I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
d
van
c
e
s
i
n
A
p
p
li
e
d
S
c
ie
n
c
e
s
(
I
JA
A
S
)
V
ol
.
15
, N
o.
1
,
M
a
r
c
h
20
26
, pp.
281
~
292
I
S
S
N
:
2252
-
8814
,
D
O
I
:
10.11591/
ij
a
a
s
.
v15.
i
1
.
pp
281
-
292
281
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
aas
.i
ae
s
c
or
e
.c
om
D
C
N
N
V
A
:
a d
e
e
p
c
on
vol
u
t
i
on
al
n
e
u
r
al
n
e
t
w
or
k
f
or
vol
c
an
i
c
ac
t
i
vi
t
y c
l
ass
i
f
i
c
at
i
on
u
si
n
g sat
e
l
l
i
t
e
i
m
age
r
y
Y
as
ir
H
u
s
s
e
in
S
h
ak
ir
1
, R
e
e
m
A
li
M
u
t
la
g
1
,
E
s
h
aq
A
z
iz
A
w
a
d
h
A
L
M
an
d
h
ar
i
2
,
M
oh
am
e
d
S
h
ab
b
ir
A
b
d
u
ln
ab
i
3
1
C
ol
l
e
ge
of
G
r
a
dua
t
e
S
t
udi
e
s
, U
ni
ve
r
s
i
t
i
T
e
na
ga
N
a
s
i
ona
l
, K
a
j
a
ng, M
a
l
a
ys
i
a
2
G
r
a
dua
t
e
S
c
hool
of
T
e
c
hnol
ogy
,
A
s
i
a
P
a
c
i
f
i
c
U
ni
ve
r
s
i
t
y of
T
e
c
hnol
ogy a
nd I
nnova
t
i
on
,
K
ua
l
a
L
um
pur
,
M
a
l
a
y
s
i
a
3
S
c
hool
of
T
e
c
hnol
ogy, A
s
i
a
P
a
c
i
f
i
c
U
ni
ve
r
s
i
t
y of
T
e
c
hnol
ogy a
nd I
nnova
t
i
on,
K
ua
l
a
L
um
pur
, M
a
l
a
y
s
i
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
S
e
p 4, 2025
R
e
vi
s
e
d
D
e
c
23, 2025
A
c
c
e
pt
e
d
J
a
n 1, 2026
Monitoring
and
classifying
volcanic
activity
are
a
critical
task
for
disaster
risk
reduction
and
hazard
management.
Recent
discoveries
i
n
m
achine
learning
and
deep
learning
have
proved
excellen
t
satellite
image
classi
fi
cation
and
volcanic
anomaly
identi
fi
cation
capabilities,
y
et
the
majority
of
existing
methods
suffer
from
small
datasets,
particula
rly
on
solitary
data
modalities
or
particular
cases
,
merely
as
examples.
I
n
this
research
work,
we
put
forward
develop
deep
convoluti
onal
neural
n
etwork
for
volcanic
activity
(DCNNVA)
classi
fi
cation
specifically
design
ed
for
satellite
imagery
on
volcanic
activity.
We
rigorously
bench
marked
DCNNVA
model'
s
strength
against
a
total
of
eight
state
-
of
-
the
-
art
t
ransfer
learning
models
:
ResNet50
,
NASNetLa
rge,
DenseN
et121,
Mob
ileNet,
InceptionV3,
Xception,
VGG19,
and
VGG16.
Comparativ
e
experi
mental
results
show
that
proposed
DCNNVA
framework
'
s
overall
perfor
mance
si
gnificantly surpasses its competitors with
an accurac
y of
99.33%, pr
ecision
of
100%,
recall
of
98.67%,
and
F1
-
score
of
99.33%,
significantly
beating
existin
g
state
-
of
-
the
-
art
methods.
Also,
we
create
a
deployable
gr
aphical
user
interface
(GUI)
system
that
is
capable
of
real
-
time
monitori
ng
on
volcanic
activity
and
generate
s
multi
-
modal
alert
processing
that
can
make
this
research
directly
applicable
for
practical
use
on
disaster
manage
ment
as
well
as
i
n
early
warning
system
s
.
This
research
contributes
a
sc
a
lable,
strong,
as
well
as
practical
solution
towards
volcanic
ha
zard
identi
fi
ca
tion
as
well
as
a
baseline
system
toward
developing
future
multi
-
modal
as
well
as
real
-
time geohaz
ard trac
king system fr
ameworks.
K
e
y
w
o
r
d
s
:
D
C
N
N
V
A
D
e
e
p l
e
a
r
ni
ng
M
ode
l
de
pl
oym
e
nt
M
ul
ti
-
m
oda
l
a
le
r
t
s
ys
te
m
S
a
te
ll
it
e
da
ta
c
la
s
s
if
ic
a
ti
on
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
M
oha
m
e
d S
ha
bbi
r
A
bdul
na
bi
S
c
hool
of
T
e
c
hnol
ogy, As
ia
P
a
c
if
ic
U
ni
ve
r
s
it
y of
T
e
c
hnol
ogy a
nd I
nnova
ti
on
K
ua
la
L
um
pur
57000, M
a
la
ys
ia
E
m
a
il
:
m
oha
m
e
d.s
ha
bbi
r
@
a
pu.e
du.my
1.
I
N
T
R
O
D
U
C
T
I
O
N
M
oni
to
r
in
g
a
nd
c
la
s
s
if
yi
ng
vol
c
a
ni
c
e
v
e
nt
s
a
r
e
w
id
e
ly
r
e
c
og
ni
z
e
d
a
s
one
of
th
e
m
o
s
t
c
r
it
ic
a
l
ye
t
c
ha
ll
e
ngi
ng
r
e
s
pon
s
ib
il
it
ie
s
f
a
c
e
d
vi
a
obs
e
r
va
to
r
ie
s
a
nd
ge
o
s
c
ie
nt
if
ic
in
s
ti
tu
ti
ons
a
r
ound
th
e
w
or
ld
.
T
h
e
vol
c
a
ni
c
a
c
ti
vi
ty
pr
e
s
e
nt
s
not
onl
y
a
s
ig
ni
f
ic
a
nt
na
tu
r
a
l
ha
z
a
r
d
to
hum
a
n
popula
ti
ons
a
nd
in
f
r
a
s
tr
uc
tu
r
e
but
a
ls
o
a
f
unda
m
e
nt
a
l
a
r
e
a
of
s
tu
dy
f
or
unde
r
s
t
a
ndi
ng
e
a
r
th
dyna
m
ic
s
ys
te
m
s
. T
r
a
di
ti
ona
ll
y,
th
e
obs
e
r
va
ti
on
a
nd
a
na
ly
s
is
of
vol
c
a
ni
c
a
c
ti
vi
ty
ha
ve
r
e
li
e
d
he
a
vi
ly
on
gr
oun
d
-
ba
s
e
d
m
oni
to
r
in
g
s
ys
te
m
s
,
m
a
nua
l
im
a
ge
in
te
r
pr
e
ta
ti
on,
a
nd
c
onve
nt
io
na
l
s
ta
ti
s
ti
c
a
l
m
ode
ls
.
H
ow
e
ve
r
,
th
e
s
e
a
ppr
oa
c
he
s
w
e
r
e
of
te
n
li
m
it
e
d
by
te
m
por
a
l
de
la
ys
r
e
s
tr
ic
te
d
c
ove
r
a
ge
,
a
nd
th
e
hi
gh
c
os
ts
a
s
s
oc
ia
te
d
w
it
h
on
-
s
it
e
m
oni
to
r
in
g.
T
he
r
a
pi
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
. 15, No. 1, M
a
r
c
h 2026
:
281
-
292
282
a
dva
nc
e
m
e
nt
of
r
e
m
ot
e
s
e
n
s
in
g
te
c
hnol
ogi
e
s
,
e
s
pe
c
i
a
ll
y
th
e
gr
ow
in
g
a
va
il
a
bi
li
ty
of
hi
gh
-
r
e
s
ol
ut
io
n
m
ul
ti
s
pe
c
tr
a
l
a
nd
th
e
r
m
a
l
s
a
te
ll
it
e
im
a
ge
r
y
,
ha
s
gr
e
a
tl
y
e
xpa
nde
d
th
e
a
bi
li
ty
of
r
e
s
e
a
r
c
he
r
s
to
m
oni
to
r
vol
c
a
ni
c
a
c
ti
vi
ty
a
t
bot
h
lo
c
a
l
a
nd
gl
ob
a
l
s
c
a
le
s
.
S
uppor
te
d
b
y
in
c
r
e
a
s
e
d
c
om
put
a
ti
ona
l
pow
e
r
,
th
e
s
e
to
ol
s
now
a
ll
ow
l
a
r
ge
-
s
c
a
le
da
ta
pr
oc
e
s
s
in
g a
nd r
e
a
l
-
ti
m
e
m
oni
to
r
in
g
c
a
pa
bi
li
ti
e
s
t
ha
t
w
e
r
e
pr
e
vi
ous
ly
una
tt
a
in
a
bl
e
us
in
g
tr
a
di
ti
ona
l
a
ppr
oa
c
he
s
[
1]
.
A
t
th
e
s
a
m
e
ti
m
e
,
a
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
ha
s
e
m
e
r
ge
d
a
s
a
tr
a
ns
f
or
m
a
ti
ve
a
ppr
oa
c
h
f
or
ha
ndl
in
g
c
om
pl
e
x
hi
gh
-
di
m
e
ns
io
na
l
d
a
ta
s
e
ts
,
a
nd
w
it
h
m
a
c
hi
ne
le
a
r
ni
ng
a
nd
de
e
p
le
a
r
ni
ng
te
c
hni
que
s
ha
ve
s
how
n
e
xc
e
pt
io
na
l
e
f
f
e
c
ti
ve
ne
s
s
in
s
ol
vi
n
g
pr
obl
e
m
s
r
e
la
te
d
to
im
a
ge
c
la
s
s
if
ic
a
ti
on
,
a
nom
a
ly
de
te
c
ti
on, a
nd pr
e
di
c
ti
ve
m
ode
li
ng
[
2]
, [
3]
.
A
m
a
jo
r
r
e
s
e
a
r
c
h
f
oc
u
s
w
a
s
th
e
c
la
s
s
if
ic
a
ti
on
of
s
a
t
e
ll
it
e
im
a
ge
r
y
,
w
hi
c
h
pl
a
y
s
a
c
r
uc
ia
l
r
ol
e
in
de
te
c
ti
ng
s
ubt
le
th
e
r
m
a
l
a
nom
a
li
e
s
,
ga
s
e
m
is
s
io
ns
,
l
a
nd
c
ove
r
s
hi
f
ts
,
a
nd
ot
he
r
in
di
c
a
to
r
s
of
vol
c
a
ni
c
a
c
ti
vi
ty
.
R
e
c
e
nt
w
or
k
ha
s
s
how
n
th
a
t
im
m
une
-
in
s
pi
r
e
d
m
a
c
hi
ne
l
e
a
r
ni
ng
a
lg
or
it
hm
s
a
nd
ot
he
r
m
ode
ls
c
a
n
s
ig
ni
f
ic
a
nt
ly
boos
t
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
e
ve
n
w
h
e
n
da
ta
s
e
ts
a
r
e
noi
s
y
or
li
m
it
e
d
[
4]
.
A
m
ong
de
e
p
le
a
r
ni
ng
m
ode
ls
,
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
(
C
N
N
s
)
s
ta
nd
out
a
s
one
of
th
e
m
os
t
e
f
f
e
c
ti
ve
te
c
hni
que
s
f
or
r
e
m
ot
e
s
e
ns
in
g a
ppl
ic
a
ti
ons
.
