I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
14
, N
o.
4
,
A
ugus
t
2025
, pp.
3160
~
3171
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
4
.pp
3160
-
3171
3160
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
D
e
e
p
t
r
an
sf
e
r
l
e
a
r
n
i
n
g f
or
c
l
ass
i
f
i
c
at
i
on
of
E
C
G
si
gn
al
s a
n
d
l
i
p
i
m
age
s i
n
m
u
l
t
i
m
o
d
al
b
i
o
m
e
t
r
i
c
au
t
h
e
n
t
i
c
at
i
on
sys
t
e
m
s
L
at
h
a K
r
is
h
n
am
oor
t
h
y, A
m
m
as
an
d
r
a S
ad
as
h
iv
ai
ah
R
aj
u
D
e
pa
r
t
m
e
nt
o
f
B
i
o
-
M
e
di
c
a
l
E
ngi
ne
e
r
i
n
g, S
r
i
S
i
d
dha
r
t
ha
I
ns
t
i
t
ut
e
o
f
T
e
c
hno
l
ogy,
S
r
i
S
i
ddha
r
t
ha
A
c
a
de
m
y o
f
H
i
ghe
r
E
duc
a
t
i
on U
n
i
ve
r
s
i
t
y,
T
um
a
kur
u, I
ndi
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
M
a
r
26, 2024
R
e
vi
s
e
d
M
a
r
28, 2025
A
c
c
e
pt
e
d
J
un 8, 2025
Authentication
plays
an
essential
role
in
diverse
kinds
of
applicati
on
that
requires
security.
Several
authenticati
on
methods
have
been
develop
ed,
but
biometric
authentication
has
gained
huge
attention
from
the
re
search
communi
ty
and
indust
ries
due
to
its
reliabilit
y
and
robustness
.
This
study
investigates
multimodal
authentica
tion
techniques
utilizing
electroc
ardiogra
m
(ECG)
signals
and
face
lip
images.
Leveraging
transfer
learning
fro
m
pre
-
trained
ResNet
and
VGG16
models,
ECG
signals
and
photos
of
the
lip
area
of
the
face
are
used
to
extract
characteristics.
Subsequently,
a
convol
utional
neural
network
(CNN)
classifi
er
is
employed
for
classifi
cation
based
on
the
extracted
features.
The
dataset
used
in
this
study
comprises
ECG
sign
als
and
face
lip
images,
representing
distinct
biometric
modalities.
Throu
gh
the
integration
of
transfer
learning
and
CNN
classifica
tion,
improving
the
reliabilit
y
and
precision
of
multim
odal
authenticati
on
systems
is
the
p
rimary
objective of the study. Verification results
show th
at the
suggested me
thod is
successful
in
producing
trustworthy
authentication
using
multi
modal
biometric
traits.
The
experimental
analysis
shows
that
the
proposed
deep
transfer
learning
-
based
model
has
reported
the
average
accuracy,
F1
-
score,
precision, an
d recall as 0.962, 0.970, 0.965, and 0.966, respe
ctively.
K
e
y
w
o
r
d
s
:
B
io
m
e
tr
ic
a
ut
he
nt
ic
a
ti
on
C
la
s
s
if
ic
a
ti
on
D
e
e
p l
e
a
r
ni
ng
E
le
c
tr
oc
a
r
di
ogr
a
m
M
ul
ti
m
oda
l
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
:
L
a
th
a
K
r
is
hna
m
oor
th
y
D
e
pa
r
tm
e
nt
of
B
io
-
M
e
di
c
a
l
E
ngi
ne
e
r
in
g, S
r
i
S
id
dha
r
th
a
I
ns
ti
tu
t
e
of
T
e
c
hnol
ogy
S
r
i
S
id
dha
r
th
a
A
c
a
de
m
y of
H
ig
he
r
E
duc
a
ti
on U
ni
ve
r
s
it
y
T
um
a
kur
u, I
ndi
a
E
m
a
il
:
la
th
a
k@
s
s
it
.e
du.i
n
1.
I
N
T
R
O
D
U
C
T
I
O
N
I
n
r
e
c
e
nt
ti
m
e
s
,
a
s
in
te
r
ne
t
of
th
in
gs
(
I
oT
)
te
c
hnol
ogy
c
ont
in
ue
s
to
a
dva
nc
e
,
th
e
ut
il
iz
a
ti
on
of
c
lo
ud
s
e
r
vi
c
e
s
ha
s
be
c
om
e
pr
e
va
le
nt
.
V
a
r
io
us
de
vi
c
e
s
a
r
e
now
e
qui
ppe
d
w
it
h
ne
twor
ki
ng
c
a
pa
bi
li
ti
e
s
to
f
a
c
il
it
a
te
c
om
m
uni
c
a
ti
on
be
twe
e
n
m
a
c
hi
ne
s
a
nd
hum
a
ns
.
C
on
s
e
que
nt
l
y,
e
ns
ur
in
g
in
f
or
m
a
ti
on
s
e
c
ur
it
y
ha
s
be
c
om
e
pa
r
a
m
ount
,
pa
r
ti
c
ul
a
r
ly
a
s
us
e
r
d
a
ta
i
s
e
m
pl
oye
d
to
gove
r
n
a
m
ul
ti
tu
de
of
de
vi
c
e
s
[
1]
.
C
onf
id
e
nt
ia
li
ty
a
nd
in
te
gr
it
y a
r
e
t
he
e
s
s
e
nt
ia
l
c
om
pone
nt
s
of
i
nf
or
m
a
ti
on s
e
c
ur
it
y.
U
s
e
r
a
ut
he
nt
ic
a
ti
on
pl
a
ys
im
por
ta
nt
r
ol
e
in
th
is
c
ont
e
xt
to
m
a
i
nt
a
in
th
e
c
onf
id
e
nt
ia
li
ty
a
nd
in
te
gr
it
y
of
th
e
da
ta
.
S
e
v
e
r
a
l
m
e
th
ods
ha
ve
be
e
n
in
tr
oduc
e
d
to
im
pr
ove
th
e
r
e
li
a
bi
li
ty
of
a
ut
he
nt
ic
a
ti
on
s
uc
h
a
s
pa
s
s
w
or
d
-
ba
s
e
d
a
ut
he
nt
ic
a
ti
on,
m
ul
ti
-
f
a
c
to
r
a
ut
he
nt
ic
a
ti
on
(
M
F
A
)
,
one
-
t
im
e
pa
s
s
w
or
d,
s
m
a
r
t
c
a
r
ds
a
nd
to
ke
n
s
.
T
he
s
e
m
e
th
ods
ha
ve
be
e
n
a
dopt
e
d
w
id
e
ly
in
va
r
io
us
a
ppl
ic
a
ti
ons
but
u
s
e
r
’
s
li
ve
li
ne
s
s
is
not
c
ons
id
e
r
e
d
in
th
e
s
e
w
or
ks
.
T
he
r
e
f
or
e
,
r
e
s
e
a
r
c
he
r
s
h
a
ve
d
e
ve
lo
pe
d
bi
om
e
tr
ic
a
ut
he
nt
ic
a
ti
o
n
s
ys
te
m
w
hi
c
h
us
e
s
s
pe
c
ia
l
bi
om
e
tr
ic
da
ta
to
id
e
nt
if
y
a
us
e
r
,
li
ke
f
in
ge
r
pr
in
ts
,
f
a
c
e
s
c
a
ns
,
ir
is
s
c
a
ns
,
or
voi
c
e
r
e
c
ogni
ti
on.
T
he
m
a
in
a
dva
nt
a
g
e
s
of
th
is
s
y
s
te
m
a
r
e
t
ha
t
it
i
s
di
f
f
ic
ul
t
to
s
poof
or
r
e
pl
ic
a
te
, of
f
e
r
s
a
hi
gh l
e
ve
l
of
s
e
c
ur
it
y, a
nd e
li
m
in
a
te
s
t
he
n
e
e
d t
o r
e
m
e
m
be
r
pa
s
s
w
or
ds
.
M
or
e
ove
r
,
th
e
bi
om
e
tr
ic
a
ut
he
nt
ic
a
ti
on
s
ys
te
m
s
ut
il
iz
e
hum
a
n
c
ha
r
a
c
te
r
is
ti
c
s
f
or
r
e
c
ogni
ti
on
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
D
e
e
p t
r
ans
fe
r
l
e
ar
ni
ng f
o
r
c
la
s
s
if
ic
at
io
n of
E
C
G
s
ig
nal
s
and li
p
i
m
age
s
i
n
…
(
L
at
ha K
r
is
hna
m
oor
th
y
)
3161
in
c
lu
di
ng
be
ha
vi
or
a
l
a
nd
phys
ic
a
l
c
ha
r
a
c
te
r
is
ti
c
s
.
T
he
be
ha
vi
o
r
a
l
c
ha
r
a
c
te
r
is
ti
c
a
na
ly
s
is
in
c
lu
de
s
voi
c
e
,
ga
it
,
a
nd
e
le
c
tr
oc
a
r
di
ogr
a
m
(
E
C
G
)
,
w
hi
le
p
hys
ic
a
l
f
e
a
tu
r
e
s
in
c
lu
de
th
e
f
a
c
e
,
f
in
ge
r
pr
in
ts
,
a
nd
ir
is
.
s
e
ve
r
a
l
w
or
ks
ha
ve
be
e
n
de
ve
lo
p
e
d
ba
s
e
d
on
th
e
s
e
bi
om
e
tr
ic
m
oda
li
ti
e
s
s
uc
h
a
s
f
in
ge
r
pr
in
t
[
2]
,
f
in
ge
r
ve
in
-
ba
s
e
d
a
ut
he
nt
ic
a
ti
on s
ys
te
m
[
1]
, f
a
c
e
[
3]
, a
nd voic
e
[
4]
. H
ow
e
ve
r
, t
r
a
di
ti
ona
l
bi
om
e
tr
ic
s
f
a
c
e
s
nume
r
ous
c
ha
ll
e
ng
e
s
,
w
hi
c
h ha
ve
s
u
s
c
e
pt
ib
il
it
y t
o s
poof
in
g or
f
or
ge
r
y
[
5]
.
M
or
e
ove
r
,
th
e
bi
om
e
tr
ic
s
ys
te
m
s
r
e
ly
on
th
e
uni
que
phys
io
lo
gi
c
a
l
a
nd
be
ha
vi
or
a
l
tr
a
it
s
of
th
e
us
e
r
s
th
e
r
e
f
or
e
it
pl
a
ys
c
r
uc
ia
l
r
ol
e
in
e
nha
nc
in
g
th
e
s
e
c
ur
it
y
a
nd
m
it
ig
a
ti
ng
th
e
vul
ne
r
a
bi
li
ti
e
s
.
A
m
ong
th
e
num
e
r
ous
bi
om
e
tr
ic
m
oda
li
ti
e
s
, c
a
r
di
a
c
ba
s
e
d bi
om
e
tr
ic
s
ys
te
m
s
h
a
ve
ga
i
ne
d huge
a
tt
e
nt
io
n i
n huma
n i
de
nt
if
ic
a
ti
on a
nd
s
e
c
ur
it
y
e
nha
nc
e
m
e
nt
.
T
h
e
in
tr
in
s
ic
e
le
c
tr
ic
a
l
a
c
ti
vi
ty
of
th
e
hum
a
n
he
a
r
t,
a
s
c
a
pt
ur
e
d
by
E
C
G
,
phot
opl
e
th
ys
m
ogr
a
m
(
P
P
G
)
,
a
nd
“
phonoc
a
r
di
ogr
a
m
(
P
C
G
)
”
s
ig
na
ls
[
6]
,
[
7]
,
o
f
f
e
r
s
a
va
lu
a
bl
e
s
our
c
e
of
in
f
or
m
a
ti
on
th
a
t
c
a
n
be
ut
il
iz
e
d
f
or
s
e
c
ur
e
a
ut
he
nt
ic
a
ti
on.
M
or
e
o
ve
r
,
a
dopt
in
g
E
C
G
s
ig
na
l
ove
r
ot
he
r
bi
om
e
tr
ic
m
oda
li
ti
e
s
ha
s
s
e
v
e
r
a
l
a
dva
nt
a
ge
s
ov
e
r
P
P
G
a
nd
P
C
G
s
ig
na
ls
.
T
he
r
e
f
or
e
,
E
C
G
is
c
ons
id
e
r
e
d
a
s
uni
que
a
nd
hi
ghl
y
in
di
v
id
ua
li
s
ti
c
bi
om
a
r
ke
r
in
m
e
di
c
a
l
dom
a
in
[
4]
be
c
a
us
e
it
pr
ovi
de
s
in
tr
ic
a
te
e
le
c
tr
ic
a
l
a
c
ti
vi
ty
pa
tt
e
r
n
of
he
a
r
t
w
hi
c
h i
s
be
ne
f
ic
ia
l
in
s
e
c
ur
it
y a
nd a
ut
h
e
nt
ic
a
ti
on s
ys
t
e
m
s
.
