BLIND SOURCE SEPARATION OF HYPERSPECTRAL DATA IN PLANETARY REMOTE SENSING: ENDMEMBER EXTRACTION AND VALIDATION1. ACCES au laboratoire GIPSA-lab
Feuille de route
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
1
GIPSA-lab
2. Mars
observed
by
Viking
Orbiters
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
2
3. Mars
observed
by
Viking
Orbiters
Geographical
linear
mixture
Mineral
dust
Pixel
size
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
3
4. Mars
observed
by
Viking
Orbiters
Geographical
linear
mixture
Mineral
dust
Pixel
size
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
4
6. HiRISE@MRO
(snapshot
HiRISE@MRO
25
cm/pix)
THEMIS
~100
m/pix
40
km
20
km
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
6
7. HiRISE@MRO
(snapshot
HiRISE@MRO
25
cm/pix)
THEMIS
~100
m/pix
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
7
8. HiRISE@MRO
(25
cm/pix)
CRISM
pixel
footprint
Mars
Reconnaissance
Orbiter
CRISM
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
8
10. Raw
image
Artifact
cleaning
Photometric
correction
Atmospheric
correction
Clean
image
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
10
12. BPSS endmember spectra BPSS Dark source
0.25
0.5
0.4 0.2
REFF value
0.3
0.2 0.15
0.1
A 0.1
B
0 e1 e2 e3 e4 e5 e6 e5’
0 50 100 150 200 250 0 50 100 150 200 250
0.5
BPSS Strong bright source
0.4
mn : spectral
signature
of
BPSS Weak bright source
0.4
BPSS endmember 1
0.35 endmember
n
BPSS endmember 2
0.3
REFF value
0.3 0.5
0.45
0.25
0.2 0.4
0.2
0.4
0.1 C 0.35
0.15
D
0.3
0.1
e1’ e2’ e3’ e4’ e6’ 0.3
0 0.25
0.05
0 50 100 150 200 250 0 50 100 150 200 250
CRISM spectral band 0.2 CRISM spectral band
0.2
0.15
0.1 0.1
0.05
BPSS endmember 4 sn : abundance
map
of
BPSS endmember 5
0.6
endmember
n
0.8
0.5
0.7
0.4 0.6
0.5
0.3
0.4
0.2 0.3
0.2
0.1
0.1
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
12
14. VCA
MVC-‐NMF
spatial-‐VCA
BPSS
[Nascimento’05]
[Miao’07]
[Zortea’09]
[Moussaoui’06]
Geometric
method
First
Minimum
volume
Incorporation
of
Statistical
with
pure
pixel
principles:
constraint
spatial
information
approach
assumption
-‐
Fast
&
efficient
-‐
Bayesian
-‐
Less-‐prevalent
-‐
Homogeneous
Advantages:
-‐
Endmembers
are
endmembers
endmembers
framework
physical
-‐
Error
bars
-‐
Non-‐physical
-‐
Spatially-‐
-‐
Impact
of
noise
spectra
-‐
Non-‐physical
confined
and
less-‐
Drawbacks:
-‐
Less-‐prevalent
spectra
prevalent
-‐
High
endmembers
computational
endmembers
time
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
14
15. VCA
MVC-‐NMF
spatial-‐VCA
BPSS
[Nascimento’05]
[Miao’07]
[Zortea’09]
[Moussaoui’06]
Geometric
method
First
Minimum
volume
Incorporation
of
Statistical
with
pure
pixel
principles:
constraint
spatial
information
approach
assumption
-‐
Fast
&
efficient
-‐
Bayesian
-‐
Less-‐prevalent
-‐
Homogeneous
Advantages:
-‐
Endmembers
are
endmembers
endmembers
framework
physical
-‐
Error
bars
-‐
Non-‐physical
-‐
Spatially-‐
-‐
Impact
of
noise
spectra
-‐
Non-‐physical
confined
and
less-‐
Drawbacks:
-‐
Less-‐prevalent
spectra
prevalent
-‐
High
endmembers
computational
endmembers
time
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
15
16. VCA
MVC-‐NMF
spatial-‐VCA
BPSS
[Nascimento’05]
[Miao’07]
[Zortea’09]
[Moussaoui’06]
Geometric
method
First
Minimum
volume
Incorporation
of
Statistical
with
pure
pixel
principles:
constraint
spatial
information
approach
assumption
-‐
Fast
&
efficient
-‐
Bayesian
-‐
Less-‐prevalent
-‐
Homogeneous
Advantages:
-‐
Endmembers
are
endmembers
endmembers
framework
physical
-‐
Error
bars
-‐
Non-‐physical
-‐
Spatially-‐
-‐
Impact
of
noise
spectra
-‐
Non-‐physical
confined
and
less-‐
Drawbacks:
-‐
Less-‐prevalent
spectra
prevalent
-‐
High
endmembers
computational
endmembers
time
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
16
17. VCA
MVC-‐NMF
spatial-‐VCA
BPSS
[Nascimento’05]
[Miao’07]
[Zortea’09]
[Moussaoui’06]
Geometric
method
First
Minimum
volume
Incorporation
of
Statistical
with
pure
pixel
principles:
constraint
spatial
information
approach
assumption
-‐
Fast
&
efficient
-‐
Bayesian
-‐
Less-‐prevalent
-‐
Homogeneous
Advantages:
-‐
Endmembers
are
endmembers
endmembers
framework
physical
-‐
Error
bars
-‐
Non-‐physical
-‐
Spatially-‐
-‐
Impact
of
noise
spectra
-‐
Non-‐physical
confined
and
less-‐
Drawbacks:
-‐
Less-‐prevalent
spectra
prevalent
-‐
High
endmembers
computational
endmembers
time
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
17
20. 0.6
Apparent reflectance 0.4 0.5
0.4
0.3
0.3
0.2
0.2
0.1
A! B!