C
N
N
s
ha
ve
a
uni
que
c
a
pa
bi
li
ty
to
a
ut
om
a
ti
c
a
ll
y
e
xt
r
a
c
t
hi
e
r
a
r
c
hi
c
a
l
a
nd
s
p
a
ti
a
l
f
e
a
tu
r
e
s
f
r
om
im
a
ge
s
,
w
hi
c
h
a
ll
ow
s
th
e
m
to
s
ur
pa
s
s
tr
a
di
ti
ona
l
m
e
t
hods
th
a
t
r
e
ly
on
ha
nd
c
r
a
f
te
d
f
e
a
tu
r
e
s
[
5]
.
T
hi
s
s
tr
e
ngt
h
is
e
s
pe
c
ia
ll
y
im
por
ta
nt
in
vol
c
a
ni
c
m
oni
to
r
in
g,
w
he
r
e
de
te
c
ti
ng
s
ubt
le
s
p
a
ti
ot
e
m
por
a
l
pa
tt
e
r
ns
c
a
n
be
c
r
it
ic
a
l
f
or
id
e
nt
if
yi
ng
e
a
r
ly
s
ig
ns
of
a
c
ti
vi
ty
.
I
n
a
ddi
ti
on,
th
e
us
e
of
tr
a
ns
f
e
r
le
a
r
ni
ng
ha
s
f
ur
th
e
r
im
pr
ove
d C
N
N
pe
r
f
or
m
a
nc
e
. M
ode
ls
t
ha
t
ha
ve
be
e
n pr
e
-
tr
a
in
e
d on la
r
ge
-
s
c
a
le
da
ta
s
e
ts
s
uc
h
a
s
I
m
a
ge
N
e
t
c
a
n
be
f
in
e
-
tu
ne
d
f
or
vo
lc
a
ni
c
im
a
ge
c
la
s
s
if
ic
a
ti
on.
T
hi
s
a
ppr
oa
c
h
s
ig
ni
f
ic
a
nt
ly
r
e
duc
e
s
th
e
r
e
qui
r
e
m
e
nt
f
or
la
r
ge
a
m
ount
s
of
doma
in
-
s
pe
c
if
ic
t
r
a
in
in
g
da
ta
w
hi
le
s
im
ul
ta
ne
ous
ly
boos
ti
ng c
la
s
s
if
ic
a
ti
on a
c
c
ur
a
c
y
[
6]
, [
7]
.
S
uc
h
a
ppr
oa
c
he
s
a
r
e
pa
r
ti
c
ul
a
r
ly
va
lu
a
bl
e
in
vol
c
a
ni
c
m
oni
to
r
in
g
,
w
he
r
e
a
nnot
a
te
d
da
ta
s
e
ts
a
r
e
s
c
a
r
c
e
a
nd
im
ba
la
nc
e
d.
D
e
s
pi
te
s
ig
ni
f
ic
a
nt
pr
ogr
e
s
s
,
e
xi
s
ti
ng
m
e
th
ods
f
a
c
e
c
ha
ll
e
nge
s
r
e
la
te
d
to
s
c
a
la
bi
li
ty
,
ge
ne
r
a
li
z
a
ti
on
a
c
r
os
s
di
f
f
e
r
e
nt
vol
c
a
noe
s
,
a
nd
th
e
in
te
gr
a
ti
on
of
m
ul
ti
m
oda
l
s
a
te
ll
it
e
da
ta
.
A
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
m
e
th
ods
a
nd
pi
x
e
l
-
ba
s
e
d
c
la
s
s
if
ie
r
s
ha
ve
s
how
n
pot
e
nt
ia
l
,
but
th
e
ir
a
ppl
ic
a
ti
on
to
vol
c
a
ni
c
a
c
ti
vi
ty
r
e
m
a
in
s
l
im
it
e
d.
T
he
c
ont
r
ib
ut
io
ns
a
r
e
t
hus
c
on
c
lu
de
d t
o be
a
s
f
ol
lo
w
s
:
i)
I
nt
r
oduc
e
s
a
de
e
p
le
a
r
ni
ng
-
ba
s
e
d
f
r
a
m
e
w
or
k f
or
vol
c
a
ni
c
a
c
ti
vi
ty
c
la
s
s
if
ic
a
ti
on us
in
g a
c
u
s
to
m
-
de
s
ig
ne
d
de
ve
lo
p de
e
p c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k f
or
vol
c
a
ni
c
a
c
ti
vi
t
y (
D
C
N
N
V
A
)
a
r
c
hi
te
c
tu
r
e
.
ii)
C
om
pr
e
he
ns
iv
e
m
ode
l
e
va
lu
a
ti
on
vi
a
e
ig
ht
s
ta
te
-
of
-
th
e
-
a
r
t
tr
a
n
s
f
e
r
le
a
r
ni
ng
m
ode
ls
,
s
uc
h
a
s
R
e
s
N
e
t5
0,
N
A
S
N
e
tL
a
r
ge
, D
e
ns
e
N
e
t1
21,
M
obi
le
N
e
t,
I
nc
e
pt
io
nV
3, X
c
e
pt
io
n, V
G
G
19, a
nd V
G
G
16.
iii)
E
nha
nc
e
d
m
ode
l
r
obus
tn
e
s
s
vi
a
da
ta
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
a
nd
e
xt
e
ns
iv
e
e
xpe
r
im
e
nt
a
l
va
li
da
ti
on
us
e
s
m
ul
ti
pl
e
pe
r
f
or
m
a
nc
e
m
e
tr
ic
s
.
iv
)
D
e
pl
oya
bl
e
gr
a
phi
c
a
l
us
e
r
in
te
r
f
a
c
e
(
G
U
I
)
s
ys
te
m
th
a
t
pr
ovi
de
s
r
e
a
l
-
ti
m
e
vol
c
a
ni
c
a
c
ti
vi
ty
m
oni
to
r
in
g
w
it
h m
ul
ti
-
m
oda
l
a
le
r
t
c
a
pa
bi
li
ti
e
s
,
m
a
ki
ng t
he
r
e
s
e
a
r
c
h pr
a
c
ti
c
a
ll
y a
ppl
ic
a
bl
e
f
or
di
s
a
s
te
r
m
a
na
ge
m
e
nt
.
T
he
s
tr
uc
tu
r
e
of
th
is
p
a
pe
r
is
out
li
ne
d
a
s
f
ol
lo
w
s
:
s
e
c
ti
on
2
pr
e
s
e
nt
s
th
e
r
e
la
t
e
d
w
or
k,
r
e
vi
e
w
in
g
pr
e
vi
ous
s
tu
di
e
s
a
nd
r
e
c
e
nt
de
ve
lo
pm
e
nt
s
r
e
le
va
nt
to
th
is
r
e
s
e
a
r
c
h
a
r
e
a
.
S
e
c
ti
on
3
th
e
m
a
te
r
ia
ls
a
nd
m
e
th
od
s
,
pr
ovi
d
es
de
ta
il
s
of
th
e
a
r
c
hi
te
c
tu
r
e
,
m
e
th
odol
ogy,
da
ta
s
e
ts
,
pr
e
-
pr
oc
e
s
s
in
g
,
di
vi
di
ng
da
ta
,
in
ve
s
ti
ga
te
d
m
ode
ls
,
a
nd
e
v
a
lu
a
ti
on
m
e
tr
ic
s
.
S
e
c
ti
on
4
r
e
s
ul
ts
a
nd
di
s
c
u
s
s
i
on
w
it
h
c
om
pa
r
is
ons
to
r
e
la
t
e
d
s
tu
di
e
s
in
th
e
li
te
r
a
tu
r
e
. F
in
a
ll
y,
s
e
c
t
io
n
5
pr
e
s
e
nt
s
t
he
c
on
c
lu
s
io
ns
.
2.
R
E
L
A
T
E
D
WORK
R
e
c
e
nt
r
e
s
e
a
r
c
h
ha
s
e
xpl
or
e
d
th
e
us
e
of
m
a
c
hi
ne
le
a
r
ni
ng
a
nd
de
e
p
le
a
r
ni
ng
f
or
s
a
te
ll
it
e
im
a
ge
c
la
s
s
if
ic
a
ti
on a
nd
vol
c
a
ni
c
a
c
ti
vi
ty
m
oni
to
r
in
g.
E
a
r
ly
s
tu
di
e
s
a
p
pl
ie
d
pi
xe
l
-
ba
s
e
d
m
a
c
hi
ne
le
a
r
ni
ng
;
s
im
il
a
r
ly
,
E
br
a
hi
m
y
a
nd
Z
ha
ng
[
8]
in
c
r
e
a
s
e
d
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
by
a
ppl
yi
ng
va
r
io
us
e
xt
r
e
m
e
le
a
r
ni
ng
m
a
c
h
in
e
c
la
s
s
if
ie
r
s
to
ge
th
e
r
.
O
u
c
hr
a
e
t
al
.
[
9]
c
om
pa
r
e
d
s
upe
r
vi
s
e
d
a
nd
uns
upe
r
vi
s
e
d
m
a
c
hi
ne
le
a
r
ni
ng
te
c
hni
qu
e
s
of
ur
ba
n
la
nd
c
ove
r
in
g
c
la
s
s
if
ic
a
ti
on
by
e
m
pl
oyi
ng
L
a
nds
a
t
8
im
a
ge
r
y
a
nd
e
m
pha
s
i
z
e
d
m
e
th
odol
ogi
c
a
l
di
f
f
e
r
e
nc
e
s
a
ppl
ie
d
to
f
e
a
tu
r
e
e
xt
r
a
c
ti
on.
I
n
vol
c
a
ni
c
c
ont
e
xt
s
,
C
a
r
ie
ll
o
e
t
al
.
[
10]
s
how
e
d
th
e
a
ppl
ic
a
ti
on
of
m
a
c
hi
ne
le
a
r
ni
ng
to
S
e
nt
in
e
l
-
2
im
a
ge
r
y
to
tr
a
c
k
vol
c
a
ni
c
th
e
r
m
a
l
a
nom
a
li
e
s
,
w
hi
le
B
ut
ta
r
a
nd
S
a
c
ha
n
[
11]
ut
il
iz
e
d
R
e
s
N
e
t
-
152
to
c
la
s
s
if
y
a
nd
ge
o
-
im
a
ge
im
a
ge
s
w
it
h
a
f
oc
us
on
th
e
r
e
qui
r
e
m
e
nt
of
a
ut
om
a
te
d
f
e
a
tu
r
e
e
xt
r
a
c
ti
on.
C
or
r
a
di
no
e
t
al
.
[
12]
u
ti
li
z
e
d
U
-
N
E
T
to
a
na
ly
z
e
21
ye
a
r
s
of
a
dva
nc
e
d
s
pa
c
e
bor
ne
th
e
r
m
a
l
e
m
is
s
io
n
a
nd
r
e
f
le
c
ti
on
r
a
di
om
e
te
r
(
A
S
T
E
R
)
gl
oba
l
th
e
r
m
a
l
i
nf
r
a
r
e
d
(
T
I
R
)
im
a
ge
r
y
of
f
iv
e
vol
c
a
noe
s
a
nd
a
tt
a
in
e
d
93%
e
f
f
e
c
ti
ve
ne
s
s
of
a
nom
a
ly
de
t
e
c
ti
on.
S
im
il
a
r
ly
,
S
hul
tz
[
13]
pr
e
s
e
nt
e
d
th
e
C
N
N
-
ba
s
e
d
f
r
a
m
e
,
hot
s
pot
le
a
r
ni
ng
a
nd
id
e
nt
if
ic
a
ti
on
ne
twor
k
(
H
ot
L
I
N
K
)
,
te
s
te
d
a
nd
pr
ove
d
w
it
h
m
ode
r
a
te
r
e
s
ol
ut
io
n
im
a
gi
ng
s
pe
c
tr
or
a
di
om
e
te
r
(
M
O
D
I
S
)
a
nd
vi
s
ib
le
in
f
r
a
r
e
d
im
a
gi
ng
r
a
di
om
e
te
r
s
ui
te
(
V
I
I
R
S
)
da
ta
c
ol
le
c
ti
ons
,
a
nd
a
tt
a
in
e
d ove
r
95%
a
c
c
ur
a
c
y of
hot
s
pot
i
de
nt
if
ic
a
ti
on.