I
n
or
d
e
r
t
o
a
ddr
e
s
s
th
e
i
s
s
u
e
s
of
tr
a
di
t
io
n
a
l
a
u
th
e
nt
ic
a
t
io
n
s
ys
te
m
s
s
e
v
e
r
a
l
a
ut
h
or
s
h
a
v
e
r
e
por
te
d
th
e
a
dv
a
n
ta
ge
s
of
c
om
b
in
i
ng
m
ul
ti
m
od
a
l
a
u
th
e
nt
ic
a
t
io
n
s
y
s
t
e
m
s
.
S
e
v
e
r
a
l
m
od
e
l
s
ha
ve
b
e
e
n
in
tr
odu
c
e
d
b
a
s
e
d
o
n
m
ul
t
im
o
da
l
a
u
th
e
nt
ic
a
ti
on
s
y
s
t
e
m
s
u
c
h
a
s
Z
h
a
n
g
e
t
a
l.
[
8]
u
s
e
d
f
a
c
e
a
nd
vo
ic
e
m
o
de
l
s
t
o
d
e
v
e
lo
p a
n
dr
o
id
b
a
s
e
d
a
ut
he
nt
i
c
a
ti
on
s
y
s
t
e
m
.
E
l
-
R
a
hi
e
m
e
t
al
.
[
1]
u
s
e
d
E
C
G
a
nd
f
in
ge
r
v
e
in
m
od
a
li
t
ie
s
. C
h
a
n
uk
ya
a
nd
T
hi
v
a
k
a
r
a
n
[
9]
us
e
d
c
om
bi
na
ti
on
of
f
in
ge
r
pr
in
t
a
nd
e
a
r
m
oda
li
ti
e
s
.
H
ow
e
ve
r
,
a
c
hi
e
vi
ng
th
e
a
c
c
ur
a
c
y
r
e
m
a
in
s
c
ha
ll
e
ngi
ng
ta
s
k
due
t
o huge
va
r
ia
ti
ons
i
n t
he
m
ul
ti
m
oda
l.
I
n or
de
r
t
o s
ol
ve
t
hi
s
pr
obl
e
m
, w
e
i
nt
r
oduc
e
a
nove
l
m
e
th
od f
or
us
e
r
a
ut
he
nt
ic
a
ti
on
th
a
t
ta
ke
s
in
to
a
c
c
o
unt
E
C
G
a
nd
li
p
e
xt
r
a
c
ti
on
f
r
om
f
a
c
ia
l
pi
c
tu
r
e
s
.
S
e
c
ti
on
2
pr
ovi
de
s
a
br
ie
f
ove
r
vi
e
w
of
th
e
r
e
l
e
va
nt
li
te
r
a
tu
r
e
,
s
e
c
ti
on
3
de
s
c
r
ib
e
s
th
e
d
e
e
p
t
r
a
ns
f
e
r
le
a
r
ni
ng
-
ba
s
e
d
m
ode
l
th
a
t
w
il
l
be
us
e
d,
s
e
c
ti
on
4
c
om
pa
r
e
s
a
nd
c
ont
r
a
s
ts
th
e
s
ol
ut
io
ns
th
a
t
ha
ve
b
e
e
n
e
xp
lo
r
e
d,
a
nd
s
e
c
ti
on
5
c
onc
lu
de
s
w
it
h
s
ugge
s
ti
ons
.
2.
L
I
T
E
R
A
T
U
R
E
S
U
R
V
E
Y
A
br
ie
f
li
te
r
a
tu
r
e
ove
r
vi
e
w
of
c
ur
r
e
nt
a
ppr
oa
c
he
s
f
or
E
C
G
,
f
a
c
ia
l,
a
nd
li
p
a
ut
h
e
nt
ic
a
ti
on
a
nd
c
a
te
gor
iz
a
ti
on
a
r
e
gi
ve
n
in
th
is
s
e
c
ti
on.
T
o
in
te
gr
a
te
m
a
ny
m
oda
li
ti
e
s
in
a
bi
om
e
tr
ic
a
ut
he
nt
ic
a
ti
on
s
ys
te
m
,
H
a
m
m
a
d
e
t
al
.
[
10]
a
tt
e
m
pt
e
d
to
in
te
gr
a
te
“
c
onvolut
io
na
l
n
e
ur
a
l
ne
twor
ks
(
C
N
N
s
)
”
w
it
h
“
Q
-
G
a
us
s
ia
n
m
ul
ti
-
s
uppor
t
ve
c
to
r
m
a
c
hi
ne
s
(
Q
G
-
M
S
V
M
s
)
”
. S
e
ve
r
a
l
f
us
io
n l
e
ve
ls
a
r
e
us
e
d by thi
s
m
ode
l.
T
o e
xt
r
a
c
t
f
e
a
tu
r
e
s
f
or
c
e
r
ta
in
m
oda
li
ti
e
s
,
C
N
N
s
a
r
e
e
m
pl
oye
d.
I
n
th
is
s
ta
ge
,
w
e
c
ho
s
e
t
w
o
C
N
N
la
ye
r
s
th
a
t
ga
v
e
us
th
e
be
s
t
a
c
c
ur
a
c
y.
E
a
c
h
f
e
a
tu
r
e
de
s
c
r
ip
ti
on
is
tr
e
a
t
e
d
a
s
a
n
in
de
pe
nde
nt
la
ye
r
.
W
e
th
e
n
m
e
r
ge
th
e
f
e
a
tu
r
e
de
s
c
r
ip
to
r
s
w
it
h
th
e
pr
opos
e
d
in
te
r
na
l
f
us
io
n
a
ppr
oa
c
h.
I
n
a
ddi
ti
on,
on
e
of
th
e
c
a
n
c
e
ll
a
bl
e
bi
om
e
tr
ic
a
ppr
oa
c
h
e
s
i
s
th
e
n
u
s
e
d
to
f
ur
th
e
r
s
tr
e
ngt
he
n t
he
s
e
c
ur
it
y of
t
he
pr
opos
e
d s
ys
te
m
a
nd t
he
s
e
t
e
m
pl
a
te
s
. D
ur
in
g t
he
a
ut
he
nt
ic
a
ti
on s
te
p, t
he
pe
r
f
or
m
a
nc
e
i
s
i
m
pr
ove
d by us
in
g Q
G
-
M
S
V
M
a
s
a
n a
ut
he
nt
ic
a
ti
on c
la
s
s
if
ie
r
.
A
ha
m
e
d
e
t
al
.
[
11]
c
om
bi
ne
d
E
C
G
a
nd
P
P
G
s
ig
na
ls
to
bui
ld
a
bi
om
e
tr
ic
s
ys
te
m
to
s
uppor
t
in
di
vi
dua
li
z
e
d
he
a
lt
hc
a
r
e
s
y
s
te
m
s
.
T
hi
s
s
tr
a
te
gy
c
om
pr
is
e
d
ti
m
e
-
dom
a
in
a
nd
c
om
bi
ne
d
ti
m
e
-
f
r
e
que
nc
y
dom
a
in
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
m
e
th
ods
th
a
t
a
r
e
ba
s
e
d
on
a
ut
or
e
gr
e
s
s
iv
e
c
oe
f
f
ic
ie
nt
s
,
th
e
S
ha
nnon
e
nt
r
opy,
a
nd
th
e
w
a
ve
le
t
pa
c
ke
t
tr
a
ns
f
or
m
.
U
s
in
g
th
e
r
e
tr
ie
ve
d
in
f
or
m
a
ti
on,
a
C
N
N
-
lo
ng
s
hor
t
-
te
r
m
m
e
m
or
y
(
L
S
T
M
)
c
la
s
s
if
ie
r
is
tr
a
in
e
d
s
ubs
e
que
nt
ly
.
I
ta
ni
e
t
al
.
[
12]
poi
nt
e
d
out
th
a
t
bi
om
e
tr
ic
s
ys
te
m
s
r
e
ly
in
g
on
f
a
c
ia
l
f
e
a
tu
r
e
s
ha
v
e
pr
obl
e
m
s
w
he
n
pe
opl
e
a
r
e
w
e
a
r
in
g
m
a
s
ks
,
a
nd
a
ut
he
nt
ic
a
ti
on
m
e
th
ods
ba
s
e
d
on
f
in
ge
r
pr
in
ts
ha
ve
p
r
obl
e
m
s
w
he
n
us
e
r
s
'
ha
nds
ge
t
da
m
p. I
n r
e
s
pons
e
t
o t
he
s
e
c
onc
e
r
n
s
, t
he
w
r
it
e
r
s
pr
opos
e
d a
n e
a
r
a
ut
he
nt
ic
a
ti
on me
th
od.
S
im
il
a
r
ly
,
P
u
r
ohi
t
a
nd
A
jm
e
r
a
[
13
]
pr
opos
e
d
a
m
ul
ti
m
oda
l
a
ut
he
nt
ic
a
ti
on
s
ys
te
m
e
s
ta
bl
i
s
he
d
on
pa
lm
,
f
in
ge
r
pr
in
t,
a
nd
e
a
r
bi
om
e
t
r
ic
s
.
G
a
bor
f
e
a
tu
r
e
s
a
r
e
us
e
d
f
or
h
a
nd
im
a
ge
s
,
th
e
hum
a
n
m
ic
r
os
tr
uc
tu
r
e
ba
s
e
d
(
H
M
S
B
)
a
dm
in
is
tr
a
to
r
f
or
f
in
ge
r
pr
in
ts
, a
nd H
M
S
B
a
nd
m
ul
ti
pl
e
r
e
gul
a
r
gr
a
di
e
nt
(
M
R
G
)
f
or
e
a
r
bi
om
e
tr
ic
s
i
n
th
is
m
e
th
odol
ogy,
w
hi
c
h
e
m
pl
oye
d
te
xt
ur
e
a
nd
f
or
m
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
a
ppr
oa
c
he
s
.
I
n
a
ddi
ti
on,
to
e
ns
ur
e
e
f
f
ic
ie
nt
f
e
a
tu
r
e
s
e
le
c
ti
on,
a
n
a
dve
r
s
a
r
ia
l
gr
a
y
w
ol
f
opt
im
iz
a
ti
on
a
ppr
oa
c
h
i
s
ut
il
iz
e
d,
f
ol
lo
w
e
d
by
th
e
ut
il
iz
a
ti
on of
a
m
ul
ti
-
k
er
ne
l
s
uppor
t
ve
c
to
r
m
a
c
hi
ne
(
S
V
M
)
c
la
s
s
if
ie
r
f
or
r
e
c
ogni
ti
on.
T
he
a
ut
hor
s
in
[
14]
in
t
r
oduc
e
d
a
ne
w
m
e
th
od
o
f
a
ut
he
nt
ic
a
ti
on
us
in
g
da
ta
f
r
om
f
a
c
ia
l
im
a
ge
s
.
T
he
di
m
e
ns
io
n
s
of
f
e
a
tu
r
e
ve
c
to
r
s
e
xt
r
a
c
te
d
f
r
om
f
a
c
e
im
a
ge
s
a
r
e
us
ua
ll
y
hi
gh,
s
o
th
e
y
ha
v
e
to
be
r
e
duc
e
d.
T
he
bi
om
e
tr
ic
ve
r
if
ic
a
ti
on
s
y
s
te
m
th
e
y
unve
il
e
d
us
e
d
di
gi
ta
l
s
ig
na
tu
r
e
s
a
nd
f
a
c
ia
l
r
e
c
ogni
ti
on
s
of
twa
r
e
.
T
he
y
a
c
hi
e
ve
th
i
s
by
e
m
pl
oyi
ng
a
f
us
io
n
f
e
a
tu
r
e
ve
c
to
r
,
w
hi
c
h
in
c
or
por
a
te
s
f
e
a
tu
r
e
s
r
e
tr
ie
ve
d
f
r
om
bot
h
m
oda
li
ti
e
s
.
T
he
y
pr
opos
e
d
to
us
e
a
“
m
odi
f
ie
d
c
ont
e
xt
-
a
w
a
r
e
(
M
C
A
)
”
a
ppr
oa
c
h
to
ge
ne
r
a
te
a
f
e
a
tu
r
e
ve
c
to
r
a
nd
e
m
pl
oy
a
“
t
a
nge
nt
ia
l
di
s
c
r
im
in
a
ti
on
a
na
ly
s
is
(
T
D
A
)
”
a
lg
or
it
hm
to
r
e
duc
e
th
e
di
m
e
ns
io
na
li
ty
of
th
e
f
e
a
tu
r
e
s
w
it
hi
n
f
a
c
ia
l
phot
ogr
a
phs
.
