I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
5
,
No
.
4
,
A
u
g
u
s
t 2
0
2
6
,
p
p
.
3
2
5
2
~
3
2
6
8
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
2
5
2
-
3
2
6
8
3252
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Bio
metric
a
uth
en
ticatio
n using
d
ua
l
-
mo
da
l deep
l
ea
r
ning
ba
sed
-
o
n
elec
troc
a
rdio
g
ra
m
a
nd ear featu
res
M
o
ha
m
ed
S.
K
ha
la
f
1
,
Sa
id F
a
t
hy
Al
-
Z
o
g
hd
y
1
,
M
a
ria
na
B
a
rso
um
1
,
I
bra
him
O
ma
ra
2,
3
1
D
e
p
a
r
t
me
n
t
o
f
M
a
t
h
e
m
a
t
i
c
s
a
n
d
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
,
M
e
n
o
u
f
i
a
U
n
i
v
e
r
s
i
t
y
,
S
h
e
b
i
n
El
-
K
o
m
,
E
g
y
p
t
2
D
e
p
a
r
t
me
n
t
o
f
C
y
b
e
r
se
c
u
r
i
t
y
,
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
a
n
d
I
n
f
o
r
ma
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
B
u
r
a
y
d
a
h
P
r
i
v
a
t
e
C
o
l
l
e
g
e
s
,
B
u
r
a
y
d
a
h
,
S
a
u
d
i
A
r
a
b
i
a
3
D
e
p
a
r
t
me
n
t
o
f
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
F
a
c
u
l
t
y
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
M
e
n
o
u
f
i
a
U
n
i
v
e
r
s
i
t
y
,
S
h
e
b
i
n
El
-
K
o
m,
Eg
y
p
t
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Oct
1
1
,
2
0
2
5
R
ev
is
ed
J
u
n
1
2
,
2
0
2
6
Acc
ep
ted
J
u
l 9
,
2
0
2
6
Bio
m
e
tri
c
a
u
t
h
e
n
ti
c
a
ti
o
n
s
y
ste
m
s
a
re
e
ss
e
n
ti
a
l
fo
r
se
c
u
re
a
c
c
e
ss
c
o
n
tr
o
l;
h
o
we
v
e
r
u
n
imo
d
a
l
sy
ste
m
s
a
re
v
u
l
n
e
ra
b
le
to
sp
o
o
fi
n
g
a
n
d
e
n
v
i
ro
n
m
e
n
tal
v
a
riatio
n
s.
Th
is
st
u
d
y
p
ro
p
o
se
s
a
d
u
a
l
-
m
o
d
a
l
b
i
o
m
e
tri
c
sy
ste
m
c
o
m
b
in
i
n
g
e
lec
tro
c
a
rd
io
g
ra
m
(ECG
)
sig
n
a
ls
a
n
d
e
a
r
fe
a
tu
re
s
t
o
e
n
h
a
n
c
e
se
c
u
rit
y
a
n
d
a
c
c
u
ra
c
y
.
D
e
e
p
lea
rn
i
n
g
a
rc
h
it
e
c
tu
re
s
in
c
lu
d
in
g
VGG
-
v
e
ry
d
e
e
p
1
6
a
n
d
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
5
(
CNN
5
)
a
re
e
v
a
lu
a
ted
fo
r
fe
a
tu
re
e
x
trac
ti
o
n
a
n
d
fu
si
o
n
.
Ex
p
e
rime
n
tal
re
su
lt
s
sh
o
w
th
a
t
h
y
b
rid
m
o
d
e
l
(VG
G
-
v
e
ry
d
e
e
p
1
6
+
CNN
5
)
a
c
h
iev
e
s
a
ro
u
n
d
9
8
%
a
c
c
u
ra
c
y
,
o
u
tp
e
rfo
rm
i
n
g
i
n
d
iv
i
d
u
a
l
m
o
d
e
ls.
Th
e
sy
ste
m
a
d
a
p
ts
to
d
a
tas
e
t
siz
e
,
u
si
n
g
CNN
5
fo
r
lar
g
e
d
a
tas
e
ts
a
n
d
VG
G
-
v
e
ry
d
e
e
p
1
6
f
o
r
sm
a
ll
e
r
o
n
e
s.
Th
is
a
p
p
ro
a
c
h
o
ffe
rs
a
ro
b
u
st,
e
ffi
c
ien
t,
a
n
d
sc
a
lab
le so
lu
ti
o
n
f
o
r
re
a
l
-
wo
rl
d
b
i
o
m
e
tri
c
a
u
th
e
n
t
ica
ti
o
n
.
K
ey
w
o
r
d
s
:
B
io
m
etr
ic
au
th
en
ticatio
n
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
5
Dee
p
lear
n
in
g
E
ar
r
ec
o
g
n
itio
n
E
lectr
o
ca
r
d
io
g
r
am
Mu
lti
-
m
o
d
al
f
u
s
io
n
VGG
-
v
er
y
d
ee
p
1
6
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
I
b
r
ah
im
Om
a
r
a
Dep
ar
tm
en
t o
f
C
y
b
er
s
ec
u
r
ity
,
C
o
lleg
e
o
f
E
n
g
i
n
ee
r
in
g
an
d
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
B
u
r
ay
d
ah
Priv
ate
C
o
lleg
es
B
u
r
ay
d
ah
,
Sau
d
i A
r
ab
ia
E
m
ail:
i_
o
m
ar
a8
4
@
ai.
m
en
o
f
i
a.
ed
u
.
eg
1.
I
NT
RO
D
UCT
I
O
N
B
io
m
etr
ic
au
th
en
ticatio
n
s
y
s
tem
s
h
av
e
b
ec
o
m
e
ess
en
tial
f
o
r
s
ec
u
r
e
ac
ce
s
s
co
n
tr
o
l
in
a
p
p
licatio
n
s
r
an
g
in
g
f
r
o
m
p
er
s
o
n
al
d
ev
ice
s
to
lar
g
e
-
s
ca
le
o
r
g
a
n
izatio
n
al
s
y
s
tem
s
[
1
]
.
Un
im
o
d
al
s
y
s
tem
s
,
wh
ich
r
el
y
o
n
a
s
in
g
le
b
io
m
etr
ic
m
o
d
ality
s
u
ch
as
f
in
g
e
r
p
r
in
ts
o
r
f
ac
ial
r
ec
o
g
n
itio
n
,
o
f
f
er
s
im
p
licity
b
u
t
f
ac
e
s
ig
n
if
ica
n
t
lim
itatio
n
s
[
2
]
.
T
h
ese
in
clu
d
e
v
u
ln
e
r
ab
ilit
y
to
s
p
o
o
f
i
n
g
atta
ck
s
(
e.
g
.
,
f
ak
e
f
in
g
er
p
r
i
n
ts
o
r
p
h
o
to
s
)
,
s
en
s
itiv
ity
to
en
v
ir
o
n
m
en
tal
f
ac
to
r
s
lik
e
lig
h
tin
g
o
r
m
o
is
tu
r
e,
an
d
h
ig
h
er
er
r
o
r
r
ates
in
s
itu
atio
n
s
,
s
u
ch
as
p
ar
tial
o
cc
lu
s
io
n
o
r
s
en
s
o
r
wea
r
[
3
]
.
T
o
ad
d
r
ess
th
ese
lim
itatio
n
s
,
r
ec
en
t
s
tu
d
ies
h
av
e
ex
p
l
o
r
ed
m
u
ltimo
d
al
b
i
o
m
etr
ic
s
y
s
tem
s
[
4
]
–
[
6
]
.
Sev
e
r
al
ap
p
r
o
ac
h
es
co
m
b
in
e
m
o
d
alities
s
u
ch
as
f
ac
e
an
d
v
o
ice
r
ec
o
g
n
itio
n
[
5
]
,
ir
is
an
d
f
in
g
er
p
r
in
t
au
th
en
ticatio
n
[
7
]
,
o
r
g
ait
an
d
p
alm
v
ein
p
atter
n
s
[
8
]
.
Alth
o
u
g
h
th
ese
s
y
s
tem
s
im
p
r
o
v
e
r
o
b
u
s
tn
ess
,
th
ey
m
ay
s
till
s
u
f
f
er
f
r
o
m
s
p
o
o
f
in
g
u
s
in
g
h
ig
h
-
q
u
ality
r
ep
li
ca
s
o
r
m
an
ip
u
lated
v
id
eo
s
[
9
]
.
I
n
ad
d
itio
n
,
m
an
y
m
u
ltimo
d
al
s
o
lu
tio
n
s
r
e
q
u
ir
e
s
u
b
s
tan
tial
co
m
p
u
tatio
n
al
r
eso
u
r
ce
s
,
lim
itin
g
th
eir
s
u
ita
b
ilit
y
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
s
o
r
d
ev
ices w
ith
lim
ited
p
r
o
ce
s
s
in
g
ca
p
a
b
ilit
y
[
1
0
]
.
E
lectr
o
ca
r
d
io
g
r
am
(
E
C
G
)
s
ig
n
als
h
av
e
r
ec
en
tly
em
er
g
ed
a
s
p
r
o
m
is
in
g
b
io
m
et
r
ic
m
o
d
ali
ty
d
u
e
to
th
eir
in
tr
in
s
ic
lin
k
to
p
h
y
s
io
lo
g
ical
an
d
b
eh
av
io
r
al
tr
aits
,
m
a
k
in
g
th
em
d
if
f
ic
u
lt
to
r
e
p
licate
o
r
s
p
o
o
f
[
6
]
,
[
1
1
]
.
