cover classification by Shaunik De
et al
. (2014).
A hybrid polarimetric
SAR
is one in which the transmit-
ted field is circularly polarized, and the resulting backscatter
is received in two mutually coherent linear polarizations. A
polarized wave experiences a change in its state of polariza-
tion when it interacts with a target. There are many ways in
which the state of polarized electromagnetic wave can be rep-
resented mathematically. Using Stoke’s parameters is one such
method containing all the information of a polarized wave.
Decomposition helps in interpreting the scattering informa-
tion embedded in the polarization of a backscattered wave.
Several coherent and incoherent decomposition methods have
been developed for fully polarimetric data. The well-known
Coherent Target Decomposition methods are the Pauli and
SDH
(sphere, diplane, and helix) and the Incoherent Target Decom-
position methods are by Freeman, Vanzyl, and Yamaguchi. A
review of these decompositions is available in (Freeman and
Durden, 1992 and 1998; Lee and Pottier, 2009; Alberga
et al
.,
2008; Yamaguchi
et al
., 2005; van Zyl, 1989; Yamaguchi
et al
.,
2006). Raney proposed different decompositions on hybrid
polarimetric data by using stoke parameters (Raney, 2007 and
2013; Raney
et al
., 2012). This decomposition has been used
in the current study for classification of rice and other classes.
Methodology
Study Area and Data
West Godavari district is situated on the bank of River Goda-
vari and is known as rice granary of Andhra Pradesh, India.
Rice is the major crop in West Godavari district in both
kharif
(rainy) and
rabi
(post-rainy) seasons and is grown under ir-
rigated systems. The study area measures over 432 km
2
from
16° 25' 49" to 16° 45' 49" North latitude and from 80° 37' 36"
to 81° 55' 60" East longitude in the part of West Godavari
district. The
RISAT
-1 C-band (5.35
GhZ
) single date (28 March
2013) hybrid Polarization (
RH
,
RV
) Fine Resolution Strip Map
1 (
FRS
-1) data is acquired with an incidence angle of 37° and
a spatial resolution of 3 meters. In order to compare data
from the optical sensor
LISS
-IV from Resourcesat-2, the 2 ×2
multi-look was applied on the
FRS
data. The nearest available
date for optical Linear Imaging Self Scanner multispectral
(
LISS
-IV
MX
) data from Resourcesat-2 satellite acquired was on
11 April 2013, which was used for assessing the accuracy of
the classification. The Resourcesat-2 satellite was launched
20 April 2011 by Indian Space Research Organization (ISRO).
It carries three sensors Linear Image Self Scanner (
LISS
)-III,
IV and Advance Wide field Sensor (
AWiFS
) with a resolution
of 23.5, 5.8 and 56 meters, respectively. The
LISS
-IV sensor
images the Earth in three spectral bands, Green from 0.52
-0.59 microns, Red from 0.62-0.68 microns, and Near Infrared
from 0.77-0.86 microns. The
LISS
-IV data has also been used
for georectification. Due to difference in path orientation of
LISS
-IV and
RISAT
-1 imagery, a subset was selected (Achanta
Mandal which represents an administrative unit) for compari-
son purposes.
Most of the rice crop in the study area was transplanted
during the second fortnight of January and the crop was about
60 to 75 days after transplantation and in its reproductive
stage. Ground information for 148 samples were collected
during field campaign for parameters such as crop type, stage,
and height synchronous to the
SAR
data acquisition. The ac-
curacy of classifications from
RISAT
-1 and
LISS
-IV data were
assessed using the ground observation data (Congalton, 1991).
Data Processing
The processing of
RISAT
-1 Hybrid Polarimetric single look
complex (
SLC
) data was carried out by using
ENVI
and Pol
SAR-
Pro image analysis software. The
SLC
data was downloaded,
converted into amplitude format, and by applying enhanced
Lee (Lee
et al
., 1991) speckle filter with a 5*5 window, speck-
le noise was removed. A Sigma naught image was generated
from the amplitude
DN
image by using the following formula
(Mayank
et al
., 2014; Manab Chakraborty
et al
., 2013; Iyyap-
pan
et al
., 2014).
σ
0
(
dB
) = 20 *
log
10(
DN
p
) –
K
dB
+ 10*
log
10(sin(
i
p
)/sin(
i
center
)) (1)
where,
σ
0
(
dB
) = The Radar Backscatter coefficient,
DN
p
= Digi-
tal number or the image pixel gray-level count for pixel p,
K
dB
= Calibration constant in dB,
i
p
= Incidence angle for the pixel
position
p
, and
i
center
= incidence angle at the center.
To understand the state of the polarized electromagnetic
wave, Stokes parameters were computed from filter-applied
data to obtain the six decomposition parameters, i.e., Degree
of polarization (
m
), Relative phase (
δ
), Degree of circularity
(
χ
), Double, Odd, and Diffuse scattering mechanisms.
The degree of polarization (
m
) can be derived from stokes
parameters and is defined as the ratio of polarized power to
the total power of the electromagnetic wave. This parameter
is a significant distinguisher for characterizing the polarity of
the backscatter. The value of
m
always lies between 0 and 1
(Raney, 2007):
m
S S S
S
=
+ +
.
1
2
2
2
3
2
0
(2)
Relative phase (
δ
) is the angular difference of phase be-
tween two components of the electric field vector. Relative
phase is an indicator of double bounce scattering, and can be
derived using Stokes parameters (Raney, 2007):
δ
=
tan
–1
(
S
3
/
S
2
)
(3)
The degree of circularity (
χ
) or ellipticity parameter is
a shape parameter and is governed by the magnitudes and
relative phase between horizontal and vertical components of
electric field vector. It takes values between +45° and -45°;
χ
,
which is a sensitive indicator of even versus odd bounce scat-
tering (Raney, 2007):
sin
S
mS
2
3
0
χ
=
.
(4)
From the above parameters, the physical information of the
target in terms of even, odd, and volume scattering mecha-
nisms can be extracted. Raney (2007) proposed even, odd,
and volume scattering on hybrid polarimetric data. These
scattering properties can be derived using stokes vector (S
0
),
relative phase, degree of polarization, and degree of circular-
ity shown in Equations 5, 6, and 7:
S m
sin
or S m
sin
=
−
(
)
−
(
0
0
1 2
2
1
* *
* *
χ
δ
)
2
Even or Double bounce scattering
(5)
S m
sin
or S m
sin
=
+
(
)
+
(
* *
* *
0
0
1 2
2
1
χ
δ
)
2
)
Odd or Single bounce scattering
(6)
Diffuseor Volume scattering S m
*
.
=
−
(
)
0
1
(7)
All these parameters have been used as a primary indicator
for representing scattering mechanism of rice crop identifi-
cation. The polarimetric decompositions and sigma naught
image were geometrically corrected using
LISS
-
IV
data of the
558
July 2015
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