N
e
xt
,
th
e
y
tr
a
in
a
m
odi
f
ie
d
m
ix
e
d
s
e
q
ue
nc
e
de
e
p
ne
ur
a
l
n
e
twor
k
(
M
M
S
-
D
N
N
)
us
in
g
th
e
f
us
io
n f
e
a
tu
r
e
ve
c
to
r
.
S
in
gh
a
nd
T
iwa
r
i
[
15]
de
ve
lo
pe
d
a
m
ul
ti
m
oda
l
a
ut
he
nt
ic
a
ti
on
s
ys
te
m
.
T
he
w
or
k
pr
opos
e
d
in
vol
ve
s
in
te
gr
a
ti
on
of
th
r
e
e
uni
m
oda
l
bi
om
e
tr
ic
s
ys
te
m
s
to
f
or
m
two
m
ul
ti
m
oda
l
bi
om
e
tr
ic
s
ys
te
m
s
.
F
or
th
e
pur
pos
e
of
th
is
s
tu
dy,
E
C
G
,
s
c
le
r
a
,
a
nd
f
in
ge
r
pr
in
t
a
r
e
c
hos
e
n
a
s
uni
m
oda
l
s
ys
te
m
.
I
n
th
e
f
ir
s
t
m
ul
ti
m
oda
l
bi
o
m
e
tr
ic
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
4
,
A
ugus
t
2025
:
3160
-
3171
3162
s
ys
te
m
w
e
a
dopt
a
s
e
qu
e
nt
ia
l
m
ode
l
a
ppr
oa
c
h
w
it
h
th
e
“
w
ha
le
o
pt
im
iz
a
ti
on
a
lg
or
it
hm
-
a
r
ti
f
ic
ia
l
ne
ur
a
l
ne
twor
k
(
W
O
A
-
A
N
N
)
”
de
c
is
io
n l
e
ve
l
f
us
io
n. M
e
a
nw
hi
le
, t
he
s
e
c
ond mul
ti
m
oda
l
bi
om
e
tr
ic
s
ys
te
m
e
m
pl
oys
a
pa
r
a
ll
e
l
m
ode
l
a
ppr
oa
c
h,
e
m
pl
oyi
ng
s
c
or
e
-
le
ve
l
f
us
io
n
ba
s
e
d
on
“
s
a
lp
s
w
a
r
m
a
lg
or
it
hm
-
de
e
p
be
li
e
f
ne
twor
k
(
SSA
-
D
B
N
)
”
. T
he
bi
om
e
tr
ic
a
ut
he
nt
ic
a
ti
on pr
oc
e
s
s
e
nc
om
pa
s
s
e
s
pr
e
pr
oc
e
s
s
in
g, f
e
a
tu
r
e
e
xt
r
a
c
ti
on, ma
tc
hi
ng,
a
nd s
c
or
in
g f
or
e
a
c
h i
ndi
vi
dua
l
uni
m
oda
l
s
ys
t
e
m
. M
a
tc
hi
ng
s
c
o
r
e
s
a
nd i
ndi
vi
dua
l
a
c
c
ur
a
c
y
f
or
e
a
c
h
bi
om
e
tr
ic
a
tt
r
ib
ut
e
a
r
e
e
nc
r
ypt
e
d
in
de
pe
nd
e
nt
ly
.
A
f
us
io
n
pr
oc
e
dur
e
ba
s
e
d
on
m
a
tc
he
r
pe
r
f
or
m
a
nc
e
is
e
m
pl
oye
d
f
or
th
e
th
r
e
e
bi
om
e
tr
ic
t
r
a
it
s
, a
s
t
he
m
a
tc
he
r
s
pr
oduc
e
v
a
r
ie
d va
lu
e
s
a
c
r
os
s
t
he
s
e
a
tt
r
ib
ut
e
s
.
C
he
r
if
i
e
t
al
.
[
16]
in
t
r
oduc
e
d
m
ul
ti
m
oda
l
a
ut
he
nt
ic
a
ti
on
s
ys
te
m
by
us
in
g
a
r
m
ge
s
tu
r
e
a
nd
e
a
r
s
ha
pe
.
T
he
e
a
r
f
e
a
tu
r
e
e
xt
r
a
c
ti
on c
on
s
id
e
r
s
l
oc
a
l
pha
s
e
qua
nt
iz
a
ti
on
m
e
c
ha
ni
s
m
w
hi
c
h i
s
u
s
e
d t
o ha
ndl
e
t
he
po
s
e
a
nd
il
lu
m
in
a
ti
on
va
r
ia
ti
ons
.
S
im
il
a
r
ly
,
f
or
a
r
m
ge
s
tu
r
e
a
ls
o
s
ta
ti
s
ti
c
a
l
f
e
a
tu
r
e
s
a
r
e
e
xt
r
a
c
te
d.
F
in
a
ll
y,
th
e
obt
a
in
e
d
f
e
a
tu
r
e
s
a
r
e
c
om
bi
ne
d
on
s
c
or
e
l
e
ve
l
by
c
ons
id
e
r
in
g
w
e
ig
ht
e
d
s
um
.
K
a
ul
e
t
al
.
[
17]
pr
e
s
e
nt
e
d
E
C
G
ba
s
e
d
bi
om
e
tr
ic
a
ut
he
nt
ic
a
ti
on
s
ys
te
m
w
he
r
e
non
-
f
id
uc
ia
l
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
a
ppr
oa
c
h
is
in
tr
oduc
e
d.
T
hi
s
a
ppr
oa
c
h
is
c
ons
tr
uc
te
d
by
th
e
c
om
bi
na
ti
on
of
di
s
c
r
e
te
c
o
s
in
e
tr
a
ns
f
or
m
a
nd
a
ut
oc
or
r
e
la
ti
on.
F
ur
th
e
r
,
th
e
obt
a
in
e
d
f
e
a
tu
r
e
s
a
r
e
th
e
n
f
e
d
in
to
th
e
ne
ur
a
l
ne
twor
k
m
ode
l
w
he
r
e
m
ul
ti
la
ye
r
pe
r
c
e
pt
r
on
a
nd
r
a
di
a
l
ba
s
is
f
unc
ti
ons
m
odul
e
s
a
r
e
us
e
d t
o t
r
a
in
t
he
m
ode
l.
K
im
e
t
al
.
[
18
]
us
e
d
e
le
c
tr
om
yogr
a
m
s
ig
na
l
be
c
a
us
e
th
e
s
e
s
ig
na
ls
c
a
nnot
be
f
or
a
ge
d
a
nd
th
e
r
e
f
or
e
s
ugge
s
te
d
a
n
a
ut
h
e
nt
ic
a
ti
on
a
ppr
oa
c
h.
A
c
c
or
di
ng
to
th
i
s
a
ppr
o
a
c
h,
ti
m
e
dom
a
in
a
tt
r
ib
ut
e
s
a
r
e
e
xt
r
a
c
te
d
f
r
om
th
e
pr
e
-
pr
oc
e
s
s
e
d
s
ig
na
l
la
te
r
L
S
T
M
i
s
us
e
d
to
m
a
tc
h
th
e
ge
s
t
ur
e
.
F
in
a
ll
y,
C
N
N
-
L
S
T
M
is
us
e
d
to
obt
a
in
th
e
f
in
a
l
c
la
s
s
if
ic
a
ti
on.
G
r
a
c
e
e
t
al
.
[
19]
bui
lt
a
n
E
E
G
a
ut
he
nt
ic
a
ti
on
s
ys
te
m
th
a
t
a
c
c
om
pl
is
he
d
s
ig
na
l
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
th
r
ough
D
W
T
be
f
or
e
us
in
g
f
e
e
d
f
or
w
a
r
d
ne
ur
a
l
ne
twor
k
f
or
t
r
a
in
in
g
th
e
e
xt
r
a
c
te
d
f
e
a
tu
r
e
s
.
T
he
c
om
pl
e
te
s
ys
te
m
a
c
c
ur
a
c
y of
t
hi
s
m
e
th
od r
e
a
c
h
e
d 87.7%
.
T
he
m
ul
ti
m
oda
l
bi
om
e
tr
ic
a
ut
he
nt
ic
a
ti
on
s
ys
te
m
pr
e
s
e
nt
e
d
by
Y
ouni
s
a
nd
A
buha
m
m
a
d
[
20]
e
m
pl
oys
R
e
s
ne
t1
01,
R
e
s
ne
t
-
I
nc
e
pt
io
nv2,
D
e
ns
e
n
e
t2
01,
A
le
xN
e
t,
a
nd
I
nc
e
pt
io
nv2
de
e
p
tr
a
ns
f
e
r
le
a
r
ni
ng
m
ode
l
t
o
c
om
bi
ne
ove
r
s
iz
e
d
ha
nd
c
r
a
f
te
d
f
e
a
tu
r
e
s
de
r
iv
e
d
f
r
om
H
og
f
e
a
tu
r
e
de
s
c
r
ip
to
r
.
T
he
f
us
io
n
ta
s
k
a
ppl
ie
s
di
s
c
r
im
in
a
nt
c
or
r
e
la
ti
on
a
na
ly
s
is
(
D
C
A
)
a
nd
c
a
noni
c
a
l
c
or
r
e
la
ti
on
a
na
ly
s
is
(
C
C
A
)
to
c
om
pl
e
te
it
.
T
hi
s
te
c
hni
que
a
c
hi
e
ve
d
a
r
e
c
ogni
ti
on
r
a
te
of
96.6%
in
it
s
r
e
s
ul
ts
.
S
ia
m
e
t
al
.
[
21]
de
m
ons
tr
a
te
d
a
f
r
a
m
e
w
or
k
f
o
r
a
ut
he
nt
ic
a
ti
on
w
it
h
E
C
G
a
nd
P
P
G
s
ig
na
ls
th
a
t
pr
ovi
de
s
c
a
nc
e
la
bl
e
bi
om
e
tr
ic
s
.
T
hi
s
w
or
k
pr
e
s
e
nt
e
d
a
te
m
pl
a
te
ge
ne
r
a
ti
on
s
ol
ut
io
n
f
or
in
di
vi
dua
l
us
e
r
s
th
r
ough
uni
f
ic
a
ti
on
te
c
hni
que
s
.
T
he
e
xt
r
a
c
ti
on
of
f
e
a
tu
r
e
s
u
s
e
s
M
e
l
-
f
r
e
que
nc
y
c
e
ps
tr
a
l
c
oe
f
f
ic
ie
nt
s
(
M
F
C
C
s
)
a
s
it
s
f
r
a
m
e
w
or
k.
T
he
pe
r
f
or
m
a
nc
e
out
c
om
e
s
of
c
la
s
s
if
ic
a
ti
on
de
pe
nd
on
th
e
ut
il
iz
a
ti
on
of
m
ul
ti
-
la
ye
r
pe
r
c
e
pt
r
on
(
M
L
P
)
a
nd
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n
c
la
s
s
if
ie
r
.
J
e
ong
e
t
al
.
[
22]
de
ve
lo
pe
d
D
e
m
oI
D
a
s
a
ne
w
a
ut
h
e
nt
ic
a
ti
on
a
ppr
oa
c
h
th
a
t
us
e
s
f
a
c
e
to
ge
th
e
r
w
it
h
voi
c
e
bi
om
e
tr
ic
s
f
or
a
ut
he
nt
ic
a
ti
on
pur
pos
e
s
.
A
le
id
a
n
e
t
al
.
[
23]
de
ve
lo
pe
d
a
ut
he
nt
i
c
a
ti
on
th
r
ough
m
oni
to
r
in
g
s
im
ul
ta
ne
ous
E
C
G
,
PPG
,
a
nd
P
C
G
s
ig
na
ls
us
in
g
de
e
p
le
a
r
ni
ng
te
c
hni
que
s
.
T
he
pr
op
os
e
d
de
e
p
le
a
r
ni
ng
s
ys
te
m
im
pl
e
m
e
nt
s
tr
a
ns
f
e
r
le
a
r
ni
ng
to
ge
th
e
r
w
it
h
L
S
T
M
a
r
c
hi
te
c
tu
r
e
f
or
f
e
a
tu
r
e
opt
im
iz
a
ti
on.
T
he
a
tt
r
ib
ut
e
s
unde
r
go
c
la
s
s
if
ic
a
ti
on
th
r
ough
a
n
im
pl
e
m
e
nt
a
ti
on
of
boos
ti
ng
m
e
c
ha
ni
s
m
.