Similar
ly
,
ea
r
im
ag
es
r
ep
r
esen
t
a
s
tab
le
b
io
m
etr
ic
tr
ait,
as
ea
r
f
ea
tu
r
es
ar
e
u
n
iq
u
e
an
d
r
elati
v
ely
u
n
a
f
f
ec
ted
b
y
f
ac
ial
ex
p
r
ess
io
n
s
o
r
lig
h
tin
g
co
n
d
itio
n
s
[
3
]
,
[
1
2
]
,
[
1
3
]
.
Ho
wev
er
,
s
tan
d
alo
n
e
E
C
G
-
o
r
ea
r
-
b
ased
s
y
s
tem
s
m
a
y
s
till
ex
p
er
ien
ce
lim
itatio
n
s
in
ac
cu
r
ac
y
an
d
r
eliab
ilit
y
u
n
d
er
d
iv
er
s
e
co
n
d
itio
n
s
[
1
4
]
.
I
n
teg
r
atin
g
th
ese
two
m
o
d
alities
ca
n
ex
p
lo
it th
eir
co
m
p
lem
en
tar
y
s
tr
en
g
th
s
wh
ile
r
ed
u
cin
g
i
n
d
iv
id
u
al
wea
k
n
ess
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
B
io
metric a
u
th
en
tica
tio
n
u
s
in
g
d
u
a
l
-
mo
d
a
l
d
ee
p
lea
r
n
in
g
b
a
s
ed
-
o
n
E
C
G
a
n
d
…
(
Mo
h
a
m
ed
S
.
K
h
a
la
f
)
3253
Mu
ltimo
d
al
b
io
m
etr
ic
s
y
s
tem
s
co
m
b
in
in
g
ea
r
an
d
E
C
G
ca
p
tu
r
e
b
o
th
an
ato
m
ical
an
d
p
h
y
s
io
lo
g
ical
ch
ar
ac
ter
is
tics
.
T
h
e
ea
r
p
r
o
v
i
d
es
s
tr
u
ctu
r
al
s
tab
ilit
y
,
wh
ile
E
C
G
en
s
u
r
es
liv
elin
es
s
an
d
s
tr
o
n
g
r
esis
tan
ce
to
s
p
o
o
f
in
g
.
R
ec
en
t
r
esear
ch
h
as
u
tili
ze
d
d
ee
p
lear
n
in
g
tech
n
iq
u
es
f
o
r
f
ea
tu
r
e
ex
tr
a
ctio
n
an
d
f
u
s
io
n
.
Kh
alaf
et
a
l.
[
6
]
p
r
o
p
o
s
ed
a
m
u
ltimo
d
al
s
y
s
tem
in
teg
r
atin
g
ea
r
im
ag
in
g
a
n
d
E
C
G
s
ig
n
als
u
s
in
g
s
ep
ar
ate
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NNs)
,
ac
h
iev
in
g
9
9
.
4
7
%
ac
cu
r
ac
y
.
E
l
-
R
ah
m
an
an
d
Allu
h
aid
an
[
1
5
]
d
ev
elo
p
e
d
a
s
in
g
le
-
ea
r
wea
r
a
b
le
with
in
-
ea
r
elec
tr
o
d
es
an
d
an
o
p
tim
ized
d
ee
p
m
atc
h
ed
f
ilter
alg
o
r
ith
m
th
a
t
ac
h
iev
ed
9
5
%
r
ec
all
an
d
ap
p
r
o
x
im
ately
9
1
%
p
r
ec
is
io
n
f
o
r
R
-
p
ea
k
d
etec
tio
n
ac
r
o
s
s
3
6
s
u
b
jects.
Mo
h
am
ed
et
a
l.
[
1
6
]
in
t
r
o
d
u
c
ed
a
C
NN
-
b
ased
s
y
s
tem
tr
ain
ed
o
n
m
ath
em
atica
l
an
al
y
s
is
o
f
im
ag
es
(
AM
I
)
an
d
ea
r
Vietn
am
1
.
0
(
E
ar
NV1
.
0
)
d
atasets
,
ac
h
iev
in
g
9
8
%
ea
r
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
.
Su
m
alath
a
et
a
l.
[
1
7
]
p
r
o
p
o
s
ed
a
tu
n
ab
le
f
ilter
b
an
k
an
d
d
ee
p
f
ea
tu
r
e
f
u
s
io
n
f
r
am
ewo
r
k
co
m
b
in
i
n
g
E
C
G
an
d
ea
r
m
o
d
alities
u
s
in
g
v
is
u
al
g
eo
m
etr
y
g
r
o
u
p
(
VGG
)
-
v
er
y
d
ee
p
1
6
f
o
r
f
ea
t
u
r
e
ex
tr
ac
tio
n
,
ac
h
iev
in
g
ab
o
u
t
9
8
.
7
5
%
ac
cu
r
ac
y
with
h
ig
h
er
c
o
m
p
u
tatio
n
al
tim
e.
T
o
id
en
tify
th
e
m
o
s
t
s
u
itab
le
m
o
d
el,
th
is
s
tu
d
y
ev
alu
ates
s
ev
er
al
d
ee
p
lear
n
in
g
ar
c
h
itectu
r
es
in
clu
d
in
g
VGG
-
v
e
r
y
d
ee
p
1
6
[
1
8
]
,
an
d
m
u
ltip
le
C
NN
m
o
d
els
(
C
NN5
–
C
NN8
)
in
ter
m
s
o
f
ac
cu
r
ac
y
,
p
r
o
ce
s
s
in
g
tim
e,
an
d
s
ca
lab
ilit
y
.
W
h
ile
VGG
-
v
er
y
d
ee
p
1
6
p
r
o
v
id
es
h
ig
h
ac
c
u
r
ac
y
,
it
is
c
o
m
p
u
tatio
n
ally
in
ten
s
iv
e
[
1
9
]
.
C
NN5
o
f
f
er
s
f
aster
p
er
f
o
r
m
a
n
ce
with
s
lig
h
tly
r
ed
u
ce
d
ac
cu
r
ac
y
.
T
h
e
r
ef
o
r
e
,
a
h
y
b
r
id
f
r
a
m
ewo
r
k
co
m
b
i
n
in
g
VGG
-
v
er
y
d
ee
p
1
6
an
d
C
NN5
is
p
r
o
p
o
s
ed
to
b
alan
ce
ac
cu
r
a
cy
an
d
e
f
f
icien
cy
d
ep
e
n
d
in
g
o
n
d
ataset
s
ize.
T
h
e
co
n
tr
ib
u
tio
n
s
o
f
th
is
wo
r
k
ar
e
th
r
ee
f
o
ld
:
‒
Dev
elo
p
m
en
t
o
f
a
r
o
b
u
s
t
d
u
al
-
m
o
d
al
b
io
m
etr
ic
au
th
e
n
ticatio
n
s
y
s
tem
c
o
m
b
in
in
g
E
C
G
s
i
g
n
als
an
d
ea
r
f
ea
tu
r
es to
o
v
e
r
co
m
e
t
h
e
lim
itatio
n
s
o
f
tr
ad
itio
n
al
u
n
im
o
d
al
an
d
s
o
m
e
ex
is
tin
g
m
u
lti
-
m
o
d
a
l sy
s
tem
s
.
‒
E
v
alu
atio
n
o
f
d
ee
p
lear
n
i
n
g
a
r
ch
itectu
r
es
(
VGG
-
v
er
y
d
ee
p
1
6
an
d
o
th
e
r
C
NN
m
o
d
els)
to
d
eter
m
in
e
th
e
m
o
s
t e
f
f
ec
tiv
e
m
o
d
el
f
o
r
th
e
p
r
o
p
o
s
ed
d
u
al
-
m
o
d
al
ap
p
r
o
ac
h
.
‒
Pro
p
o
s
in
g
a
h
y
b
r
id
s
y
s
tem
th
at
b
alan
ce
s
s
p
ee
d
an
d
ac
cu
r
ac
y
,
en
s
u
r
in
g
s
ca
lab
ilit
y
,
an
d
ad
ap
tab
ilit
y
ac
r
o
s
s
d
iv
er
s
e
ap
p
licatio
n
s
ce
n
ar
io
s
.
T
h
is
p
ap
er
is
o
r
g
an
ized
as
f
o
llo
ws
.
Sectio
n
2
p
r
o
v
id
es
an
o
v
er
v
iew
o
f
r
ec
e
n
t
wo
r
k
s
in
b
io
m
etr
ic
au
th
en
ticatio
n
,
h
i
g
h
lig
h
tin
g
th
eir
ad
v
an
ce
m
e
n
ts
an
d
lim
itatio
n
s
.
Sectio
n
3
o
u
tlin
es
th
e
p
r
o
p
o
s
ed
d
u
al
-
m
o
d
al
s
y
s
tem
an
d
m
o
d
el
d
esig
n
.
Sec
tio
n
4
p
r
o
v
id
es
in
s
ig
h
ts
in
to
p
er
f
o
r
m
a
n
ce
m
etr
ics
an
d
co
m
p
ar
ativ
e
ev
alu
atio
n
s
.
