I
nte
rna
t
io
na
l J
o
urna
l o
f
Adv
a
nces in Applie
d Science
s
(
I
J
AAS)
Vo
l.
15
,
No
.
1
,
Ma
r
ch
20
26
,
p
p
.
2
8
1
~
2
9
2
I
SS
N:
2252
-
8
8
1
4
,
DOI
:
1
0
.
1
1
5
9
1
/ijaas
.
v15.
i
1
.
pp
281
-
2
9
2
281
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
a
s
.
ia
esco
r
e.
co
m
DCNNV
A:
a deep
conv
o
lutiona
l ne
ura
l net
wo
rk
f
o
r
v
o
lca
nic
a
ctivity
clas
sifica
t
io
n using
sa
tellit
e
ima
g
ery
Ya
s
ir
H
us
s
ei
n Sha
k
ir
1
,
Ree
m
Ali Mu
t
la
g
1
,
E
s
ha
q Az
iz
Awa
dh
AL
M
a
nd
ha
ri
2
,
M
o
ha
m
ed
Sh
a
bb
ir
A
bd
ul
na
bi
3
1
C
o
l
l
e
g
e
o
f
G
r
a
d
u
a
t
e
S
t
u
d
i
e
s,
U
n
i
v
e
r
s
i
t
i
Te
n
a
g
a
N
a
s
i
o
n
a
l
,
K
a
j
a
n
g
,
M
a
l
a
y
s
i
a
2
G
r
a
d
u
a
t
e
S
c
h
o
o
l
o
f
T
e
c
h
n
o
l
o
g
y
,
A
s
i
a
P
a
c
i
f
i
c
U
n
i
v
e
r
si
t
y
o
f
Te
c
h
n
o
l
o
g
y
a
n
d
I
n
n
o
v
a
t
i
o
n
,
K
u
a
l
a
L
u
m
p
u
r
,
M
a
l
a
y
si
a
3
S
c
h
o
o
l
o
f
T
e
c
h
n
o
l
o
g
y
,
A
s
i
a
P
a
c
i
f
i
c
U
n
i
v
e
r
si
t
y
o
f
Te
c
h
n
o
l
o
g
y
a
n
d
I
n
n
o
v
a
t
i
o
n
,
K
u
a
l
a
L
u
mp
u
r
,
M
a
l
a
y
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Sep
4
,
2
0
2
5
R
ev
is
ed
Dec
2
3
,
2
0
2
5
Acc
ep
ted
J
an
1
,
2
0
2
6
M
o
n
i
to
ri
n
g
a
n
d
c
las
sify
in
g
v
o
lca
n
ic
a
c
ti
v
it
y
a
re
a
c
rit
ica
l
tas
k
f
o
r
d
isa
ste
r
risk
re
d
u
c
ti
o
n
a
n
d
h
a
z
a
rd
m
a
n
a
g
e
m
e
n
t.
Re
c
e
n
t
d
isc
o
v
e
ries
i
n
m
a
c
h
in
e
lea
rn
in
g
a
n
d
d
e
e
p
lea
rn
in
g
h
a
v
e
p
ro
v
e
d
e
x
c
e
ll
e
n
t
sa
telli
t
e
ima
g
e
c
las
si
fi
c
a
ti
o
n
a
n
d
v
o
lca
n
ic
a
n
o
m
a
ly
id
e
n
ti
fi
c
a
ti
o
n
c
a
p
a
b
il
it
ies
,
y
e
t
th
e
m
a
jo
rit
y
o
f
e
x
ist
in
g
m
e
th
o
d
s
s
u
ffe
r
fro
m
sm
a
ll
d
a
tas
e
ts,
p
a
rti
c
u
larly
o
n
so
li
tary
d
a
ta
m
o
d
a
li
ti
e
s
o
r
p
a
rti
c
u
lar
c
a
se
s
,
m
e
re
ly
a
s
e
x
a
m
p
le
s.
In
t
h
is
re
se
a
rc
h
wo
rk
,
we
p
u
t
fo
rwa
rd
d
e
v
e
lo
p
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
fo
r
v
o
lca
n
ic
a
c
ti
v
it
y
(DCN
NV
A)
c
las
si
fi
c
a
ti
o
n
sp
e
c
ifi
c
a
ll
y
d
e
sig
n
e
d
fo
r
sa
telli
te
ima
g
e
ry
o
n
v
o
lca
n
ic
a
c
ti
v
it
y
.
We
ri
g
o
r
o
u
sl
y
b
e
n
c
h
m
a
rk
e
d
DCN
NV
A
m
o
d
e
l'
s
st
re
n
g
th
a
g
a
i
n
st
a
t
o
tal
o
f
e
ig
h
t
sta
te
-
of
-
th
e
-
a
rt
tran
sfe
r
lea
rn
in
g
m
o
d
e
ls
:
Re
sN
e
t5
0
,
NA
S
Ne
tL
a
rg
e
,
De
n
se
Ne
t1
2
1
,
M
o
b
i
leN
e
t,
In
c
e
p
ti
o
n
V3
,
Xc
e
p
ti
o
n
,
VG
G
1
9
,
a
n
d
VG
G
1
6
.
Co
m
p
a
ra
ti
v
e
e
x
p
e
rime
n
tal
re
su
lt
s
sh
o
w
th
a
t
p
r
o
p
o
se
d
DC
NN
VA
fra
m
e
wo
rk
'
s
o
v
e
ra
ll
p
e
rfo
rm
a
n
c
e
si
g
n
ifi
c
a
n
t
ly
su
rp
a
ss
e
s
it
s
c
o
m
p
e
t
it
o
rs
wit
h
a
n
a
c
c
u
ra
c
y
o
f
9
9
.
3
3
%
,
p
re
c
isio
n
o
f
1
0
0
%
,
re
c
a
ll
o
f
9
8
.
6
7
%
,
a
n
d
F
1
-
sc
o
re
o
f
9
9
.
3
3
%
,
si
g
n
ifi
c
a
n
t
ly
b
e
a
ti
n
g
e
x
isti
n
g
sta
te
-
of
-
t
h
e
-
a
rt
m
e
th
o
d
s
.
Also
,
we
c
re
a
te
a
d
e
p
l
o
y
a
b
le
g
ra
p
h
ica
l
u
se
r
in
terfa
c
e
(G
UI)
sy
ste
m
th
a
t
is
c
a
p
a
b
le
o
f
re
a
l
-
ti
m
e
m
o
n
i
to
rin
g
o
n
v
o
lca
n
ic
a
c
ti
v
i
ty
a
n
d
g
e
n
e
ra
te
s
m
u
lt
i
-
m
o
d
a
l
a
lert
p
ro
c
e
ss
in
g
th
a
t
c
a
n
m
a
k
e
th
is
re
se
a
rc
h
d
irec
tl
y
a
p
p
li
c
a
b
le
f
o
r
p
ra
c
ti
c
a
l
u
se
o
n
d
isa
ste
r
m
a
n
a
g
e
m
e
n
t
a
s
we
ll
a
s
i
n
e
a
rly
wa
rn
in
g
sy
ste
m
s
.
Th
is
re
se
a
rc
h
c
o
n
tri
b
u
tes
a
sc
a
lab
le,
stro
n
g
,
a
s
we
ll
a
s
p
ra
c
ti
c
a
l
so
lu
ti
o
n
to
wa
rd
s v
o
lca
n
ic
h
a
z
a
rd
id
e
n
ti
fi
c
a
ti
o
n
a
s
we
ll
a
s
a
b
a
se
li
n
e
s
y
ste
m
to
wa
rd
d
e
v
e
l
o
p
i
n
g
fu
t
u
re
m
u
l
ti
-
m
o
d
a
l
a
s
we
ll
a
s
re
a
l
-
ti
m
e
g
e
o
h
a
z
a
rd
trac
k
in
g
sy
st
e
m
fra
m
e
wo
rk
s.
K
ey
w
o
r
d
s
:
DC
NNV
A
Dee
p
lear
n
in
g
Mo
d
el
d
ep
lo
y
m
en
t
Mu
lti
-
m
o
d
al
aler
t sy
s
tem
Satellite d
ata
clas
s
if
icatio
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Mo
h
am
ed
Sh
a
b
b
ir
Ab
d
u
ln
ab
i
Sch
o
o
l o
f
T
ec
h
n
o
lo
g
y
,
Asi
a
Pacif
ic
Un
iv
er
s
ity
o
f
T
ec
h
n
o
lo
g
y
an
d
I
n
n
o
v
atio
n
Ku
ala
L
u
m
p
u
r
5
7
0
0
0
,
Ma
lay
s
ia
E
m
ail:
m
o
h
am
ed
.
s
h
ab
b
ir
@
ap
u
.
ed
u
.
m
y
1.
I
NT
RO
D
UCT
I
O
N
Mo
n
ito
r
in
g
a
n
d
class
if
y
in
g
v
o
lcan
ic
ev
en
ts
ar
e
wid
ely
r
ec
o
g
n
ized
as
o
n
e
o
f
th
e
m
o
s
t
c
r
itical
y
et
ch
allen
g
in
g
r
esp
o
n
s
ib
ilit
ies
f
ac
ed
v
ia
o
b
s
er
v
ato
r
ies
a
n
d
g
eo
s
cien
tific
in
s
titu
tio
n
s
ar
o
u
n
d
th
e
wo
r
ld
.
T
h
e
v
o
lcan
ic
ac
tiv
ity
p
r
esen
ts
n
o
t
o
n
ly
a
s
ig
n
i
f
ican
t
n
at
u
r
al
h
a
za
r
d
to
h
u
m
an
p
o
p
u
latio
n
s
an
d
in
f
r
astru
ctu
r
e
b
u
t
also
a
f
u
n
d
am
en
tal
ar
ea
o
f
s
tu
d
y
f
o
r
u
n
d
er
s
tan
d
i
n
g
ea
r
th
d
y
n
am
ic
s
y
s
tem
s
.
T
r
ad
itio
n
ally
,
th
e
o
b
s
er
v
atio
n
an
d
an
aly
s
is
o
f
v
o
lcan
ic
ac
tiv
ity
h
av
e
r
elied
h
ea
v
ily
o
n
g
r
o
u
n
d
-
b
ased
m
o
n
ito
r
in
g
s
y
s
tem
s
,
m
an
u
a
l
im
ag
e
in
ter
p
r
etatio
n
,
an
d
co
n
v
en
tio
n
al
s
tatis
t
ical
m
o
d
els.
Ho
w
ev
er
,
th
ese
ap
p
r
o
ac
h
es
wer
e
o
f
ten
lim
ited
b
y
tem
p
o
r
al
d
elay
s
r
estricte
d
co
v
er
ag
e,
a
n
d
th
e
h
ig
h
c
o
s
ts
ass
o
ciate
d
with
o
n
-
s
ite
m
o
n
ito
r
in
g
.
T
h
e
r
ap
id
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
1
,
Ma
r
c
h
2
0
2
6
:
281
-
2
9
2
282
ad
v
an
ce
m
e
n
t
o
f
r
em
o
te
s
en
s
in
g
tech
n
o
lo
g
ies
,
e
s
p
ec
ially
th
e
g
r
o
win
g
av
ailab
ilit
y
o
f
h
ig
h
-
r
eso
lu
tio
n
m
u
ltis
p
ec
tr
al
an
d
th
er
m
al
s
atellite
im
ag
er
y
,
h
as
g
r
ea
tly
ex
p
an
d
e
d
th
e
ab
ilit
y
o
f
r
esear
ch
er
s
to
m
o
n
ito
r
v
o
lcan
ic
ac
tiv
ity
at
b
o
th
lo
ca
l
an
d
g
lo
b
al
s
ca
les.