M
e
r
gi
ng
f
a
c
e
a
nd
pa
lm
pr
in
t
a
nd
ir
is
bi
om
e
tr
ic
s
c
ons
ti
tu
te
s
th
e
hybr
id
bi
om
e
tr
ic
a
ut
he
nt
ic
a
ti
on
s
ys
te
m
d
e
s
c
r
ib
e
d
in
[
24]
.
V
a
r
ia
ti
ons
of
gr
oup
s
e
a
r
c
h
opt
im
iz
a
ti
on
(
M
G
S
O
)
a
ppr
oa
c
h
opt
im
iz
e
th
e
pr
oc
e
s
s
of
e
xt
r
a
c
ti
ng
f
e
a
tu
r
e
s
.
A
te
a
c
he
r
le
a
r
ni
ng
ba
s
e
d
de
e
p
le
a
r
ni
ng
m
ode
l
s
e
r
ve
s
to
pe
r
f
or
m
t
he
l
a
s
t
s
ta
ge
c
la
s
s
if
ic
a
ti
on of
f
e
a
tu
r
e
s
. M
oda
k
a
nd J
ha
[
25]
bui
lt
a
m
ul
ti
m
oda
l
a
ut
he
nt
ic
a
ti
on
s
ys
te
m
by
c
om
bi
ni
ng
f
a
c
e
a
nd
e
ye
a
lo
ngs
id
e
f
in
ge
r
pr
in
t
m
oda
ls
.
T
hi
s
a
ppr
oa
c
h
V
io
la
-
J
one
s
a
ppr
oa
c
h
f
or
f
a
c
e
s
e
gm
e
nt
a
ti
on,
la
te
r
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
is
pe
r
f
or
m
e
d
a
nd
obt
a
in
e
d
f
e
a
tu
r
e
s
a
r
e
opt
im
iz
e
d
by
us
in
g
c
ha
os
-
ba
s
e
d
s
a
lp
s
w
a
r
m
a
lg
or
it
hm
(
C
S
S
A
)
.
F
in
a
ll
y,
r
ul
e
-
ba
s
e
d
a
da
pt
iv
e
ne
ur
o
-
f
uz
z
y
in
f
e
r
e
nc
e
s
ys
te
m
(R
-
A
N
F
I
S
)
a
lg
or
it
hm
is
in
tr
oduc
e
d
f
or
c
la
s
s
if
ic
a
ti
on.
A
ls
o,
de
e
p
le
a
r
ni
ng
m
ode
ls
in
te
gr
a
te
d
w
it
h
tr
a
ns
f
e
r
le
a
r
ni
ng
ba
s
e
d
r
e
c
e
nt
s
tu
di
e
s
[
26]
–
[
28]
ha
ve
s
how
n
gr
e
a
t
im
pr
ove
m
e
nt
in
e
nha
nc
in
g
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
w
hi
le
a
ppl
yi
ng
on i
m
a
ge
pr
oc
e
s
s
in
g t
a
s
ks
.
3.
P
R
O
P
O
S
E
D
M
O
D
E
L
M
a
ny
a
ppl
ic
a
ti
ons
ut
il
iz
e
de
e
p
le
a
r
ni
ng
te
c
hni
que
s
b
e
c
a
u
s
e
th
e
s
e
m
e
th
od
s
of
f
e
r
e
xc
e
pt
io
na
l
a
bi
li
ti
e
s
to
di
s
c
ove
r
in
tr
ic
a
te
pa
tt
e
r
ns
in
pr
oc
e
s
s
in
g
s
uc
h
a
s
im
a
g
e
s
a
nd
vi
de
os
to
g
e
th
e
r
w
it
h
bi
om
e
di
c
a
l
im
a
ge
s
.
T
he
tr
a
ns
f
e
r
le
a
r
ni
ng
te
c
hni
que
f
in
ds
e
xt
e
ns
iv
e
u
s
e
a
c
r
os
s
m
ul
ti
pl
e
a
ppl
ic
a
ti
ons
s
in
c
e
it
le
ve
r
a
ge
s
de
e
p
le
a
r
ni
ng
m
ode
ls
tr
a
in
e
d
f
or
in
it
ia
l
pur
pos
e
s
to
e
xe
c
ut
e
or
r
e
tr
a
in
th
e
m
f
or
s
im
il
a
r
-
r
e
la
te
d
ta
s
ks
.
O
ur
r
e
s
e
a
r
c
h
us
e
s
th
i
s
m
ode
l
s
tr
uc
tu
r
e
f
or
c
onduc
ti
ng E
C
G
a
na
ly
s
is
a
nd
l
ip
i
m
a
ge
e
va
lu
a
ti
on.
T
he
c
la
s
s
if
ic
a
ti
on
of
E
C
G
s
ig
na
ls
be
c
om
e
s
po
s
s
ib
le
th
r
ough
u
s
a
ge
of
pr
e
-
tr
a
in
e
d
s
ig
na
l
pr
oc
e
s
s
in
g
m
ode
ls
f
r
om
di
s
ti
nc
t
d
if
f
e
r
e
nt
pr
oc
e
s
s
in
g
ta
s
ks
.
T
he
de
e
p
le
a
r
ni
ng
m
ode
ls
tr
a
in
e
d
w
it
h
ge
ne
r
a
l
ti
m
e
-
s
e
r
ie
s
in
f
or
m
a
ti
on
c
a
n
s
uc
c
e
s
s
f
ul
ly
pe
r
f
or
m
c
la
s
s
if
ic
a
ti
on
dut
ie
s
in
E
C
G
a
ppl
ic
a
ti
ons
.
T
he
pr
e
-
tr
a
in
e
d
m
ode
l
e
na
bl
e
s
le
a
r
ni
ng of
vi
ta
l
f
e
a
tu
r
e
s
c
ont
a
in
in
g t
e
m
por
a
l
pa
tt
e
r
ns
a
nd f
r
e
qu
e
nc
y c
ha
r
a
c
te
r
is
ti
c
s
. T
he
pr
e
-
e
xi
s
ti
ng f
e
a
tu
r
e
s
unde
r
go
c
us
to
m
iz
a
ti
on
be
f
or
e
e
xe
c
ut
in
g
s
pe
c
if
ic
c
la
s
s
if
ic
a
ti
on
pr
oc
e
dur
e
s
.
T
he
pr
e
-
t
r
a
in
e
d
m
ode
ls
de
m
ons
tr
a
te
f
unc
ti
ona
li
ty
in
th
e
pr
oc
e
s
s
in
g
of
f
a
c
e
a
nd
li
p
im
a
ge
s
to
o.
T
he
pr
e
-
tr
a
in
e
d
m
ode
ls
c
a
n
e
xt
r
a
c
t
vi
s
ua
l
f
e
a
tu
r
e
s
r
e
la
te
d
to
li
p
m
ove
m
e
nt
or
e
xpr
e
s
s
io
n
by
ut
il
iz
in
g
m
ode
ls
f
r
om
I
m
a
ge
N
e
t
or
R
e
s
N
e
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
D
e
e
p t
r
ans
fe
r
l
e
ar
ni
ng f
o
r
c
la
s
s
if
ic
at
io
n of
E
C
G
s
ig
nal
s
and li
p
i
m
age
s
i
n
…
(
L
at
ha K
r
is
hna
m
oor
th
y
)
3163
T
he
r
e
s
e
a
r
c
h
a
d
a
pt
s
th
e
s
e
m
ode
l
s
to
e
xt
r
a
c
t
c
ha
r
a
c
te
r
is
ti
c
s
f
r
om
bot
h
da
ta
ty
pe
s
be
f
or
e
im
pl
e
m
e
nt
in
g
a
d
e
e
p
le
a
r
ni
ng
c
la
s
s
if
ic
a
ti
on
m
e
th
od
to
id
e
nt
if
y
us
e
r
s
.
W
e
h
a
ve
in
c
lu
de
d
th
e
a
ut
he
nt
ic
it
y
m
e
a
s
ur
e
th
r
ough
th
e
de
c
is
io
n
-
m
a
ki
ng
pr
oc
e
s
s
.
3.1.
D
at
a au
gm
e
n
t
at
io
n
T
he
d
a
ta
a
ugm
e
nt
a
ti
on
pl
a
ys
im
por
ta
nt
r
ol
e
in
d
e
e
p
l
e
a
r
ni
ng
a
nd
m
a
c
hi
ne
l
e
a
r
ni
ng
ba
s
e
d
ta
s
ks
.
I
t
is
w
id
e
ly
a
dopt
e
d
m
e
c
ha
ni
s
m
in
or
de
r
to
w
id
e
n
tr
a
in
in
g
da
ta
a
nd
e
xpa
nd
da
ta
s
e
t
s
iz
e
by
a
ppl
yi
ng
di
f
f
e
r
e
nt
m
e
c
ha
ni
s
m
s
.
I
n
th
e
c
ont
e
xt
of
E
C
G
,
da
ta
a
ugm
e
nt
a
ti
on
c
a
n
be
pa
r
ti
c
ul
a
r
ly
us
e
f
ul
f
or
im
p
r
ovi
ng
m
ode
l
ge
ne
r
a
li
z
a
ti
on
a
nd
r
obu
s
tn
e
s
s
,
e
s
pe
c
ia
ll
y
w
h
e
n
de
a
li
ng
w
it
h
li
m
it
e
d
da
ta
. T
he
da
t
a
a
ugm
e
nt
a
ti
on
s
c
h
e
m
e
s
a
r
e
a
s
f
ol
lo
w
s
:
‒
A
ddi
ng
th
e
noi
s
e
:
a
ddi
ng
r
a
ndom
noi
s
e
to
E
C
G
s
ig
na
ls
c
a
n
s
im
u
la
te
noi
s
e
pr
e
s
e
nt
in
r
e
a
l
-
w
or
ld
r
e
c
or
di
ngs
,
m
a
ki
ng
th
e
m
ode
l
m
or
e
r
obus
t
to
noi
s
e
.
V
a
r
io
us
ki
nds
of
noi
s
e
,
s
uc
h
a
s
G
a
us
s
ia
n
noi
s
e
,
w
hi
te
noi
s
e
,
or
e
ve
n s
pe
c
if
ic
t
ype
s
of
i
nt
e
r
f
e
r
e
nc
e
noi
s
e
, c
a
n b
e
a
dde
d t
o t
he
E
C
G
s
ig
na
ls
.
‒
T
e
m
por
a
l
va
r
ia
ti
ons
:
th
e
E
C
G
s
ig
na
ls
ha
ve
huge
im
pa
c
t
of
te
m
por
a
l
va
r
ia
ti
ons
th
e
r
e
f
or
e
w
e
in
c
or
por
a
te
.
T
hi
s
he
lp
s
t
h
e
m
ode
l
le
a
r
n t
o be
i
nva
r
ia
nt
t
o s
li
ght
t
e
m
por
a
l
s
hi
f
ts
i
n t
he
s
ig
na
l.
‒
A
m
pl
it
ude
s
c
a
li
ng:
it
c
a
n
s
im
ul
a
te
va
r
ia
ti
ons
in
s
ig
na
l
s
tr
e
n
gt
h
or
e
le
c
tr
ode
pl
a
c
e
m
e
nt
th
us
he
lp
s
to
ge
ne
r
a
li
z
e
be
tt
e
r
t
o va
r
ia
ti
ons
i
n s
ig
na
l
in
te
n
s
it
y
‒
B
a
s
e
li
ne
w
a
nde
r
:
i
nt
r
oduc
in
g
ba
s
e
li
ne
w
a
nde
r
by
a
ddi
ng
a
lo
w
-
f
r
e
que
nc
y
s
in
us
oi
d
a
l
c
om
pone
nt
to
th
e
E
C
G
s
ig
na
l
c
a
n
s
im
ul
a
te
va
r
ia
ti
ons
in
ba
s
e
li
ne
dr
if
t.
T
hi
s
he
lp
s
th
e
m
ode
l
le
a
r
n
to
de
te
c
t
a
nd
c
la
s
s
if
y
E
C
G
f
e
a
tu
r
e
s
a
c
c
ur
a
te
ly
i
n t
he
pr
e
s
e
nc
e
of
ba
s
e
li
ne
dr
if
t.
S
im
il
a
r
ly
,
w
e
a
ppl
y
da
ta
a
ugm
e
nt
a
ti
on
on
li
p
im
a
ge
da
ta
s
e
t.
T
he
a
ugm
e
nt
a
ti
on
on
im
a
ge
da
ta
in
c
lu
de
s
s
e
ve
r
a
l
s
ta
ge
s
s
uc
h
a
s
:
r
ot
a
ti
on,
f
li
p,
b
r
ig
ht
ne
s
s
,
c
ont
r
a
s
t
a
dj
us
tm
e
nt
,
noi
s
e
a
ddi
ti
on,
a
nd
c
r
oppi
ng.