Sectio
n
5
h
ig
h
lig
h
ts
th
e
p
r
ac
ti
ca
l
im
p
licatio
n
s
o
f
th
e
p
r
o
p
o
s
ed
h
y
b
r
i
d
s
y
s
tem
.
Sectio
n
6
co
n
clu
d
es
th
e
p
ap
er
with
th
e
f
u
tu
r
e
wo
r
k
.
2.
RE
L
AT
E
D
WO
RK
S
B
io
m
etr
ic
au
th
en
ticatio
n
s
y
s
tem
s
h
av
e
ev
o
lv
e
d
s
ig
n
if
ic
an
tly
in
r
ec
en
t
y
ea
r
s
with
n
u
m
e
r
o
u
s
ad
v
an
ce
m
e
n
ts
aim
ed
at
im
p
r
o
v
in
g
s
ec
u
r
ity
,
ac
c
u
r
ac
y
,
an
d
r
o
b
u
s
tn
ess
[
2
0
]
.
T
r
ad
itio
n
al
u
n
im
o
d
al
s
y
s
tem
s
s
u
ch
as
f
in
g
er
p
r
in
t,
f
ac
ial,
o
r
i
r
is
r
ec
o
g
n
itio
n
h
av
e
b
ee
n
wid
ely
s
tu
d
ied
an
d
im
p
lem
en
ted
d
u
e
to
th
eir
ea
s
e
o
f
u
s
e
an
d
ac
ce
s
s
ib
ilit
y
.
Ho
wev
er
,
th
ese
s
y
s
tem
s
h
av
e
m
u
ch
v
u
ln
er
ab
ilit
y
,
s
u
c
h
as
s
u
s
ce
p
tib
ilit
y
to
s
p
o
o
f
in
g
,
en
v
ir
o
n
m
en
tal
f
ac
to
r
s
,
a
n
d
li
m
ited
ac
cu
r
ac
y
in
ch
allen
g
in
g
co
n
d
itio
n
s
(
e
.
g
.
,
p
ar
tial
o
cc
lu
s
io
n
an
d
lo
w
-
q
u
ality
s
en
s
o
r
s
)
.
T
o
ad
d
r
ess
th
ese
s
h
o
r
tco
m
in
g
s
,
t
h
e
r
esear
ch
er
s
h
av
e
in
cr
ea
s
in
g
ly
t
u
r
n
ed
to
m
u
lti
-
m
o
d
al
b
io
m
etr
i
c
s
y
s
tem
s
,
wh
ich
in
teg
r
ate
m
u
ltip
le
m
o
d
alities
to
im
p
r
o
v
e
au
t
h
en
ticatio
n
r
eliab
ilit
y
.
T
h
e
f
o
l
lo
win
g
s
u
b
s
ec
tio
n
s
d
escr
ib
e
u
n
im
o
d
al
an
d
m
u
lti
-
m
o
d
al
b
io
m
etr
ic
s
y
s
tem
s
b
ased
o
n
E
C
G
an
d
ea
r
f
ea
t
u
r
es.
2
.
1
.
E
lect
ro
c
a
rdio
g
ra
m
-
ba
s
ed
bio
m
et
ric
a
uthent
ica
t
io
n
E
C
G
-
b
ased
b
io
m
etr
ic
au
th
en
ti
ca
tio
n
h
as
attr
ac
ted
in
c
r
ea
s
in
g
atten
tio
n
d
u
e
to
th
e
u
n
i
q
u
en
e
s
s
o
f
E
C
G
p
atter
n
s
,
wh
ich
ar
e
clo
s
ely
r
elate
d
to
an
in
d
iv
id
u
al’
s
p
h
y
s
io
lo
g
ical
an
d
b
e
h
av
io
r
al
ch
a
r
a
cter
is
tics
[
6
]
,
[
2
1
]
.
T
h
ese
s
ig
n
als
p
r
o
v
id
e
a
h
ig
h
lev
el
o
f
in
d
iv
id
u
ality
a
n
d
ar
e
d
if
f
icu
lt
to
r
ep
licate
o
r
s
p
o
o
f
,
m
ak
in
g
th
em
a
p
r
o
m
is
in
g
m
o
d
ality
f
o
r
s
ec
u
r
e
au
th
en
ticatio
n
s
y
s
tem
s
[
2
2
]
,
[
2
3
]
.
R
ec
en
t
s
tu
d
ies
h
av
e
ex
p
lo
r
ed
m
ac
h
in
e
lear
n
in
g
an
d
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es,
p
ar
ticu
la
r
ly
C
NNs,
t
o
class
if
y
E
C
G
s
ig
n
als
f
o
r
u
s
e
r
id
en
tific
atio
n
an
d
v
er
if
icatio
n
with
h
ig
h
ac
cu
r
ac
y
u
n
d
er
co
n
tr
o
lled
co
n
d
itio
n
s
[
2
2
]
,
[
2
3
]
.
Sev
er
al
s
tu
d
ies
h
av
e
d
em
o
n
s
tr
ated
th
e
ef
f
ec
tiv
en
ess
o
f
E
C
G
-
b
ased
b
io
m
etr
ic
s
y
s
tem
s
u
s
in
g
d
if
f
er
en
t
s
ig
n
al
p
r
o
ce
s
s
in
g
an
d
d
ee
p
lear
n
in
g
tech
n
iq
u
es.
Fo
r
ex
am
p
le,
Me
l
zi
et
a
l.
[
2
2
]
ev
alu
ated
E
C
G
b
io
m
etr
ics
in
b
o
th
v
er
if
icatio
n
an
d
id
en
tific
atio
n
task
s
,
ac
h
iev
in
g
v
er
y
lo
w
e
r
r
o
r
r
ates
in
s
in
g
le
-
an
d
m
u
lt
i
-
s
ess
io
n
s
ce
n
ar
io
s
.
Fatim
ah
et
a
l.
[
2
4
]
u
tili
ze
d
Fo
u
r
ier
d
ec
o
m
p
o
s
itio
n
m
eth
o
d
(
FDM)
an
d
p
h
ase
tr
an
s
f
o
r
m
(
PT)
tech
n
iq
u
es
to
ac
h
iev
e
h
ig
h
id
en
tific
atio
n
ac
cu
r
ac
y
ac
r
o
s
s
m
u
ltip
le
E
C
G
d
atasets
.
T
r
an
s
f
er
lear
n
in
g
ap
p
r
o
ac
h
es
u
s
in
g
p
r
e
-
tr
ain
e
d
C
NN
m
o
d
els
wer
e
also
ex
p
lo
r
e
d
b
y
Ku
m
a
r
an
d
Pu
r
u
s
h
o
ttam
a
[
2
5
]
d
em
o
n
s
tr
atin
g
im
p
r
o
v
ed
r
ec
o
g
n
itio
n
p
e
r
f
o
r
m
a
n
ce
th
r
o
u
g
h
f
ea
tu
r
e
f
u
s
io
n
.
Ad
d
itio
n
ally
,
f
ea
tu
r
e
s
elec
tio
n
tech
n
iq
u
es
co
m
b
in
ed
with
m
ac
h
in
e
lear
n
in
g
class
if
ier
s
h
av
e
b
ee
n
u
s
ed
to
en
h
an
ce
r
ec
o
g
n
itio
n
p
er
f
o
r
m
a
n
ce
.
Mo
r
e
ad
v
a
n
ce
d
d
ee
p
lear
n
in
g
m
o
d
els,
in
clu
d
in
g
s
iam
ese
n
etwo
r
k
s
an
d
ca
p
s
u
le
n
etwo
r
k
s
,
h
av
e
f
u
r
th
er
im
p
r
o
v
ed
E
C
G
-
b
ased
b
io
m
etr
ic
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
ac
r
o
s
s
s
ev
er
al
b
en
ch
m
ar
k
d
atasets
[
2
6
]
.
A
co
m
p
ar
is
o
n
o
f
th
ese
s
tu
d
ies is
s
u
m
m
ar
ized
in
T
a
b
le
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
5
2
-
3
2
6
8
3254
T
ab
le
1
.
Su
m
m
a
r
y
o
f
r
ec
e
n
t d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es
in
E
C
G
r
ec
o
g
n
itio
n
S
t
u
d
y
M
e
t
h
o
d
o
l
o
g
y
D
a
t
a
s
e
t
(
s)
A
c
c
u
r
a
c
y
(
%)
M
e
l
z
i
e
t
a
l
.
[
2
2
]
M
u
l
t
i
-
sess
i
o
n
Tr
a
d
i
t
i
o
n
a
l
/
u
ser
-
f
r
i
e
n
d
l
y
9
6
.
4
6
–
100
F
a
t
i
m
a
h
e
t
a
l
.
[
2
4
]
F
D
M
+
P
T
C
h
e
c
k
y
o
u
r
b
i
o
si
g
n
a
l
s
h
e
r
e
i
n
i
t
i
a
t
i
v
e
(
C
Y
B
H
i
)
,
M
I
T
-
B
I
H
,
a
n
d
E
C
G
-
ID
9
1
.
0
7
/
9
7
.
9
2
/
9
8
.
4
5
K
u
mar
e
t
a
l
.