O
th
e
r
a
ppr
oa
c
he
s
ha
ve
us
e
d
C
N
N
s
in
non
-
im
a
ge
dom
a
in
s
.
O
ña
te
e
t
al
.
[
14]
f
or
e
c
a
s
te
d
m
ic
r
o
-
e
a
r
th
qua
ke
s
us
in
g
m
a
ni
f
ol
d
le
a
r
ni
ng
a
nd
a
udi
o
-
dr
iv
e
n
f
e
a
tu
r
e
s
,
a
nd
a
c
hi
e
ve
d
ove
r
94%
a
c
c
ur
a
c
y.
N
unna
r
i
a
nd
C
a
lv
a
r
i
[
15]
c
ont
r
a
s
te
d
e
ig
ht
C
N
N
m
ode
ls
f
or
e
r
upt
iv
e
a
c
ti
vi
ty
m
oni
to
r
in
g
of
M
ount
E
tn
a
a
nd
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
D
C
N
N
V
A
:
a de
e
p c
onv
ol
ut
io
nal
ne
ur
al
ne
tw
o
r
k
f
or
v
ol
c
ani
c
ac
ti
v
it
y
c
la
s
s
if
ic
at
io
n
…
(
Y
as
ir
H
u
s
s
e
in
Shak
ir
)
283
e
s
ta
bl
is
he
d
tr
a
ns
f
e
r
le
a
r
ni
ng
a
s
s
upe
r
io
r
.
C
he
n
e
t
al
.
[
16]
pr
opo
s
e
d
t
r
a
ns
f
e
r
le
a
r
ni
ng
-
ba
s
e
d
V
G
G
(
T
V
G
G
)
f
o
r
r
e
m
ot
e
s
e
ns
in
g
im
a
ge
c
l
a
s
s
if
ic
a
ti
on
w
it
h
99.18%
a
c
c
ur
a
c
y
ba
s
e
d
on
a
V
G
G
-
ba
s
e
d
tr
a
ns
f
e
r
le
a
r
ni
ng
te
c
hni
que
.
M
oha
n
e
t
al
.
[
17]
in
tr
oduc
e
d
H
ot
s
pot
te
r
,
a
n
e
nd
-
to
-
e
nd
s
ys
te
m
de
s
ig
ne
d
to
a
ut
om
a
ti
c
a
ll
y
de
te
c
t
s
ubt
le
vol
c
a
ni
c
th
e
r
m
a
l
a
nom
a
li
e
s
in
s
a
t
e
ll
it
e
im
a
ge
r
y
w
hi
le
a
ls
o
ge
ne
r
a
ti
ng
ke
y
th
e
r
m
a
l
s
ta
ti
s
ti
c
s
.
E
a
r
li
e
r
m
e
th
ods
f
or
a
ut
om
a
te
d
vol
c
a
ni
c
th
e
r
m
a
l
f
e
a
tu
r
e
(
V
T
F
)
d
e
te
c
ti
on
w
e
r
e
li
m
it
e
d
by
s
m
a
ll
d
a
ta
s
e
ts
a
nd
na
r
r
ow
ge
ogr
a
phi
c
c
ove
r
a
ge
.
T
h
e
R
e
e
d
-
X
ia
ol
i
a
lg
or
it
hm
(
L
R
X
)
+
C
N
N
a
ppr
oa
c
h,
w
hi
c
h
tr
a
in
e
d
C
N
N
f
or
50
e
poc
hs
on
th
e
L
R
X
s
c
or
e
im
a
ge
s
,
pr
oduc
e
d
a
n
F
1
-
s
c
or
e
of
88.4%
a
nd
a
te
s
t
a
c
c
ur
a
c
y
of
90.3%
.
H
ue
r
ta
s
e
t
al
.
[
18
]
de
ve
lo
pe
d
V
G
G
16
a
nd
I
nc
e
pt
io
n
C
N
N
m
ode
ls
to
a
na
ly
z
e
m
ud
vol
c
a
no
im
a
ge
r
y
a
nd
f
in
di
ngs
s
how
th
a
t
bot
h
tr
a
ns
f
e
r
le
a
r
ni
ng
a
nd
c
us
to
m
C
N
N
s
a
r
e
w
e
ll
-
s
ui
te
d
f
or
ha
ndl
in
g
th
e
c
ha
ll
e
nge
s
of
th
is
ty
pe
of
da
ta
.
T
he
V
G
G
16
m
ode
l
a
c
hi
e
ve
d
te
s
t
a
c
c
ur
a
c
y
of
93%
,
w
it
h
pr
e
c
is
io
n
a
nd
r
e
c
a
ll
s
c
or
e
s
of
93%
a
nd
94%
f
or
th
e
“
m
ud”
c
la
s
s
a
nd
93%
a
nd
93%
f
o
r
th
e
“
no
m
ud”
c
la
s
s
.
T
h
e
I
nc
e
pt
io
n
m
ode
l
a
ls
o
de
m
ons
tr
a
te
d
s
ta
bl
e
pe
r
f
or
m
a
nc
e
w
it
h only m
in
or
va
r
ia
ti
on.
I
n
s
um
m
a
r
y,
th
e
s
tu
di
e
s
to
d
a
te
ha
v
e
ut
il
iz
e
d
s
e
ve
r
a
l
v
a
r
io
us
da
ta
s
e
ts
a
nd
va
r
ie
ty
of
m
a
c
hi
ne
le
a
r
ni
ng
a
nd
de
e
p
le
a
r
ni
ng
a
lg
or
it
hm
s
te
c
hni
que
s
to
vol
c
a
ni
c
m
oni
to
r
in
g
a
nd
s
a
te
ll
it
e
im
a
g
e
c
la
s
s
if
ic
a
ti
on.
W
hi
le
th
e
s
e
a
ppr
oa
c
he
s
de
m
ons
tr
a
te
s
ig
ni
f
ic
a
nt
pr
ogr
e
s
s
,
m
o
s
t
r
e
m
a
in
a
t
th
e
e
xp
e
r
im
e
nt
a
l
s
ta
ge
a
nd
la
c
k
pr
a
c
ti
c
a
l
de
pl
oym
e
nt
.
I
n
pa
r
ti
c
ul
a
r
no
e
xi
s
ti
ng
r
e
s
e
a
r
c
h
h
a
s
pr
opos
e
d
a
c
om
pr
e
he
n
s
iv
e
f
r
a
m
e
w
or
k
th
a
t
in
te
gr
a
te
s
a
de
pl
oy
a
bl
e
G
U
I
w
it
h
r
e
a
l
-
ti
m
e
vol
c
a
ni
c
a
c
ti
vi
ty
m
oni
to
r
in
g
a
nd
m
ul
ti
-
m
oda
l
a
le
r
ti
ng
c
a
pa
bi
li
ti
e
s
f
e
a
tu
r
e
s
th
a
t
a
r
e
e
s
s
e
nt
ia
l
f
or
e
f
f
e
c
ti
ve
di
s
a
s
te
r
m
a
na
ge
m
e
nt
a
ppl
ic
a
ti
ons
.
F
ur
th
e
r
m
or
e
,
m
a
ny
s
tu
di
e
s
a
r
e
c
ons
tr
a
in
e
d
by
li
m
it
e
d
da
ta
s
e
ts
a
nd
in
c
on
s
is
te
nt
pe
r
f
or
m
a
nc
e
a
c
r
os
s
di
f
f
e
r
e
nt
c
ont
e
xt
s
,
hi
ghl
ig
ht
in
g
th
e
n
e
e
d
f
or
m
or
e
r
obus
t
,
s
c
a
la
bl
e
,
a
nd
pr
a
c
ti
c
a
ll
y a
ppl
ic
a
bl
e
s
ol
ut
io
ns
. A
c
om
pa
r
a
ti
ve
a
na
ly
s
is
of
th
e
pr
e
vi
ous
r
e
la
te
d
w
or
k i
s
pr
e
s
e
nt
e
d i
n t
a
bul
a
r
f
or
m
i
n T
a
bl
e
1.
T
a
bl
e
1
.
S
um
m
a
r
y of
r
e
c
e
nt
m
a
c
hi
ne
le
a
r
ni
ng
a
nd de
e
p l
e
a
r
ni
n
g f
or
vol
c
a
ni
c
s
a
te
ll
it
e
i
m
a
ge
c
la
s
s
if
ic
a
ti
on
R
e
f
e
r
e
nc
e
/
y
ear
D
a
t
a
s
e
t
M
e
t
hods
R
e
s
ul
t
s
(%)
W
e
a
kne
s
s
C
or
r
a
di
no
e
t
al
.
[
12]
,
2024
21 ye
a
r
s
of
A
S
T
E
R
T
I
R
da
t
a
(
5
vol
c
a
noe
s
)
C
N
N
(
U
-
N
E
T
a
r
c
hi
t
e
c
t
ur
e
)
A
c
c
ur
a
c
y =93
L
i
m
i
t
e
d t
o A
S
T
E
R
T
I
R
i
m
a
ge
r
y,
m
ode
l
ge
ne
r
a
l
i
z
a
bi
l
i
t
y t
o ot
he
r
s
e
ns
or
s
not
t
e
s
t
e
d.
S
hul
t
z
[
13]
, 2024
V
I
I
R
S
a
nd M
O
D
I
S
i
m
a
ge
r
y (
A
l
a
s
ka
n
vol
c
a
noe
s
)
C
N
N
(
H
ot
L
I
N
K
)
A
c
c
ur
a
c
y =98
D
a
t
a
s
e
t
ge
ogr
a
phi
c
a
l
l
y r
e
s
t
r
i
c
t
e
d
(
A
l
a
s
ka
)
,
l
i
m
i
t
e
d va
l
i
da
t
i
on on
di
ve
r
s
e
vol
c
a
ni
c
s
e
t
t
i
ng
s
.
O
ña
t
e
e
t
al
.
[
14]
,
2024
S
e
i
s
m
i
c
da
t
a
(
C
ot
opa
xi
a
nd
L
l
a
i
m
a
)
A
udi
o
f
e
a
t
ur
e
s
+ps
yc
hoa
c
ou
s
t
i
c
s
c
a
l
e
s
+m
a
ni
f
ol
d l
e
a
r
ni
ng
A
c
c
ur
a
c
y =94.44
–
95.45
F
oc
us
e
d on s
e
i
s
m
i
c
da
t
a
onl
y,
l
a
c
ks
m
ul
t
i
-
m
oda
l
i
nt
e
gr
a
t
i
on
w
i
t
h s
a
t
e
l
l
i
t
e
i
m
a
ge
r
y.
N
unna
r
i
a
nd
C
a
l
va
r
i
[
15]
, 2024
G
r
ound
-
ba
s
e
d
t
he
r
m
a
l
i
m
a
ge
s
(
M
t
.
E
t
na
)
C
om
pa
r
i
s
on of
8 C
N
N
s
(
S
que
e
z
e
N
e
t
, G
oogl
e
N
e
t
,
D
e
ns
e
N
e
t
201,
R
e
s
N
e
t
18,
S
huf
f
l
e
N
e
t
, D
a
r
kN
e
t
19,
A
l
e
xN
e
t
, V
G
G
-
16)
A
c
c
ur
a
c
y =94.07
R
e
s
ul
t
s
r
e
s
t
r
i
c
t
e
d t
o one
vol
c
a
no, c
om
put
a
t
i
ona
l
l
y
e
xpe
ns
i
ve
due
t
o m
ul
t
i
pl
e
C
N
N
c
om
pa
r
i
s
ons
.
C
he
n
e
t
al
.
[
16]
,
2024
T
i
a
ngong
-
2 r
e
m
ot
e
s
e
ns
i
ng da
t
a
s
e
t
T
V
G
G
(
f
r
om
I
m
a
ge
N
e
t
)
A
c
c
ur
a
c
y =99.18
R
e
c
a
l
l
=99.17
R
e
l
i
e
s
he
a
vi
l
y on t
r
a
ns
f
e
r
l
e
a
r
ni
ng
a
nd
l
a
c
ks
t
e
s
t
i
ng on
vol
c
a
ni
c
da
t
a
s
e
t
s
.