A
br
ie
f
di
s
c
us
s
io
n
a
bout
t
he
s
e
a
ugm
e
nt
a
ti
ons
i
s
gi
ve
n
a
s
f
ol
lo
w
s
:
‒
R
ot
a
ti
on:
i
t
pe
r
f
or
m
s
r
ot
a
ti
on on the
or
ig
in
a
l
im
a
ge
t
o i
nt
r
oduc
e
t
he
va
r
ia
bi
li
ty
w
hi
c
h he
lp
s
t
o i
m
pr
ove
t
he
r
obus
tn
e
s
s
t
o di
f
f
e
r
e
nt
va
r
ia
ti
ons
.
‒
I
m
a
ge
f
li
ppi
ng:
f
li
p
th
e
f
a
c
e
im
a
ge
s
hor
iz
ont
a
ll
y
to
ge
ne
r
a
te
le
f
t
-
r
ig
ht
m
ir
r
o
r
im
a
ge
s
,
w
hi
c
h
c
a
n
he
lp
th
e
m
ode
l
ge
ne
r
a
li
z
e
be
tt
e
r
t
o f
a
c
e
s
w
it
h di
f
f
e
r
e
nt
or
ie
nt
a
ti
ons
.
‒
B
r
ig
ht
ne
s
s
a
nd
c
ont
r
a
s
t
a
dj
us
tm
e
nt
:
th
is
he
lp
s
to
c
ons
id
e
r
th
e
va
r
yi
ng
li
ght
ni
ng
a
nd
il
lu
m
in
a
ti
on
c
ondi
ti
ons
.
‒
N
oi
s
e
a
ddi
ti
on
:
in
tr
oduc
e
r
a
ndom
noi
s
e
to
th
e
f
a
c
e
/l
ip
im
a
ge
s
to
s
im
ul
a
te
noi
s
e
in
im
a
ge
a
c
qui
s
it
io
n
de
vi
c
e
s
or
e
nvi
r
onm
e
nt
a
l
c
ondi
ti
ons
.
‒
C
r
op
a
nd
z
oom
:
c
r
op
a
nd
z
oom
th
e
f
a
c
e
im
a
ge
s
to
f
oc
us
on
di
f
f
e
r
e
nt
r
e
gi
ons
of
in
te
r
e
s
t,
s
uc
h
a
s
th
e
li
ps
,
to
he
lp
t
he
m
ode
l
le
a
r
n i
nva
r
ia
nt
r
e
pr
e
s
e
nt
a
ti
ons
.
3.2.
F
e
at
u
r
e
e
xt
r
ac
t
io
n
W
e
us
e
th
e
a
ugm
e
nt
e
d
E
C
G
a
nd
l
ip
im
a
g
e
da
ta
to
e
xt
r
a
c
t
th
e
f
e
a
tu
r
e
s
.
T
he
f
ir
s
t
s
ubs
e
c
ti
on
pr
e
s
e
nt
s
th
e
pr
oc
e
s
s
of
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
f
r
om
E
C
G
da
ta
a
nd
s
e
c
ond
s
ta
ge
pr
e
s
e
nt
s
th
e
out
c
om
e
of
li
p
im
a
ge
f
e
a
tu
r
e
e
xt
r
a
c
ti
on. He
r
e
, R
e
s
N
e
t5
0 pr
e
-
tr
a
in
e
d m
ode
l
is
us
e
d t
o e
xt
r
a
c
t
th
e
de
e
p f
e
a
tu
r
e
s
f
r
om
s
c
a
lo
gr
a
m
i
m
a
ge
s
.
3.2.1.
E
C
G
f
e
at
u
r
e
e
xt
r
a
c
t
io
n
T
hi
s
s
e
c
ti
on
d
e
s
c
r
ib
e
s
th
e
s
ugge
s
te
d
m
ode
l
f
or
E
C
G
f
e
a
tu
r
e
e
xt
r
a
c
ti
on.
I
n
th
is
w
or
k,
w
e
ha
ve
c
ons
id
e
r
e
d
w
a
ve
le
t
tr
a
ns
f
or
m
ba
s
e
m
e
th
od
to
e
xt
r
a
c
t
th
e
f
e
a
tu
r
e
s
f
r
om
in
put
E
C
G
s
ig
na
l.
W
a
ve
le
t
tr
a
n
s
f
or
m
s
a
r
e
w
id
e
ly
a
dopt
e
d
a
s
t
he
e
xt
e
n
s
io
n of
c
onve
nt
io
na
l
F
our
ie
r
t
r
a
ns
f
or
m
m
ode
l.
H
ow
e
ve
r
, t
he
F
our
ie
r
t
r
a
ns
f
or
m
ope
r
a
te
s
on
a
s
in
gul
a
r
f
r
e
que
nc
y
or
s
c
a
le
,
w
h
e
r
e
a
s
w
a
ve
le
ts
ope
r
a
te
a
c
r
os
s
m
ul
ti
pl
e
s
c
a
le
s
of
f
r
e
que
nc
ie
s
.
W
a
ve
le
t
a
na
ly
s
is
in
vol
ve
s
d
e
c
om
pos
in
g
a
ny s
ig
na
l
in
to
va
r
io
us
ve
r
s
io
ns
w
it
h
di
f
f
e
r
e
nt
s
hi
f
ts
a
nd
s
c
a
le
s
f
r
om
th
e
or
ig
in
a
l
w
a
ve
le
ts
.
I
n
our
pr
opos
e
d
m
e
th
odol
ogy,
w
e
pr
im
a
r
il
y
us
e
th
e
“
c
ont
in
uous
w
a
ve
le
t
tr
a
n
s
f
or
m
(
C
W
T
)
”
.
W
e
m
a
p
th
e
s
ig
na
l
ont
o a
ti
m
e
s
c
a
le
dom
a
in
,
w
he
r
e
e
a
c
h
ti
m
e
s
c
a
le
in
de
xe
s
a
s
p
e
c
if
ic
s
ub
s
e
t
of
th
e
f
r
e
que
nc
y doma
in
. T
he
C
W
T
of
a
ny give
n
s
ig
na
l
(
)
is
obt
a
in
e
d ba
s
e
d on a
n i
nt
e
gr
a
l
of
(
)
a
s
f
ol
lo
w
s
:
(
,
)
=
1
√
∫
(
)
∗
,
(
−
)
∞
−
∞
(
1)
W
he
r
e
c
ha
r
a
c
te
r
is
e
s
th
e
m
ot
he
r
w
a
ve
l
e
t.
T
he
s
hi
f
ti
ng
a
nd
s
c
a
li
ng
ope
r
a
ti
on
on
m
ot
he
r
w
a
ve
le
t
pr
oduc
e
da
ught
e
r
w
a
ve
le
t
w
hi
c
h
is
r
e
pr
e
s
e
nt
e
d
a
s
,
w
he
r
e
a
nd
r
e
pr
e
s
e
nt
s
th
e
s
c
a
li
ng
a
nd
s
hi
f
ti
ng
f
a
c
to
r
s
,
r
e
s
pe
c
ti
ve
ly
.
T
he
C
W
T
g
e
ne
r
a
te
s
s
e
v
e
r
a
l
w
a
ve
le
t
c
oe
f
f
ic
ie
nt
s
.
T
he
C
W
T
m
odul
e
is
th
e
pr
e
-
pr
oc
e
s
s
e
d
f
il
te
r
e
d
s
ig
na
l.
S
pe
c
if
ic
C
W
T
w
a
v
e
le
t
c
oe
f
f
ic
ie
nt
s
a
r
e
de
r
iv
e
d us
in
g t
he
C
W
T
. T
h
e
n, t
he
s
e
r
ie
s
of
c
ont
in
uous
w
a
v
e
le
t
f
il
te
r
ba
nks
a
r
e
a
ppl
ie
d
on
th
e
s
e
c
oe
f
f
ic
ie
nt
s
.
A
two
-
di
m
e
ns
io
na
l
im
a
ge
of
th
e
C
W
T
c
oe
f
f
ic
ie
nt
s
f
or
e
a
c
h
E
C
G
r
e
c
or
d
is
pr
oduc
e
d
a
s
th
e
out
c
om
e
.
F
ig
u
r
e
1
il
lu
s
tr
a
te
s
th
e
s
c
a
lo
gr
a
m
of
a
ut
he
nt
ic
us
e
r
a
nd
im
pos
te
r
w
he
r
e
F
ig
ur
e
1(
a
)
i
ll
us
tr
a
te
s
t
he
s
c
a
lo
gr
a
m
of
a
ut
he
nt
ic
us
e
r
a
nd F
ig
ur
e
1(
b)
s
how
s
t
he
s
c
a
lo
gr
a
m
of
i
m
pos
te
r
.
T
r
a
ns
f
e
r
le
a
r
ni
ng
-
ba
s
e
d
m
ode
ls
pr
ovi
de
th
e
e
xt
r
a
c
ti
on
o
f
r
o
bus
t
f
e
a
tu
r
e
s
by
pr
oc
e
s
s
in
g
th
e
da
ta
th
r
ough
de
e
p
le
a
r
ni
ng
m
ode
l
s
.
T
he
m
ode
l
obt
a
in
s
de
e
p
f
e
a
tu
r
e
s
f
r
om
s
c
a
lo
gr
a
m
im
a
g
e
s
by
e
m
pl
oyi
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
4
,
A
ugus
t
2025
:
3160
-
3171
3164
R
e
s
N
e
t5
0
a
s
it
s
pr
e
-
tr
a
in
e
d
a
r
c
hi
te
c
tu
r
e
.
T
h
e
ne
twor
k
s
tr
uc
tu
r
e
na
m
e
d
R
e
s
N
e
t
-
50
f
unc
ti
ons
a
s
a
d
e
e
p
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k (
D
C
N
N
)
a
r
c
hi
te
c
tu
r
e
.
T
he
R
e
s
N
e
t
a
r
c
hi
te
c
tu
r
e
pr
ovi
de
s
t
r
a
in
in
g c
a
pa
bi
li
ti
e
s
f
or
de
e
p ne
ur
a
l
ne
twor
k
s
a
nd t
hi
s
i
s
one
of
i
ts
va
r
ia
nt
ve
r
s
io
n
s
. R
e
s
N
e
t
-
50 r
e
pr
e
s
e
nt
s
a
50
-
la
ye
r
ne
twor
k
s
tr
uc
tu
r
e
th
a
t
ope
r
a
te
s
th
r
ough
bl
oc
ks
c
ont
a
in
in
g c
onvolut
io
na
l
ope
r
a
ti
o
ns
a
lt
hough R
e
s
N
e
t
-
110
de
m
ons
tr
a
te
s
th
e
le
a
s
t
va
r
ia
ti
on
be
twe
e
n
th
is
m
ode
l
a
nd
th
e
s
ta
nd
a
r
d
D
e
ns
e
N
e
t
im
pl
e
m
e
nt
a
ti
on.
T
he
m
a
in
br
e
a
kt
hr
ough
of
R
e
s
N
e
t
s
e
r
ve
s
a
s
t
he
f
ounda
ti
on f
or
i
n
tr
oduc
in
g
r
e
s
id
ua
l
c
onne
c
ti
ons
w
hi
c
h e
xt
e
nd de
e
p ne
twor
k t
r
a
in
in
g
c
a
pa
bi
li
ti
e
s
by
s
ol
vi
ng
th
e
va
ni
s
hi
ng
gr
a
di
e
nt
pr
obl
e
m
.
T
he
r
e
s
id
ua
l
li
nks
c
onne
c
t
s
tr
a
ig
ht
to
th
e
c
onvolut
io
na
l
la
ye
r
s
f
or
a
ddi
ng
th
e
ir
e
x
tr
a
in
f
or
m
a
ti
on
to
th
e
out
put
f
e
a
tu
r
e
s
th
a
t
e
xi
s
t
in
th
e
c
onvolut
io
na
l
la
ye
r
s
a
nd
th
e
or
ig
in
a
l
ne
twor
k i
nput
. T
he
ne
twor
k t
r
a
in
in
g m
e
c
ha
ni
s
m
l
e
a
r
ns
r
e
s
id
ua
l
f
unc
ti
ons
be
c
a
u
s
e
r
e
s
id
ua
l
c
onn
e
c
ti
ons
e
na
bl
e
th
is
le
a
r
ni
ng
m
e
th
od
in
s
te
a
d
of
p
e
r
f
or
m
in
g
di
r
e
c
t
m
a
p
pi
ng.