[
2
5
]
A
l
e
x
N
e
t
+
G
o
o
g
l
e
N
e
t
f
u
si
o
n
P
r
i
v
a
t
e
D
B
9
6
.
6
P
a
t
r
o
e
t
a
l
.
[
2
7
]
G
e
n
e
t
i
c
a
l
g
o
r
i
t
h
m (
GA
)
+
e
l
a
st
i
c
n
e
t
(
EN
)
+
r
a
n
d
o
m f
o
r
e
s
t
(
RF
)
P
r
i
v
a
t
e
D
B
9
5
.
3
0
/
9
4
.
9
0
P
r
a
k
a
s
h
e
t
a
l
.
[
2
6
]
M
o
d
i
f
i
e
d
s
i
a
m
e
se
n
e
t
w
o
r
k
EC
G
-
ID
9
9
.
8
5
B
o
u
j
n
o
u
n
i
e
t
a
l
.
[
2
3
]
C
a
p
su
l
e
n
e
t
w
o
r
k
s
P
TB
d
i
a
g
n
o
si
s
EC
G
(
P
TB
)
P
T
B
,
M
I
T
-
B
I
H
,
M
I
T
-
B
I
H
S
T
c
h
a
n
g
e
(
S
TD
B
)
,
a
n
d
M
I
T
-
B
I
H
n
o
r
mal
si
n
u
s
r
h
y
t
h
m
(
N
S
R
D
B
)
9
9
.
5
–
1
0
0
2
.
2
.
E
a
r
r
ec
o
g
nitio
n
E
ar
b
io
m
etr
ics
in
v
o
lv
e
th
e
a
n
aly
s
is
o
f
th
e
o
u
ter
ea
r
'
s
s
h
ap
e,
s
ize,
an
d
u
n
iq
u
e
f
ea
tu
r
es
[
2
8
]
,
[
2
9
]
,
C
o
m
p
ar
ed
with
f
ac
ial
r
ec
o
g
n
itio
n
,
ea
r
f
ea
tu
r
es
r
em
ain
r
ela
tiv
ely
s
tab
le
o
v
er
tim
e
an
d
a
r
e
less
af
f
ec
ted
b
y
f
ac
ial
ex
p
r
ess
io
n
s
,
ag
in
g
,
o
r
m
o
d
er
ate
e
n
v
ir
o
n
m
en
tal
v
a
r
ia
tio
n
s
.
R
ec
en
t
s
tu
d
ies
h
av
e
s
h
o
wn
th
at
ea
r
-
b
ased
au
th
en
ticatio
n
s
y
s
tem
s
ca
n
ac
h
iev
e
h
ig
h
ac
cu
r
ac
y
,
e
s
p
ec
ially
wh
en
co
m
b
i
n
ed
with
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es
[
3
0
]
–
[
3
2
]
.
Xu
et
a
l.
[
3
1
]
p
r
o
p
o
s
ed
a
n
ea
r
r
ec
o
g
n
itio
n
ap
p
r
o
ac
h
b
ased
o
n
th
e
r
esid
u
al
n
etwo
r
k
(
R
esNet
)
-
1
8
d
ee
p
r
esid
u
al
n
etwo
r
k
.
T
h
e
m
o
d
el
e
x
p
lo
its
r
esid
u
al
co
n
n
ec
tio
n
s
to
allev
iate
th
e
v
a
n
is
h
in
g
g
r
ad
ien
t
p
r
o
b
lem
a
n
d
im
p
r
o
v
e
f
ea
tu
r
e
ex
tr
ac
tio
n
ca
p
ab
ilit
y
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
wa
s
ev
alu
ated
o
n
th
e
Un
iv
er
s
ity
o
f
Scien
ce
an
d
T
ec
h
n
o
lo
g
y
B
eijin
g
(
USTB)
I
I
I
e
ar
d
ataset
an
d
ac
h
iev
ed
a
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
o
f
9
8
.
2
7
%,
o
u
tp
e
r
f
o
r
m
in
g
s
ev
er
al
co
n
v
e
n
tio
n
al
ea
r
r
ec
o
g
n
iti
o
n
an
d
d
ee
p
lear
n
in
g
a
p
p
r
o
a
ch
es.
T
h
ese
r
esu
lts
d
em
o
n
s
tr
ate
th
e
ef
f
ec
tiv
e
n
ess
o
f
r
esid
u
al
lear
n
in
g
f
o
r
r
o
b
u
s
t
ea
r
b
io
m
etr
ic
r
ec
o
g
n
itio
n
.
I
n
s
tu
d
ies
[
3
3
]
,
[
3
4
]
in
v
esti
g
ated
f
in
e
-
tu
n
ed
u
s
in
g
cu
s
to
m
-
s
ized
in
p
u
ts
d
eter
m
i
n
ed
s
p
ec
if
ically
f
o
r
ea
ch
C
NN
ar
ch
itectu
r
e,
an
d
d
ee
p
r
esid
u
al
n
etwo
r
k
s
r
esp
ec
tiv
ely
,
a
s
in
g
le
R
esNeXt1
0
1
m
o
d
el
ac
h
iev
es
a
r
a
n
k
-
1
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
o
f
9
3
.
4
5
%,
an
d
ac
h
iev
es
th
e
b
e
s
t
r
an
k
-
1
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
o
f
9
5
.
8
5
%
u
s
in
g
an
en
s
em
b
le
o
f
f
in
e
-
tu
n
ed
R
esNeX
t1
0
1
m
o
d
el.
Mo
r
e
o
v
er
,
th
ey
o
b
tain
ed
b
y
av
er
ag
i
n
g
en
s
em
b
les
o
f
f
in
e
-
tu
n
ed
n
e
two
r
k
s
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
o
f
9
9
.
6
4
%,
9
8
.
5
7
%,
8
1
.
8
9
%,
an
d
6
7
.
2
5
%
o
n
th
e
AM
I
,
AM
I
cr
o
p
p
ed
(
AM
I
C
)
,
W
es
t
Po
m
er
an
ian
Un
iv
er
s
ity
o
f
T
ec
h
n
o
lo
g
y
(
W
PUT)
,
an
d
an
n
o
tated
web
ea
r
(
AW
E
)
d
atab
ases
,
r
esp
ec
tiv
ely
.
He
e
t
a
l
.
[
3
5
]
f
u
r
t
h
e
r
a
d
v
a
n
c
e
d
t
h
e
f
i
e
l
d
b
y
i
n
t
r
o
d
u
c
i
n
g
c
o
n
t
r
a
s
ti
v
e
l
o
s
s
l
e
a
r
n
i
n
g
,
a
c
h
ie
v
i
n
g
9
9
.
5
9
%
a
c
c
u
r
a
c
y
,
a
n
d
P
r
i
y
a
d
h
a
r
s
h
i
n
i
e
t
a
l
.
[
3
6
]
d
e
s
i
g
n
e
d
s
i
x
-
l
a
y
e
r
C
N
N
t
o
a
u
t
o
m
a
ti
c
a
ll
y
l
e
a
r
n
d
i
s
c
r
i
m
i
n
at
i
v
e
e
a
r
f
e
a
t
u
r
e
s
w
i
t
h
o
u
t
r
el
y
i
n
g
o
n
h
a
n
d
c
r
a
f
t
e
d
d
e
s
c
r
i
p
t
o
r
s
.
T
h
e
m
o
d
e
l
w
a
s
e
v
a
l
u
a
te
d
o
n
t
h
e
I
n
d
i
a
n
I
n
s
t
it
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
D
e
l
h
i
-
I
I
(
IITD
-
II
)
a
n
d
A
M
I
e
a
r
d
a
t
a
s
e
t
s
,
a
c
h
i
e
v
in
g
r
e
c
o
g
n
i
t
i
o
n
a
c
c
u
r
a
c
i
e
s
o
f
9
7
.
3
6
%
a
n
d
9
6
.
9
9
%
,
r
e
s
p
e
c
t
i
v
el
y
.
M
e
h
t
a
e
t
a
l
.
[
3
7
]
p
r
o
p
o
s
ed
an
e
n
s
em
b
le
-
b
ased
h
y
b
r
id
tr
an
s
f
er
lear
n
in
g
f
r
a
m
ewo
r
k
f
o
r
2
D
ea
r
r
ec
o
g
n
itio
n
u
s
in
g
p
r
e
-
tr
ain
e
d
VGG1
6
an
d
VGG1
9
C
NN
s
.
Dee
p
f
ea
tu
r
es
ex
tr
ac
ted
f
r
o
m
b
o
th
n
etwo
r
k
s
wer
e
co
m
b
in
ed
t
h
r
o
u
g
h
an
e
n
s
em
b
le
s
tr
ateg
y
to
im
p
r
o
v
e
t
h
e
d
is
cr
im
in
ativ
e
ca
p
ab
ilit
y
o
f
t
h
e
r
ec
o
g
n
itio
n
s
y
s
tem
.
L
ei
et
a
l.
[
3
8
]
p
r
o
p
o
s
ed
a
li
g
h
tweig
h
t
ea
r
r
ec
o
g
n
itio
n
f
r
a
m
ewo
r
k
b
ased
o
n
an
atten
tio
n
m
ec
h
an
is
m
an
d
f
ea
tu
r
e
f
u
s
io
n
s
tr
ateg
y
.