M
oha
n
e
t
al
.
[
17]
,
2025
T
he
r
m
a
l
s
a
t
e
l
l
i
t
e
da
t
a
L
R
X
+C
N
N
(
50 e
poc
hs
)
A
c
c
ur
a
c
y =90.30
F1
-
s
c
or
e
=88.40
D
a
t
a
s
e
t
r
e
l
a
t
i
ve
l
y s
m
a
l
l
,
pe
r
f
or
m
a
nc
e
l
ow
e
r
t
ha
n ot
he
r
C
N
N
-
ba
s
e
d m
ode
l
s
.
H
ue
r
t
a
s
e
t
al
.
[
18]
,
2025
I
N
G
V
(
m
ud
v
ol
c
a
no
i
m
a
ge
r
y
)
V
G
G
16 a
nd I
nc
e
pt
i
on
C
N
N
A
c
c
ur
a
c
y =93
P
r
e
c
i
s
i
on =93
R
e
c
a
l
l
=94
F
oc
us
l
i
m
i
t
e
d t
o m
ud vol
c
a
noe
s
;
pe
r
f
or
m
a
nc
e
s
t
a
bi
l
i
t
y a
c
r
os
s
br
oa
de
r
da
t
a
s
e
t
s
not
a
s
s
e
s
s
e
d.
3.
M
A
T
E
R
I
A
L
S
A
N
D
M
E
T
H
O
D
S
T
hi
s
s
e
c
ti
on
c
ont
a
in
s
th
e
m
a
te
r
ia
ls
a
nd
m
e
th
od
s
f
r
a
m
e
w
or
k
us
e
d
in
th
is
s
tu
dy
,
w
hi
c
h
i
s
vi
s
ua
ll
y
s
um
m
a
r
iz
e
d
in
F
ig
ur
e
1.
T
he
m
e
th
odol
ogy
is
s
tr
uc
tu
r
e
d
in
to
s
ix
m
a
in
s
te
ps
:
da
ta
a
c
qui
s
it
io
n,
da
ta
pr
e
pr
oc
e
s
s
in
g, da
ta
di
vi
s
io
n, m
ode
l
d
e
ve
lo
pm
e
nt
a
nd t
r
a
in
in
g,
e
va
lu
a
ti
on, a
nd de
pl
oym
e
nt
.
3.1. Dat
a ac
q
u
is
it
io
n
T
hi
s
ki
nd
of
da
ta
s
e
t
c
on
s
is
ts
of
th
e
im
a
ge
s
of
th
e
vol
c
a
ni
c
a
c
ti
vi
ty
,
a
nd
th
is
ha
s
b
e
e
n
got
te
n
f
r
om
s
a
te
ll
it
e
im
a
ge
s
.
T
he
f
ol
lo
w
in
g
f
ig
ur
e
s
of
s
e
ve
r
a
l
im
a
ge
s
ta
ke
n
f
r
om
s
e
ve
r
a
l
c
la
s
s
e
s
of
vol
c
a
noe
s
(
Y
e
s
A
c
ti
vi
ty
)
a
nd
(
N
oA
c
ti
vi
ty
)
a
r
e
pr
e
s
e
nt
e
d
in
F
ig
u
r
e
2.
T
he
c
om
pl
e
te
da
ta
s
e
t
us
e
d
f
or
th
is
e
xpe
r
im
e
nt
is
publ
ic
ly
a
va
il
a
bl
e
onl
in
e
[
19]
.
3.2. Dat
a p
r
e
p
r
oc
e
s
s
in
g
T
o
e
ns
ur
e
th
a
t
th
e
da
t
a
w
a
s
s
ta
nda
r
di
z
e
d
a
nd
opt
im
iz
e
d
f
or
de
e
p
le
a
r
ni
ng
,
s
e
ve
r
a
l
pr
e
pr
oc
e
s
s
in
g
s
te
ps
w
e
r
e
a
ppl
ie
d
to
th
e
im
a
ge
s
,
a
nd
s
te
ps
w
e
r
e
a
im
e
d
a
t
m
a
ki
ng
th
e
da
ta
s
e
t
m
or
e
uni
f
or
m
,
r
e
duc
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
. 15, No. 1, M
a
r
c
h 2026
:
281
-
292
284
va
r
ia
bi
li
ty
,
a
nd
a
ll
ow
in
g
th
e
m
ode
l
to
ge
ne
r
a
li
z
e
m
or
e
to
ne
w
da
ta
.
P
r
e
pr
oc
e
s
s
in
g
pi
pe
li
ne
s
in
vol
ve
d
r
e
s
c
a
li
ng
a
nd
r
e
s
iz
in
g
tr
a
n
s
f
or
m
a
ti
ons
,
nor
m
a
li
z
a
ti
on
of
th
e
pi
xe
ls
,
a
nd
th
e
a
ppl
ic
a
ti
on
of
da
ta
a
ugm
e
nt
a
ti
on
s
tr
a
te
gi
e
s
. D
a
ta
a
ugm
e
nt
a
ti
on w
a
s
pa
r
ti
c
ul
a
r
ly
i
m
por
ta
nt
, a
s
i
t
e
f
f
e
c
ti
ve
ly
i
nc
r
e
a
s
e
d t
he
s
iz
e
of
t
he
da
ta
s
e
t
a
nd
in
tr
oduc
e
d
va
r
ia
ti
ons
th
a
t
s
im
ul
a
te
r
e
a
l
-
w
or
ld
c
ondi
ti
ons
(
di
f
f
e
r
e
nt
or
ie
nt
a
ti
ons
,
s
c
a
le
s
,
a
nd
di
s
to
r
ti
ons
)
.
T
hi
s
he
lp
s
to
pr
e
ve
nt
ove
r
f
it
t
in
g
a
nd
im
p
r
ove
r
obus
tn
e
s
s
,
a
s
s
how
n
in
F
ig
ur
e
3.
T
he
a
ugm
e
nt
a
ti
on
w
a
s
pe
r
f
or
m
e
d
us
in
g t
he
A
ugm
e
nt
or
l
ib
r
a
r
y a
nd t
he
s
pe
c
if
ic
t
r
a
ns
f
or
m
a
ti
ons
,
a
nd t
he
ir
pr
oba
bi
li
ti
e
s
a
r
e
s
how
n i
n T
a
bl
e
2.
F
ig
ur
e
1. P
r
opos
e
d
f
r
a
m
e
w
or
k f
or
vo
lc
a
ni
c
a
c
ti
vi
ty
c
la
s
s
if
ic
a
ti
o
n us
in
g s
a
te
ll
it
e
i
m
a
ge
r
y
F
ig
ur
e
2.
S
a
m
pl
e
s
a
te
ll
it
e
i
m
a
ge
s
f
r
om
t
he
'
ye
s
a
c
ti
vi
ty
'
a
nd '
no
a
c
ti
vi
ty
'
vol
c
a
ni
c
da
ta
s
e
t
c
la
s
s
e
s
F
ig
ur
e
3. I
ll
us
tr
a
ti
on of
da
ta
a
ugm
e
nt
a
ti
on t
e
c
hni
que
s
a
ppl
ie
d t
o or
ig
in
a
l
vol
c
a
ni
c
i
m
a
ge
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
D
C
N
N
V
A
:
a de
e
p c
onv
ol
ut
io
nal
ne
ur
al
ne
tw
o
r
k
f
or
v
ol
c
ani
c
ac
ti
v
it
y
c
la
s
s
if
ic
at
io
n
…
(
Y
as
ir
H
u
s
s
e
in
Shak
ir
)
285
T
a
bl
e
2
. D
a
ta
pr
e
pr
oc
e
s
s
in
g
a
nd a
ugm
e
nt
a
ti
on pa
r
a
m
e
te
r
s
f
or
m
ode
l
tr
a
in
in
g
P
r
e
pr
oc
e
s
s
i
ng
s
t
e
p
D
e
t
a
i
l
s
R
e
s
c
a
l
i
ng
1/
255
.
S
i
z
e
(
224, 224)
.
D
a
t
a
a
ugm
e
nt
a
t
i
on
T
e
c
hni
que
.
F
l
i
p
l
e
f
t
r
i
ght
:
p
r
oba
bi
l
i
t
y
of
0.3
.
F
l
i
p
t
op bot
t
om
:
pr
oba
bi
l
i
t
y
of
0.5
.
R
ot
a
t
e
:
pr
oba
bi
l
i
t
y
of
0.5, w
i
t
h a
m
a
xi
m
um
l
e
f
t
a
nd r
i
ght
r
ot
a
t
i
on of
5 de
gr
e
e
s
.
Z
oom
:
pr
oba
bi
l
i
t
y o
f
0.3, w
i
t
h a
z
oom
f
a
c
t
or
be
t
w
e
e
n 1.1 a
nd 1.2
.
R
a
ndom
di
s
t
or
t
i
on:
pr
oba
bi
l
i
t
y of
1, w
i
t
h a
gr
i
d w
i
dt
h a
nd he
i
ght
o
f
3 a
nd a
m
a
gni
t
ude
of
5.
3.3.
D
at
a
d
iv
is
io
n
T
o
pr
e
pa
r
e
th
e
da
ta
s
e
t
f
or
tr
a
in
in
g
a
nd
e
va
lu
a
ti
on,
w
e
di
vi
de
d
i
t
in
to
tr
a
in
in
g,
va
li
da
ti
on,
a
nd
te
s
ti
ng
s
ubs
e
ts
.
T
he
s
e
t
w
a
s
di
vi
de
d
to
be
75%
f
or
tr
a
in
in
g
(
2,250
i
m
a
ge
s
)
,
15%
f
or
va
li
da
ti
on
(
350
im
a
ge
s
)
,
a
nd
20%
f
or
te
s
ti
ng
(
600
im
a
ge
s
)
;
a
ll
of
th
e
s
e
s
e
ts
ha
ve
two
c
l
a
s
s
e
s
.
T
he
ove
r
a
ll
s
e
t
h
a
d
3,000
im
a
ge
s
a
c
r
os
s
two
c
a
te
gor
ie
s
,
a
s
s
how
n i
n
T
a
bl
e
3.
T
a
bl
e
3
.
D
a
ta
s
e
t
pa
r
ti
ti
oni
ng f
or
m
ode
l
tr
a
in
in
g, va
li
da
ti
on, a
nd t
e
s
ti
ng
S
ubs
e
t
N
um
be
r
of
i
m
a
ge
s
P
e
r
c
e
nt
a
ge
(%)
N
um
be
r
of
c
l
a
s
s
es
T
r
a
i
ni
ng
2,250
75
2
V
a
l
i
da
t
i
on
350
15
2
T
e
s
t
i
ng
600
20
2
T
ot
a
l
3,000
100
2
3.4.
D
e
e
p
le
ar
n
in
g m
od
e
ls
an
d
ar
c
h
it
e
c
t
u
r
e
C
ur
r
e
nt
br
e
a
kt
hr
oughs
in
de
e
p
le
a
r
ni
ng
ha
ve
im
m
e
ns
e
ly
pr
op
e
ll
e
d
th
e
pr
ogr
e
s
s
of
s
a
te
ll
it
e
im
a
ge
pr
oc
e
s
s
in
g
,
s
pe
c
if
ic
a
ll
y
r
e
ga
r
di
ng
th
e
id
e
nt
if
ic
a
ti
on
of
na
tu
r
a
l
phe
nom
e
na
li
ke
vol
c
a
ni
c
a
c
ti
vi
ty
.
E
m
pl
oyi
ng
C
N
N
s
a
nd
tr
a
ns
f
e
r
le
a
r
ni
ng
f
a
c
il
it
a
te
s
qui
c
k
f
e
a
tu
r
e
de
te
c
ti
on
f
r
om
da
ta
of
hi
gh
di
m
e
ns
io
na
li
ty
,
w
hi
le
a
t
th
e
s
a
m
e
ti
m
e
ove
r
c
om
in
g
di
f
f
ic
ul
ti
e
s
s
te
m
m
in
g
f
r
om
li
m
it
e
d
da
ta
s
e
ts
w
it
h
a
nnot
a
ti
on
s
.