Su
p
p
o
r
te
d
b
y
i
n
cr
ea
s
ed
co
m
p
u
tatio
n
al
p
o
wer
,
th
e
s
e
to
o
ls
n
o
w
allo
w
lar
g
e
-
s
ca
le
d
ata
p
r
o
ce
s
s
in
g
an
d
r
ea
l
-
tim
e
m
o
n
ito
r
in
g
ca
p
a
b
ilit
ies
th
at
wer
e
p
r
e
v
io
u
s
ly
u
n
attain
ab
le
u
s
in
g
tr
ad
itio
n
al
a
p
p
r
o
ac
h
es
[
1
]
.
At
th
e
s
am
e
tim
e,
ar
tific
ial
in
tellig
en
ce
h
as
em
er
g
e
d
as
a
tr
an
s
f
o
r
m
ativ
e
ap
p
r
o
ac
h
f
o
r
h
an
d
lin
g
c
o
m
p
l
ex
h
ig
h
-
d
im
e
n
s
io
n
al
d
atasets
,
an
d
with
m
ac
h
i
n
e
lear
n
in
g
an
d
d
ee
p
lear
n
in
g
tech
n
iq
u
es
h
a
v
e
s
h
o
wn
ex
c
ep
tio
n
al
ef
f
ec
tiv
e
n
ess
in
s
o
lv
in
g
p
r
o
b
lem
s
r
elate
d
to
i
m
ag
e
class
if
icatio
n
,
an
o
m
aly
d
etec
tio
n
,
an
d
p
r
ed
ic
tiv
e
m
o
d
elin
g
[
2
]
,
[
3
]
.
A
m
ajo
r
r
esear
ch
f
o
cu
s
was
th
e
class
if
icatio
n
o
f
s
atellite
im
ag
er
y
,
wh
ic
h
p
lay
s
a
cr
u
ci
al
r
o
le
in
d
etec
tin
g
s
u
b
tle
th
er
m
al
an
o
m
alies,
g
as
em
is
s
io
n
s
,
lan
d
co
v
er
s
h
if
ts
,
an
d
o
th
er
in
d
icato
r
s
o
f
v
o
lcan
ic
ac
tiv
ity
.
R
ec
en
t
wo
r
k
h
as
s
h
o
w
n
th
at
im
m
u
n
e
-
in
s
p
ir
ed
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
a
n
d
o
th
er
m
o
d
els
ca
n
s
ig
n
if
ican
tly
b
o
o
s
t
class
if
icatio
n
ac
c
u
r
ac
y
ev
e
n
wh
e
n
d
atasets
ar
e
n
o
is
y
o
r
lim
ited
[4
]
.
A
m
o
n
g
d
ee
p
lear
n
in
g
m
o
d
els
,
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NNs)
s
tan
d
o
u
t
as
o
n
e
o
f
th
e
m
o
s
t
ef
f
ec
tiv
e
tec
h
n
iq
u
es
f
o
r
r
em
o
te
s
en
s
in
g
ap
p
licatio
n
s
.
C
NN
s
h
av
e
a
u
n
iq
u
e
ca
p
ab
ilit
y
to
au
t
o
m
atica
lly
ex
tr
ac
t
h
ier
ar
ch
ical
an
d
s
p
atial
f
ea
tu
r
es
f
r
o
m
im
ag
es,
w
h
ich
allo
ws
th
em
to
s
u
r
p
ass
tr
ad
itio
n
al
m
eth
o
d
s
th
at
r
ely
o
n
h
an
d
c
r
af
ted
f
ea
t
u
r
es
[
5
]
.
T
h
is
s
tr
en
g
th
is
esp
ec
ial
ly
im
p
o
r
tan
t
in
v
o
lcan
ic
m
o
n
ito
r
in
g
,
wh
er
e
d
etec
tin
g
s
u
b
tle
s
p
atio
tem
p
o
r
al
p
atter
n
s
ca
n
b
e
cr
itical
f
o
r
id
en
tify
i
n
g
ea
r
ly
s
ig
n
s
o
f
ac
tiv
ity
.
I
n
ad
d
itio
n
,
th
e
u
s
e
o
f
tr
a
n
s
f
er
lear
n
in
g
h
as
f
u
r
th
er
im
p
r
o
v
e
d
C
NN
p
er
f
o
r
m
an
ce
.
Mo
d
els th
at
h
av
e
b
ee
n
p
r
e
-
tr
ai
n
ed
o
n
lar
g
e
-
s
ca
le
d
atasets
s
u
ch
as I
m
ag
eNe
t c
an
b
e
f
in
e
-
tu
n
ed
f
o
r
v
o
lcan
ic
im
a
g
e
class
if
icatio
n
.
T
h
is
a
p
p
r
o
ac
h
s
ig
n
if
ican
tly
r
ed
u
ce
s
th
e
r
eq
u
ir
em
en
t
f
o
r
lar
g
e
am
o
u
n
ts
o
f
d
o
m
ain
-
s
p
ec
if
ic
tr
ain
in
g
d
ata
wh
ile
s
im
u
ltan
eo
u
s
ly
b
o
o
s
tin
g
class
if
icatio
n
ac
c
u
r
ac
y
[
6
]
,
[
7
]
.
Su
c
h
ap
p
r
o
ac
h
es
ar
e
p
ar
ticu
lar
ly
v
alu
ab
le
in
v
o
lca
n
ic
m
o
n
ito
r
in
g
,
wh
e
r
e
an
n
o
tated
d
atasets
ar
e
s
ca
r
ce
an
d
im
b
alan
ce
d
.
Desp
ite
s
ig
n
if
i
ca
n
t
p
r
o
g
r
ess
,
ex
is
tin
g
m
e
th
o
d
s
f
ac
e
ch
allen
g
es
r
elat
ed
to
s
ca
lab
ilit
y
,
g
en
er
aliza
tio
n
ac
r
o
s
s
d
if
f
er
e
n
t
v
o
lcan
o
es
,
an
d
th
e
in
te
g
r
atio
n
o
f
m
u
ltimo
d
al
s
atellite
d
ata.
Ar
tific
ial
in
tellig
en
ce
m
eth
o
d
s
an
d
p
ix
el
-
b
ased
class
if
ier
s
h
av
e
s
h
o
wn
p
o
ten
tial
,
b
u
t
th
eir
a
p
p
li
ca
tio
n
to
v
o
lcan
ic
ac
tiv
ity
r
em
ain
s
lim
ited
.
T
h
e
co
n
tr
i
b
u
tio
n
s
ar
e
t
h
u
s
co
n
clu
d
ed
to
b
e
as f
o
llo
ws:
i)
I
n
tr
o
d
u
ce
s
a
d
ee
p
lear
n
in
g
-
b
as
ed
f
r
am
ewo
r
k
f
o
r
v
o
lcan
ic
ac
t
iv
ity
class
if
icatio
n
u
s
in
g
a
cu
s
to
m
-
d
esig
n
ed
d
ev
elo
p
d
ee
p
c
o
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
f
o
r
v
o
lca
n
ic
ac
tiv
ity
(
DC
NNVA
)
ar
ch
itectu
r
e.
ii)
C
o
m
p
r
eh
en
s
iv
e
m
o
d
el
e
v
alu
at
io
n
v
ia
eig
h
t
s
tate
-
of
-
th
e
-
ar
t
tr
an
s
f
er
lear
n
in
g
m
o
d
els
,
s
u
ch
as
R
esNet5
0
,
NASNetL
ar
g
e,
Den
s
eNe
t1
2
1
,
Mo
b
ileNet,
I
n
ce
p
tio
n
V3
,
Xce
p
tio
n
,
VGG1
9
,
a
n
d
VGG1
6
.
iii)
E
n
h
an
ce
d
m
o
d
el
r
o
b
u
s
tn
ess
v
ia
d
ata
au
g
m
en
tatio
n
tech
n
i
q
u
es
an
d
ex
ten
s
iv
e
ex
p
e
r
im
e
n
tal
v
alid
atio
n
u
s
es
m
u
ltip
le
p
er
f
o
r
m
an
ce
m
e
tr
ics.
iv
)
Dep
lo
y
ab
le
g
r
a
p
h
ical
u
s
er
in
ter
f
ac
e
(
GUI
)
s
y
s
tem
th
at
p
r
o
v
id
es
r
ea
l
-
tim
e
v
o
lcan
ic
ac
tiv
ity
m
o
n
ito
r
in
g
with
m
u
lti
-
m
o
d
al
aler
t c
a
p
ab
il
ities
,
m
ak
in
g
th
e
r
esear
ch
p
r
ac
tically
ap
p
licab
le
f
o
r
d
is
aster
m
an
ag
em
en
t.
T
h
e
s
tr
u
ctu
r
e
o
f
th
is
p
ap
er
is
o
u
tlin
ed
as
f
o
llo
ws:
s
ec
tio
n
2
p
r
esen
ts
th
e
r
elate
d
wo
r
k
,
r
ev
iewin
g
p
r
ev
io
u
s
s
tu
d
ies
an
d
r
ec
en
t
d
e
v
elo
p
m
en
ts
r
elev
a
n
t
to
th
is
r
e
s
ea
r
ch
ar
ea
.
Sectio
n
3
t
h
e
m
at
er
ials
an
d
m
eth
o
d
s
,
p
r
o
v
id
es
d
etails
o
f
t
h
e
ar
c
h
itectu
r
e,
m
eth
o
d
o
lo
g
y
,
d
atasets
,
p
r
e
-
p
r
o
ce
s
s
in
g
,
d
i
v
id
in
g
d
ata,
i
n
v
esti
g
ated
m
o
d
els,
an
d
ev
alu
atio
n
m
etr
i
cs.
Sectio
n
4
r
esu
lts
an
d
d
is
cu
s
s
i
o
n
with
co
m
p
ar
is
o
n
s
to
r
elate
d
s
tu
d
ies
in
th
e
liter
atu
r
e.
Fin
ally
,
s
ec
t
io
n
5
p
r
esen
ts
th
e
co
n
clu
s
io
n
s
.
2.
RE
L
AT
E
D
WO
RK
R
ec
en
t
r
esear
ch
h
as
ex
p
lo
r
ed
th
e
u
s
e
o
f
m
ac
h
in
e
lear
n
in
g
an
d
d
ee
p
lear
n
in
g
f
o
r
s
atellite
im
ag
e
class
if
icatio
n
an
d
v
o
lcan
ic
ac
t
iv
ity
m
o
n
ito
r
i
n
g
.
E
ar
l
y
s
tu
d
ies
ap
p
lied
p
ix
el
-
b
ased
m
ac
h
in
e
l
ea
r
n
in
g
;
s
im
ilar
ly
,
E
b
r
ah
im
y
a
n
d
Z
h
a
n
g
[
8
]
in
c
r
ea
s
ed
class
if
icatio
n
ac
cu
r
ac
y
b
y
ap
p
ly
i
n
g
v
ar
i
o
u
s
ex
tr
em
e
lear
n
in
g
m
ac
h
i
n
e
class
if
ier
s
to
g
eth
er
.
Ou
ch
r
a
et
a
l.
[
9
]
c
o
m
p
ar
e
d
s
u
p
e
r
v
is
ed
an
d
u
n
s
u
p
er
v
is
ed
m
ac
h
in
e
lea
r
n
in
g
tec
h
n
iq
u
es
o
f
u
r
b
an
lan
d
c
o
v
er
in
g
class
if
icatio
n
b
y
em
p
lo
y
in
g
L
a
n
d
s
at
8
im
ag
er
y
an
d
em
p
h
asized
m
eth
o
d
o
lo
g
ica
l
d
if
f
er
en
ce
s
a
p
p
lied
to
f
ea
tu
r
e
ex
tr
ac
tio
n
.
I
n
v
o
lcan
ic
co
n
tex
ts
,
C
ar
iello
et
a
l.