T
h
e
p
r
o
p
o
s
ed
n
etwo
r
k
e
n
h
an
ce
s
d
is
cr
im
in
ativ
e
ea
r
r
ep
r
esen
tat
io
n
s
b
y
in
teg
r
atin
g
ch
an
n
el
atten
tio
n
m
o
d
u
les
with
m
u
lti
-
lev
el
f
ea
tu
r
e
f
u
s
io
n
.
E
x
p
er
im
e
n
tal
r
esu
lts
d
e
m
o
n
s
tr
ated
th
at
th
e
p
r
o
p
o
s
ed
lig
h
tweig
h
t
ar
ch
itect
u
r
e
ac
h
iev
ed
co
m
p
etitiv
e
r
ec
o
g
n
itio
n
p
er
f
o
r
m
a
n
ce
wh
ile
s
ig
n
if
ican
tly
r
ed
u
ci
n
g
co
m
p
u
tatio
n
al
co
m
p
le
x
ity
,
m
ak
in
g
it
s
u
itab
le
f
o
r
r
ea
l
-
tim
e
b
io
m
etr
ic
ap
p
licatio
n
s
.
Desp
ite
th
ese
ad
v
an
ce
m
e
n
ts
,
ea
r
r
ec
o
g
n
itio
n
s
y
s
tem
s
s
ti
ll
f
ac
e
ch
allen
g
es
in
r
ea
l
-
wo
r
ld
s
ce
n
ar
io
s
d
u
e
t
o
v
ar
iatio
n
s
in
p
o
s
e,
o
cc
lu
s
io
n
,
lig
h
tin
g
c
o
n
d
itio
n
s
,
an
d
b
ac
k
g
r
o
u
n
d
n
o
is
e.
I
n
a
d
d
itio
n
,
m
an
y
ex
is
tin
g
ap
p
r
o
a
ch
es
ar
e
o
p
tim
ized
f
o
r
u
n
im
o
d
al
d
atasets
an
d
m
ay
n
o
t
g
en
e
r
alize
ef
f
ec
tiv
ely
in
m
u
ltimo
d
al
b
i
o
m
etr
ic
s
y
s
tem
s
.
T
ab
le
2
s
u
m
m
ar
izes
v
ar
io
u
s
d
ee
p
lear
n
in
g
tech
n
iq
u
es
em
p
l
o
y
ed
i
n
ea
r
r
ec
o
g
n
itio
n
,
h
ig
h
lig
h
tin
g
th
eir
ar
ch
itectu
r
e,
d
atasets
,
an
d
ac
h
iev
ed
ac
c
u
r
ac
y
.
T
o
o
v
er
c
o
m
e
th
e
lim
itatio
n
s
o
f
u
n
im
o
d
al
ap
p
r
o
ac
h
es,
r
ese
ar
ch
er
s
h
av
e
ex
p
lo
r
ed
h
y
b
r
id
b
io
m
etr
ic
s
y
s
tem
s
th
at
co
m
b
in
e
E
C
G
an
d
ea
r
f
ea
tu
r
es.
T
h
ese
d
u
al
-
m
o
d
al
s
y
s
tem
s
ex
p
lo
it
t
h
e
co
m
p
lem
en
tar
y
ch
ar
ac
ter
is
tics
o
f
ea
ch
m
o
d
ality
to
im
p
r
o
v
e
au
th
en
ticatio
n
ac
cu
r
ac
y
an
d
s
ec
u
r
ity
.
Pre
v
io
u
s
s
tu
d
ies
h
av
e
s
h
o
wn
th
at
in
teg
r
atin
g
E
C
G
with
o
th
er
b
i
o
m
etr
ic
m
o
d
alities
,
s
u
ch
as f
ac
e
o
r
f
in
g
er
p
r
in
t r
ec
o
g
n
itio
n
,
e
n
h
an
ce
s
r
o
b
u
s
tn
ess
an
d
r
ed
u
ce
s
f
alse
r
ejec
tio
n
r
ates
in
r
ea
l
-
wo
r
l
d
en
v
ir
o
n
m
e
n
ts
[
3
9
]
,
[
4
0
]
.
Dee
p
l
ea
r
n
in
g
tech
n
iq
u
es,
p
ar
ticu
lar
ly
C
NNs,
h
av
e
s
ig
n
if
ican
tly
ad
v
a
n
ce
d
b
i
o
m
etr
i
c
au
th
en
ticatio
n
b
y
en
ab
lin
g
au
to
m
ated
f
ea
tu
r
e
ex
tr
ac
tio
n
a
n
d
ef
f
icien
t
class
if
icatio
n
.
Ar
c
h
itectu
r
es
s
u
ch
a
s
VGG
-
v
er
y
d
ee
p
1
6
,
R
esNet,
an
d
Alex
'
s
n
etwo
r
k
(
Alex
Net
)
h
av
e
b
ee
n
wid
ely
ap
p
lied
to
b
io
m
etr
ic
d
ata
p
r
o
ce
s
s
in
g
[
4
1
]
.
H
o
wev
er
,
t
h
ese
m
o
d
els
ar
e
o
f
ten
co
m
p
u
tatio
n
ally
ex
p
e
n
s
iv
e,
lim
itin
g
th
eir
s
u
itab
ilit
y
f
o
r
r
ea
l
-
tim
e
o
r
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
ap
p
licatio
n
s
.
C
o
n
s
eq
u
en
tly
,
r
ec
e
n
t
r
esear
ch
h
as
f
o
cu
s
ed
o
n
lig
h
tweig
h
t
o
r
h
y
b
r
id
a
r
ch
itectu
r
es
th
at
b
alan
ce
ac
cu
r
ac
y
an
d
co
m
p
u
tatio
n
al
e
f
f
icien
cy
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
B
io
metric a
u
th
en
tica
tio
n
u
s
in
g
d
u
a
l
-
mo
d
a
l
d
ee
p
lea
r
n
in
g
b
a
s
ed
-
o
n
E
C
G
a
n
d
…
(
Mo
h
a
m
ed
S
.
K
h
a
la
f
)
3255
T
ab
le
2
.
Su
m
m
a
r
y
o
f
r
ec
e
n
t d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es in
ea
r
r
ec
o
g
n
itio
n
S
t
u
d
y
A
p
p
r
o
a
c
h
D
a
t
a
s
e
t
A
c
c
u
r
a
c
y
(
%)
O
mara
e
t
a
l
.
[
3
0
]
C
N
N
+
d
i
s
c
r
i
m
i
n
a
n
t
c
o
r
r
e
l
a
t
i
o
n
a
n
a
l
y
si
s
(
D
C
A
)
U
S
TB
a
n
d
I
I
TD
d
a
t
a
se
t
s
9
9
.
0
0
a
n
d
9
9
.
5
0
X
u
e
t
a
l
.
[
3
1
]
R
e
sN
e
t
-
18
-
b
a
se
d
a
p
p
r
o
a
c
h
U
S
TB
I
I
I
d
a
t
a
se
t
9
8
.
2
7
A
l
sh
a
z
l
y
e
t
a
l
.
[
3
3
]
F
i
n
e
-
t
u
n
e
d
R
e
sN
e
X
t
1
0
1
e
n
sem
b
l
e
Ea
r
V
N
1
.
0
d
a
t
a
s
e
t
9
5
.
8
5
A
l
sh
a
z
l
y
e
t
a
l
.
[
3
4
]
En
se
mb
l
e
o
f
f
i
n
e
-
t
u
n
e
d
C
N
N
s
A
M
I
a
n
d
A
W
E
d
a
t
a
s
e
t
s
9
9
.
6
4
a
n
d
6
7
.
2
5
He
e
t
a
l
.
[
3
5
]
C
o
n
t
r
a
st
i
v
e
l
o
ss
l
e
a
r
n
i
n
g
N
o
t
s
p
e
c
i
f
i
e
d
9
9
.
5
9
P
r
i
y
a
d
h
a
r
s
h
i
n
i
e
t
a
l
.
[
3
6
]
C
N
N
s
I
I
TD
-
I
I
a
n
d
A
M
I
d
a
t
a
s
e
t
s
9
7
.
3
6
M
e
h
t
a
e
t
a
l
.
[
3
7
]
H
y
b
r
i
d
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
(
V
G
G
1
6
+
V
G
G
1
9
e
n
s
e
mb
l
e
)
I
I
TD
d
a
t
a
se
t
9
9
.
3
5
Le
i
e
t
a
l
.
[
3
8
]
A
t
t
e
n
t
i
o
n
mec
h
a
n
i
sm
a
n
d
f
e
a
t
u
r
e
f
u
s
i
o
n
U
S
TB
a
n
d
E
a
r
V
N
1
.
0
d
a
t
a
s
e
t
s
1
0
0
a
n
d
8
8
.
3
0
2
.
3
.
M
ulti
-
m
o
da
l
bio
m
et
ric
s
y
s
t
em
s
ba
s
ed
o
n
elec
t
ro
ca
rdio
g
ra
m
a
nd
ea
r
R
ec
en
t
s
tu
d
ies
h
av
e
ex
p
lo
r
ed
co
m
b
in
in
g
d
if
f
er
e
n
t
b
io
m
etr
ic
m
o
d
alities
to
o
v
er
co
m
e
th
e
lim
itatio
n
s
o
f
in
d
iv
id
u
al
s
y
s
tem
s
.
