 Pc
T = RP( ) T RP( )†=
T11 T12 T13
T21 T22 T23
T31 T32 T33
Rp ( ) =1 0 00 cos 2 sin 20 – sin 2 cos 2
= 14
tan- 1 2 Re T23
T22 T33
T =T11 T12 T13
T21 T22 T23
T31 T32 T33
= 1n k p k p
†n
C = T12 ( ) + 16
Pv
Coherency matrix rotation in imaging window
Four-component decomposition
Algorithm is given in terms of
coherency matrix elements only
Ps Pv
Pd
volume scattering
double bounce
surface scattering
Kyoto, JapanBefore After rotation
ALOS-PALSAR Quad Pol data
Patch A
Patch B
Forest area
Angle distribution
Kyoto
Power distributions
Original
Rotation
Niigata
Pine trees Oblique urban Orthogonal urban
Angle distributionNiigata
Patch E ° Patch D -20°
Angle distribution
Pi-SAR-XNiigata Japan
+4 -20
L-band Pi-SAR data: Downtown Niigata, JAPAN2007-10-04 (5*5 window)
Before After rotation
Before After T33 rotationBeijing, China
Conclusion
for fully polarimetric data sets
Four-component decomposition with
provides better classification result
Ps Pv
Pd