[
1
0
]
s
h
o
we
d
th
e
a
p
p
licatio
n
o
f
m
ac
h
in
e
lear
n
in
g
to
Sen
tin
el
-
2
im
ag
er
y
to
t
r
ac
k
v
o
lcan
ic
th
er
m
al
an
o
m
alies,
wh
ile
B
u
ttar
an
d
Sach
an
[
1
1
]
u
tili
ze
d
R
esNet
-
1
5
2
to
class
if
y
an
d
g
eo
-
im
ag
e
im
a
g
es
with
a
f
o
c
u
s
o
n
t
h
e
r
e
q
u
ir
em
en
t
o
f
au
to
m
ated
f
ea
tu
r
e
ex
tr
ac
tio
n
.
C
o
r
r
ad
in
o
et
a
l.
[
1
2
]
u
tili
ze
d
U
-
NE
T
to
an
al
y
ze
2
1
y
ea
r
s
o
f
ad
v
an
ce
d
s
p
ac
eb
o
r
n
e
th
e
r
m
al
em
is
s
io
n
an
d
r
ef
lectio
n
r
ad
i
o
m
eter
(
ASTE
R
)
g
lo
b
al
th
er
m
al
in
f
r
ar
ed
(
T
I
R
)
im
ag
er
y
o
f
f
iv
e
v
o
lcan
o
es
an
d
attain
ed
9
3
%
e
f
f
ec
tiv
en
ess
o
f
an
o
m
aly
d
etec
tio
n
.
Similar
l
y
,
Sh
u
ltz
[
1
3
]
p
r
esen
te
d
th
e
C
NN
-
b
ased
f
r
am
e,
h
o
ts
p
o
t
lear
n
i
n
g
a
n
d
id
en
tific
at
io
n
n
etwo
r
k
(
H
o
tLI
NK
)
,
tes
ted
an
d
p
r
o
v
ed
with
m
o
d
er
ate
r
eso
lu
tio
n
im
ag
in
g
s
p
ec
tr
o
r
ad
io
m
eter
(
MO
DI
S)
an
d
v
is
ib
le
in
f
r
ar
e
d
im
ag
i
n
g
r
ad
io
m
eter
s
u
ite
(
VI
I
R
S
)
d
ata
co
llectio
n
s
,
an
d
attain
ed
o
v
er
9
5
% a
cc
u
r
a
cy
o
f
h
o
ts
p
o
t id
en
tific
atio
n
.
Oth
er
ap
p
r
o
ac
h
es
h
av
e
u
s
e
d
C
NNs
in
n
o
n
-
im
ag
e
d
o
m
ain
s
.
Oñ
ate
et
a
l.
[
1
4
]
f
o
r
ec
asted
m
icr
o
-
ea
r
th
q
u
ak
es
u
s
in
g
m
an
if
o
ld
lear
n
in
g
an
d
au
d
io
-
d
r
iv
en
f
ea
tu
r
es
,
an
d
ac
h
iev
ed
o
v
er
9
4
%
ac
cu
r
ac
y
.
Nu
n
n
ar
i
an
d
C
alv
ar
i
[
1
5
]
co
n
tr
asted
eig
h
t
C
NN
m
o
d
els
f
o
r
er
u
p
tiv
e
ac
tiv
ity
m
o
n
ito
r
i
n
g
o
f
Mo
u
n
t
E
tn
a
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l
Sci
I
SS
N:
2252
-
8
8
1
4
DC
N
N
V
A
:
a
d
ee
p
co
n
vo
lu
tio
n
a
l n
eu
r
a
l n
etw
o
r
k
fo
r
vo
lca
n
ic
a
ctivity
cla
s
s
ifica
tio
n
…
(
Ya
s
i
r
Hu
s
s
ein
S
h
a
kir)
283
estab
lis
h
ed
tr
an
s
f
er
lear
n
in
g
a
s
s
u
p
er
io
r
.
C
h
en
et
a
l.
[
1
6
]
p
r
o
p
o
s
ed
t
r
an
s
f
er
lear
n
i
n
g
-
b
ased
VGG
(
T
VGG
)
f
o
r
r
em
o
te
s
en
s
in
g
im
ag
e
clas
s
if
icatio
n
with
9
9
.
1
8
%
ac
cu
r
ac
y
b
ased
o
n
a
VGG
-
b
ase
d
tr
an
s
f
er
lear
n
in
g
tech
n
iq
u
e.
Mo
h
a
n
et
a
l.
[
1
7
]
in
tr
o
d
u
ce
d
Ho
ts
p
o
tter
,
an
en
d
-
to
-
en
d
s
y
s
tem
d
esig
n
e
d
to
a
u
to
m
atica
lly
d
etec
t
s
u
b
tle
v
o
lcan
ic
t
h
er
m
al
a
n
o
m
alies
in
s
atellite
im
ag
er
y
wh
i
le
also
g
en
er
atin
g
k
e
y
th
e
r
m
al
s
tatis
tic
s
.
E
ar
lier
m
eth
o
d
s
f
o
r
au
t
o
m
ated
v
o
lca
n
ic
th
er
m
al
f
ea
tu
r
e
(
VT
F)
d
et
ec
tio
n
wer
e
lim
ited
b
y
s
m
all
d
atasets
an
d
n
ar
r
o
w
g
eo
g
r
a
p
h
ic
co
v
er
a
g
e
.
T
h
e
R
ee
d
-
Xiao
li
alg
o
r
ith
m
(
L
R
X)
+CNN
ap
p
r
o
ac
h
,
wh
ich
tr
ain
e
d
C
NN
f
o
r
5
0
ep
o
ch
s
o
n
th
e
L
R
X
s
co
r
e
im
ag
es,
p
r
o
d
u
ce
d
a
n
F1
-
s
co
r
e
o
f
8
8
.
4
%
an
d
a
test
ac
cu
r
ac
y
o
f
9
0
.
3
%.
Hu
er
tas
et
a
l.
[
1
8
]
d
ev
elo
p
e
d
VGG1
6
an
d
I
n
ce
p
t
i
o
n
C
NN
m
o
d
els
to
an
aly
ze
m
u
d
v
o
lcan
o
im
a
g
er
y
an
d
f
in
d
in
g
s
s
h
o
w
th
at
b
o
th
tr
an
s
f
er
lear
n
in
g
an
d
c
u
s
to
m
C
NNs
ar
e
well
-
s
u
ited
f
o
r
h
an
d
lin
g
th
e
c
h
allen
g
es
o
f
t
h
is
ty
p
e
o
f
d
ata.
T
h
e
VGG1
6
m
o
d
el
ac
h
iev
e
d
test
ac
cu
r
ac
y
o
f
9
3
%,
with
p
r
ec
is
io
n
an
d
r
ec
all
s
co
r
es
o
f
93%
an
d
9
4
%
f
o
r
th
e
“m
u
d
”
class
an
d
9
3
%
a
n
d
9
3
%
f
o
r
th
e
“n
o
m
u
d
”
class
.
T
h
e
I
n
ce
p
tio
n
m
o
d
el
also
d
em
o
n
s
tr
ated
s
tab
le
p
er
f
o
r
m
an
ce
with
o
n
ly
m
in
o
r
v
ar
iatio
n
.
I
n
s
u
m
m
ar
y
,
th
e
s
tu
d
ies
to
d
ate
h
av
e
u
tili
ze
d
s
ev
er
al
v
ar
io
u
s
d
atasets
an
d
v
ar
iety
o
f
m
ac
h
in
e
lear
n
i
n
g
an
d
d
ee
p
lear
n
in
g
al
g
o
r
ith
m
s
tec
h
n
iq
u
es
to
v
o
lca
n
ic
m
o
n
ito
r
in
g
an
d
s
atellite
i
m
ag
e
class
if
icatio
n
.
W
h
ile
th
ese
ap
p
r
o
ac
h
es
d
em
o
n
s
tr
ate
s
ig
n
if
ican
t
p
r
o
g
r
ess
,
m
o
s
t
r
em
ain
at
th
e
ex
p
er
im
e
n
tal
s
tag
e
an
d
lack
p
r
ac
tical
d
ep
lo
y
m
en
t.
I
n
p
ar
ticu
lar
n
o
ex
is
tin
g
r
esear
c
h
h
as
p
r
o
p
o
s
ed
a
co
m
p
r
eh
e
n
s
iv
e
f
r
am
ew
o
r
k
th
at
in
teg
r
ates
a
d
ep
lo
y
a
b
le
GUI
with
r
ea
l
-
tim
e
v
o
lcan
ic
ac
tiv
it
y
m
o
n
ito
r
in
g
an
d
m
u
lti
-
m
o
d
al
aler
tin
g
ca
p
ab
ilit
i
es
f
ea
tu
r
es
th
at
ar
e
ess
en
tial
f
o
r
ef
f
ec
tiv
e
d
is
aster
m
an
a
g
em
e
n
t
ap
p
licatio
n
s
.
Fu
r
th
e
r
m
o
r
e
,
m
an
y
s
tu
d
ies
a
r
e
co
n
s
tr
ain
ed
b
y
lim
ited
d
atasets
an
d
in
c
o
n
s
is
ten
t
p
er
f
o
r
m
an
c
e
ac
r
o
s
s
d
if
f
e
r
en
t
c
o
n
tex
ts
,
h
i
g
h
lig
h
tin
g
th
e
n
ee
d
f
o
r
m
o
r
e
r
o
b
u
s
t
,
s
ca
lab
le
,
an
d
p
r
ac
ti
ca
lly
ap
p
licab
le
s
o
lu
tio
n
s
.
A
co
m
p
ar
ativ
e
an
aly
s
is
o
f
t
h
e
p
r
ev
io
u
s
r
elate
d
wo
r
k
is
p
r
esen
ted
in
tab
u
lar
f
o
r
m
in
T
ab
le
1
.
T
ab
le
1
.
Su
m
m
a
r
y
o
f
r
ec
e
n
t m
ac
h
in
e
lear
n
in
g
an
d
d
ee
p
lea
r
n
in
g
f
o
r
v
o
lca
n
ic
s
atellite
im
ag
e
class
if
icatio
n
Ref
e
r
e
n
c
e
/
y
e
a
r
D
a
t
a
s
e
t
M
e
t
h
o
d
s
R
e
s
u
l
t
s
(
%)
W
e
a
k
n
e
ss
C
o
r
r
a
d
i
n
o
e
t
a
l
.
[
1
2
]
,
2
0
2
4
2
1
y
e
a
r
s
o
f
A
S
TE
R
TI
R
d
a
t
a
(
5
v
o
l
c
a
n
o
e
s)
C
N
N
(
U
-
N
ET
a
r
c
h
i
t
e
c
t
u
r
e
)
A
c
c
u
r
a
c
y
=
9
3
Li
mi
t
e
d
t
o
A
S
TE
R
TI
R
i
m
a
g
e
r
y
,
mo
d
e
l
g
e
n
e
r
a
l
i
z
a
b
i
l
i
t
y
t
o
o
t
h
e
r
sen
s
o
r
s
n
o
t
t
e
s
t
e
d
.
S
h
u
l
t
z
[
1
3
]
,
2
0
2
4
V
I
I
R
S
a
n
d
M
O
D
I
S
i
ma
g
e
r
y
(
A
l
a
s
k
a
n
v
o
l
c
a
n
o
e
s)
C
N
N
(
H
o
t
LI
N
K
)
A
c
c
u
r
a
c
y
=
9
8
D
a
t
a
s
e
t
g
e
o
g
r
a
p
h
i
c
a
l
l
y
r
e
st
r
i
c
t
e
d
(
A
l
a
sk
a
)
,
l
i
m
i
t
e
d
v
a
l
i
d
a
t
i
o
n
o
n
d
i
v
e
r
se
v
o
l
c
a
n
i
c
s
e
t
t
i
n
g
s.
O
ñ
a
t
e
e
t
a
l
.