Fo
r
ex
am
p
le,
f
ac
e
an
d
ea
r
r
ec
o
g
n
itio
n
s
y
s
tem
s
h
av
e
b
ee
n
r
esear
ch
ed
to
im
p
r
o
v
e
ac
cu
r
ac
y
an
d
r
o
b
u
s
tn
ess
,
esp
ec
ially
in
s
ce
n
ar
io
s
wh
er
e
f
ac
ial
r
ec
o
g
n
itio
n
alo
n
e
is
n
o
t
s
u
f
f
icien
t
(
e.
g
.
,
u
n
d
er
lo
w
lig
h
tin
g
o
r
with
a
g
in
g
f
ac
es)
[
4
2
]
,
[
4
3
]
,
p
r
in
cip
al
co
m
p
o
n
en
t
an
aly
s
is
(
PC
A)
,
an
d
in
d
ep
en
d
en
t
co
m
p
o
n
en
t
an
al
y
s
is
(
I
C
A)
alg
o
r
ith
m
s
to
en
h
a
n
ce
f
ea
t
u
r
e
r
ep
r
esen
tatio
n
an
d
s
elec
tio
n
.
Similar
ly
,
E
C
G,
ir
is
,
an
d
f
in
g
er
p
r
in
t
-
b
ased
s
y
s
tem
s
h
av
e
b
ee
n
p
r
o
p
o
s
ed
f
o
r
h
ig
h
e
r
s
ec
u
r
ity
an
d
r
eliab
ilit
y
,
as
th
ese
two
m
o
d
alities
p
r
o
v
id
e
co
m
p
le
m
en
tar
y
f
ea
tu
r
es
[
4
4
]
,
[
4
5
]
.
R
ec
en
tly
,
ea
r
an
d
E
C
G
s
ig
n
als
h
av
e
e
m
e
r
g
ed
as
p
r
o
m
is
in
g
co
m
p
lem
en
tar
y
b
io
m
etr
ic
m
o
d
alities
.
Sev
er
al
s
tu
d
ies
h
av
e
in
v
esti
g
ated
th
e
in
teg
r
atio
n
o
f
th
ese
m
o
d
alities
u
s
in
g
d
if
f
e
r
en
t
f
ea
tu
r
e
ex
tr
ac
t
io
n
an
d
f
u
s
io
n
tech
n
iq
u
es.
Dee
p
lear
n
in
g
ap
p
r
o
ac
h
es,
p
ar
tic
u
lar
ly
C
NNs,
h
av
e
b
ee
n
wid
ely
u
s
ed
to
ex
tr
ac
t
d
is
cr
im
in
ativ
e
f
ea
tu
r
es
f
r
o
m
b
o
th
ea
r
im
ag
es
an
d
E
C
G
s
ig
n
als
[
4
4
]
,
[
4
5
]
.
Oth
er
s
tu
d
ies
h
av
e
in
c
o
r
p
o
r
ated
s
ig
n
al
p
r
o
ce
s
s
in
g
m
eth
o
d
s
as
v
ar
i
o
u
s
lo
ca
l
b
in
ar
y
p
atter
n
(
L
B
P)
to
en
h
an
ce
f
ea
tu
r
e
r
ep
r
esen
tatio
n
a
n
d
s
elec
tio
n
[
4
6
]
.
Dif
f
er
en
t
f
u
s
io
n
s
tr
ateg
ie
s
,
in
clu
d
in
g
f
ea
tu
r
e
-
lev
el
a
n
d
s
co
r
e
-
lev
el
f
u
s
io
n
,
h
av
e
also
b
ee
n
a
p
p
lied
to
in
teg
r
ate
th
e
i
n
f
o
r
m
atio
n
f
r
o
m
b
o
th
m
o
d
alities
,
r
esu
ltin
g
i
n
h
ig
h
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
ac
r
o
s
s
s
ev
er
al
d
atasets
[
4
3
]
–
[
4
7
]
.
A
co
m
p
a
r
is
o
n
o
f
th
ese
ap
p
r
o
ac
h
es
is
s
u
m
m
ar
ized
in
T
ab
le
3
.
Desp
ite
th
ese
p
r
o
m
is
in
g
r
esu
lts
,
m
an
y
ex
is
tin
g
m
u
ltimo
d
al
s
y
s
tem
s
s
u
f
f
er
f
r
o
m
lim
ited
d
at
aset
d
iv
er
s
ity
,
an
d
r
ed
u
ce
d
a
d
ap
tab
ilit
y
to
d
i
f
f
er
en
t
d
ataset
s
izes
a
s
s
h
o
wn
in
T
ab
le
3
.
T
o
a
d
d
r
ess
th
ese
lim
itatio
n
s
,
th
is
s
tu
d
y
p
r
o
p
o
s
es
a
h
y
b
r
id
d
u
al
-
m
o
d
al
au
th
en
ticatio
n
s
y
s
tem
th
at
d
y
n
am
ically
s
elec
ts
b
etwe
en
VGG
-
v
er
y
d
ee
p
1
6
a
n
d
C
NN5
b
ased
o
n
d
ataset
s
ca
l
e,
b
alan
cin
g
ac
cu
r
ac
y
an
d
co
m
p
u
tatio
n
al
ef
f
icien
cy
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
is
ev
alu
ated
u
s
in
g
p
u
b
licly
av
ai
lab
le
d
atasets
(
E
C
G
-
I
D
an
d
AM
I
)
an
d
em
p
lo
y
s
ef
f
icien
t
f
u
s
io
n
tech
n
iq
u
es
to
p
r
o
v
id
e
a
s
ca
lab
le
s
o
lu
tio
n
f
o
r
r
ea
l
-
wo
r
ld
b
io
m
etr
ic
au
t
h
en
ti
ca
tio
n
.
T
ab
le
3
.
Su
m
m
a
r
y
o
f
r
ec
e
n
t m
u
lti
-
m
o
d
al
b
io
m
etr
ic
s
y
s
tem
s
co
m
b
in
in
g
ea
r
an
d
E
C
G
si
g
n
a
ls
S
t
u
d
y
M
o
d
a
l
i
t
i
e
s
F
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
F
u
si
o
n
me
t
h
o
d
D
a
t
a
s
e
t
s
i
z
e
A
c
c
u
r
a
c
y
(
%)
K
h
a
l
a
f
e
t
a
l
.
[
6
]
Ea
r
+
EC
G
V
G
G
1
6
F
e
a
t
u
r
e
-
l
e
v
e
l
f
u
s
i
o
n
9
0
su
b
j
e
c
t
s
9
9
.
4
7
Th
a
r
e
w
a
l
e
t
a
l
.
[
4
3
]
3
D
(
f
a
c
e
+
e
a
r
)
P
C
A
+
I
C
P
S
c
o
r
e
-
l
e
v
e
l
f
u
s
i
o
n
F
a
c
e
r
e
c
o
g
n
i
t
i
o
n
g
r
a
n
d
c
h
a
l
l
e
n
g
e
(
F
R
G
C
)
+
U
n
i
v
e
r
si
t
y
o
f
N
o
t
r
e
D
a
me
(
UND
)
c
o
l
l
e
c
t
i
o
n
F
9
9
.
2
5
K
a
i
l
a
s
e
t
a
l
.
[
4
4
]
EC
G
+
i
r
i
s
h
e
a
r
t
r
a
t
e
v
a
r
i
a
b
i
l
i
t
y
(
H
R
V
)
+
C
N
N
+
d
e
e
p
r
e
si
d
u
a
l
n
e
t
w
o
r
k
s
(
D
e
e
p
R
e
sN
e
t
)
F
e
a
t
u
r
e
-
l
e
v
e
l
f
u
s
i
o
n
N
o
t
s
p
e
c
i
f
i
e
d
9
7
.
0
0
G
a
r
g
e
t
a
l
.
[
4
5
]
F
i
n
g
e
r
p
r
i
n
t
+
i
r
i
s
+
E
C
G
S
w
i
n
t
r
a
n
sf
o
r
mer
+
C
N
N
F
e
a
t
u
r
e
-
l
e
v
e
l
f
u
s
i
o
n
C
o
m
b
i
n
e
d
mu
l
t
i
m
o
d
a
l
d
a
t
a
se
t
9
9
.
0
0
R
e
g
o
u
i
d
e
t
a
l
.
[
4
6
]
EC
G
+
e
a
r
+
i
r
i
s
1D
-
LB
P
,
S
h
i
f
t
e
d
1
D
-
LB
P
,
1
D
-
MR
-
L
B
P
F
e
a
t
u
r
e
-
l
e
v
e
l
f
u
s
i
o
n
M
u
l
t
i
p
l
e
p
u
b
l
i
c
d
a
t
a
s
e
t
s
1
0
0
G
u
v
e
n
e
t
a
l
.
[
4
7
]
EC
G
+
sp
e
e
c
h
1D
-
C
N
N
V
o
t
i
n
g
-
b
a
se
d
f
u
si
o
n
9
0
su
b
j
e
c
t
s
1
0
0
S
u
ma
l
a
t
h
a
e
t
a
l
.