H
e
r
e
in
,
w
e
put
f
or
w
a
r
d
a
ne
w
D
C
N
N
V
A
a
nd
e
ig
ht
pr
e
-
e
xi
s
ti
ng
tr
a
ns
f
e
r
le
a
r
ni
ng
m
o
de
ls
,
R
e
s
N
e
t5
0,
N
A
S
N
e
tL
a
r
ge
,
D
e
ns
e
N
e
t1
21,
M
obi
le
N
e
t,
I
nc
e
pt
io
nV
3,
X
c
e
pt
io
n,
V
G
G
19,
a
nd
V
G
G
16
,
t
o
bui
ld
a
c
om
pl
e
te
c
la
s
s
if
ic
a
ti
on
s
ys
te
m
of
vol
c
a
ni
c
s
a
te
ll
it
e
i
m
a
ge
s
.
3.4.1
.
D
e
e
p
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
f
or
vol
c
an
ic
ac
t
iv
it
y
V
ol
c
a
ni
c
a
c
ti
vi
ty
(
D
C
N
N
V
A
)
is
a
c
us
to
m
iz
e
d
a
r
c
hi
te
c
tu
r
e
de
ve
lo
pe
d
onl
y
f
or
vol
c
a
ni
c
a
c
ti
vi
ty
c
la
s
s
if
ic
a
ti
on us
in
g s
a
te
ll
it
e
s
,
a
nd t
he
ne
twor
k i
s
s
o de
s
ig
ne
d t
ha
t
it
m
a
in
ta
in
s
a
ba
la
nc
e
be
twe
e
n c
om
put
a
ti
on
e
f
f
ic
ie
nc
y
a
nd
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
,
a
nd
th
e
r
e
f
or
e
is
us
a
bl
e
f
or
ne
a
r
r
e
a
l
-
ti
m
e
m
oni
to
r
in
g.
I
t
is
a
c
om
bi
na
ti
on
of
s
pa
c
e
f
e
a
tu
r
e
s
,
e
xt
r
a
c
ti
ng
c
onvolut
io
n
a
nd
p
ool
in
g
ope
r
a
ti
ons
,
a
nd
f
ul
ly
c
onn
e
c
te
d
la
y
e
r
s
ut
il
iz
e
d a
t
th
e
t
im
e
of
c
la
s
s
if
ic
a
ti
on (
≈1.19
M
t
r
a
in
a
bl
e
pa
r
a
m
e
t
e
r
s
)
.
i)
C
onvolut
io
na
l
la
ye
r
s
:
th
e
c
onvolut
io
na
l
ope
r
a
ti
on
e
xt
r
a
c
ts
hi
e
r
a
r
c
hi
c
a
l
r
e
pr
e
s
e
nt
a
ti
ons
f
r
om
in
pu
t
im
a
ge
s
by
a
ppl
yi
ng
le
a
r
na
bl
e
ke
r
ne
ls
.
M
a
th
e
m
a
ti
c
a
ll
y,
th
e
c
o
nvol
ut
io
n
a
t
la
ye
r
c
a
n
be
e
xpr
e
s
s
e
d
a
s
pr
e
s
e
nt
e
d i
n (
1)
.
,
(
)
=
∑
∑
+
,
+
(
−
1
)
−
1
=
0
−
1
=
0
.
,
(
)
+
(
)
(
1)
W
he
r
e
(
−
1
)
r
e
pr
e
s
e
nt
s
i
nput
f
r
om
t
he
pr
e
vi
ous
l
a
ye
r
,
(
)
is
a
c
onvolut
i
ona
l
ke
r
ne
l,
(
)
is
bi
a
s
ed
, a
nd
,
(
)
is
th
e
f
e
a
tu
r
e
m
a
p
a
t
pos
it
io
n
(
,
)
.
T
he
non
-
li
ne
a
r
a
c
ti
va
ti
on
f
unc
ti
on
R
e
L
U
(
(
)
=
(
0
,
)
)
is
a
ppl
ie
d t
o i
nt
r
oduc
e
non
-
li
ne
a
r
it
y.
ii)
P
ool
in
g
la
ye
r
s
:
to
r
e
duc
e
s
pa
ti
a
l
di
m
e
ns
io
ns
w
hi
le
pr
e
s
e
r
vi
ng
e
s
s
e
nt
ia
l
f
e
a
tu
r
e
s
,
m
a
x
pool
in
g
is
us
e
pool
in
g ope
r
a
ti
on i
s
de
f
in
e
d
in
(
2)
.
,
=
(
,
)
∈
(
+
,
+
)
(
2)
W
he
r
e
de
not
e
s
th
e
pool
in
g
r
e
gi
on
ope
r
a
ti
on
de
c
r
e
a
s
e
s
c
o
m
put
a
ti
ona
l
c
om
pl
e
xi
ty
a
nd
c
ont
r
ol
s
ove
r
f
it
ti
ng via
i
nt
r
oduc
in
g t
r
a
ns
la
ti
ona
l
in
va
r
ia
nc
e
.
iii)
F
ul
ly
c
onne
c
te
d
la
ye
r
s
:
th
e
e
xt
r
a
c
te
d
f
e
a
tu
r
e
s
a
r
e
f
la
tt
e
ne
d
a
n
d
pa
s
s
e
d
to
f
ul
ly
c
onne
c
te
d
de
ns
e
la
ye
r
s
f
or
c
la
s
s
if
ic
a
ti
on. T
he
t
r
a
ns
f
or
m
a
ti
on i
s
gi
ve
n
in
(
3)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
. 15, No. 1, M
a
r
c
h 2026
:
281
-
292
286
=
(
+
)
(
3)
W
he
r
e
a
nd
de
not
e
th
e
w
e
ig
ht
s
a
nd
bi
a
s
e
s
of
de
n
s
e
la
ye
r
,
r
e
s
p
e
c
ti
ve
ly
,
a
nd
r
e
pr
e
s
e
nt
s
th
e
R
e
L
U
or
s
of
tm
a
x
a
c
ti
va
ti
on
f
unc
ti
on
,
de
pe
ndi
ng
on
l
a
ye
r
.
T
h
e
f
in
a
l
s
of
tm
a
x
c
la
s
s
if
ie
r
pr
oduc
e
s
a
pr
oba
bi
li
ty
ove
r
t
w
o output
c
la
s
s
e
s
,
w
he
r
e
=
2 c
or
r
e
s
ponds
t
o t
he
numbe
r
o
f
c
la
s
s
e
s
pr
e
s
e
nt
e
d i
n
(
4)
.
(
=
/
)
=
(
)
∑
(
)
=
1
,
{
1
,
2
}
(
4)
iv
)
D
r
opout
a
nd
opt
im
iz
a
ti
on:
to
e
nha
n
c
e
ge
n
e
r
a
li
z
a
ti
on
,
a
dr
opout
la
ye
r
w
it
h
r
a
te
p
=
0.5 wa
s
in
c
or
por
a
te
d
,
w
hi
c
h r
a
ndoml
y de
a
c
ti
va
te
s
ne
ur
ons
dur
in
g t
r
a
in
in
g. T
he
m
ode
l
is
opt
im
iz
e
d us
in
g t
he
A
da
m
opt
im
iz
e
r
,
w
hi
c
h
a
da
pt
iv
e
ly
upda
te
s
le
a
r
ni
ng
r
a
te
s
f
or
e
a
c
h
pa
r
a
m
e
t
e
r
.
T
h
e
c
a
te
gor
ic
a
l
c
r
os
s
-
e
nt
r
opy
lo
s
s
f
unc
ti
on
is
e
m
pl
oye
d
, a
s
pr
e
s
e
nt
e
d i
n
(
5)
.
ℒ
=
−
∑
∑
,
̂
,
=
1
=
1
(
5)
W
he
r
e
,
is
t
he
gr
ound tr
ut
h l
a
be
l
a
nd
̂
,
is
t
he
pr
e
di
c
te
d pr
oba
bi
li
ty
f
or
c
la
s
s
.
T
he
D
C
N
N
V
A
c
ons
i
s
ts
of
f
iv
e
c
onvolut
io
na
l
la
ye
r
s
(
64
-
256
f
il
te
r
s
)
,
e
a
c
h
of
w
hi
c
h
is
f
ol
lo
w
e
d
by
m
a
x
-
pool
in
g
f
oc
us
in
g
on
di
m
e
ns
io
n
r
e
duc
ti
on,
a
f
ul
ly
c
onne
c
te
d
256
-
ne
ur
on
la
ye
r
,
a
dr
opout
la
ye
r
,
a
nd
a
f
in
a
l
out
put
c
la
s
s
if
ic
a
ti
on
s
of
tm
a
x
l
a
ye
r
.
I
t
ha
s
r
oughly
1.19
m
il
li
on
tr
a
in
a
bl
e
pa
r
a
m
e
te
r
s
,
c
or
r
e
s
ponding
to
a
li
ght
but
de
e
p
a
r
c
hi
te
c
tu
r
e
f
r
ie
ndl
y
to
s
c
a
la
bi
li
ty
a
nd
e
f
f
ic
ie
nc
y
.
T
he
m
ode
l
w
a
s
opt
im
iz
e
d
us
in
g
th
e
A
da
m
opt
im
iz
e
r
w
it
h
a
le
a
r
ni
ng
r
a
te
of
0.001,
a
nd
tr
a
in
e
d
w
it
h
th
e
c
a
te
gor
ic
a
l
c
r
os
s
-
e
nt
r
opy
lo
s
s
f
unc
ti
on.
T
r
a
in
in
g
w
a
s
c
ondu
c
te
d
w
it
h
a
ba
tc
h
s
iz
e
of
32
ov
e
r
10
e
poc
h
s
,
u
s
in
g
th
e
G
oogl
e
C
ol
la
bor
a
to
r
y
pl
a
tf
or
m
w
it
h
a
n
N
V
I
D
I
A
T
e
s
la
T
4 G
P
U
(
16 G
B
)
.
F
ig
ur
e
4
s
how
s
th
e
D
C
N
N
V
A
a
r
c
hi
te
c
tu
r
e
.
F
ig
ur
e
4.
A
r
c
hi
te
c
tu
r
a
l
di
a
gr
a
m
of
t
he
pr
opos
e
d
D
C
N
N
V
A
3.4.2
.
R
e
s
N
e
t
50
I
nt
r
oduc
e
d
by
H
e
e
t
al
.
[
20]
,
R
e
s
N
e
t5
0
in
tr
oduc
e
s
r
e
s
id
ua
l
le
a
r
ni
ng
th
r
ough
id
e
nt
it
y
s
hor
tc
ut
c
onne
c
ti
ons
.
T
hi
s
m
it
ig
a
te
s
th
e
pr
obl
e
m
of
va
ni
s
hi
ng
g
r
a
di
e
nt
s
in
de
e
pe
r
ne
twor
ks
.
B
y
s
ta
c
ki
ng
c
onvolut
io
na
l
bl
oc
ks
w
it
h
r
e
s
id
ua
l
li
nks
,
th
e
m
ode
l
e
n
a
bl
e
s
s
ta
bl
e
tr
a
in
in
g
a
nd
im
pr
ove
d
f
e
a
tu
r
e
e
xt
r
a
c
ti
on,
m
a
ki
ng i
t
w
e
ll
-
s
ui
te
d f
or
c
om
pl
e
x i
m
a
ge
c
la
s
s
if
ic
a
ti
on t
a
s
k
s
s
u
c
h a
s
vol
c
a
ni
c
a
c
ti
vi
ty
r
e
c
ogni
ti
on.
3.4.3
.
N
A
S
N
e
t
L
ar
ge
N
A
S
N
e
tL
a
r
ge
,
in
tr
oduc
e
d
by
Z
oph
e
t
al
.