T
he
te
c
hni
que
e
na
bl
e
s
th
e
tr
a
in
in
g
of
de
e
p
ne
twor
ks
by
he
lp
in
g
opt
im
iz
a
ti
on
pr
oc
e
s
s
e
s
.
T
h
e
R
e
s
N
e
t5
0
m
ode
l
hou
s
e
s
50
s
e
que
nt
ia
l
la
y
e
r
s
f
or
it
s
s
tr
uc
tu
r
e
.
T
he
c
om
pl
e
te
m
ode
l
s
tr
uc
tu
r
e
a
ppe
a
r
s
i
n F
ig
ur
e
2.
(
a
)
(
b)
F
ig
ur
e
1.
S
c
a
lo
gr
a
m
r
e
pr
e
s
e
nt
a
ti
ons
of
E
C
G
s
ig
na
ls
:
(
a
)
a
ut
he
nt
ic
E
C
G
a
nd (
b)
i
m
pos
te
r
E
C
G
F
ig
ur
e
2. R
e
s
N
e
t
m
ode
l
T
he
de
ta
il
e
d di
s
c
us
s
io
n i
s
pr
e
s
e
nt
e
d
a
s
f
ol
lo
w
s
:
a)
I
nput
la
ye
r
:
th
e
i
nput
la
ye
r
a
c
c
e
pt
s
t
he
i
nput
i
m
a
ge
da
ta
. I
n R
e
s
N
e
t,
i
nput
im
a
ge
s
a
r
e
t
ypi
c
a
ll
y r
e
s
iz
e
d t
o a
f
ix
e
d s
iz
e
, of
te
n 224
×
224 pixe
ls
,
a
nd nor
m
a
li
z
e
d.
b)
C
onvolut
io
na
l
la
ye
r
s
(
c
onv
bl
oc
ks
)
:
th
e
f
ir
s
t
la
ye
r
i
s
a
C
N
N
l
a
y
e
r
w
it
h 64 f
il
te
r
s
of
s
iz
e
7
×
7, f
ol
lo
w
e
d
by a
m
a
x
-
pool
in
g l
a
ye
r
. T
hi
s
l
a
ye
r
s
e
r
ve
s
a
s
t
he
i
ni
ti
a
l
f
e
a
tu
r
e
e
xt
r
a
c
to
r
.
c)
R
e
s
id
ua
l
bl
oc
ks
:
R
e
s
N
e
t
-
50
c
ons
is
t
s
of
16
r
e
s
id
ua
l
bl
oc
ks
,
or
g
a
ni
z
e
d
in
to
di
f
f
e
r
e
nt
s
ta
ge
s
.
E
a
c
h
r
e
s
id
ua
l
bl
oc
k c
ont
a
in
s
s
e
ve
r
a
l
c
onvolut
io
na
l
la
ye
r
s
a
nd
s
hor
tc
ut
c
onne
c
ti
ons
.
‒
S
ta
ge
1
(
C
onv2_x)
:
th
e
f
ir
s
t
s
ta
g
e
c
ont
a
in
s
one
r
e
s
id
ua
l
bl
oc
k
w
it
h
two
c
onvolut
io
na
l
la
y
e
r
s
.
T
he
out
put
of
t
hi
s
s
ta
ge
i
s
pa
s
s
e
d t
hr
ough a
c
ti
va
ti
on a
nd b
a
tc
h nor
m
a
li
z
a
ti
on l
a
ye
r
s
.
‒
S
ta
ge
2
(
C
onv3_x)
:
th
e
s
e
c
ond
s
ta
g
e
c
ont
a
in
s
th
r
e
e
r
e
s
id
ua
l
b
lo
c
ks
,
e
a
c
h
w
it
h
s
e
ve
r
a
l
c
onvolut
io
na
l
la
ye
r
s
. T
he
out
put
of
e
a
c
h bl
o
c
k i
s
pa
s
s
e
d t
hr
ough a
c
ti
va
ti
on a
n
d ba
tc
h nor
m
a
li
z
a
ti
on l
a
ye
r
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
D
e
e
p t
r
ans
fe
r
l
e
ar
ni
ng f
o
r
c
la
s
s
if
ic
at
io
n of
E
C
G
s
ig
nal
s
and li
p
i
m
age
s
i
n
…
(
L
at
ha K
r
is
hna
m
oor
th
y
)
3165
‒
S
ta
ge
3
(
C
onv4_x)
:
th
e
th
ir
d
s
ta
ge
c
ont
a
in
s
f
our
r
e
s
id
ua
l
bl
oc
k
s
,
e
a
c
h
w
it
h
s
e
ve
r
a
l
c
onvolut
io
na
l
l
a
ye
r
s
.
S
im
il
a
r
to
th
e
pr
e
vi
ous
s
t
a
ge
s
,
th
e
out
put
of
e
a
c
h
bl
oc
k
is
pa
s
s
e
d
th
r
ough
a
c
ti
va
ti
on
a
nd
b
a
tc
h
nor
m
a
li
z
a
ti
on l
a
ye
r
s
.
‒
S
ta
ge
4
(
C
onv5_x)
:
th
e
f
ou
r
th
s
ta
ge
c
ont
a
in
s
s
ix
r
e
s
id
ua
l
bl
oc
ks
,
e
a
c
h
w
it
h
s
e
ve
r
a
l
c
onvolut
io
na
l
la
ye
r
s
.
T
he
out
put
of
e
a
c
h bl
oc
k i
s
p
a
s
s
e
d t
hr
ough a
c
ti
va
ti
on a
nd ba
tc
h
nor
m
a
li
z
a
ti
on l
a
ye
r
s
.
d)
G
lo
ba
l
a
ve
r
a
ge
pool
in
g:
th
e
out
put
f
e
a
tu
r
e
m
a
p
is
pa
s
s
e
d
th
r
ough
a
gl
oba
l
a
ve
r
a
ge
pool
in
g
la
ye
r
a
f
te
r
th
e
la
s
t
r
e
s
id
ua
l
bl
oc
k.
I
t
c
r
ow
ds
th
e
s
pa
ti
a
l
di
m
e
ns
io
n
of
th
e
f
e
a
tu
r
e
m
a
p
dow
n
to
1
×
1
w
hi
le
ke
e
pi
ng
th
e
s
a
m
e
num
be
r
of
c
ha
nne
ls
.
e)
F
ul
ly
c
onne
c
te
d
la
ye
r
(
out
put
la
ye
r
)
:
th
e
n
,
th
e
f
in
a
l
out
put
is
pr
oduc
e
d
by
a
ddi
ng
a
f
ul
ly
c
onne
c
te
d
la
y
e
r
w
it
h a
c
ti
va
ti
on. W
he
n de
a
li
ng w
it
h i
m
a
ge
c
la
s
s
if
ic
a
ti
on, t
he
ou
tp
ut
i
s
t
he
pr
oba
bi
li
ti
e
s
t
ha
t
th
e
i
m
a
ge
i
s
of
a
c
la
s
s
.
3.2.2.
L
ip
i
m
age
e
xt
r
ac
t
io
n
T
hi
s
s
ubs
e
c
ti
on de
s
c
r
ib
e
s
t
he
pr
opos
e
d t
r
a
ns
f
e
r
l
e
a
r
ni
ng
-
ba
s
e
d
m
ode
l
f
or
l
ip
i
m
a
ge
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
f
or
a
ut
he
nt
ic
a
ti
on.
F
o
r
th
is
ta
s
k,
w
e
ha
ve
us
e
d
V
G
G
16
f
e
a
tu
r
e
e
xt
r
a
c
ti
on.
M
a
ny
c
om
put
e
r
vi
s
io
n
ta
s
ks
ha
ve
m
a
de
e
xt
e
ns
iv
e
us
e
of
V
G
G
16,
s
uc
h
a
s
f
e
a
tu
r
e
e
xt
r
a
c
ti
on,
obj
e
c
t
de
te
c
ti
on,
a
nd
im
a
ge
c
a
te
gor
iz
a
ti
on.
I
t
ha
s
a
ls
o
s
e
r
ve
d
a
s
a
b
a
s
e
m
ode
l
f
or
tr
a
ns
f
e
r
le
a
r
ni
ng
in
m
a
ny
a
ppl
i
c
a
ti
ons
,
w
he
r
e
th
e
pr
e
-
tr
a
in
e
d
V
G
G
16
w
e
ig
ht
s
a
r
e
f
in
e
-
tu
ne
d
on
s
pe
c
if
ic
da
t
a
s
e
t
s
f
or
c
e
r
ta
in
ta
s
k
s
.
W
it
h
nu
m
e
r
ous
c
onvolut
io
na
l
la
y
e
r
s
f
ol
lo
w
e
d
by
m
a
x
pool
in
g
la
ye
r
s
,
V
G
G
16'
s
s
ix
te
e
n
la
ye
r
s
a
r
e
gr
oupe
d
in
to
f
iv
e
gr
oups
.
F
ul
ly
c
onne
c
te
d
c
a
te
gor
iz
a
ti
on
la
ye
r
s
m
a
ke
up
th
e
la
s
t
s
e
t
of
la
ye
r
s
.
V
G
G
16
is
c
ha
r
a
c
te
r
iz
e
d
by
it
s
r
e
pe
ti
ti
ve
c
onvolut
io
na
l
bl
oc
ks
, e
a
c
h
c
ont
a
in
in
g
m
ul
ti
pl
e
3
×
3 c
onvolut
io
na
l
la
ye
r
s
f
ol
lo
w
e
d by a
m
a
x pooli
ng l
a
ye
r
. S
pe
c
if
ic
a
ll
y, t
he
a
r
c
hi
te
c
tu
r
e
c
ons
is
t
s
:
‒
C
onvolut
io
na
l
la
ye
r
s
:
s
hor
t
3
×
3
f
il
te
r
s
w
it
h
a
1
s
tr
id
e
a
nd
pa
ddi
ng
a
r
e
us
e
d
by
th
e
c
onvolut
io
na
l
la
ye
r
s
.t
o
m
a
in
ta
in
t
he
s
pa
ti
a
l
di
m
e
ns
io
ns
of
t
he
i
nput
f
e
a
tu
r
e
m
a
ps
. T
h
e
s
e
l
a
ye
r
s
a
r
e
r
e
s
pons
ib
le
f
or
l
e
a
r
ni
ng s
pa
ti
a
l
hi
e
r
a
r
c
hi
e
s
of
f
e
a
tu
r
e
s
i
n t
he
i
nput
i
m
a
ge
s
.
‒
R
e
c
ti
f
ie
d
li
ne
a
r
uni
t
(
R
e
L
U
)
a
c
ti
va
ti
on:
to
m
a
ke
th
e
ne
twor
k
non
-
li
ne
a
r
,
R
e
L
U
a
c
ti
va
ti
on
f
e
a
tu
r
e
s
a
r
e
a
dde
d
a
f
te
r
e
ve
r
y c
onvolut
io
na
l
la
ye
r
.
‒
M
a
x
pool
in
g
la
ye
r
s
:
m
a
x
pool
in
g
la
y
e
r
s
f
ol
lo
w
e
a
c
h
gr
o
up
of
c
onvolut
io
na
l
la
ye
r
s
to
de
c
r
e
a
s
e
c
om
put
a
ti
ona
l
c
om
pl
e
xi
ty
by downs
a
m
pl
in
g f
e
a
tu
r
e
m
a
ps
'
s
p
a
ti
a
l
di
m
e
ns
io
ns
a
nd c
ont
r
ol
li
ng ove
r
f
it
ti
ng.
A
f
te
r
t
he
c
onvolut
io
na
l
bl
oc
ks
, V
G
G
16
in
c
lu
de
s
t
hr
e
e
f
ul
ly
c
on
ne
c
te
d l
a
ye
r
s
f
ol
lo
w
e
d by a
S
of
tM
a
x
out
put
l
a
ye
r
. T
he
a
r
c
hi
te
c
tu
r
e
of
t
hi
s
m
ode
l
is
de
pi
c
te
d i
n
F
ig
ur
e
3. T
he
s
e
l
a
ye
r
s
s
e
r
ve
a
s
t
he
c
la
s
s
if
ie
r
f
or
t
h
e
ne
twor
k, m
a
ppi
ng t
he
e
xt
r
a
c
te
d f
e
a
tu
r
e
s
t
o c
la
s
s
pr
oba
bi
li
ti
e
s
.
‒
F
la
tt
e
n
la
ye
r
:
th
e
f
e
a
tu
r
e
m
a
ps
a
r
e
f
ir
s
t
f
la
tt
e
ne
d
in
to
a
1D
ve
c
to
r
be
f
or
e
be
in
g
f
e
d
in
to
th
e
de
ns
e
la
ye
r
s
,
w
hi
c
h a
r
e
l
oc
a
te
d pr
e
c
e
di
ng t
he
f
ul
ly
c
onne
c
te
d l
a
ye
r
s
.