[
1
7
]
EC
G
+
f
i
n
g
e
r
p
r
i
n
t
+
F
K
P
S
i
a
mes
e
n
e
u
r
a
l
n
e
t
w
o
r
k
s
W
e
i
g
h
t
e
d
s
c
o
r
e
-
l
e
v
e
l
f
u
s
i
o
n
9
0
su
b
j
e
c
t
s
9
9
.
8
0
A
sh
w
i
n
i
e
t
a
l
.
[
4
8
]
EC
G
+
i
r
i
s
En
se
mb
l
e
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
F
e
a
t
u
r
e
-
l
e
v
e
l
f
u
s
i
o
n
N
o
t
s
p
e
c
i
f
i
e
d
9
5
.
6
5
2
.
4
.
L
im
it
a
t
io
ns
a
nd
re
s
ea
rc
h g
a
ps
Desp
ite
t
h
e
p
r
o
g
r
ess
m
a
d
e
i
n
m
u
lti
-
m
o
d
a
l
an
d
d
ee
p
le
ar
n
i
n
g
-
b
ase
d
b
io
m
e
tr
ic
a
u
t
h
e
n
ti
ca
ti
o
n
s
y
s
t
em
s
,
s
ev
er
al
c
h
al
le
n
g
es
r
em
ai
n
.
T
h
ese
in
cl
u
d
e
t
h
e
h
i
g
h
c
o
m
p
u
ta
ti
o
n
al
d
em
a
n
d
s
f
o
r
d
ee
p
le
ar
n
i
n
g
m
o
d
els
t
h
at
n
ee
d
lar
g
e
a
n
d
d
i
v
e
r
s
e
d
atas
ets
to
t
r
ai
n
ac
cu
r
a
te
m
o
d
els
,
s
o
m
e
o
f
p
r
e
v
i
o
u
s
p
u
b
lis
h
e
d
p
ap
er
s
[
4
5
]
–
[
4
8
]
,
b
ased
o
n
u
n
k
n
o
wn
a
n
d
p
r
iv
ate
d
atasets
.
Ad
d
itio
n
ally
,
r
ea
l
-
wo
r
ld
f
ac
to
r
s
s
u
ch
as
n
o
is
e,
s
en
s
o
r
ca
lib
r
atio
n
,
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
5
2
-
3
2
6
8
3256
en
v
ir
o
n
m
en
tal
v
ar
iatio
n
s
co
n
t
in
u
e
to
af
f
ec
t
th
e
p
er
f
o
r
m
an
c
e
o
f
b
i
o
m
etr
ic
s
y
s
tem
s
.
T
h
er
e
is
a
n
ee
d
f
o
r
m
o
r
e
ef
f
icien
t
m
o
d
els
th
at
b
ala
n
ce
ac
cu
r
ac
y
with
s
p
ee
d
a
n
d
s
ca
la
b
ilit
y
,
p
ar
ticu
lar
l
y
f
o
r
lar
g
e
-
s
c
ale
ap
p
licatio
n
s
o
r
s
y
s
tem
s
with
lim
ited
r
eso
u
r
ce
s
.
T
h
is
p
ap
er
p
r
o
p
o
s
es
a
h
y
b
r
id
d
u
al
-
m
o
d
al
au
t
h
en
ticatio
n
s
y
s
tem
th
at
c
o
m
b
in
es
E
C
G
s
ig
n
a
ls
an
d
ea
r
f
ea
tu
r
es
to
e
n
h
an
ce
s
ec
u
r
ity
wh
ile
ad
d
r
ess
in
g
th
e
lim
it
atio
n
s
o
f
e
x
is
tin
g
ap
p
r
o
ac
h
es.
B
y
lev
er
ag
in
g
d
ee
p
lear
n
in
g
m
o
d
els,
in
clu
d
i
n
g
VGG
-
v
er
y
d
ee
p
1
6
a
n
d
C
NN5
,
th
is
s
tu
d
y
aim
s
to
en
h
an
ce
b
o
th
ac
cu
r
ac
y
a
n
d
p
r
o
ce
s
s
in
g
tim
e,
o
f
f
er
i
n
g
an
ad
ap
tab
le
s
o
lu
tio
n
f
o
r
r
ea
l
-
w
o
r
ld
ap
p
licatio
n
s
.
A
k
ey
s
tr
e
n
g
th
o
f
m
u
ltimo
d
al
co
m
p
lex
ity
lies
in
co
m
b
in
in
g
m
u
ltip
le
m
o
d
alities
in
a
way
th
at
m
ax
im
izes
th
e
ad
v
an
tag
es
o
f
ea
ch
m
o
d
ality
wh
ile
m
in
im
izin
g
co
m
p
u
tatio
n
al
o
v
er
h
ea
d
.
Fu
r
th
er
m
o
r
e,
t
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
s
ee
k
s
to
b
r
id
g
e
th
e
g
ap
b
etwe
en
p
er
f
o
r
m
an
ce
an
d
co
m
p
u
tatio
n
al
e
f
f
icie
n
cy
,
en
a
b
lin
g
s
ca
lab
le
an
d
r
o
b
u
s
t
b
io
m
etr
ic
au
th
en
ticatio
n
s
y
s
tem
s
.
3.
M
E
T
H
O
DS A
ND
M
A
T
E
R
I
AL
S
T
h
e
m
eth
o
d
o
lo
g
ical
m
o
d
el
o
f
th
is
r
esear
ch
is
d
esig
n
ed
to
ad
d
r
ess
th
e
co
m
p
lex
ch
allen
g
es
o
f
m
u
lti
-
m
o
d
al
b
io
m
etr
ic
au
t
h
en
ticatio
n
th
r
o
u
g
h
a
r
ig
o
r
o
u
s
an
d
s
y
s
tem
atic
ex
p
er
im
en
tal
s
y
s
tem
.
B
y
i
n
t
e
g
r
a
t
i
n
g
E
C
G
s
i
g
n
a
ls
a
n
d
e
a
r
f
e
a
t
u
r
e
an
a
l
y
s
is
,
t
h
is
s
t
u
d
y
a
i
m
s
t
o
d
ev
e
l
o
p
a
r
o
b
u
s
t
a
u
t
h
e
n
t
i
c
at
i
o
n
s
y
s
te
m
a
s
s
h
o
w
n
i
n
F
i
g
u
r
e
1
.
T
h
e
m
eth
o
d
o
l
o
g
y
en
co
m
p
ass
es
co
m
p
r
e
h
en
s
iv
e
d
ata
co
llectio
n
,
s
o
p
h
is
ticated
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
es,
ad
v
an
ce
d
d
ee
p
lear
n
in
g
a
r
ch
itectu
r
e,
an
d
c
o
m
p
r
eh
en
s
iv
e
p
e
r
f
o
r
m
an
ce
ev
al
u
a
tio
n
s
tr
ateg
ies.
T
h
e
f
o
llo
win
g
s
u
b
-
s
ec
tio
n
s
d
escr
ib
e
ea
ch
p
h
ase
o
f
th
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
.
Fig
u
r
e
1
.
Pro
p
o
s
ed
d
u
al
-
m
o
d
a
l b
io
m
etr
ic
au
th
e
n
ticatio
n
s
y
s
tem
3
.
1
.
Da
t
a
s
et
p
re
pa
ra
t
io
n
3
.
1
.
1
.
Da
t
a
a
cquis
it
io
n
T
h
is
s
tu
d
y
em
p
lo
y
ed
p
u
b
licl
y
av
ailab
le
d
ata
s
ets
f
o
r
b
o
th
E
C
G
s
ig
n
als
an
d
ea
r
im
ag
es
to
en
s
u
r
e
r
ep
r
o
d
u
cib
ilit
y
an
d
en
ab
le
b
en
ch
m
ar
k
in
g
ag
ain
s
t
estab
lis
h
ed
d
ata
s
o
u
r
ce
s
.
Sp
ec
if
ically
,
th
e
E
C
G
-
I
D
d
ataset
[
4
9
]
,
[
5
0
]
was
u
s
ed
f
o
r
E
C
G
s
ig
n
als,
an
d
t
h
e
AM
I
d
ataset
[
5
1
]
,
[
5
2
]
was
u
s
ed
f
o
r
ea
r
im
ag
es.
T
h
ese
d
atasets
p
r
o
v
id
e
d
iv
er
s
e
a
n
d
h
ig
h
-
q
u
ality
s
am
p
les
th
at
a
r
e
wid
ely
u
s
ed
in
b
io
m
et
r
ic
au
th
en
ticatio
n
r
esear
ch
.
T
h
e
E
C
G
-
I
D
d
ataset
in
clu
d
es
r
ec
o
r
d
i
n
g
s
f
r
o
m
9
0
p
ar
ticip
a
n
ts
,
ea
ch
co
n
t
r
ib
u
tin
g
m
u
ltip
l
e
E
C
G
r
ec
o
r
d
in
g
s
.
T
h
e
s
ig
n
als
wer
e
co
llected
u
s
in
g
a
s
in
g
le
-
lead
E
C
G
s
y
s
tem
at
a
5
0
0
Hz
s
am
p
lin
g
r
ate,
w
ith
ea
ch
r
ec
o
r
d
in
g
s
ess
io
n
las
tin
g
ap
p
r
o
x
i
m
ately
2
0
s
ec
o
n
d
s
—
p
r
o
v
id
in
g
s
u
f
f
icien
t
tem
p
o
r
al
r
eso
lu
tio
n
f
o
r
r
eliab
le
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
m
o
d
elin
g
.