[
21]
,
is
a
ne
ur
a
l
a
r
c
hi
te
c
tu
r
e
s
e
a
r
c
h
(
N
A
S
)
-
di
s
c
ove
r
e
d
a
r
c
hi
te
c
tu
r
e
th
a
t
is
a
im
e
d
a
t
opt
im
iz
in
g
ne
twor
k
s
tr
uc
tu
r
e
s
a
ut
onomous
ly
f
or
hi
ghe
r
pe
r
f
o
r
m
a
nc
e
.
I
t
is
a
m
odul
a
r
a
r
c
hi
te
c
tu
r
e
th
a
t
ut
il
iz
e
s
r
e
duc
ti
on
a
nd
nor
m
a
l
c
e
ll
s
.
T
he
m
ode
l
c
a
n
a
c
hi
e
ve
s
c
a
la
bl
e
de
pt
h
a
nd
w
id
th
w
it
h a
c
c
ur
a
c
y a
nd c
om
put
a
ti
ona
l
e
f
f
ic
ie
nc
y.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
D
C
N
N
V
A
:
a de
e
p c
onv
ol
ut
io
nal
ne
ur
al
ne
tw
o
r
k
f
or
v
ol
c
ani
c
ac
ti
v
it
y
c
la
s
s
if
ic
at
io
n
…
(
Y
as
ir
H
u
s
s
e
in
Shak
ir
)
287
3.4.4
.
D
e
n
s
e
N
e
t
121
D
e
ns
e
N
e
t1
21
,
i
nt
r
oduc
e
d
by
H
ua
ng
e
t
al
.
[
22]
,
c
onne
c
ts
e
a
c
h
ne
twor
k
la
ye
r
w
it
h
a
ll
ot
he
r
ne
twor
k
la
ye
r
s
f
e
e
d
-
f
or
w
a
r
dl
y
a
nd
pr
om
ot
e
s
f
e
a
tu
r
e
r
e
us
e
a
nd
e
f
f
ic
ie
nt
gr
a
di
e
nt
f
lo
w
.
T
hi
s
de
ns
e
c
onn
e
c
ti
on
m
in
im
iz
e
s
r
e
dunda
nc
y
a
nd
im
pr
ove
s
le
a
r
ni
ng
e
f
f
ic
ie
nc
y
.
I
t
a
ls
o
pr
om
ot
e
s
th
e
a
bi
li
ty
o
f
th
e
m
ode
l
to
id
e
nt
if
y
s
ubt
le
vol
c
a
ni
c
f
e
a
tu
r
e
s
b
a
s
e
d on
s
a
te
ll
it
e
da
ta
.
3.4.5
.
M
ob
il
e
N
e
t
M
obi
le
N
e
t
,
pr
e
s
e
nt
e
d by How
a
r
d
e
t
al
.
[
23]
,
i
s
a
l
ig
ht
w
e
ig
ht
C
N
N
a
r
c
hi
te
c
tu
r
e
de
s
ig
ne
d s
pe
c
if
ic
a
ll
y
f
or
m
obi
le
a
nd
e
m
be
dde
d
vi
s
io
n
a
ppl
ic
a
ti
ons
.
B
y
us
in
g
de
pt
h
w
is
e
s
e
pa
r
a
bl
e
c
onvolut
io
ns
a
nd
s
ub
s
ta
nt
ia
ll
y
de
c
r
e
a
s
e
s
c
om
put
a
ti
on
a
nd
m
e
m
or
y
r
e
qui
r
e
m
e
nt
s
.
It
m
a
in
ta
i
ns
a
c
c
ur
a
c
y
,
m
a
ki
ng
it
a
li
ke
ly
c
a
ndi
da
te
f
or
r
e
a
l
-
ti
m
e
vol
c
a
ni
c
a
c
ti
vi
ty
m
oni
to
r
in
g.
3.4.6
.
I
n
c
e
p
t
io
n
V
3
I
nc
e
pt
io
nV
3
,
pr
opos
e
d
by
S
z
e
ge
dy
e
t
al
.
[
24]
,
im
p
r
ove
s
th
e
e
f
f
ic
ie
nc
y
of
C
N
N
s
us
in
g
f
a
c
to
r
iz
e
d
c
onvolut
io
ns
a
nd dim
e
ns
io
n r
e
duc
ti
on me
th
ods
of
t
he
in
c
e
pt
io
n m
odul
e
s
. I
ts
a
r
c
hi
te
c
tu
r
e
e
na
bl
e
s
t
he
ne
twor
k
to
a
c
hi
e
ve
m
ul
ti
-
s
c
a
le
f
e
a
tu
r
e
c
a
pt
ur
e
a
t
onc
e
.
I
t
a
ls
o
e
nha
nc
e
s
vol
c
a
ni
c
im
a
g
e
di
ve
r
s
it
y
-
ba
s
e
d
r
e
c
ogni
ti
on pe
r
f
or
m
a
nc
e
.
3.4.7
.
X
c
e
p
t
io
n
X
c
e
pt
io
n
,
s
ta
te
d
by
C
hol
le
t
[
25]
,
e
xt
e
nds
th
e
I
nc
e
pt
io
n
a
r
c
hi
te
c
tu
r
e
vi
a
r
e
pl
a
c
in
g
in
c
e
pt
io
n
m
odul
e
s
w
it
h de
pt
hw
is
e
s
e
pa
r
a
bl
e
c
onvolut
io
ns
. T
hi
s
s
tr
uc
tu
r
e
de
c
oupl
e
s
s
pa
ti
a
l
a
nd c
ha
nne
l
-
w
is
e
f
il
te
r
in
g
. I
t
le
a
d
s
to
im
pr
ove
d r
e
pr
e
s
e
nt
a
ti
ona
l
c
a
pa
c
it
y a
nd
e
f
f
ic
ie
nt
t
r
a
in
in
g.
3.4.8
.
V
G
G
19 an
d
V
G
G
16
V
G
G
s
tr
uc
tu
r
e
d
by
S
im
onya
n
a
nd
Z
is
s
e
r
m
a
n
[
26]
(
V
G
G
16
a
nd
V
G
G
19)
a
r
e
di
s
ti
ngui
s
he
d
by
th
e
i
r
pl
a
in
ne
s
s
a
nd c
ons
is
te
nt
a
r
c
hi
te
c
tu
r
e
a
nd
a
r
e
ba
s
e
d
on
s
uc
c
e
s
s
i
ve
c
onvolut
io
na
l
la
ye
r
s
w
it
h
ti
ny
(
3×
3)
f
il
te
r
s
.
A
lt
hough
de
e
p
,
th
e
s
e
m
ode
ls
a
r
e
pow
e
r
f
ul
in
a
ll
im
a
ge
c
la
s
s
if
ic
a
ti
on
ta
s
ks
.
T
he
ir
s
im
pl
e
a
r
c
hi
te
c
tu
r
e
a
ls
o
m
a
ke
s
t
r
a
ns
f
e
r
l
e
a
r
ni
ng f
e
a
s
ib
le
f
or
de
te
c
ti
ng volc
a
ni
c
a
c
ti
vi
ty
.
3.
5.
E
val
u
at
io
n
p
e
r
f
or
m
an
c
e
T
he
tr
a
in
e
d
m
ode
ls
w
e
r
e
a
s
s
e
s
s
e
d
on
th
e
te
s
ti
ng
da
ta
s
e
t
us
in
g
s
e
ve
r
a
l
pe
r
f
or
m
a
nc
e
m
e
a
s
ur
e
s
,
in
c
lu
di
ng
a
c
c
ur
a
c
y,
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
F
1
-
s
c
or
e
.
T
h
e
f
or
m
ul
a
s
f
or
a
ll
pe
r
f
or
m
a
nc
e
v
a
lu
e
s
a
r
e
pr
e
s
e
nt
e
d
in
(
6)
-
(
9
)
. T
he
s
e
va
lu
e
s
w
e
r
e
c
om
put
e
d u
s
in
g t
he
c
onf
us
io
n m
a
tr
i
x s
how
n i
n
T
a
bl
e
4
.
=
+
+
+
+
(
6)
=
+
(
7)
=
+
(
8)
1
−
=
2
×
×
+
(
9)
T
a
bl
e
4
.
S
tr
uc
tu
r
e
of
a
c
onf
us
io
n m
a
tr
ix
f
or
bi
na
r
y c
la
s
s
if
ic
a
ti
on
A
c
t
ua
l
pos
i
t
i
ve
A
c
t
ua
l
ne
ga
t
i
ve
P
r
e
di
c
t
e
d
pos
i
t
i
ve
T
r
ue
pos
i
t
i
v
e
(
T
P
)
F
a
l
s
e
pos
i
t
i
v
e
(
F
P
)
P
r
e
di
c
t
e
d
ne
ga
t
i
ve
F
a
l
s
e
ne
ga
t
i
ve
(
F
N
)
T
r
ue
ne
ga
t
i
ve
(
T
N
)
3.6.
D
e
p
lo
ym
e
n
t
m
od
e
l
T
o
f
a
c
il
it
a
te
pr
a
c
ti
c
a
l
ut
il
iz
a
ti
on
a
nd
de
m
ons
tr
a
te
th
e
r
e
a
l
-
w
or
ld
a
ppl
ic
a
bi
li
ty
of
th
e
pr
opos
e
d
D
C
N
N
V
A
m
ode
l
,
a
s
ta
nda
lo
ne
de
s
kt
op
a
ppl
ic
a
ti
on
w
a
s
de
ve
lo
pe
d.
T
hi
s
s
ys
te
m
to
ol
br
id
ge
s
th
e
ga
p
be
twe
e
n
e
xpe
r
im
e
nt
a
l
va
li
da
ti
on
a
nd
th
e
e
nd
-
us
e
r
a
ppl
ic
a
ti
on
.
I
t
pr
ovi
de
s
a
n
in
tu
it
iv
e
pl
a
tf
or
m
f
or
vo
lc
a
ni
c
a
c
ti
vi
ty
a
s
s
e
s
s
m
e
nt
.
4.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
he
e
xp
e
r
im
e
nt
a
l
r
e
s
ul
ts
c
l
e
a
r
ly
in
di
c
a
te
th
a
t
th
e
ne
w
D
C
N
N
V
A
m
ode
l
gr
e
a
tl
y
s
ur
pa
s
s
e
d
a
ll
tr
a
ns
f
e
r
le
a
r
ni
ng
ne
twor
ks
w
it
h
a
m
a
xi
m
um
a
c
c
ur
a
c
y
99.33
%
a
nd
a
lm
os
t
pe
r
f
e
c
t
pr
e
c
i
s
io
n
100%
,
w
hi
c
h
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
. 15, No. 1, M
a
r
c
h 2026
:
281
-
292
288
c
le
a
r
ly
s
how
s
th
a
t
it
c
a
n
a
c
c
ur
a
te
ly
id
e
nt
if
y
vol
c
a
ni
c
a
c
ti
vi
ty
.
I
ts
s
tr
ong
a
bi
li
ty
to
s
uppr
e
s
s
f
a
ls
e
ne
ga
ti
ve
s
i
s
f
ur
th
e
r
c
onf
ir
m
e
d
by
it
s
hi
gh
r
e
c
a
ll
va
lu
e
98.67%
,
w
hi
c
h
is
ve
r
y
im
por
ta
nt
in
e
a
r
ly
w
a
r
ni
ngs
r
e
ga
r
di
ng
vol
c
a
ni
c
ha
z
a
r
ds
.
H
ow
e
ve
r
,
w
e
c
a
n
s
e
e
th
a
t
R
e
s
N
e
t5
0
pe
r
f
or
m
e
d
poor
ly
w
it
h
a
v
e
r
y
lo
w
a
c
c
ur
a
c
y
le
v
e
l
of
onl
y
68%
,
w
hi
c
h
w
e
a
tt
r
ib
u
te
to
th
e
f
a
c
t
th
a
t
it
s
de
e
pe
r
r
e
s
id
ua
l
uni
ts
a
r
e
not
opt
im
a
l
f
o
r
th
is
da
ta
s
e
t.
