Classification of sea ice types

strangerwineΤεχνίτη Νοημοσύνη και Ρομποτική

19 Οκτ 2013 (πριν από 4 χρόνια και 20 μέρες)

87 εμφανίσεις

Nansen International Environmental

and Remote Sensing Centre

Classification of sea ice types
in ENVISAT Synthetic Aperture
Radar images

Corrected SAR image

B


Incidence
angle

Incidence angle

Preprocessing of
ENVISAT
SAR
data
:

1.

Backscattering
Coefficients as a Function
of Incidence Angle
(angular correction)

Incidence
angle range:
17



42

Original SAR image

A

Swath

Satellite

flight
direction

a)
original ENVISAT WSM,
Kara Sea on April 18, 2005 at 06:22 UTC.
σ
°

for different ice
types at 23
°
.
b) Estimated σ
°

values for various ice types derived from calibrated ENVISAT
WSM
at 23
°
.

Preprocessing of
ENVISAT
SAR data

2.

Backscattering Coefficients for Different Ice Types

Nilas

FY
deformed

MY floes

FY level

Analysis of SAR Image Texture Features

3 3 4 4 4

4 4 4 6 5

5 4 4 5 5

4 4 3 4 5

4 5 4 4 4

an
element of GLCM,

number of neighbor pixel pairs,

Texture features

Correlation




energy



…………….

GLCM calculation

Sea ice classification

Original
Envisat

ASAR WSM image

ASAR image absolute
calibration and angular correction

Image features calculation: mean backscatter,
texture characteristics

Image classification

Neural Network

Image classification

Bayesian approach

Chart
composition

in conditional colors

Satellite SAR sea ice automatic
classification using Neural Network
algorithm

ENVISAT ASAR image and two

variants of its
classification with different of ice type detailing

a) Fragment of original ENVISAT
Wideswath

SAR
image,
taken over the northern part of Kara Sea on
February 04, 2008 at
11:00;
b) fragment of SAR image after normalization of σ
°

to incidence angle of
25
°

and analysis by an ice expert from
AARI.
c
)
the result of NN classification; d) ice chart provided by
AARI for February 06, 2008.
e
) manually colored image according to sea ice expert’s visual analysis
.

LFYI


level first
-
year ice

DFYI


deformed first
-
year ice

MYI


multiyear ice

А
-

LFYI, where ridges cover less
20% of ice
surface,

B
-

FYI, where ridges cover more
than 40% of ice surface,

C
-

MYI area
,

D
-

thick FYI and second
-
year ice

Satellite SAR sea ice automatic
classification using Bayesian
algorithm

Probability density functions of
σ

, derived from
calibrated
ENVISAT ASAR
images in the Central Arctic
for incidence angles of 25


for MYI
,
LFYI, and
DFYI.

Location of ENVISAT ASAR
images
.


A posteriori probability
p(
ω
j
/
x
i
)

for various sea ice
classes is calculated
:

p(
ω
j
)



a priori probability
,
p(x
i
/
ω
j
)



conditional
probability of
x
i

for
ω
j
class
.

Conditional probabilities
were assessed from calibrated ENVISAT
Wideswath

SAR images, where areas
typical for each ice type were delineated visually
.

Bayesian algorithm

Satellite SAR sea ice automatic classification using
Bayesian algorithm

M
AIN

DIAGONAL

OF

THE

CONFUSION

MATRIXES

FOR

COMPARISON

OF

CLASSIFICATION

RESULTS
,
APPLIED

TO

ENVISAT ASAR
IMAGES
,
USED

FOR

NN
TRAINING
/
TESTING

AND

B
AYESIAN

APPROACH
,
WITH

AARI
ICE

CHARTS

This study is summarized in

Zakhvatkina
, N. Y.;
Alexandrov
, V. Y.;
Johannessen
, O. M.;
Sandven
, S.;
Frolov
, I. Y.
Classification of Sea Ice Types in ENVISAT Synthetic Aperture Radar Images. IEEE
Transactions on Geoscience and Remote Sensing.
10.1109/TGRS.2012.2212445

(in
press).

C
ONFUSION
M
ATRICES FOR
C
OMPARISON OF
NN
AND
B
AYESIAN
C
LASSIFICATION
R
ESULTS
, A
PPLIED TO
ENVISAT ASAR I
MAGES
,
W
ITH
V
ISUAL
E
XPERT
E
STIMATION
. (
A
) J
ANUARY
18, 2008
,
AND
(
B
)
F
EBRUARY
4, 2008

Sea ice
\
open water edge

Neural Networks
:
ENVISAT ASAR + AMSR
-
E data
(SSM/I)


ROI
:
Fram

Strait


resolution
:
525

m

METno:

OW SeaIce unclass

OW


90.3

6.4


3.2

SeaIce


7.
3


90.4


2.3

unclass


9.
7


9.
6



0.0
1

Sea ice

OW ( < 15 %)

Sea ice

OW

METno sea ice concentration data were ranged into
2 classes: OW


concentration values from 0 to 15%;
sea ice


from 15 to 100 %.

NNet classification: sea ice as sea ice; OW


OWcalm + OW rough.

Matrix of agreement between manual ice
concentration product (METno) and automated
SAR sea ice product

7 February 2011

Ice concentration provided by
Met.no Institute

ENVISAT ASAR WSM image

classification result: OW
\
ice edge

White color



sea ice

Black

color



land mask

Blue

color



rough water

Dark blue



calm OW or
nilas

Green color



unclassified
pixels

30 January

2011

4

December

2011

White color



sea ice

Black

color



land mask

Blue

color



rough water

Dark blue



calm OW or
nilas

Green color



unclassified
pixels

Satellite SAR sea ice automatic
classification:


Radarsat

2

HH

HV

26 March 2012,

HH

HV

Satellite SAR sea ice automatic
classification:


Radarsat

2

7 November 2012,
Fram

Straight

Resolution 50 m, HH
-
pol



Resolution 1600 m

а) Fragment of original ENVISAT
Wideswath

SAR
image,
taken over north of the Kara Sea on January
18, 2008 at 09:55 UTC; b) fragment of SAR image with normalized σ
°

to incidence angle of 25o and
analysis by an ice expert from AARI: А is LFYI, where ridges cover less then 20% of ice surface, B is
FYI, where ridges cover more than 40% of ice surface, C is MYI area, and D is ON.
c
) result of the NN
classification of four ice types where the class ON is included; d) ice chart provided by AARI for
January 16
-
18, 2008
.
e) manually colored image according to sea ice expert’s visual analysis.

Nilas

FY
deformed

MY floes

FY level

Analysis of SAR Image Texture Features

3 3 4 4 4

4 4 4 6 5

5 4 4 5 5

4 4 3 4 5

4 5 4 4 4

an
element of GLCM,

α
is fixed vector directions or orientation (0

, 45

, 90

, or 135

349
),

d
is the co
-
occurrence distance (
between pixels
),

K
is the number of quantized gray levels
in the
image

number of neighbor pixel pairs,

Texture features





…………….

GLCM calculation