[
1
4
]
,
2
0
2
4
S
e
i
smi
c
d
a
t
a
(
C
o
t
o
p
a
x
i
a
n
d
Ll
a
i
ma)
A
u
d
i
o
f
e
a
t
u
r
e
s+
p
s
y
c
h
o
a
c
o
u
st
i
c
sca
l
e
s+m
a
n
i
f
o
l
d
l
e
a
r
n
i
n
g
A
c
c
u
r
a
c
y
=
9
4
.
4
4
–
9
5
.
4
5
F
o
c
u
se
d
o
n
se
i
smi
c
d
a
t
a
o
n
l
y
,
l
a
c
k
s m
u
l
t
i
-
mo
d
a
l
i
n
t
e
g
r
a
t
i
o
n
w
i
t
h
sa
t
e
l
l
i
t
e
i
m
a
g
e
r
y
.
N
u
n
n
a
r
i
a
n
d
C
a
l
v
a
r
i
[
1
5
]
,
2
0
2
4
G
r
o
u
n
d
-
b
a
s
e
d
t
h
e
r
mal
i
m
a
g
e
s (
M
t
.
Et
n
a
)
C
o
m
p
a
r
i
so
n
o
f
8
C
N
N
s
(
S
q
u
e
e
z
e
N
e
t
,
G
o
o
g
l
e
N
e
t
,
D
e
n
seN
e
t
2
0
1
,
R
e
sN
e
t
1
8
,
S
h
u
f
f
l
e
N
e
t
,
D
a
r
k
N
e
t
1
9
,
A
l
e
x
N
e
t
,
V
G
G
-
16)
A
c
c
u
r
a
c
y
=
9
4
.
0
7
R
e
s
u
l
t
s res
t
r
i
c
t
e
d
t
o
o
n
e
v
o
l
c
a
n
o
,
c
o
mp
u
t
a
t
i
o
n
a
l
l
y
e
x
p
e
n
si
v
e
d
u
e
t
o
m
u
l
t
i
p
l
e
C
N
N
c
o
m
p
a
r
i
so
n
s.
C
h
e
n
e
t
a
l
.
[
1
6
]
,
2
0
2
4
Ti
a
n
g
o
n
g
-
2
r
e
mo
t
e
sen
s
i
n
g
d
a
t
a
s
e
t
TV
G
G
(
f
r
o
m I
mag
e
N
e
t
)
A
c
c
u
r
a
c
y
=
9
9
.
1
8
R
e
c
a
l
l
=
9
9
.
1
7
R
e
l
i
e
s
h
e
a
v
i
l
y
o
n
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
a
n
d
l
a
c
k
s
t
e
s
t
i
n
g
o
n
v
o
l
c
a
n
i
c
d
a
t
a
set
s
.
M
o
h
a
n
e
t
a
l
.
[
1
7
]
,
20
25
Th
e
r
m
a
l
sat
e
l
l
i
t
e
d
a
t
a
LR
X
+
C
N
N
(
5
0
e
p
o
c
h
s)
A
c
c
u
r
a
c
y
=
9
0
.
3
0
F1
-
sc
o
r
e
=
8
8
.
4
0
D
a
t
a
s
e
t
r
e
l
a
t
i
v
e
l
y
sm
a
l
l
,
p
e
r
f
o
r
m
a
n
c
e
l
o
w
e
r
t
h
a
n
o
t
h
e
r
C
N
N
-
b
a
s
e
d
mo
d
e
l
s.
H
u
e
r
t
a
s
e
t
a
l
.
[
1
8
]
,
2
0
2
5
I
N
G
V
(
mu
d
v
o
l
c
a
n
o
i
ma
g
e
r
y
)
V
G
G
1
6
a
n
d
I
n
c
e
p
t
i
o
n
C
N
N
A
c
c
u
r
a
c
y
=
9
3
P
r
e
c
i
s
i
o
n
=
9
3
R
e
c
a
l
l
=
9
4
F
o
c
u
s
l
i
m
i
t
e
d
t
o
m
u
d
v
o
l
c
a
n
o
e
s
;
p
e
r
f
o
r
m
a
n
c
e
st
a
b
i
l
i
t
y
a
c
r
o
ss
b
r
o
a
d
e
r
d
a
t
a
se
t
s
n
o
t
a
ssess
e
d
.
3.
M
AT
E
R
I
AL
S AN
D
M
E
T
H
O
DS
T
h
is
s
ec
tio
n
co
n
tain
s
th
e
m
ater
ials
an
d
m
eth
o
d
s
f
r
a
m
ewo
r
k
u
s
ed
in
th
is
s
tu
d
y
,
wh
ich
i
s
v
is
u
ally
s
u
m
m
ar
ized
in
Fig
u
r
e
1
.
T
h
e
m
eth
o
d
o
l
o
g
y
is
s
tr
u
ct
u
r
e
d
in
to
s
ix
m
ain
s
tep
s
:
d
ata
ac
q
u
is
itio
n
,
d
ata
p
r
ep
r
o
ce
s
s
in
g
,
d
ata
d
iv
is
io
n
,
m
o
d
el
d
e
v
elo
p
m
e
n
t a
n
d
tr
ain
i
n
g
,
ev
al
u
atio
n
,
a
n
d
d
e
p
lo
y
m
e
n
t.
3
.
1
.
Da
t
a
a
cquis
it
io
n
T
h
is
k
in
d
o
f
d
ataset
co
n
s
is
ts
o
f
th
e
im
ag
es
o
f
th
e
v
o
lcan
ic
ac
tiv
ity
,
an
d
t
h
is
h
as
b
ee
n
g
o
tten
f
r
o
m
s
atellite
im
ag
es.
T
h
e
f
o
llo
win
g
f
ig
u
r
es
o
f
s
ev
e
r
al
im
ag
es
tak
en
f
r
o
m
s
ev
er
al
cl
ass
es
o
f
v
o
lcan
o
es
(
YesActiv
ity
)
an
d
(
No
Activ
ity
)
ar
e
p
r
esen
te
d
in
Fig
u
r
e
2
.
T
h
e
co
m
p
lete
d
ataset
u
s
ed
f
o
r
th
is
ex
p
er
im
en
t
is
p
u
b
licly
av
ailab
le
o
n
lin
e
[1
9]
.
3
.
2
.
Da
t
a
prepro
ce
s
s
ing
T
o
en
s
u
r
e
th
at
th
e
d
ata
was
s
tan
d
ar
d
ized
a
n
d
o
p
tim
ized
f
o
r
d
ee
p
lear
n
in
g
,
s
ev
er
al
p
r
e
p
r
o
ce
s
s
in
g
s
tep
s
wer
e
ap
p
lied
to
th
e
im
ag
es
,
an
d
s
tep
s
wer
e
aim
ed
at
m
ak
in
g
th
e
d
ataset
m
o
r
e
u
n
if
o
r
m
,
r
e
d
u
cin
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
1
,
Ma
r
c
h
2
0
2
6
:
281
-
2
9
2
284
v
ar
iab
ilit
y
,
an
d
allo
win
g
th
e
m
o
d
el
to
g
en
er
alize
m
o
r
e
to
n
ew
d
ata.
Pre
p
r
o
ce
s
s
in
g
p
ip
elin
es
in
v
o
lv
e
d
r
escalin
g
an
d
r
esizin
g
tr
a
n
s
f
o
r
m
atio
n
s
,
n
o
r
m
aliza
tio
n
o
f
th
e
p
ix
els
,
an
d
th
e
ap
p
licatio
n
o
f
d
ata
au
g
m
en
tatio
n
s
tr
ateg
ies.
Data
au
g
m
en
tatio
n
was p
ar
ticu
lar
ly
im
p
o
r
tan
t,
as
it e
f
f
ec
tiv
ely
in
cr
ea
s
ed
th
e
s
ize
o
f
th
e
d
ataset
an
d
in
tr
o
d
u
ce
d
v
ar
iatio
n
s
th
at
s
im
u
late
r
ea
l
-
wo
r
l
d
co
n
d
itio
n
s
(
d
i
f
f
er
en
t
o
r
ien
tatio
n
s
,
s
ca
les,
an
d
d
is
to
r
tio
n
s
)
.
T
h
is
h
elp
s
to
p
r
e
v
en
t
o
v
er
f
itti
n
g
a
n
d
im
p
r
o
v
e
r
o
b
u
s
tn
ess
,
as
s
h
o
wn
in
Fig
u
r
e
3
.
T
h
e
au
g
m
en
ta
tio
n
was
p
er
f
o
r
m
ed
u
s
in
g
th
e
Au
g
m
en
to
r
li
b
r
ar
y
a
n
d
th
e
s
p
ec
if
ic
tr
a
n
s
f
o
r
m
atio
n
s
,
an
d
th
eir
p
r
o
b
a
b
ilit
ies ar
e
s
h
o
wn
in
T
ab
le
2
.
Fig
u
r
e
1
.
Pro
p
o
s
ed
f
r
am
ewo
r
k
f
o
r
v
o
lcan
ic
ac
tiv
ity
class
if
icatio
n
u
s
in
g
s
atellite
im
ag
er
y
Fig
u
r
e
2
.
Sam
p
le
s
atellite
im
ag
es f
r
o
m
th
e
'
y
es a
ctiv
ity
'
an
d
'
n
o
ac
tiv
ity
'
v
o
lcan
ic
d
ataset
class
es
Fig
u
r
e
3
.
I
ll
u
s
tr
atio
n
o
f
d
ata
a
u
g
m
en
tatio
n
tech
n
iq
u
es a
p
p
lie
d
to
o
r
i
g
in
al
v
o
lca
n
ic
im
ag
es
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l
Sci
I
SS
N:
2252
-
8
8
1
4
DC
N
N
V
A
:
a
d
ee
p
co
n
vo
lu
tio
n
a
l n
eu
r
a
l n
etw
o
r
k
fo
r
vo
lca
n
ic
a
ctivity
cla
s
s
ifica
tio
n
…
(
Ya
s
i
r
Hu
s
s
ein
S
h
a
kir)
285
T
ab
le
2
.
Data
p
r
ep
r
o
ce
s
s
in
g
a
n
d
au
g
m
en
tatio
n
p
ar
am
eter
s
f
o
r
m
o
d
el
tr
ain
in
g
P
r
e
p
r
o
c
e
ss
i
n
g
s
t
e
p
D
e
t
a
i
l
s
R
e
sc
a
l
i
n
g
1
/
2
5
5
.
S
i
z
e
(
2
2
4
,
2
2
4
)
.
D
a
t
a
a
u
g
m
e
n
t
a
t
i
o
n
Te
c
h
n
i
q
u
e
.
F
l
i
p
l
e
f
t
r
i
g
h
t
:
p
r
o
b
a
b
i
l
i
t
y
o
f
0
.
3
.
F
l
i
p
t
o
p
b
o
t
t
o
m
:
p
r
o
b
a
b
i
l
i
t
y
o
f
0
.
5
.
R
o
t
a
t
e
:
p
r
o
b
a
b
i
l
i
t
y
of
0
.
5
,
w
i
t
h
a
m
a
x
i
mu
m
l
e
f
t
a
n
d
r
i
g
h
t
r
o
t
a
t
i
o
n
o
f
5
d
e
g
r
e
e
s
.
Zo
o
m:
p
r
o
b
a
b
i
l
i
t
y
o
f
0
.
3
,
w
i
t
h
a
z
o
o
m
f
a
c
t
o
r
b
e
t
w
e
e
n
1
.
1
a
n
d
1
.
2
.
R
a
n
d
o
m
d
i
st
o
r
t
i
o
n
:
p
r
o
b
a
b
i
l
i
t
y
o
f
1
,
w
i
t
h
a
g
r
i
d
w
i
d
t
h
a
n
d
h
e
i
g
h
t
o
f
3
a
n
d
a
ma
g
n
i
t
u
d
e
o
f
5
.
3
.
3
.