‒
D
e
ns
e
l
a
ye
r
s
:
th
e
f
ul
ly
c
onne
c
te
d
la
ye
r
s
c
ont
a
in
a
la
r
ge
num
be
r
of
ne
ur
ons
,
e
na
bl
in
g
th
e
ne
twor
k
to
a
c
qui
r
e
knowle
dge
of
ge
ne
r
a
l
c
ha
r
a
c
te
r
is
ti
c
s
a
nd ge
ne
r
a
t
e
f
or
e
c
a
s
ts
u
s
i
ng t
he
r
e
tr
ie
ve
d r
e
pr
e
s
e
nt
a
ti
ons
.
‒
S
of
tM
a
x
a
c
ti
va
ti
on:
th
e
S
of
tM
a
x
a
c
ti
va
ti
on
f
unc
ti
on
in
th
e
o
ut
put
la
ye
r
c
onve
r
ts
th
e
r
a
w
out
put
of
th
e
ne
twor
k i
nt
o c
la
s
s
pr
oba
bi
li
ti
e
s
, e
na
bl
in
g m
ul
ti
-
c
la
s
s
c
la
s
s
if
ic
a
t
io
n.
F
ig
ur
e
3. V
G
G
16 a
r
c
hi
te
c
tu
r
e
3.2.3. Clas
s
if
i
c
at
io
n
T
hi
s
s
e
c
ti
on s
how
s
t
he
c
la
s
s
if
ic
a
ti
on mode
l
us
e
d f
or
bot
h E
C
G
a
nd l
ip
i
m
a
ge
da
ta
. T
hi
s
m
e
th
od us
e
s
C
N
N
ba
s
e
d
c
l
a
s
s
if
ic
a
ti
on
m
ode
l.
T
hi
s
c
la
s
s
if
ic
a
ti
on
m
ode
l
in
put
la
ye
r
,
c
onvolut
io
n
la
ye
r
,
m
a
x
pool
in
g
la
ye
r
,
f
la
tt
e
n
la
ye
r
,
de
n
s
e
la
y
e
r
,
dr
op
out
la
y
e
r
a
nd
o
ut
put
la
ye
r
.
F
ig
u
r
e
4
s
how
s
th
e
a
r
c
hi
te
c
tu
r
e
of
C
N
N
c
la
s
s
if
ie
r
.
A
r
c
hi
te
c
tu
r
e
:
in
put
(
im
a
ge
)
-
c
onvolut
io
na
l
la
ye
r
(
e
.g.,
3
×
3
f
il
te
r
s
,
R
e
L
U
a
c
ti
va
ti
on)
-
m
a
x
pool
in
g
la
ye
r
(
e
.g.,
2
×
2
pool
s
iz
e
)
-
c
onvolut
io
na
l
la
ye
r
-
m
a
x
pool
in
g
la
ye
r
-
f
la
tt
e
n
la
ye
r
-
de
ns
e
la
y
e
r
(
e
.g.,
128
ne
ur
ons
,
R
e
L
U
a
c
ti
va
ti
on)
,
dr
opout
la
ye
r
(
opt
io
na
l,
f
or
r
e
gul
a
r
iz
a
ti
on)
,
out
put
la
ye
r
(
e
.g.,
S
of
tM
a
x
a
c
ti
va
ti
on
f
or
m
ul
ti
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
4
,
A
ugus
t
2025
:
3160
-
3171
3166
F
ig
ur
e
4. A
r
c
hi
te
c
tu
r
e
of
C
N
N
c
la
s
s
if
ie
r
4.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
hi
s
s
e
c
ti
on
c
ons
is
t
s
of
di
f
f
e
r
e
nt
s
ubs
e
c
ti
on
w
he
r
e
s
ub
s
e
c
ti
on
4.1
pr
e
s
e
nt
s
th
e
da
ta
s
e
t
de
ta
il
s
,
s
e
c
ti
on
4.2
pr
e
s
e
nt
s
th
e
de
ta
il
s
of
pe
r
f
or
m
a
nc
e
m
e
a
s
ur
e
m
e
nt
pa
r
a
m
e
te
r
s
,
s
e
c
ti
on
4.3
pr
e
s
e
nt
s
th
e
r
e
s
ul
t
of
pr
opos
e
d m
ode
l
a
nd c
om
pa
r
e
s
i
t
pe
r
f
or
m
a
nc
e
w
it
h c
ur
r
e
nt
c
la
s
s
if
ic
a
ti
on s
c
he
m
e
s
.
4
.1.
D
at
as
e
t
d
e
t
ai
ls
T
he
da
ta
s
e
t
de
ve
lo
pm
e
nt
in
c
lu
de
d
c
ol
le
c
ti
ng
20
E
C
G
s
ig
na
ls
a
nd
20
f
a
c
e
im
a
ge
s
f
r
om
e
a
c
h
of
th
e
10
pa
r
ti
c
ip
a
nt
s
in
th
e
s
tu
dy.
T
he
a
c
qui
s
it
io
n
of
E
C
G
s
ig
na
ls
r
e
qui
r
e
s
a
f
il
te
r
in
g
pr
oc
e
s
s
be
c
a
u
s
e
th
e
s
ig
na
l
s
te
nd
to
be
c
om
e
c
ont
a
m
in
a
te
d.
T
he
r
e
s
e
a
r
c
h
in
ve
s
ti
ga
t
e
s
m
ul
ti
pl
e
f
or
m
s
of
E
C
G
s
ig
na
l
in
te
r
f
e
r
e
nc
e
w
hi
c
h
c
ons
is
t
of
w
hi
te
noi
s
e
a
nd
c
ol
or
noi
s
e
a
lo
n
g
w
it
h
m
ot
io
n
a
r
ti
f
a
c
ts
e
le
c
tr
ode
a
r
ti
f
a
c
ts
a
nd
b
a
s
e
li
ne
w
a
nde
r
.
T
o
e
xt
r
a
c
t
th
e
f
a
c
e
li
p
r
e
gi
on
f
r
om
im
a
ge
s
th
e
"
dl
ib
"
P
yt
ho
n
li
br
a
r
y
de
te
c
ts
f
a
c
ia
l
la
ndm
a
r
ks
.
T
hi
s
w
or
k
e
xc
lu
de
d
s
c
e
n
a
r
io
a
na
ly
s
is
f
or
obs
tr
uc
te
d
r
e
gi
ons
be
c
a
us
e
th
e
m
a
in
r
e
s
e
a
r
c
h
obj
e
c
ti
ve
w
a
s
th
e
e
xt
r
a
c
ti
on
of
li
p
a
r
e
a
s
.
T
h
e
te
a
m
a
ppl
ie
d
da
ta
a
ugm
e
nt
a
ti
on
m
e
th
ods
to
ge
ne
r
a
te
va
r
io
us
da
ta
s
e
t
s
f
or
e
a
c
h
us
e
r
be
c
a
us
e
it
im
pr
ove
d
th
e
r
e
li
a
bi
li
ty
of
e
xt
r
a
c
te
d
f
e
a
tu
r
e
s
.
T
he
E
C
G
s
ig
na
l
s
a
m
pl
e
in
F
ig
ur
e
5
di
s
pl
a
ys
two
di
m
e
n
s
io
ns
w
he
r
e
t
im
e
f
lo
w
s
a
c
r
os
s
t
he
hor
iz
ont
a
l
a
xi
s
a
nd a
m
pl
it
ude
r
is
e
s
on t
he
ve
r
ti
c
a
l
a
xi
s
.
T
he
unf
il
te
r
e
d
s
ig
na
ls
a
nd
th
e
ir
f
il
te
r
e
d
c
ount
e
r
pa
r
ts
a
r
e
s
how
n
in
F
ig
ur
e
6.
W
hi
te
noi
s
e
is
s
how
n
in
F
ig
ur
e
6(
a
)
,
c
ol
or
noi
s
e
in
F
ig
ur
e
6(
b)
,
m
ot
io
n
a
r
ti
f
a
c
t
in
F
ig
ur
e
6(
c
)
,
e
le
c
tr
ode
a
r
ti
f
a
c
t
in
F
ig
ur
e
6(
d)
,
a
nd
ba
s
e
li
ne
w
a
nd
e
r
in
F
ig
ur
e
6(
e
)
.
T
he
il
lu
s
t
r
a
ti
ons
of
e
a
c
h
ki
nd
of
noi
s
e
a
nd
f
il
te
r
e
d
s
ig
na
l
s
how
n
in
F
ig
ur
e
6.
T
he
in
put
s
ig
na
l,
w
hi
te
noi
s
e
,
c
ol
or
noi
s
e
,
a
r
ti
f
a
c
ts
c
a
u
s
e
d
b
y
m
ot
io
n,
a
r
ti
f
a
c
ts
c
a
us
e
d
by
e
le
c
tr
ode
s
,
a
nd
ba
s
e
li
ne
w
a
nd
e
r
,
a
s
w
e
ll
a
s
th
e
f
il
te
r
e
d
s
ig
na
ls
u
s
e
d
f
or
bi
om
e
tr
ic
a
ut
he
nt
ic
a
t
io
n
w
e
r
e
s
how
n
in
F
ig
ur
e
6.
W
e
ha
ve
a
l
s
o
ut
il
iz
e
d
f
a
c
ia
l
la
ndm
a
r
k
de
t
e
c
ti
on
to
e
xt
r
a
c
t
th
e
li
p
r
e
gi
on
f
r
om
f
a
c
e
phot
os
in
a
s
im
il
a
r
ve
in
.
T
he
f
a
c
e
im
a
ge
,
e
xt
r
a
c
te
d
li
p
im
a
ge
a
nd
e
nha
nc
e
d
im
a
ge
a
r
e
gi
ve
n
in
F
ig
ur
e
7.
L
a
ndm
a
r
k
id
e
nt
if
ic
a
ti
on
pr
oc
e
dur
e
is
us
e
d
to
g
e
t
c
oor
di
na
te
s
of
th
e
r
e
duc
e
d
a
nd e
nha
nc
e
d
li
p
r
e
gi
on
a
nd
a
ppl
y
hor
iz
ont
a
l
a
nd
ve
r
ti
c
a
l
im
a
ge
f
li
ppi
ng.
F
ig
ur
e
5
. S
a
m
pl
e
E
C
G
s
ig
na
l
f
o
r
i
nput
4.2. P
e
r
f
or
m
an
c
e
m
e
as
u
r
e
m
e
n
t
p
ar
a
m
e
t
e
r
s
E
va
lu
a
ti
on
of
th
e
pr
opos
e
d
m
e
th
od
de
pe
nds
on
c
onf
us
io
n
m
a
tr
ix
c
a
lc
ul
a
ti
ons
.
T
h
e
c
onf
us
io
n
m
a
tr
ix
c
r
e
a
ti
on
de
pe
nds
on
tr
ue
pos
it
iv
e
(
T
P
)
,
f
a
ls
e
pos
it
iv
e
(
F
P
)
,
f
a
ls
e
ne
ga
ti
ve
(
F
N
)
,
a
nd
tr
ue
ne
ga
ti
ve
(
T
N
)
.
T
a
bl
e
1 i
ll
us
tr
a
te
s
t
hi
s
c
la
s
s
e
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
D
e
e
p t
r
ans
fe
r
l
e
ar
ni
ng f
o
r
c
la
s
s
if
ic
at
io
n of
E
C
G
s
ig
nal
s
and li
p
i
m
age
s
i
n
…
(
L
at
ha K
r
is
hna
m
oor
th
y
)
3167
(
a
)
(
b)
(
c
)
(
d)
(
e
)
F
ig
ur
e
6
. N
oi
s
e
a
dde
d E
C
G
s
ig
na
l
a
nd f
il
te
r
e
d s
ig
na
ls
f
or
bi
om
e
tr
ic
a
ut
he
nt
ic
a
ti
on (
a
)
w
hi
te
noi
s
e
,
(
b)
c
ol
or
noi
s
e
, (
c
)
m
ot
io
n a
r
ti
f
a
c
ts
, (
d)
e
le
c
tr
ode
a
r
ti
f
a
c
ts
, a
nd (
e
)
ba
s
e
li
ne
w
a
nde
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
4
,
A
ugus
t
2025
:
3160
-
3171
3168
T
a
bl
e
1. C
onf
us
io
n
m
a
tr
ix
c
la
s
s
e
s
A
c
t
ua
l
c
l
a
s
s
P
r
e
di
c
t
e
d c
l
a
s
s
A
ut
he
nt
i
c
I
m
pos
t
e
r
A
ut
he
nt
i
c
TP
FN
I
m
pos
t
e
r
FP
TN
W
e
c
a
lc
ul
a
t
e
a
c
c
ur
a
c
y
a
nd
pr
e
c
is
io
n
a
nd
F
1
-
s
c
or
e
by
us
in
g
our
pr
opos
e
d
a
ppr
oa
c
h
ba
s
e
d
on
th
e
c
onf
us
io
n
m
a
tr
ix
da
ta
.