M
obi
le
N
e
t,
D
e
ns
e
N
e
t1
21, a
nd I
nc
e
pt
io
nV
3 pe
r
f
or
m
e
d e
qua
ll
y
w
e
ll
(
95
-
96%
a
c
c
ur
a
te
)
, but
f
a
il
e
d t
o r
e
a
c
h t
h
e
a
c
c
ur
a
c
y
of
th
e
D
C
N
N
V
A
.
O
th
e
r
ol
de
r
de
s
ig
ns
,
s
uc
h
a
s
V
G
G
16
a
nd
V
G
G
19
,
pe
r
f
or
m
e
d
lo
w
e
r
in
a
c
c
ur
a
c
y
le
ve
ls
c
om
pa
r
e
d t
o ne
w
a
r
c
hi
te
c
tu
r
e
de
s
ig
ns
.
A
s
um
m
a
r
y of
t
he
r
e
s
ul
ts
i
s
pr
ovi
de
d i
n F
ig
ur
e
5.
T
o
be
tt
e
r
il
lu
s
tr
a
te
th
e
c
la
s
s
if
ic
a
ti
on
pe
r
f
or
m
a
nc
e
,
F
ig
ur
e
6
pl
ot
s
th
e
c
onf
us
io
n
m
a
tr
ic
e
s
of
a
ll
a
lg
or
it
hm
s
.
T
he
s
e
c
ont
a
in
a
gr
a
phi
c
a
l
r
e
pr
e
s
e
nt
a
ti
on
of
tr
ue
pos
it
iv
e
s
,
tr
ue
ne
ga
ti
ve
s
,
f
a
ls
e
pos
it
iv
e
s
,
a
nd
f
a
ls
e
ne
ga
ti
ve
s
w
it
hi
n
th
e
“
vol
c
a
ni
c
a
c
ti
vi
ty
”
a
nd
“
no
a
c
ti
vi
t
y
”
c
la
s
s
e
s
.
A
na
ly
s
is
of
th
e
c
onf
us
io
n
m
a
tr
ix
c
onf
ir
m
s
th
e
s
upe
r
io
r
it
y
of
th
e
D
C
N
N
V
A
m
ode
l
,
w
hi
c
h
c
or
r
e
c
tl
y
c
la
s
s
if
ie
d
a
lm
os
t
a
ll
s
a
m
pl
e
s
but
c
ont
a
in
e
d
a
ne
gl
ig
ib
le
num
be
r
of
f
a
ls
e
ne
ga
ti
ve
4
c
a
s
e
s
of
vol
c
a
ni
c
a
c
ti
vi
ty
m
is
c
la
s
s
if
ie
d
a
s
no
a
c
ti
vi
ty
.
C
ont
r
a
r
y
to
th
is
,
num
e
r
ous
m
is
c
la
s
s
if
ic
a
ti
ons
w
e
r
e
w
it
ne
s
s
e
d
in
R
e
s
N
e
t5
0, pr
im
a
r
il
y
th
e
a
bs
e
nc
e
of
de
t
e
c
ti
on
of
vol
c
a
ni
c
a
c
ti
vi
ty
in
a
m
a
jo
r
it
y
of
c
a
s
e
s
.
M
obi
le
N
e
t,
D
e
n
s
e
N
e
t1
21,
a
nd
I
nc
e
pt
io
nV
3
pe
r
f
or
m
e
d
out
s
ta
ndi
ngl
y
but
w
it
h
a
s
li
ght
ly
hi
ghe
r
m
is
c
la
s
s
if
ic
a
ti
on
r
a
te
c
om
pa
r
e
d
to
D
C
N
N
V
A
.
T
he
gr
a
ph
of
tr
a
in
in
g
a
nd
v
a
li
da
ti
on
a
c
c
ur
a
c
y
of
D
C
N
N
V
A
f
or
te
n
e
po
c
hs
i
s
s
ho
w
n
in
F
ig
ur
e
7.
F
r
om
th
e
gr
a
ph,
th
e
a
c
c
ur
a
c
y
of
th
e
m
ode
l
in
c
r
e
a
s
e
s
ve
r
y
f
a
s
t
a
t
e
a
c
h e
po
c
h
,
but
s
lo
w
s
dow
n f
r
om
e
poc
h 4 up to t
he
l
a
s
t
e
poc
h.
F
ig
ur
e
5.
A
c
c
ur
a
c
y
c
om
pa
r
is
on of
t
he
pr
opos
e
d
D
C
N
N
V
A
m
o
de
l
a
ga
in
s
t
s
ta
t
e
-
of
-
th
e
-
a
r
t
F
ig
ur
e
6. C
onf
us
io
n
m
a
tr
ic
e
s
c
om
pa
r
in
g t
he
c
la
s
s
if
ic
a
ti
on pe
r
f
or
m
a
nc
e
of
D
C
N
N
V
A
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
D
C
N
N
V
A
:
a de
e
p c
onv
ol
ut
io
nal
ne
ur
al
ne
tw
o
r
k
f
or
v
ol
c
ani
c
ac
ti
v
it
y
c
la
s
s
if
ic
at
io
n
…
(
Y
as
ir
H
u
s
s
e
in
Shak
ir
)
289
F
ig
ur
e
7.
D
C
N
N
V
A
tr
a
in
in
g a
nd va
li
da
ti
on pe
r
f
or
m
a
nc
e
ove
r
10
e
poc
hs
T
he
s
upe
r
io
r
pe
r
f
or
m
a
nc
e
of
t
he
D
C
N
N
V
A
m
ode
l
w
a
r
r
a
nt
e
d i
t
s
t
r
a
ns
it
io
n f
r
om
a
r
e
s
e
a
r
c
h pr
ot
ot
ype
to
a
pr
a
c
ti
c
a
l
to
ol
.
T
o
th
is
e
nd,
a
f
unc
ti
ona
l
de
s
kt
op
a
ppl
ic
a
ti
on
w
a
s
de
ve
lo
pe
d
a
nd
de
pl
oye
d.
T
hi
s
a
ppl
ic
a
ti
on
pr
ovi
de
s
a
us
e
r
-
f
r
ie
ndl
y
in
te
r
f
a
c
e
th
a
t
a
ll
ow
s
e
nd
-
u
s
e
r
s
,
s
uc
h
a
s
g
e
ol
ogi
s
ts
or
m
oni
to
r
in
g
s
ta
ti
on
pe
r
s
onne
l,
to
pe
r
f
or
m
r
e
a
l
-
ti
m
e
vol
c
a
ni
c
a
c
ti
vi
ty
a
s
s
e
s
s
m
e
nt
s
.
T
he
de
pl
oym
e
nt
s
uc
c
e
s
s
f
ul
ly
de
m
ons
tr
a
te
s
th
e
m
ode
l'
s
ope
r
a
ti
ona
l
vi
a
bi
li
ty
.
A
s
s
how
n
in
th
e
a
ppl
ic
a
ti
on
i
nt
e
r
f
a
c
e
F
ig
ur
e
8,
u
s
e
r
s
c
a
n
upl
oa
d
s
a
te
ll
it
e
im
a
ge
r
y,
a
nd
th
e
s
ys
te
m
r
e
tu
r
ns
a
n
in
s
ta
nt
a
ne
ou
s
c
la
s
s
if
ic
a
ti
o
n
(
"
a
c
ti
ve
"
or
"
no
a
c
ti
ve
"
)
a
c
c
om
pa
ni
e
d
by
a
c
onf
id
e
nc
e
s
c
or
e
.
T
he
in
te
r
f
a
c
e
c
le
a
r
ly
di
s
pl
a
ys
th
e
pr
e
di
c
ti
on,
f
or
in
s
ta
nc
e
,
"
pr
e
di
c
ti
on:
no
a
c
ti
ve
|
c
onf
id
e
nc
e
:
99.30%
"
(
F
ig
ur
e
8(
a
)
)
or
"
pr
e
di
c
ti
on:
a
c
ti
ve
|
c
onf
id
e
nc
e
:
99.90%
"
(
F
ig
ur
e
8
(
b)
)
,
pr
ovi
di
ng
tr
a
ns
pa
r
e
nt
a
nd
im
m
e
di
a
te
r
e
s
ul
ts
to
th
e
op
e
r
a
to
r
.
C
r
uc
ia
ll
y,
th
e
a
ppl
ic
a
ti
on
in
c
or
por
a
te
s
a
m
ul
ti
-
m
oda
l
a
le
r
t
s
ys
te
m
th
a
t,
upon
de
te
c
ti
ng
"
a
c
ti
ve
"
vol
c
a
ni
c
a
c
ti
vi
ty
,
tr
ig
ge
r
s
a
c
le
a
r
,
s
ynt
he
s
iz
e
d
voi
c
e
w
a
r
ni
ng:
"
w
a
r
ni
ng!
vol
c
a
ni
c
a
c
ti
vi
ty
de
t
e
c
te
d
."
T
hi
s
f
e
a
tu
r
e
i
s
de
s
ig
ne
d
to
c
a
pt
ur
e
th
e
ope
r
a
to
r
'
s
a
tt
e
nt
io
n
im
m
e
di
a
te
ly
,
w
hi
c
h
is
pa
r
a
m
ount
in
hi
gh
-
s
ta
ke
s
m
oni
to
r
in
g
e
nvi
r
onm
e
nt
s
.
T
he
s
uc
c
e
s
s
f
ul
in
te
gr
a
ti
on
of
th
e
hi
gh
-
a
c
c
ur
a
c
y
D
C
N
N
V
A
m
ode
l
in
to
th
is
de
pl
oya
bl
e
s
ys
te
m
und
e
r
s
c
or
e
s
a
nd
a
ls
o
r
e
a
di
ne
s
s
f
or
us
e
in
qua
s
i
-
r
e
a
l
-
ti
m
e
de
c
is
io
n
-
s
uppor
t
s
c
e
na
r
io
s
,
e
f
f
e
c
ti
ve
ly
br
id
gi
ng
th
e
ga
p
b
e
twe
e
n
th
e
or
e
ti
c
a
l
m
ode
l
pe
r
f
or
m
a
nc
e
a
nd
pr
a
c
ti
c
a
l,
on
-
th
e
-
gr
ound uti
li
ty
.
(
a
)
(
b)
F
ig
ur
e
8. D
C
N
N
V
A
a
ppl
ic
a
ti
on i
nt
e
r
f
a
c
e
f
or
vol
c
a
ni
c
a
c
ti
vi
ty
c
la
s
s
if
ic
a
ti
on
of
(
a
)
"
no a
c
ti
ve
"
(
b)
"
a
c
ti
ve
"
F
ur
th
e
r
m
or
e
,
pe
r
f
or
m
a
nc
e
a
nd
pr
a
c
ti
c
a
l
im
pl
e
m
e
nt
a
ti
on
of
th
e
pr
opos
e
d
D
C
N
N
V
A
m
od
e
l
a
r
e
c
om
pa
r
e
d
to
c
ont
e
m
por
a
r
y
li
te
r
a
tu
r
e
r
e
vi
e
w
in
T
a
bl
e
5
.
T
he
a
na
ly
s
is
r
e
ve
a
l
s
th
a
t
,
how
e
ve
r
va
r
io
us
s
tu
di
e
s
ha
ve
a
c
hi
e
v
e
d
hi
gh
a
c
c
ur
a
c
y
,
our
w
or
k
di
s
ti
ngui
s
he
s
two
c
r
it
ic
a
l
a
r
e
a
s
:
i
)
a
c
hi
e
vi
ng
th
e
b
e
s
t
ov
e
r
a
ll
pe
r
f
or
m
a
nc
e
a
c
r
os
s
m
ul
ti
pl
e
m
e
tr
ic
s
a
nd
ii
)
s
uc
c
e
s
s
f
ul
ly
br
id
gi
ng
th
e
ga
p
to
a
pr
a
c
ti
c
a
l
de
pl
oya
bl
e
to
ol
.