Da
t
a
div
is
io
n
T
o
p
r
e
p
ar
e
th
e
d
ataset
f
o
r
tr
ai
n
in
g
an
d
ev
alu
atio
n
,
we
d
iv
id
ed
it
in
to
tr
ain
in
g
,
v
alid
atio
n
,
an
d
test
in
g
s
u
b
s
ets.
T
h
e
s
et
was
d
iv
id
ed
to
b
e
7
5
%
f
o
r
tr
ain
in
g
(
2
,
2
5
0
im
ag
es),
1
5
%
f
o
r
v
alid
atio
n
(
3
5
0
im
ag
es),
a
n
d
2
0
%
f
o
r
test
in
g
(
6
0
0
im
a
g
es)
;
all
o
f
th
ese
s
ets
h
av
e
two
class
es.
T
h
e
o
v
er
all
s
et
h
ad
3
,
0
0
0
im
ag
es
ac
r
o
s
s
two
ca
teg
o
r
ies
,
as sh
o
wn
in
T
a
b
le
3
.
T
ab
le
3
.
Data
s
et
p
ar
titi
o
n
in
g
f
o
r
m
o
d
el
tr
ain
in
g
,
v
alid
atio
n
,
an
d
test
in
g
S
u
b
s
e
t
N
u
mb
e
r
o
f
i
m
a
g
e
s
P
e
r
c
e
n
t
a
g
e
(
%)
N
u
mb
e
r
o
f
c
l
a
ss
es
Tr
a
i
n
i
n
g
2
,
2
5
0
75
2
V
a
l
i
d
a
t
i
o
n
3
5
0
15
2
Te
st
i
n
g
6
0
0
20
2
To
t
a
l
3
,
0
0
0
1
0
0
2
3
.
4
.
Dee
p
lea
rning
mo
dels
a
nd
a
rc
hite
ct
ure
C
u
r
r
en
t
b
r
ea
k
th
r
o
u
g
h
s
in
d
ee
p
lear
n
in
g
h
av
e
im
m
en
s
ely
p
r
o
p
elled
th
e
p
r
o
g
r
ess
o
f
s
atel
lite
im
ag
e
p
r
o
ce
s
s
in
g
,
s
p
ec
if
ically
r
eg
a
r
d
in
g
t
h
e
id
e
n
tific
atio
n
o
f
n
atu
r
al
p
h
e
n
o
m
e
n
a
lik
e
v
o
lca
n
ic
ac
tiv
ity
.
E
m
p
l
o
y
in
g
C
NNs
an
d
tr
an
s
f
er
lear
n
in
g
f
a
cilitates
q
u
ick
f
ea
tu
r
e
d
etec
tio
n
f
r
o
m
d
ata
o
f
h
ig
h
d
im
e
n
s
io
n
ality
,
wh
ile
at
th
e
s
am
e
tim
e
o
v
er
co
m
in
g
d
if
f
ic
u
lties
s
tem
m
in
g
f
r
o
m
lim
ited
d
atasets
wi
th
an
n
o
tatio
n
s
.
Her
e
in
,
we
p
u
t
f
o
r
war
d
a
n
ew
DC
NNVA
an
d
eig
h
t
p
r
e
-
ex
is
tin
g
tr
an
s
f
er
lear
n
in
g
m
o
d
els
,
R
esNet5
0
,
NASNetL
ar
g
e,
Den
s
eNe
t1
2
1
,
Mo
b
ileNet,
I
n
ce
p
tio
n
V3
,
Xc
ep
tio
n
,
VGG1
9
,
an
d
VGG1
6
,
to
b
u
ild
a
co
m
p
lete
class
if
icatio
n
s
y
s
tem
o
f
v
o
lcan
ic
s
atellite
im
ag
es.
3
.
4
.
1
.
Dee
p
co
nv
o
lutio
na
l ne
ura
l net
wo
rk
f
o
r
v
o
lca
nic a
c
t
iv
it
y
Vo
lcan
ic
ac
tiv
ity
(
DC
NNVA
)
is
a
cu
s
to
m
ized
ar
ch
itectu
r
e
d
ev
elo
p
e
d
o
n
ly
f
o
r
v
o
lcan
i
c
ac
tiv
ity
class
if
icatio
n
u
s
in
g
s
atellites
,
an
d
th
e
n
etwo
r
k
is
s
o
d
esig
n
e
d
th
at
it
m
ain
tain
s
a
b
ala
n
ce
b
etwe
en
co
m
p
u
tatio
n
ef
f
icien
cy
an
d
class
if
icatio
n
ac
cu
r
ac
y
,
an
d
th
er
ef
o
r
e
is
u
s
ab
le
f
o
r
n
ea
r
r
ea
l
-
tim
e
m
o
n
ito
r
in
g
.
I
t
is
a
co
m
b
in
atio
n
o
f
s
p
ac
e
f
ea
tu
r
e
s
,
ex
tr
ac
tin
g
co
n
v
o
lu
tio
n
an
d
p
o
o
lin
g
o
p
e
r
atio
n
s
,
an
d
f
u
ll
y
co
n
n
ec
ted
lay
er
s
u
tili
ze
d
at
th
e
tim
e
o
f
class
if
icatio
n
(
≈
1
.
1
9
M
t
r
ain
ab
le
p
a
r
a
m
eter
s
)
.
i)
C
o
n
v
o
lu
tio
n
al
la
y
er
s
:
th
e
co
n
v
o
lu
tio
n
al
o
p
e
r
atio
n
e
x
tr
ac
t
s
h
ier
ar
ch
ical
r
ep
r
esen
tatio
n
s
f
r
o
m
in
p
u
t
im
ag
es
b
y
ap
p
ly
in
g
lear
n
ab
le
k
er
n
els.
Ma
th
em
atica
lly
,
th
e
co
n
v
o
lu
tio
n
at
lay
e
r
ca
n
b
e
ex
p
r
ess
ed
as
p
r
esen
ted
in
(
1
)
.
,
(
)
=
∑
∑
+
,
+
(
−
1
)
−
1
=
0
−
1
=
0
.
,
(
)
+
(
)
(
1
)
W
h
er
e
(
−
1
)
r
ep
r
esen
ts
in
p
u
t
f
r
o
m
t
h
e
p
r
ev
io
u
s
lay
er
,
(
)
is
a
co
n
v
o
l
u
tio
n
al
k
e
r
n
el,
(
)
is
b
ias
ed
,
a
n
d
,
(
)
is
th
e
f
ea
tu
r
e
m
ap
at
p
o
s
itio
n
(
,
)
.
T
h
e
n
o
n
-
li
n
ea
r
ac
tiv
atio
n
f
u
n
ctio
n
R
eL
U
(
(
)
=
(
0
,
)
)
is
ap
p
lied
to
in
tr
o
d
u
ce
n
o
n
-
lin
ea
r
ity
.
ii)
Po
o
lin
g
lay
er
s
:
to
r
ed
u
ce
s
p
atial
d
im
en
s
io
n
s
wh
ile
p
r
eser
v
in
g
ess
en
tial
f
ea
tu
r
es
,
m
ax
p
o
o
lin
g
is
u
s
e
p
o
o
lin
g
o
p
er
atio
n
is
d
ef
in
e
d
in
(
2
)
.
,
=
(
,
)
∈
(
+
,
+
)
(
2
)
W
h
er
e
d
en
o
tes
th
e
p
o
o
lin
g
r
eg
io
n
o
p
er
atio
n
d
ec
r
ea
s
es
co
m
p
u
tatio
n
al
c
o
m
p
l
ex
ity
an
d
co
n
t
r
o
ls
o
v
er
f
itti
n
g
v
ia
in
tr
o
d
u
cin
g
tr
a
n
s
latio
n
al
in
v
ar
ian
ce
.
iii)
Fu
lly
co
n
n
ec
ted
lay
er
s
:
th
e
e
x
tr
ac
ted
f
ea
tu
r
es
ar
e
f
latten
ed
an
d
p
ass
ed
to
f
u
lly
co
n
n
ec
te
d
d
en
s
e
lay
er
s
f
o
r
class
if
icatio
n
.
T
h
e
tr
a
n
s
f
o
r
m
atio
n
is
g
iv
en
in
(
3
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
1
,
Ma
r
c
h
2
0
2
6
:
281
-
2
9
2
286
=
(
+
)
(
3
)
W
h
er
e
an
d
d
en
o
te
th
e
weig
h
ts
an
d
b
iases
o
f
d
e
n
s
e
lay
er
,
r
esp
ec
tiv
ely
,
an
d
r
ep
r
esen
ts
th
e
R
eL
U
o
r
s
o
f
tm
ax
a
ctiv
atio
n
f
u
n
ctio
n
,
d
e
p
en
d
in
g
o
n
lay
e
r
.
T
h
e
f
i
n
al
s
o
f
tm
ax
class
if
ier
p
r
o
d
u
ce
s
a
p
r
o
b
ab
ilit
y
o
v
er
two
o
u
tp
u
t c
lass
es
,
wh
er
e
=2
co
r
r
esp
o
n
d
s
to
th
e
n
u
m
b
er
o
f
class
es
p
r
esen
ted
in
(
4
)
.
(
=
/
)
=
(
)
∑
(
)
=
1
,
{
1
,
2
}
(
4
)
iv
)
Dr
o
p
o
u
t
an
d
o
p
tim
izatio
n
:
to
en
h
an
ce
g
en
e
r
aliza
tio
n
,
a
d
r
o
p
o
u
t
lay
er
with
r
ate
p
=0
.
5
was
in
co
r
p
o
r
ate
d
,
wh
ich
r
an
d
o
m
l
y
d
ea
ctiv
ates n
eu
r
o
n
s
d
u
r
i
n
g
tr
ain
in
g
.
T
h
e
m
o
d
el
is
o
p
tim
ized
u
s
in
g
th
e
Ad
am
o
p
tim
izer
,
wh
ich
ad
ap
tiv
ely
u
p
d
ates
lear
n
in
g
r
ates
f
o
r
ea
ch
p
a
r
am
eter
.
T
h
e
ca
teg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
f
u
n
ctio
n
is
em
p
lo
y
ed
,
as
p
r
esen
ted
in
(
5
)
.
ℒ
=
−
∑
∑
,
̂
,
=
1
=
1
(
5
)
W
h
er
e
,
is
th
e
g
r
o
u
n
d
tr
u
th
lab
el
an
d
̂
,
is
th
e
p
r
ed
icted
p
r
o
b
a
b
ilit
y
f
o
r
class
.
T
h
e
DC
NNVA
co
n
s
i
s
ts
o
f
f
iv
e
co
n
v
o
lu
tio
n
al
lay
er
s
(
6
4
-
2
5
6
f
ilter
s
)
,
ea
ch
o
f
wh
ich
is
f
o
llo
wed
b
y
m
ax
-
p
o
o
lin
g
f
o
c
u
s
in
g
o
n
d
im
en
s
io
n
r
ed
u
ctio
n
,
a
f
u
lly
co
n
n
ec
ted
2
5
6
-
n
eu
r
o
n
lay
er
,
a
d
r
o
p
o
u
t
lay
er
,
a
n
d
a
f
in
al
o
u
tp
u
t
class
if
icatio
n
s
o
f
t
m
ax
lay
er
.
I
t
h
as
r
o
u
g
h
ly
1
.
1
9
m
illi
o
n
tr
ain
ab
le
p
ar
am
ete
r
s
,
co
r
r
esp
o
n
d
in
g
to
a
lig
h
t
b
u
t
d
ee
p
a
r
ch
itectu
r
e
f
r
i
en
d
ly
to
s
ca
lab
ilit
y
an
d
ef
f
icien
cy
.
T
h
e
m
o
d
el
was
o
p
tim
iz
ed
u
s
in
g
th
e
Ad
am
o
p
tim
izer
with
a
lear
n
in
g
r
ate
o
f
0
.
0
0
1
,
an
d
tr
a
in
ed
with
th
e
ca
teg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
f
u
n
ctio
n
.