O
ut
of
a
ll
to
ta
l
num
b
e
r
of
in
s
ta
nc
e
s
a
c
c
ur
a
c
y
r
e
pr
e
s
e
nt
s
th
e
pr
opor
ti
on
of
pr
ope
r
ly
c
la
s
s
if
ie
d c
a
s
e
s
. T
he
a
c
c
ur
a
c
y d
e
te
r
m
in
a
ti
on f
ol
lo
w
s
t
hi
s
m
e
th
od.
=
+
+
+
+
(
2)
P
r
e
c
is
io
n
e
va
lu
a
ti
on
ta
ke
s
pl
a
c
e
f
or
th
e
pr
oduc
e
d
m
e
th
od.
T
he
p
r
e
c
is
io
n
e
va
lu
a
ti
on
in
c
lu
de
s
a
c
a
lc
ul
a
ti
on
us
in
g
T
P
i
ns
ta
nc
e
s
a
nd t
he
ir
c
om
bi
na
ti
on w
it
h t
r
ue
a
nd f
a
ls
e
i
ns
t
a
nc
e
s
.
=
+
(
3)
L
a
s
tl
y, t
he
c
om
put
a
ti
on of
F
-
m
e
a
s
ur
e
de
pe
nds
on t
he
s
e
ns
it
iv
it
y a
nd pr
e
c
is
io
n va
lu
e
s
:
”
=
2
×
×
+
(
4)
F
ig
ur
e
7
. F
a
c
e
i
m
a
ge
, c
or
r
e
s
ponding f
a
c
ia
l
la
ndm
a
r
ks
a
nd
a
ug
m
e
nt
a
ti
on
4.3. Com
p
ar
at
iv
e
an
al
ys
is
T
he
pr
opos
e
d
m
od
e
l
m
e
a
s
ur
e
s
it
s
pe
r
f
or
m
a
nc
e
u
s
in
g
a
c
c
ur
a
c
y, pr
e
c
is
io
n
a
lo
ngs
id
e
F
1
-
s
c
or
e
m
e
tr
ic
s
w
hi
c
h
a
r
e
a
s
s
e
s
s
e
d
a
ga
in
s
t
c
ur
r
e
nt
be
nc
hm
a
r
k
c
la
s
s
if
ic
a
ti
on
m
ode
ls
.
T
he
c
ur
r
e
nt
out
put
pe
r
f
or
m
a
nc
e
is
a
s
s
e
s
s
e
d
a
ga
in
s
t
a
ll
e
xi
s
ti
ng
m
a
c
hi
ne
le
a
r
ni
ng
a
nd
de
e
p
le
a
r
ni
ng
c
la
s
s
if
ie
r
s
.
T
he
m
a
c
hi
ne
le
a
r
ni
ng
c
la
s
s
if
ic
a
ti
on
m
ode
l
pe
r
f
or
m
a
nc
e
r
e
s
ul
ts
a
ppe
a
r
i
n T
a
bl
e
2.
T
a
bl
e
2. C
om
pa
r
a
ti
ve
a
na
ly
s
e
s
w
it
h
m
a
c
hi
ne
l
e
a
r
ni
ng
te
c
hni
que
s
P
a
r
a
m
e
t
e
r
S
V
M
N
e
ur
a
l
n
e
t
w
or
k
RF
DT
P
r
opos
e
d
A
c
c
ur
a
c
y
0.651
0.712
0.781
0.835
0.962
F1
-
s
c
or
e
0.655
0.698
0.772
0.820
0.970
P
r
e
c
i
s
i
on
0.648
0.711
0.795
0.815
0.965
R
e
c
a
l
l
0.658
0.728
0.735
0.805
0.966
S
e
ns
i
t
i
vi
t
y
0.655
0.711
0.733
0.842
0.977
S
pe
c
i
f
i
c
i
t
y
0.661
0.715
0.785
0.835
0.985
A
c
c
or
di
ng
to
th
is
e
xpe
r
im
e
nt
,
th
e
pr
opos
e
d
de
e
p
tr
a
ns
f
e
r
le
a
r
ni
n
g
m
ode
l
r
e
por
te
d
hi
ghe
r
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
a
s
96.2%
w
he
r
e
a
s
th
e
S
V
M
c
l
a
s
s
if
ie
r
r
e
por
te
d
th
e
l
ow
e
s
t
a
c
c
ur
a
c
y
a
s
65.10%
.
T
he
S
V
M
m
ode
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
D
e
e
p t
r
ans
fe
r
l
e
ar
ni
ng f
o
r
c
la
s
s
if
ic
at
io
n of
E
C
G
s
ig
nal
s
and li
p
i
m
age
s
i
n
…
(
L
at
ha K
r
is
hna
m
oor
th
y
)
3169
c
ons
id
e
r
s
s
im
pl
e
te
xt
ur
e
,
c
ol
or
a
nd
s
h
a
pe
f
e
a
tu
r
e
s
w
he
r
e
a
s
pr
op
os
e
d
m
ode
l
us
e
s
tr
a
n
s
f
e
r
le
a
r
ni
ng
m
ode
l
w
he
r
e
E
C
G
s
ig
na
l
is
c
onve
r
te
d
in
to
im
a
ge
a
nd
th
e
n
de
e
p
f
e
a
tu
r
e
s
a
r
e
e
xt
r
a
c
te
d.
F
ig
ur
e
8
de
pi
c
ts
th
e
obt
a
in
e
d
pe
r
f
or
m
a
nc
e
a
na
ly
s
is
.
F
ur
th
e
r
,
w
e
c
om
pa
r
e
th
e
obt
a
in
e
d
pe
r
f
or
m
a
nc
e
w
it
h
di
f
f
e
r
e
nt
de
e
p
le
a
r
ni
ng
-
ba
s
e
d
c
la
s
s
if
ic
a
ti
on
m
ode
ls
. T
he
obt
a
in
e
d c
om
pa
r
a
ti
ve
a
na
ly
s
is
i
s
de
pi
c
te
d i
n be
lo
w
gi
ve
n T
a
bl
e
3. A
c
c
or
di
ng t
o
t
hi
s
e
xpe
r
im
e
nt
,
th
e
pr
opos
e
d
m
ode
l
ha
s
r
e
por
te
d
hi
ghe
s
t
a
c
c
ur
a
c
y
w
he
r
e
a
s
th
e
ot
he
r
de
e
p
le
a
r
ni
ng
c
la
s
s
if
ie
r
s
ha
ve
r
e
por
te
d
0.843,
0.855,
0.858,
a
nd
0.901
by
us
in
g
C
N
N
,
L
S
T
M
,
ga
te
d
r
e
c
ur
r
e
nt
uni
t
(
G
R
U
)
,
a
nd
C
N
N
-
L
S
T
M
c
la
s
s
if
ie
r
s
,
r
e
s
pe
c
ti
ve
ly
. T
h
e
c
om
pa
r
a
ti
ve
a
na
ly
s
is
de
pi
c
t
e
d i
n F
ig
ur
e
9
.
F
ig
ur
e
8
. C
om
pa
r
a
ti
ve
a
na
ly
s
is
w
it
h di
f
f
e
r
e
nt
m
a
c
hi
ne
l
e
a
r
ni
ng
te
c
hni
que
s
T
a
bl
e
3. C
om
pa
r
a
ti
ve
a
na
ly
s
e
s
w
it
h de
e
p l
e
a
r
ni
ng t
e
c
hni
que
s
P
a
r
a
m
e
t
e
r
C
N
N
L
S
T
M
G
R
U
C
N
N
-
L
S
T
M
P
r
opos
e
d
A
c
c
ur
a
c
y
0.843
0.855
0.858
0.901
0.962
F1
-
s
c
or
e
0.855
0.850
0.875
0.911
0.970
P
r
e
c
i
s
i
on
0.835
0.856
0.855
0.902
0.965
R
e
c
a
l
l
0.865
0.842
0.835
0.908
0.966
S
e
ns
i
t
i
vi
t
y
0.860
0.855
0.833
0.911
0.977
S
pe
c
i
f
i
c
i
t
y
0.866
0.861
0.855
0.912
0.985
F
ig
ur
e
9
. C
om
pa
r
a
ti
ve
a
na
ly
s
is
w
it
h di
f
f
e
r
e
nt
de
e
p l
e
a
r
ni
ng
m
e
th
ods
5.
C
O
N
C
L
U
S
I
O
N
T
hi
s
s
tu
dy
e
xpl
or
e
d
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
m
ul
ti
m
oda
l
a
ut
he
nt
ic
a
ti
on
le
ve
r
a
gi
ng
R
e
s
N
e
t
a
nd
V
G
G
16
tr
a
ns
f
e
r
le
a
r
ni
ng
f
or
f
e
a
tu
r
e
e
xt
r
a
c
ti
on,
f
ol
lo
w
e
d
by
C
N
N
f
or
c
la
s
s
if
ic
a
ti
on.
T
he
r
e
a
l
-
ti
m
e
da
ta
s
e
t
c
om
pr
is
e
d
E
C
G
s
ig
na
ls
a
nd
f
a
c
e
li
p
im
a
ge
s
f
r
om
in
di
vi
dua
ls
,
r
e
pr
e
s
e
nt
in
g
di
s
ti
nc
t
bi
om
e
tr
ic
m
oda
li
ti
e
s
.
T
hr
ough
th
e
in
te
gr
a
ti
on
of
R
e
s
N
e
t
a
nd
V
G
G
16
tr
a
ns
f
e
r
le
a
r
ni
ng
te
c
hni
que
s
,
r
ic
h
a
nd
di
s
c
r
im
in
a
ti
ve
f
e
a
tu
r
e
s
w
e
r
e
e
xt
r
a
c
te
d
f
r
om
th
e
E
C
G
s
ig
na
ls
a
nd
f
a
c
e
li
p
im
a
ge
s
.
L
e
ve
r
a
gi
ng
th
e
pr
e
-
tr
a
in
e
d
m
ode
ls
a
ll
ow
e
d
f
or
th
e
ut
il
iz
a
ti
on
o
f
de
e
p
hi
e
r
a
r
c
hi
c
a
l
r
e
pr
e
s
e
nt
a
ti
ons
,
e
nha
n
c
in
g
th
e
r
obus
tn
e
s
s
a
nd
di
s
c
r
im
in
a
ti
ve
pow
e
r
of
th
e
e
xt
r
a
c
te
d
f
e
a
tu
r
e
s
.
S
ubs
e
que
nt
ly
,
a
C
N
N
c
la
s
s
if
ie
r
w
a
s
e
m
pl
oye
d
to
c
la
s
s
if
y
th
e
e
xt
r
a
c
te
d
f
e
a
tu
r
e
s
a
nd
a
ut
he
nt
ic
a
te
u
s
e
r
s
ba
s
e
d
0
0
.2
0
.4
0
.6
0
.8
1
1
.2
A
c
c
u
r
a
c
y
F
1
-
s
c
o
r
e
P
r
ec
i
s
i
o
n
R
ec
a
l
l
S
en
s
i
t
i
v
i
t
y
S
p
e
c
i
f
i
c
i
t
y
O
b
t
a
i
n
ed
P
er
f
o
r
m
a
n
c
e
P
a
r
a
m
et
er
s
Com
p
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
w
i
t
h
m
a
c
h
i
n
e
l
ea
r
n
i
n
g
t
ec
h
n
i
q
u
es
S
V
M
N
eu
r
a
l
N
et
w
o
r
k
RF
DT
P
r
o
p
o
s
ed
0
.7
5
0
.8
0
.8
5
0
.9
0
.9
5
1
A
c
c
u
r
a
c
y
F
1
-
s
c
o
r
e
P
r
ec
i
s
i
o
n
R
ec
a
l
l
S
en
s
i
t
i
v
i
t
y
S
p
e
c
i
f
i
c
i
t
y
O
b
t
a
i
n
ed
P
er
f
o
r
m
a
n
c
e
P
a
r
a
m
et
er
s
Com
p
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
w
i
t
h
d
ee
p
l
ea
r
n
i
n
g
t
ec
h
n
i
q
u
es
CN
N
L
S
T
M
G
R
U
CN
N
-
L
S
T
M
P
r
o
p
o
s
ed
Evaluation Warning : The document was created with Spire.PDF for Python.