T
hi
s
pr
a
c
ti
c
a
l
to
ol
f
e
a
tu
r
e
s
a
us
e
r
-
f
r
ie
ndl
y
G
U
I
f
or
r
e
a
l
-
t
im
e
a
na
ly
s
i
s
a
nd
in
c
or
por
a
te
s
a
uni
que
m
ul
ti
-
m
oda
l
a
le
r
t
s
ys
te
m
th
a
t
pr
ovi
de
s
im
m
e
di
a
te
a
udi
to
r
y
w
a
r
ni
ngs
upon
de
te
c
ti
on
of
vol
c
a
ni
c
a
c
ti
vi
ty
.
T
hi
s
c
om
bi
na
ti
on
of
s
ta
te
-
of
-
th
e
-
a
r
t
pr
e
di
c
ti
ve
pe
r
f
or
m
a
nc
e
a
nd
a
f
unc
ti
ona
l
a
tt
e
nt
i
on
-
gr
a
bbi
ng
de
pl
oym
e
nt
pl
a
tf
or
m
r
e
pr
e
s
e
nt
s
a
s
ig
ni
f
ic
a
nt
c
ont
r
ib
ut
io
n t
o t
he
f
ie
ld
of
ope
r
a
ti
ona
l
vol
c
a
ni
c
ha
z
a
r
d m
oni
to
r
in
g.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
. 15, No. 1, M
a
r
c
h 2026
:
281
-
292
290
T
a
bl
e
5
. C
om
pa
r
a
ti
ve
a
na
ly
s
is
of
vol
c
a
ni
c
a
c
ti
vi
ty
de
te
c
ti
on me
th
ods
R
e
f
e
r
e
nc
e
M
e
t
hodol
ogy
P
e
r
f
or
m
a
nc
e
(
%
)
D
e
pl
oym
e
nt
A
l
e
r
t
s
ys
t
e
m
C
or
r
a
di
no
e
t
al
.
[
12]
C
N
N
(
U
N
E
T
)
A
c
c
ur
a
c
y
=93
N
ot
r
e
por
t
e
d
N
ot
r
e
por
t
e
d
S
hul
t
z
[
13]
C
N
N
(
H
ot
L
I
N
K
)
A
c
c
ur
a
c
y
=98
N
ot
r
e
por
t
e
d
N
ot
r
e
por
t
e
d
O
ña
t
e
e
t
al
.
[
14]
A
udi
o f
e
a
t
ur
e
s
+
m
a
ni
f
ol
d l
e
a
r
ni
ng
A
c
c
ur
a
c
y
=94.44 to 95.45
N
ot
r
e
por
t
e
d
N
ot
r
e
por
t
e
d
N
unna
r
i
a
nd
C
a
l
va
r
i
[
15]
C
om
pa
r
i
s
on of
8 C
N
N
s
A
c
c
ur
a
c
y
=94.07
N
ot
r
e
por
t
e
d
N
ot
r
e
por
t
e
d
C
he
n
e
t
al
.
[
16]
T
V
G
G
A
c
c
ur
a
c
y
=99.18
Re
c
a
l
l
=99.17
N
ot
r
e
por
t
e
d
N
ot
r
e
por
t
e
d
M
oha
n
e
t
al
.
[
17]
L
R
X
+C
N
N
A
c
c
ur
a
c
y
=90.30
F1
-
s
c
or
e
=88.40
N
ot
r
e
por
t
e
d
N
ot
r
e
por
t
e
d
H
ue
r
t
a
s
e
t
al
.
[
18]
V
G
G
16 a
nd I
nc
e
pt
i
on C
N
N
A
c
c
ur
a
c
y
=93
Pr
e
c
i
s
i
on
=93
N
ot
r
e
por
t
e
d
N
ot
r
e
por
t
e
d
O
ur
w
or
k
P
r
opos
e
d D
C
N
N
V
A
A
c
c
ur
a
c
y
=99.33
Pr
e
c
i
s
i
on
=100.
Re
c
a
l
l
=98.67
F1
-
s
c
or
e
=99.33
Y
e
s
(
f
unc
t
i
ona
l
de
s
kt
op
a
ppl
i
c
a
t
i
on
)
Y
e
s
(
t
e
xt
-
to
-
s
pe
e
c
h a
udi
o
a
l
a
r
m
)
5.
C
O
N
C
L
U
S
I
O
N
I
n
th
is
s
tu
dy,
a
de
ve
lo
p
ed
D
C
N
N
V
A
c
la
s
s
if
ic
a
ti
on
w
a
s
r
ig
or
ous
ly
va
li
da
te
d
a
nd
s
uc
c
e
s
s
f
ul
ly
de
pl
oye
d.
T
he
m
ode
l
w
a
s
e
va
lu
a
t
e
d
a
ga
in
s
t
e
ig
ht
s
ta
t
e
-
of
-
th
e
-
a
r
t
tr
a
ns
f
e
r
le
a
r
ni
ng
a
r
c
hi
te
c
tu
r
e
s
,
in
c
lu
di
ng
R
e
s
N
e
t5
0,
N
A
S
N
e
tL
a
r
ge
,
D
e
ns
e
N
e
t1
21,
M
obi
le
N
e
t,
I
nc
e
pt
i
onV
3,
X
c
e
pt
io
n,
V
G
G
16,
a
nd
V
G
G
19.
T
he
e
xpe
r
im
e
nt
a
l
r
e
s
ul
ts
de
m
ons
tr
a
te
th
a
t
pr
opos
e
d
D
C
N
N
V
A
m
ode
l
c
ons
is
te
nt
ly
out
pe
r
f
or
m
e
d
a
ll
m
ode
l
ne
twor
ks
a
c
r
os
s
a
ll
e
va
lu
a
ti
on
m
e
tr
ic
s
,
a
c
hi
e
vi
ng
a
c
c
ur
a
c
y
(
99.
33%
)
,
pr
e
c
is
io
n
(
100
%
)
,
r
e
c
a
ll
(
98.67%
)
,
a
nd
F1
-
s
c
or
e
(
99.33%
)
.
T
he
c
om
pr
e
he
ns
iv
e
a
na
ly
s
is
s
uppor
te
d
by
c
onf
us
io
n
m
a
tr
ic
e
s
a
nd
tr
a
in
in
g
gr
a
phs
c
onf
ir
m
s
th
e
m
ode
l'
s
r
obus
t
c
a
pa
bi
li
ty
to
m
in
im
iz
e
bot
h
f
a
ls
e
pos
it
iv
e
s
a
nd
f
a
l
s
e
ne
g
a
ti
ve
s
,
w
it
h
a
pa
r
ti
c
ul
a
r
ly
s
tr
ong
pe
r
f
or
m
a
nc
e
in
r
e
duc
in
g
f
a
ls
e
ne
ga
ti
ve
s
,
a
c
r
it
ic
a
l
r
e
qui
r
e
m
e
nt
f
or
e
a
r
ly
w
a
r
ni
ng
s
ys
te
m
s
in
vol
c
a
ni
c
ha
z
a
r
d
m
oni
to
r
in
g.
B
e
yond
th
e
or
e
ti
c
a
l
pe
r
f
o
r
m
a
n
c
e
,
th
is
r
e
s
e
a
r
c
h
m
a
ke
s
a
s
ub
s
ta
nt
ia
l
pr
a
c
ti
c
a
l
c
ont
r
ib
ut
io
n
th
r
ough
th
e
s
uc
c
e
s
s
f
ul
de
ve
lo
pm
e
nt
a
nd
d
e
pl
oym
e
nt
of
a
n
op
e
r
a
ti
ona
l
de
s
kt
op
a
ppl
ic
a
ti
on
,
a
nd
im
pl
e
m
e
nt
a
ti
on r
e
pr
e
s
e
nt
s
a
s
ig
ni
f
ic
a
nt
a
dva
nc
e
m
e
nt
be
yond c
ur
r
e
nt
s
ta
te
-
of
-
th
e
-
a
r
t
a
ppr
oa
c
he
s
,
br
id
gi
ng t
he
ga
p
be
twe
e
n
e
xpe
r
im
e
nt
a
l
m
ode
ls
a
nd
pr
a
c
ti
c
a
l
ut
il
it
y.
T
h
e
a
ppl
ic
a
ti
on
f
e
a
tu
r
e
s
a
n
in
tu
it
iv
e
gr
a
phi
c
a
l
in
te
r
f
a
c
e
f
or
r
e
a
l
-
ti
m
e
m
oni
to
r
in
g
a
nd
in
c
or
por
a
te
s
a
pi
one
e
r
in
g
m
ul
ti
m
oda
l
a
le
r
t
s
ys
te
m
th
a
t
pr
ovi
de
s
im
m
e
di
a
te
a
udi
to
r
y
w
a
r
ni
ngs
upon
de
te
c
ti
on
of
vol
c
a
ni
c
a
c
ti
vi
ty
,
a
f
e
a
tu
r
e
a
bs
e
nt
s
in
c
om
pa
r
a
bl
e
s
tu
di
e
s
.
F
ut
ur
e
w
or
k
w
il
l
a
im
to
e
xt
e
nd
th
e
c
ur
r
e
nt
a
r
c
hi
te
c
tu
r
e
to
s
uppor
t
m
ul
ti
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on
of
vol
c
a
ni
c
a
c
ti
vi
ty
,
a
ll
ow
in
g
f
or
m
or
e
nua
nc
e
d
r
e
c
ogni
ti
on
of
di
f
f
e
r
e
nt
e
r
upt
io
n
ty
pe
s
,
a
nd
th
e
f
r
a
m
e
w
or
k
w
il
l
be
e
nha
nc
e
d
th
r
ough
th
e
in
te
gr
a
ti
on
o
f
m
ul
ti
m
oda
l
da
ta
,
c
om
bi
ni
ng
in
f
or
m
a
ti
on
f
r
om
s
our
c
e
s
s
uc
h
a
s
s
a
te
ll
it
e
s
a
nd
gr
ound
-
ba
s
e
d
s
e
is
m
ic
s
e
n
s
or
s
.
F
in
a
ll
y,
e
f
f
or
ts
w
il
l
f
oc
us
on
de
pl
oyi
ng
th
is
s
ys
te
m
in
ope
r
a
ti
ona
l
m
oni
to
r
in
g ne
twor
ks
t
o s
tr
e
ngt
he
n e
a
r
ly
w
a
r
ni
ng s
ys
te
m
s
a
nd i
m
pr
ove
di
s
a
s
te
r
r
e
s
pons
e
s
tr
a
te
gi
e
s
.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
A
ut
hor
s
s
ta
te
no f
undi
ng i
nvol
ve
d.
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
di
T
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on.
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Y
a
s
ir
H
us
s
e
in
S
ha
ki
r
✓
✓
✓
✓
✓
✓
✓
✓
✓
R
e
e
m
A
li
M
ut
la
g
✓
✓
✓
✓
✓
✓
E
s
ha
q A
z
iz
A
w
a
dh A
L
M
a
ndha
r
i
✓
✓
✓
✓
✓
✓
✓
M
oha
m
e
d S
ha
bbi
r
A
bdul
na
bi
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
l
i
z
a
t
i
on
M
:
M
e
t
hodol
ogy
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
da
t
i
on
Fo
:
Fo
r
m
a
l
a
na
l
ys
i
s
I
:
I
nve
s
t
i
ga
t
i
on
R
:
R
e
s
our
c
e
s
D
:
D
a
t
a
C
ur
a
t
i
on
O
:
W
r
i
t
i
ng
-
O
r
i
gi
na
l
D
r
a
f
t
E
:
W
r
i
t
i
ng
-
R
e
vi
e
w
&
E
di
t
i
ng
Vi
:
Vi
s
ua
l
i
z
a
t
i
on
Su
:
Su
pe
r
vi
s
i
on
P
:
P
r
oj
e
c
t
a
dm
i
ni
s
t
r
a
t
i
on
Fu
:
Fu
ndi
ng a
c
qui
s
i
t
i
on
Evaluation Warning : The document was created with Spire.PDF for Python.