T
r
ain
in
g
was
co
n
d
u
cted
with
a
b
atch
s
ize
o
f
3
2
o
v
er
1
0
e
p
o
ch
s
,
u
s
in
g
th
e
Go
o
g
le
C
o
llab
o
r
ato
r
y
p
latf
o
r
m
with
a
n
NVI
DI
A
T
esla T
4
GPU
(
1
6
G
B
)
.
Fig
u
r
e
4
s
h
o
ws
th
e
DC
NNVA
ar
ch
itectu
r
e.
Fig
u
r
e
4
.
Ar
c
h
itectu
r
al
d
iag
r
a
m
o
f
th
e
p
r
o
p
o
s
ed
DC
NNVA
3
.
4
.
2
.
ResNet
5
0
I
n
tr
o
d
u
ce
d
b
y
He
et
a
l.
[
2
0
]
,
R
esNet5
0
in
tr
o
d
u
ce
s
r
esid
u
al
lear
n
in
g
th
r
o
u
g
h
id
en
tity
s
h
o
r
tcu
t
co
n
n
ec
tio
n
s
.
T
h
is
m
itig
ates
th
e
p
r
o
b
lem
o
f
v
an
is
h
in
g
g
r
ad
ien
ts
in
d
ee
p
er
n
etw
o
r
k
s
.
B
y
s
tack
in
g
co
n
v
o
l
u
tio
n
al
b
lo
ck
s
with
r
esid
u
al
lin
k
s
,
th
e
m
o
d
el
en
a
b
les
s
tab
le
tr
ain
in
g
an
d
im
p
r
o
v
ed
f
ea
tu
r
e
ex
tr
ac
tio
n
,
m
ak
in
g
it we
ll
-
s
u
ited
f
o
r
co
m
p
lex
im
ag
e
class
if
icatio
n
task
s
s
u
ch
as v
o
lcan
ic
ac
tiv
ity
r
ec
o
g
n
itio
n
.
3
.
4
.
3
.
NASNet
L
a
rg
e
NASNetL
ar
g
e
,
in
tr
o
d
u
ce
d
b
y
Z
o
p
h
et
a
l.
[
2
1
]
,
is
a
n
eu
r
a
l
ar
ch
itectu
r
e
s
ea
r
ch
(
NAS)
-
d
is
co
v
er
ed
ar
ch
itectu
r
e
th
at
is
aim
ed
a
t
o
p
tim
izin
g
n
etwo
r
k
s
tr
u
ctu
r
e
s
au
to
n
o
m
o
u
s
ly
f
o
r
h
ig
h
er
p
er
f
o
r
m
a
n
ce
.
I
t
is
a
m
o
d
u
lar
ar
ch
itectu
r
e
th
at
u
til
izes
r
ed
u
ctio
n
a
n
d
n
o
r
m
al
ce
lls
.
T
h
e
m
o
d
el
ca
n
ac
h
iev
e
s
ca
lab
le
d
ep
th
an
d
wid
th
with
ac
cu
r
ac
y
a
n
d
c
o
m
p
u
tatio
n
al
ef
f
icien
cy
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l
Sci
I
SS
N:
2252
-
8
8
1
4
DC
N
N
V
A
:
a
d
ee
p
co
n
vo
lu
tio
n
a
l n
eu
r
a
l n
etw
o
r
k
fo
r
vo
lca
n
ic
a
ctivity
cla
s
s
ifica
tio
n
…
(
Ya
s
i
r
Hu
s
s
ein
S
h
a
kir)
287
3
.
4
.
4
.
DenseNet
1
2
1
Den
s
eN
et1
2
1
,
i
n
tr
o
d
u
ce
d
b
y
Hu
an
g
et
a
l.
[
2
2
]
,
c
o
n
n
ec
ts
ea
ch
n
etwo
r
k
lay
er
with
all
o
th
e
r
n
etwo
r
k
lay
er
s
f
ee
d
-
f
o
r
war
d
ly
an
d
p
r
o
m
o
tes
f
ea
tu
r
e
r
e
u
s
e
an
d
ef
f
icien
t
g
r
ad
ien
t
f
l
o
w.
T
h
is
d
en
s
e
co
n
n
ec
tio
n
m
in
im
izes
r
ed
u
n
d
an
cy
an
d
im
p
r
o
v
es
lear
n
in
g
e
f
f
icien
cy
.
I
t
also
p
r
o
m
o
tes
th
e
ab
ilit
y
o
f
th
e
m
o
d
el
t
o
id
en
tif
y
s
u
b
tle
v
o
lcan
ic
f
ea
tu
r
es b
ased
o
n
s
atellite
d
ata.
3
.
4
.
5
.
M
o
bil
eNe
t
Mo
b
ileNet
,
p
r
esen
ted
b
y
H
o
war
d
et
a
l.
[
2
3
]
,
is
a
lig
h
tweig
h
t
C
NN
ar
ch
itectu
r
e
d
esig
n
e
d
s
p
ec
if
ically
f
o
r
m
o
b
ile
an
d
em
b
e
d
d
ed
v
is
io
n
ap
p
licatio
n
s
.
B
y
u
s
in
g
d
e
p
th
wis
e
s
ep
ar
ab
le
c
o
n
v
o
lu
tio
n
s
an
d
s
u
b
s
tan
tially
d
ec
r
ea
s
es
co
m
p
u
tatio
n
a
n
d
m
em
o
r
y
r
e
q
u
ir
em
e
n
ts
.
It
m
ain
t
ain
s
ac
cu
r
ac
y
,
m
a
k
in
g
it
a
li
k
ely
ca
n
d
id
ate
f
o
r
r
ea
l
-
tim
e
v
o
lcan
ic
ac
tiv
ity
m
o
n
ito
r
in
g
.
3
.
4
.
6
.
I
ncept
io
nV3
I
n
ce
p
tio
n
V3
,
p
r
o
p
o
s
ed
b
y
Szeg
ed
y
et
a
l.
[
2
4
]
,
im
p
r
o
v
es
t
h
e
ef
f
icien
c
y
o
f
C
NNs
u
s
in
g
f
ac
to
r
ize
d
co
n
v
o
l
u
tio
n
s
an
d
d
im
e
n
s
io
n
r
ed
u
ctio
n
m
eth
o
d
s
o
f
t
h
e
in
ce
p
tio
n
m
o
d
u
les.
I
ts
ar
c
h
itectu
r
e
en
ab
les
th
e
n
etwo
r
k
to
ac
h
iev
e
m
u
lti
-
s
ca
le
f
ea
tu
r
e
ca
p
tu
r
e
at
o
n
ce
.
I
t
also
en
h
an
ce
s
v
o
lcan
ic
im
ag
e
d
iv
er
s
ity
-
b
ased
r
ec
o
g
n
itio
n
p
er
f
o
r
m
a
n
ce
.
3
.
4
.
7
.
Xce
ptio
n
Xce
p
tio
n
,
s
tated
b
y
C
h
o
llet
[
2
5
]
,
ex
te
n
d
s
th
e
I
n
ce
p
tio
n
ar
c
h
itectu
r
e
v
ia
r
e
p
lacin
g
i
n
ce
p
tio
n
m
o
d
u
le
s
with
d
ep
th
wis
e
s
ep
ar
ab
le
co
n
v
o
lu
tio
n
s
.
T
h
is
s
tr
u
ctu
r
e
d
ec
o
u
p
les
s
p
atial
an
d
ch
a
n
n
el
-
wis
e
f
ilter
in
g
.
I
t
lea
d
s
to
im
p
r
o
v
e
d
r
ep
r
esen
tatio
n
al
ca
p
ac
ity
an
d
ef
f
icien
t tr
ain
in
g
.
3
.
4
.
8
.
VG
G
1
9
a
nd
VG
G
1
6
VGG
s
tr
u
ctu
r
ed
b
y
Simo
n
y
a
n
an
d
Z
is
s
er
m
an
[
2
6
]
(
VGG1
6
an
d
VGG1
9
)
ar
e
d
is
tin
g
u
is
h
e
d
b
y
th
eir
p
lain
n
ess
an
d
co
n
s
is
ten
t
ar
ch
it
ec
tu
r
e
an
d
ar
e
b
ased
o
n
s
u
cc
e
s
s
iv
e
co
n
v
o
lu
tio
n
al
lay
e
r
s
with
tin
y
(
3
×
3
)
f
ilter
s
.
Alth
o
u
g
h
d
ee
p
,
t
h
ese
m
o
d
els
ar
e
p
o
wer
f
u
l
in
all
im
a
g
e
cl
ass
if
icat
io
n
task
s
.
T
h
eir
s
im
p
le
ar
ch
itectu
r
e
also
m
ak
es tr
an
s
f
er
lear
n
i
n
g
f
ea
s
ib
le
f
o
r
d
etec
tin
g
v
o
lca
n
ic
ac
tiv
ity
.
3.
5
.
E
v
a
lua
t
i
o
n
perf
o
rma
nce
T
h
e
tr
ain
e
d
m
o
d
els
wer
e
ass
ess
ed
o
n
th
e
test
in
g
d
ataset
u
s
in
g
s
ev
er
al
p
er
f
o
r
m
an
ce
m
ea
s
u
r
es,
in
clu
d
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F
1
-
s
co
r
e
.
T
h
e
f
o
r
m
u
las
f
o
r
all
p
er
f
o
r
m
a
n
ce
v
al
u
es
ar
e
p
r
esen
ted
in
(
6
)
-
(
9
)
.
T
h
ese
v
alu
es we
r
e
co
m
p
u
ted
u
s
in
g
th
e
co
n
f
u
s
io
n
m
atr
ix
s
h
o
wn
in
T
a
b
le
4
.
=
+
+
+
+
(
6
)
=
+
(
7
)
=
+
(
8
)
1
−
=
2
×
×
+
(
9
)
T
ab
le
4
.
Stru
ctu
r
e
o
f
a
co
n
f
u
s
io
n
m
atr
ix
f
o
r
b
in
ar
y
class
if
icatio
n
A
c
t
u
a
l
p
o
si
t
i
v
e
A
c
t
u
a
l
n
e
g
a
t
i
v
e
P
r
e
d
i
c
t
e
d
p
o
si
t
i
v
e
Tr
u
e
p
o
si
t
i
v
e
(
TP)
F
a
l
se
p
o
si
t
i
v
e
(
F
P
)
P
r
e
d
i
c
t
e
d
n
e
g
a
t
i
v
e
F
a
l
se
n
e
g
a
t
i
v
e
(
F
N
)
Tr
u
e
n
e
g
a
t
i
v
e
(
TN
)
3
.
6
.
Deplo
y
m
ent
m
o
del
T
o
f
ac
ilit
ate
p
r
ac
tical
u
tili
za
tio
n
an
d
d
em
o
n
s
tr
ate
th
e
r
e
al
-
wo
r
ld
a
p
p
licab
ilit
y
o
f
th
e
p
r
o
p
o
s
ed
DC
NNV
A
m
o
d
el
,
a
s
tan
d
al
o
n
e
d
esk
to
p
ap
p
licatio
n
was
d
ev
elo
p
e
d
.
T
h
is
s
y
s
tem
to
o
l
b
r
id
g
es
th
e
g
ap
b
etwe
en
ex
p
er
im
en
tal
v
alid
ati
o
n
an
d
th
e
en
d
-
u
s
er
ap
p
licatio
n
.
I
t
p
r
o
v
id
es
an
in
tu
itiv
e
p
latf
o
r
m
f
o
r
v
o
lcan
ic
ac
tiv
ity
ass
ess
m
en
t.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
ex
p
er
im
en
tal
r
esu
lts
cle
ar
ly
in
d
icate
th
at
th
e
n
ew
DC
NNV
A
m
o
d
el
g
r
ea
tly
s
u
r
p
ass
ed
all
tr
an
s
f
er
lear
n
in
g
n
etwo
r
k
s
wi
th
a
m
ax
im
u
m
ac
c
u
r
ac
y
9
9
.