Spectral
product:
spectral
signatures
1 2 3 4 5 6
0
1
Spatial
product:
abundance
maps
2 3 4 5 6
1.32 1.65 1.98 2.31 2.64 1.32 1.65 1.98 2.31 2.64
MVC NMF associated spectra spatial VCA associated spectra
R1.1
um
B2.3
um
0.4
0.4
Apparent reflectance
0.3
0.3 A! B!
0.2
0.2
0.1
0.1
C! 1.32
1 2
1.65
3 4
1.98
5
2.31
6
2.64
D! 1.32
1 2
1.65
3 4
1.98
5
2.31
6
2.64
Wavelength in microns Wavelength in microns
C! D! Dark
Strong
bright
Weak
bright
Final
product:
composite
abundance
map!
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
20
21. Sources:
Dark,
strong
bright,
weak
bright
C!
HiRISE
image
[Ceamanos
TGRS
2011]
MVC-‐NMF
composite
map
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
21
22. Sources:
Dark,
strong
bright,
weak
bright
VCA
BPSS
A! B!
MVC-‐NMF
spatial-‐VCA
C! D!
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
22
26. Dark
features
reference
abundance
map
CO2
ice
CRISM HiRISE
150 m 50 m
Detail
of
the
Russell
dune
observed
by
the
CRISM
and
the
HiRISE
instruments.
CRISM
frt42aa
in
blue,
HiRISE
PSP_002482_1255
in
green
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
26
27. Dark
features
reference
abundance
map A!
CO2
ice
C!
CRISM HiRISE
150 m 50 m
Detail
of
the
Russell
dune
observed
by
the
CRISM
and
the
HiRISE
instruments.
CRISM
frt42aa
in
blue,
HiRISE
PSP_002482_1255
in
green
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
27
28. HiRISE PSP_002482_1255 CRISM frt000042aa Registration correlation coefficient
1
0.25
m/pix
18
m/pix
Avg.
Corr.
=
0.9
29862×63004
pix
604×420
pix
0.7
0.8
0.7
0.6
HiRISE
image
0.5
0.4
0.3
1.
Registration
CRISM
image
0.2
0.1
A! B! C! 0
Reg.
HiRISE
image
2.
Classification
Classification
map
3.
Pixel
counting
Abundance
map
(ground
truth)
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
28
29. HiRISE PSP_002482_1255 CRISM frt000042aa Registration correlation coefficient
1
0.25
m/pix
18
m/pix
Avg.
Corr.
=
0.9
29862×63004
pix
604×420
pix
0.7
0.8
0.7
0.6
HiRISE
image
0.5
0.4
0.3
1.
Registration
CRISM
image
0.2
0.1
A! B! C! 0
Reg.
HiRISE
image
2.
Classification
Classification
map
Classification map Ground truth
1
A! B! 0.9
0.8
3.
Pixel
counting
0.7
0.6
0.5
Abundance
map
0.4
(ground
truth)
0.3
0.2
0.1
0
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
29
30. HiRISE PSP_002482_1255 CRISM frt000042aa Registration correlation coefficient
1
0.25
m/pix
18
m/pix
Avg.
Corr.
=
0.9
29862×63004
pix
604×420
pix
0.7
0.8
0.7
0.6
HiRISE
image
0.5
0.4
0.3
1.
Registration
CRISM
image
0.2
0.1
A! B! C! 0
Reg.
HiRISE
image
2.
Classification
Classification
map
Classification map Ground truth
1
Pixel
counting
for
two
CRISM
pixels
A! B! 0.9
0.8
3.
Pixel
counting
a(xi)=0.10
a(xj)=0.35
0.7
0.6
0.5
Abundance
map
0.4
(ground
truth)
0.3
0.2
0.1
0
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
30
31. VCA
BPSS
MVC-‐NMF
spatial-‐VCA
Ground
truth
Registration
accuracy
• 10%
error
between
ground
truth
and
unmixing
results
• MVC-‐NMF
obtains
the
best
r =
0.83
and
ε =
0.08
• BPSS
provides
accurate
abundances
• VCA
provides
underestimated
abundances
• spatial-‐VCA
does
not
extract
the
dark
source
satisfactorily
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
31
34. ACCES au laboratoire GIPSA-lab
Feuille de route
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
34
GIPSA-lab
36. Dark
Strong
bright
Weak
bright
Non-‐linear
residue
due
to
unaccurate
atmospheric
correction
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
36
37. VZA=-‐30º
Dark
Strong
bright
Weak
bright
target
Non-‐linear
residue
due
to
unaccurate
atmospheric
correction
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
37
38. VZA=30º
VZA=-‐30º
Dark
Strong
bright
Weak
bright
target
Non-‐linear
residue
due
to
unaccurate
atmospheric
correction
X.
Ceamanos.
26/07/11
IGARSS
2011
-‐
xavier.ceamanos@obs.ujf-‐grenoble.fr
38