3
3
%
an
d
alm
o
s
t
p
er
f
ec
t
p
r
ec
i
s
io
n
1
0
0
%,
wh
ich
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
1
,
Ma
r
c
h
2
0
2
6
:
281
-
2
9
2
288
clea
r
ly
s
h
o
ws
th
at
it
ca
n
ac
cu
r
ately
id
en
tify
v
o
lcan
ic
ac
tiv
ity
.
I
ts
s
tr
o
n
g
ab
ilit
y
t
o
s
u
p
p
r
es
s
f
alse
n
eg
ativ
es
is
f
u
r
th
er
c
o
n
f
ir
m
ed
b
y
its
h
ig
h
r
ec
all
v
alu
e
9
8
.
6
7
%,
wh
ic
h
is
v
er
y
im
p
o
r
tan
t
in
ea
r
ly
war
n
in
g
s
r
eg
a
r
d
in
g
v
o
lcan
ic
h
az
ar
d
s
.
Ho
wev
er
,
we
ca
n
s
ee
t
h
at
R
esNet5
0
p
e
r
f
o
r
m
ed
p
o
o
r
ly
with
a
v
er
y
l
o
w
ac
cu
r
ac
y
lev
el
o
f
o
n
ly
6
8
%,
wh
ich
we
a
ttrib
u
t
e
to
th
e
f
ac
t
th
at
its
d
ee
p
er
r
esid
u
al
u
n
its
ar
e
n
o
t
o
p
tim
a
l
f
o
r
th
is
d
ataset.
Mo
b
ileNet,
Den
s
eNe
t1
2
1
,
an
d
I
n
ce
p
tio
n
V
3
p
er
f
o
r
m
e
d
eq
u
al
ly
well
(
9
5
-
9
6
% a
cc
u
r
ate)
,
b
u
t
f
ailed
to
r
ea
ch
t
h
e
ac
cu
r
ac
y
o
f
th
e
DC
NNVA
.
Oth
er
o
ld
er
d
esig
n
s
,
s
u
ch
as
V
GG1
6
an
d
V
GG1
9
,
p
er
f
o
r
m
e
d
lo
wer
in
ac
c
u
r
ac
y
lev
els co
m
p
ar
ed
t
o
n
ew
ar
c
h
itectu
r
e
d
esig
n
s
.
A
s
u
m
m
ar
y
o
f
th
e
r
esu
lts
is
p
r
o
v
id
ed
i
n
Fig
u
r
e
5
.
T
o
b
etter
illu
s
tr
ate
th
e
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
,
Fig
u
r
e
6
p
lo
ts
th
e
co
n
f
u
s
io
n
m
atr
i
ce
s
o
f
all
alg
o
r
ith
m
s
.
T
h
ese
co
n
tain
a
g
r
ap
h
ical
r
ep
r
esen
tatio
n
o
f
tr
u
e
p
o
s
itiv
es,
tr
u
e
n
eg
ativ
es,
f
alse
p
o
s
itiv
es
,
an
d
f
alse
n
eg
ativ
es
with
in
th
e
“
v
o
lcan
ic
ac
tiv
ity
”
an
d
“
n
o
ac
t
iv
ity
”
class
es.
An
aly
s
is
o
f
th
e
co
n
f
u
s
io
n
m
atr
ix
co
n
f
ir
m
s
t
h
e
s
u
p
e
r
io
r
ity
o
f
th
e
DC
NNVA
m
o
d
el
,
wh
ic
h
c
o
r
r
ec
tly
class
if
ied
alm
o
s
t
all
s
am
p
les
b
u
t
co
n
tain
e
d
a
n
eg
lig
ib
le
n
u
m
b
er
o
f
f
alse
n
eg
ativ
e
4
ca
s
es
o
f
v
o
lcan
ic
ac
tiv
ity
m
is
class
if
ied
as
n
o
a
ctiv
ity
.
C
o
n
tr
ar
y
to
th
is
,
n
u
m
er
o
u
s
m
is
class
if
icati
o
n
s
wer
e
witn
ess
ed
in
R
esNet5
0
,
p
r
im
ar
ily
th
e
a
b
s
en
ce
o
f
d
etec
tio
n
o
f
v
o
lca
n
ic
a
ctiv
ity
in
a
m
ajo
r
ity
o
f
ca
s
es.
Mo
b
ileNet,
Den
s
eNe
t1
2
1
,
an
d
I
n
ce
p
tio
n
V3
p
er
f
o
r
m
e
d
o
u
ts
tan
d
in
g
ly
b
u
t
with
a
s
lig
h
tly
h
ig
h
er
m
is
class
if
icat
io
n
r
ate
co
m
p
a
r
ed
to
DC
NNVA
.
T
h
e
g
r
ap
h
o
f
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
o
f
DC
NNVA
f
o
r
ten
e
p
o
ch
s
is
s
h
o
wn
in
Fig
u
r
e
7
.
Fro
m
th
e
g
r
a
p
h
,
th
e
ac
c
u
r
ac
y
o
f
th
e
m
o
d
el
in
c
r
ea
s
es
v
er
y
f
ast at
ea
ch
ep
o
ch
,
b
u
t slo
ws d
o
wn
f
r
o
m
ep
o
c
h
4
u
p
to
th
e
la
s
t e
p
o
ch
.
Fig
u
r
e
5
.
Acc
u
r
ac
y
c
o
m
p
ar
is
o
n
o
f
th
e
p
r
o
p
o
s
ed
DC
NNVA
m
o
d
el
ag
ain
s
t state
-
of
-
th
e
-
ar
t
Fig
u
r
e
6
.
C
o
n
f
u
s
io
n
m
atr
ices
co
m
p
ar
in
g
th
e
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
o
f
DC
NNVA
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l
Sci
I
SS
N:
2252
-
8
8
1
4
DC
N
N
V
A
:
a
d
ee
p
co
n
vo
lu
tio
n
a
l n
eu
r
a
l n
etw
o
r
k
fo
r
vo
lca
n
ic
a
ctivity
cla
s
s
ifica
tio
n
…
(
Ya
s
i
r
Hu
s
s
ein
S
h
a
kir)
289
Fig
u
r
e
7
.
DC
NNVA
tr
ain
in
g
a
n
d
v
alid
atio
n
p
er
f
o
r
m
a
n
ce
o
v
e
r
10
e
p
o
ch
s
T
h
e
s
u
p
er
i
o
r
p
er
f
o
r
m
an
ce
o
f
t
h
e
DC
NNVA
m
o
d
el
war
r
a
n
te
d
its
tr
an
s
itio
n
f
r
o
m
a
r
esear
ch
p
r
o
t
o
ty
p
e
to
a
p
r
ac
tical
to
o
l.
T
o
th
is
en
d
,
a
f
u
n
ctio
n
al
d
esk
to
p
ap
p
licatio
n
was
d
e
v
elo
p
e
d
a
n
d
d
ep
lo
y
e
d
.
T
h
is
ap
p
licatio
n
p
r
o
v
id
es
a
u
s
er
-
f
r
i
en
d
ly
in
ter
f
ac
e
th
at
allo
ws
en
d
-
u
s
er
s
,
s
u
ch
as
g
eo
lo
g
is
ts
o
r
m
o
n
ito
r
in
g
s
tatio
n
p
er
s
o
n
n
el,
to
p
er
f
o
r
m
r
ea
l
-
tim
e
v
o
lcan
ic
ac
tiv
ity
a
s
s
es
s
m
en
ts
.
T
h
e
d
ep
lo
y
m
en
t
s
u
cc
ess
f
u
lly
d
em
o
n
s
tr
ates
th
e
m
o
d
el'
s
o
p
er
atio
n
al
v
ia
b
ilit
y
.
As
s
h
o
wn
in
t
h
e
ap
p
licatio
n
in
ter
f
ac
e
Fig
u
r
e
8
,
u
s
er
s
ca
n
u
p
l
o
ad
s
atellite
im
ag
er
y
,
an
d
th
e
s
y
s
tem
r
etu
r
n
s
an
in
s
tan
tan
eo
u
s
class
if
ica
tio
n
(
"
ac
tiv
e
"
o
r
"
n
o
ac
tiv
e
")
ac
co
m
p
an
ied
b
y
a
co
n
f
id
en
ce
s
co
r
e.
T
h
e
in
ter
f
ac
e
clea
r
ly
d
is
p
lay
s
th
e
p
r
e
d
ictio
n
,
f
o
r
in
s
tan
ce
,
"p
r
ed
i
ctio
n
:
n
o
ac
tiv
e
|
co
n
f
id
en
ce
:
9
9
.
3
0
%"
(
Fig
u
r
e
8
(
a)
)
o
r
"p
r
ed
ictio
n
:
ac
tiv
e
|
co
n
f
id
en
ce
:
9
9
.
9
0
%"
(
Fig
u
r
e
8
(
b
)
)
,
p
r
o
v
i
d
in
g
tr
an
s
p
ar
en
t
an
d
im
m
ed
iate
r
e
s
u
lts
to
th
e
o
p
er
ato
r
.
C
r
u
cially
,
th
e
ap
p
licatio
n
in
co
r
p
o
r
ates
a
m
u
lti
-
m
o
d
al
aler
t
s
y
s
tem
th
at,
u
p
o
n
d
etec
tin
g
"a
ctiv
e"
v
o
lcan
ic
ac
tiv
ity
,
tr
ig
g
e
r
s
a
clea
r
,
s
y
n
th
esized
v
o
ice
war
n
in
g
:
"wa
r
n
in
g
!
v
o
lcan
ic
ac
tiv
ity
d
etec
ted
.
"
T
h
is
f
ea
tu
r
e
is
d
esig
n
e
d
to
ca
p
t
u
r
e
th
e
o
p
er
ato
r
'
s
atten
tio
n
im
m
ed
iately
,
wh
ic
h
is
p
ar
am
o
u
n
t
in
h
ig
h
-
s
tak
es
m
o
n
ito
r
in
g
en
v
ir
o
n
m
e
n
ts
.
T
h
e
s
u
cc
ess
f
u
l
in
teg
r
atio
n
o
f
th
e
h
ig
h
-
ac
c
u
r
ac
y
DC
NNV
A
m
o
d
el
in
to
th
is
d
ep
lo
y
ab
le
s
y
s
tem
u
n
d
er
s
co
r
e
s
an
d
also
r
ea
d
in
ess
f
o
r
u
s
e
in
q
u
asi
-
r
ea
l
-
tim
e
d
ec
is
io
n
-
s
u
p
p
o
r
t
s
ce
n
ar
io
s
,
ef
f
ec
tiv
ely
b
r
id
g
in
g
th
e
g
ap
b
etwe
en
th
e
o
r
etica
l
m
o
d
el
p
er
f
o
r
m
an
ce
an
d
p
r
ac
tical,
o
n
-
t
h
e
-
g
r
o
u
n
d
u
tili
ty
.
(
a)
(
b
)
Fig
u
r
e
8
.
DC
NNVA
ap
p
licatio
n
in
ter
f
ac
e
f
o
r
v
o
lcan
ic
ac
tiv
i
ty
class
if
icatio
n
of
(
a)
"
n
o
ac
ti
v
e"
(
b
)
"a
ctiv
e"
Fu
r
th
er
m
o
r
e
,
p
er
f
o
r
m
a
n
ce
an
d
p
r
ac
tical
im
p
lem
en
tatio
n
o
f
th
e
p
r
o
p
o
s
ed
DC
NNVA
m
o
d
el
ar
e
co
m
p
ar
ed
to
co
n
tem
p
o
r
ar
y
lit
er
atu
r
e
r
ev
iew
in
T
ab
le
5
.
T
h
e
an
aly
s
is
r
ev
ea
ls
th
at
,
h
o
we
v
er
v
ar
i
o
u
s
s
tu
d
ies
h
av
e
ac
h
iev
ed
h
i
g
h
ac
cu
r
ac
y
,
o
u
r
wo
r
k
d
is
tin
g
u
is
h
es
two
cr
itica
l
ar
ea
s
:
i
)
a
ch
iev
i
n
g
th
e
b
est
o
v
er
all
p
er
f
o
r
m
an
ce
ac
r
o
s
s
m
u
ltip
le
m
etr
ics
an
d
ii
)
s
u
cc
ess
f
u
lly
b
r
id
g
in
g
th
e
g
ap
to
a
p
r
ac
tical
d
ep
lo
y
ab
le
to
o
l.
T
h
is
p
r
ac
tical
to
o
l
f
ea
tu
r
es
a
u
s
er
-
f
r
ien
d
ly
GUI
f
o
r
r
ea
l
-
tim
e
an
al
y
s
is
an
d
in
co
r
p
o
r
ates
a
u
n
iq
u
e
m
u
lti
-
m
o
d
al
a
ler
t
s
y
s
tem
th
at
p
r
o
v
id
es
im
m
ed
iate
au
d
ito
r
y
war
n
in
g
s
u
p
o
n
d
et
ec
tio
n
o
f
v
o
lcan
ic
ac
tiv
ity
.
T
h
is
co
m
b
in
atio
n
o
f
s
tate
-
of
-
th
e
-
ar
t
p
r
ed
ictiv
e
p
er
f
o
r
m
an
ce
a
n
d
a
f
u
n
ctio
n
al
atte
n
tio
n
-
g
r
ab
b
in
g
d
e
p
lo
y
m
e
n
t
p
l
atf
o
r
m
r
e
p
r
esen
ts
a
s
ig
n
if
ican
t c
o
n
tr
ib
u
tio
n
to
t
h
e
f
iel
d
o
f
o
p
er
atio
n
al
v
o
lcan
ic
h
az
ar
d
m
o
n
ito
r
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
1
,
Ma
r
c
h
2
0
2
6
:
281
-
2
9
2
290
T
ab
le
5
.
C
o
m
p
a
r
ativ
e
an
aly
s
is
o
f
v
o
lcan
ic
ac
tiv
ity
d
etec
tio
n
m
eth
o
d
s
Ref
e
r
e
n
c
e
M
e
t
h
o
d
o
l
o
g
y
P
e
r
f
o
r
ma
n
c
e
(
%)
D
e
p
l
o
y
me
n
t
A
l
e
r
t
s
y
s
t
e
m
C
o
r
r
a
d
i
n
o
e
t
a
l
.
[
1
2
]
C
N
N
(
U
N
ET)
A
c
c
u
r
a
c
y
=
9
3
N
o
t
r
e
p
o
r
t
e
d
N
o
t
r
e
p
o
r
t
e
d
S
h
u
l
t
z
[
1
3
]
C
N
N
(
H
o
t
LI
N
K
)
A
c
c
u
r
a
c
y
=
9
8
N
o
t
r
e
p
o
r
t
e
d
N
o
t
r
e
p
o
r
t
e
d
O
ñ
a
t
e
e
t
a
l
.
[
1
4
]
A
u
d
i
o
f
e
a
t
u
r
e
s +
ma
n
i
f
o
l
d
l
e
a
r
n
i
n
g
A
c
c
u
r
a
c
y
=
9
4
.
4
4
t
o
9
5
.
4
5
N
o
t
r
e
p
o
r
t
e
d
N
o
t
r
e
p
o
r
t
e
d
N
u
n
n
a
r
i
a
n
d
C
a
l
v
a
r
i
[
1
5
]
C
o
m
p
a
r
i
so
n
o
f
8
C
N
N
s
A
c
c
u
r
a
c
y
=
9
4
.
0
7
N
o
t
r
e
p
o
r
t
e
d
N
o
t
r
e
p
o
r
t
e
d
C
h
e
n
e
t
a
l
.
[
1
6
]
TV
G
G
A
c
c
u
r
a
c
y
=
9
9
.
1
8
Re
c
a
l
l
=
9
9
.
1
7
N
o
t
r
e
p
o
r
t
e
d
N
o
t
r
e
p
o
r
t
e
d
M
o
h
a
n
e
t
a
l
.
[
1
7
]
LR
X
+
C
N
N
A
c
c
u
r
a
c
y
=
9
0
.
3
0
F1
-
sc
o
r
e
=
8
8
.
4
0
N
o
t
r
e
p
o
r
t
e
d
N
o
t
r
e
p
o
r
t
e
d
H
u
e
r
t
a
s
e
t
a
l
.
[
1
8
]
V
G
G
1
6
a
n
d
I
n
c
e
p
t
i
o
n
C
N
N
A
c
c
u
r
a
c
y
=
9
3
Pr
e
c
i
s
i
o
n
=
9
3
N
o
t
r
e
p
o
r
t
e
d
N
o
t
r
e
p
o
r
t
e
d
O
u
r
w
o
r
k
P
r
o
p
o
se
d
D
C
N
N
V
A
A
c
c
u
r
a
c
y
=
9
9
.
3
3
Pr
e
c
i
s
i
o
n
=
1
0
0
.
Re
c
a
l
l
=
9
8
.
6
7
F1
-
sc
o
r
e
=
9
9
.
3
3
Y
e
s (
f
u
n
c
t
i
o
n
a
l
d
e
s
k
t
o
p
a
p
p
l
i
c
a
t
i
o
n
)
Y
e
s (
t
e
x
t
-
to
-
sp
e
e
c
h
a
u
d
i
o
a
l
a
r
m
)
5.
CO
NCLU
SI
O
N
I
n
th
is
s
tu
d
y
,
a
d
ev
el
o
p
ed
DC
NNV
A
class
if
icatio
n
was
r
ig
o
r
o
u
s
ly
v
alid
ated
a
n
d
s
u
cc
ess
f
u
lly
d
ep
lo
y
e
d
.
T
h
e
m
o
d
el
was
ev
alu
ated
ag
ain
s
t
eig
h
t
s
tate
-
of
-
th
e
-
ar
t
tr
an
s
f
er
lear
n
in
g
ar
c
h
i
tectu
r
es
,
in
clu
d
in
g
R
esNet5
0
,
NAS
NetL
ar
g
e,
Den
s
eNe
t1
2
1
,
Mo
b
ileNet,
I
n
ce
p
tio
n
V3
,
Xce
p
tio
n
,
VGG1
6
,
an
d
VGG
1
9
.
T
h
e
ex
p
er
im
en
tal
r
esu
lts
d
em
o
n
s
tr
ate
th
at
p
r
o
p
o
s
ed
DC
NNVA
m
o
d
el
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
ed
all
m
o
d
e
l
n
etwo
r
k
s
ac
r
o
s
s
all
ev
al
u
atio
n
m
etr
ics
,
ac
h
iev
in
g
ac
cu
r
ac
y
(
9
9
.
3
3
%),
p
r
ec
is
io
n
(
1
0
0
%),
r
e
ca
ll
(
9
8
.
6
7
%),
a
n
d
F1
-
s
co
r
e
(
9
9
.
3
3
%).
T
h
e
co
m
p
r
eh
en
s
iv
e
an
aly
s
i
s
s
u
p
p
o
r
ted
b
y
co
n
f
u
s
io
n
m
atr
ices
a
n
d
tr
ai
n
in
g
g
r
a
p
h
s
co
n
f
ir
m
s
t
h
e
m
o
d
el'
s
r
o
b
u
s
t
ca
p
ab
ilit
y
to
m
in
im
ize
b
o
t
h
f
alse
p
o
s
itiv
es
an
d
f
alse
n
eg
ativ
es
,
with
a
p
ar
ticu
lar
ly
s
tr
o
n
g
p
er
f
o
r
m
an
ce
in
r
ed
u
cin
g
f
alse
n
eg
ativ
es
,
a
cr
itical
r
eq
u
ir
em
en
t
f
o
r
ea
r
ly
war
n
in
g
s
y
s
tem
s
in
v
o
lcan
ic
h
az
ar
d
m
o
n
ito
r
in
g
.
B
ey
o
n
d
th
eo
r
etica
l
p
er
f
o
r
m
an
ce
,
th
is
r
esear
ch
m
ak
es
a
s
u
b
s
tan
tial
p
r
ac
tical
co
n
tr
ib
u
tio
n
th
r
o
u
g
h
th
e
s
u
cc
ess
f
u
l
d
ev
elo
p
m
en
t
an
d
d
ep
lo
y
m
en
t
o
f
an
o
p
er
atio
n
al
d
esk
to
p
ap
p
licatio
n
,
a
n
d
im
p
lem
en
tatio
n
r
ep
r
esen
ts
a
s
i
g
n
if
i
ca
n
t
a
d
v
an
ce
m
e
n
t
b
e
y
o
n
d
cu
r
r
en
t
s
tate
-
of
-
t
h
e
-
ar
t
a
p
p
r
o
ac
h
es
,
b
r
i
d
g
in
g
th
e
g
ap
b
etwe
en
ex
p
er
im
e
n
tal
m
o
d
els
an
d
p
r
ac
tical
u
tili
ty
.
T
h
e
ap
p
licatio
n
f
ea
tu
r
es
an
in
tu
itiv
e
g
r
ap
h
ical
in
ter
f
ac
e
f
o
r
r
ea
l
-
tim
e
m
o
n
it
o
r
in
g
a
n
d
in
c
o
r
p
o
r
ates
a
p
io
n
ee
r
in
g
m
u
ltimo
d
al
aler
t
s
y
s
tem
th
at
p
r
o
v
id
es
im
m
ed
iate
au
d
ito
r
y
war
n
i
n
g
s
u
p
o
n
d
et
ec
tio
n
o
f
v
o
lcan
ic
a
ctiv
ity
,
a
f
ea
tu
r
e
ab
s
en
ts
in
c
o
m
p
ar
ab
le
s
tu
d
ies.
Fu
tu
r
e
wo
r
k
will
aim
to
ex
t
en
d
th
e
c
u
r
r
e
n
t
ar
ch
itectu
r
e
to
s
u
p
p
o
r
t
m
u
lti
-
class
class
if
icatio
n
o
f
v
o
lcan
ic
ac
tiv
ity
,
allo
win
g
f
o
r
m
o
r
e
n
u
an
ce
d
r
ec
o
g
n
itio
n
o
f
d
i
f
f
e
r
en
t
er
u
p
tio
n
ty
p
es
,
an
d
th
e
f
r
am
ewo
r
k
will
b
e
en
h
an
ce
d
t
h
r
o
u
g
h
t
h
e
in
teg
r
at
io
n
o
f
m
u
ltimo
d
al
d
ata,
co
m
b
in
in
g
in
f
o
r
m
atio
n
f
r
o
m
s
o
u
r
ce
s
s
u
ch
as
s
atellite
s
an
d
g
r
o
u
n
d
-
b
ased
s
eismic
s
en
s
o
r
s
.
Fin
ally
,
ef
f
o
r
ts
will
f
o
cu
s
o
n
d
e
p
lo
y
in
g
th
is
s
y
s
tem
in
o
p
er
atio
n
al
m
o
n
ito
r
in
g
n
etwo
r
k
s
to
s
tr
en
g
th
en
ea
r
ly
war
n
in
g
s
y
s
tem
s
an
d
im
p
r
o
v
e
d
is
aster
r
esp
o
n
s
e
s
tr
ateg
ies.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT
)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Yasir
Hu
s
s
ein
Sh
ak
ir
✓
✓
✓
✓
✓
✓
✓
✓
✓
R
ee
m
Ali M
u
tlag
✓
✓
✓
✓
✓
✓
E
s
h
aq
Aziz
Awa
d
h
AL
Ma
n
d
h
ar
i
✓
✓
✓
✓
✓
✓
✓
Mo
h
am
ed
Sh
a
b
b
ir
Ab
d
u
ln
ab
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.