T
h
e
AM
I
d
ataset
co
n
tain
s
h
ig
h
-
r
eso
lu
tio
n
ea
r
im
a
g
es
f
r
o
m
1
0
0
p
a
r
ticip
an
ts
,
ca
p
tu
r
e
d
u
s
in
g
a
co
n
tr
o
lled
s
etu
p
th
at
in
clu
d
es
b
o
th
f
r
o
n
tal
an
d
an
g
led
v
iew
s
.
T
h
e
im
ag
es
o
f
f
er
co
n
s
is
ten
t
q
u
ality
,
s
to
r
ed
in
R
GB
f
o
r
m
at
with
a
r
eso
lu
tio
n
o
f
1
0
2
4
×7
6
8
p
ix
els.
T
h
e
k
e
y
p
ar
am
eter
s
o
f
b
o
t
h
d
atasets
ar
e
s
u
m
m
ar
ized
in
T
ab
le
4
,
in
clu
d
in
g
s
am
p
le
s
ize,
r
ec
o
r
d
in
g
e
q
u
ip
m
e
n
t,
s
ig
n
al
d
u
r
atio
n
,
an
d
s
am
p
lin
g
ch
a
r
ac
ter
is
tics
.
T
ab
le
4
.
Data
ch
a
r
ac
ter
is
tics
P
a
r
a
me
t
e
r
EC
G
s
i
g
n
a
l
s (E
C
G
-
I
D
d
a
t
a
s
e
t
)
Ea
r
i
m
a
g
e
s (A
M
I
d
a
t
a
se
t
)
S
a
mp
l
e
s
i
z
e
9
0
p
a
r
t
i
c
i
p
a
n
t
s
1
0
0
p
a
r
t
i
c
i
p
a
n
t
s
R
e
c
o
r
d
i
n
g
d
e
v
i
c
e
S
i
n
g
l
e
-
l
e
a
d
EC
G
sy
s
t
e
m
D
i
g
i
t
a
l
c
a
m
e
r
a
S
i
g
n
a
l
d
u
r
a
t
i
o
n
2
0
se
c
o
n
d
s
M
u
l
t
i
p
l
e
a
n
g
l
e
s
S
a
mp
l
i
n
g
r
a
t
e
5
0
0
H
z
1
0
2
4
×
7
6
8
p
i
x
e
l
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
B
io
metric a
u
th
en
tica
tio
n
u
s
in
g
d
u
a
l
-
mo
d
a
l
d
ee
p
lea
r
n
in
g
b
a
s
ed
-
o
n
E
C
G
a
n
d
…
(
Mo
h
a
m
ed
S
.
K
h
a
la
f
)
3257
3
.
1
.
2
.
Da
t
a
p
re
pro
ce
s
s
ing
Sig
n
al
an
d
im
ag
e
p
r
e
p
r
o
ce
s
s
in
g
in
v
o
l
v
ed
s
o
p
h
is
ticated
tech
n
iq
u
es
to
im
p
r
o
v
e
f
ea
tu
r
e
q
u
ality
an
d
en
h
an
ce
d
ata
r
eliab
ilit
y
,
as
illu
s
tr
ated
in
Fig
u
r
e
2
.
E
C
G
s
ig
n
als
u
n
d
er
wen
t
b
a
n
d
p
ass
f
i
lter
in
g
to
r
em
o
v
e
n
o
is
e,
u
tili
zin
g
a
tr
an
s
f
er
f
u
n
ctio
n
th
at
ef
f
ec
tiv
ely
atten
u
ates
u
n
wan
ted
f
r
eq
u
en
cies.
T
h
e
Pan
-
T
o
m
p
k
in
s
alg
o
r
ith
m
[
5
3
]
a
d
o
p
ts
f
o
r
p
r
ec
is
e
R
-
p
ea
k
d
etec
tio
n
,
en
s
u
r
in
g
ac
cu
r
ate
ca
r
d
iac
c
y
cle
id
en
tifi
ca
tio
n
.
E
ar
im
a
g
es
u
s
in
g
ad
v
a
n
ce
d
im
ag
e
n
o
r
m
aliza
tio
n
tech
n
iq
u
es
f
o
r
r
ep
r
o
ce
s
s
in
g
,
in
clu
d
in
g
h
is
to
g
r
a
m
eq
u
aliza
tio
n
an
d
co
n
tr
ast
lim
ited
ad
ap
tiv
e
h
is
to
g
r
am
eq
u
aliza
tio
n
(
C
L
AHE
)
[
5
4
]
.
A
cu
s
to
m
s
eg
m
e
n
tatio
n
a
lg
o
r
ith
m
e
x
tr
ac
ted
th
e
r
eg
io
n
o
f
i
n
ter
est
(
R
OI
)
,
f
o
cu
s
in
g
o
n
u
n
iq
u
e
ea
r
c
h
ar
ac
ter
is
tics
.
Fig
u
r
e
2
.
C
o
m
p
r
eh
en
s
iv
e
p
r
ep
r
o
ce
s
s
in
g
wo
r
k
f
l
o
w
f
o
r
E
C
G
an
d
ea
r
im
a
g
e
d
ata
3
.
2
.
Dee
p
lea
rning
a
rc
hite
ct
ures
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
em
p
lo
y
s
d
ee
p
lear
n
in
g
ar
ch
itectu
r
es
to
ex
tr
ac
t
d
is
cr
im
in
ativ
e
f
ea
t
u
r
es
f
r
o
m
E
C
G
s
ig
n
als
an
d
ea
r
im
ag
es
wh
ile
b
alan
cin
g
co
m
p
u
tatio
n
al
ef
f
icien
cy
an
d
class
if
icat
io
n
ac
cu
r
ac
y
.
T
h
e
f
r
am
ewo
r
k
f
o
llo
ws
a
m
u
lti
-
s
tag
e
f
ea
tu
r
e
ex
t
r
ac
tio
n
an
d
f
u
s
io
n
s
tr
ateg
y
to
g
en
er
ate
r
ep
r
esen
tativ
e
f
ea
tu
r
e
em
b
ed
d
in
g
s
.
T
h
e
f
ir
s
t
ar
ch
ite
ctu
r
e
is
in
s
p
ir
ed
b
y
VGG
-
v
e
r
y
d
ee
p
1
6
,
u
s
in
g
s
eq
u
en
tial
c
o
n
v
o
lu
ti
o
n
al
lay
e
r
s
with
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
e
L
U
)
ac
tiv
atio
n
an
d
m
ax
-
p
o
o
l
in
g
to
ex
tr
ac
t
h
ier
ar
ch
ical
f
e
atu
r
es,
alo
n
g
with
d
r
o
p
o
u
t
to
r
ed
u
ce
o
v
e
r
f
itti
n
g
.
T
h
e
s
ec
o
n
d
ar
c
h
itectu
r
e,
C
NN5
,
ex
ten
d
s
th
is
s
tr
u
ctu
r
e
with
an
ad
d
itio
n
al
co
n
v
o
l
u
tio
n
al
lay
er
to
ca
p
tu
r
e
m
o
r
e
co
m
p
lex
s
p
atial
p
atter
n
s
.
Featu
r
e
m
ap
s
f
r
o
m
b
o
th
m
o
d
els
ar
e
f
latten
ed
an
d
p
r
o
ce
s
s
ed
t
h
r
o
u
g
h
f
u
lly
c
o
n
n
ec
ted
lay
er
s
to
g
en
er
ate
e
m
b
ed
d
in
g
s
.
A
tr
ip
let
n
etwo
r
k
f
r
am
ewo
r
k
is
u
s
ed
d
u
r
in
g
tr
ain
in
g
,
wh
er
e
an
ch
o
r
,
p
o
s
itiv
e,
an
d
n
eg
ativ
e
s
am
p
le
s
p
ass
th
r
o
u
g
h
a
s
h
ar
ed
f
ea
tu
r
e
ex
tr
ac
to
r
t
o
lear
n
em
b
ed
d
in
g
s
th
at
m
in
im
ize
in
tr
a
-
class
d
is
tan
ce
an
d
m
ax
im
ize
in
ter
-
class
s
ep
ar
atio
n
.
T
h
e
tr
ip
let
lo
s
s
is
ap
p
lied
s
ep
ar
ately
f
o
r
ea
ch
m
o
d
ality
.
Fo
r
E
C
G
s
ig
n
als,
th
e
an
ch
o
r
is
a
r
ef
er
en
ce
E
C
G
s
eg
m
en
t
f
r
o
m
an
en
r
o
l
led
s
u
b
ject,
th
e
p
o
s
itiv
e
is
a
n
o
th
er
s
eg
m
en
t
f
r
o
m
th
e
s
am
e
s
u
b
ject,
a
n
d
t
h
e
n
eg
ativ
e
is
a
s
eg
m
en
t
f
r
o
m
a
d
if
f
er
en
t
s
u
b
ject.
Similar
ly
,
f
o
r
ea
r
im
ag
es,
th
e
an
c
h
o
r
is
a
r
ef
er
en
ce
ea
r
im
a
g
e,
th
e
p
o
s
itiv
e
is
an
o
th
er
im
ag
e
o
f
th
e
s
am
e
ea
r
,
an
d
th
e
n
e
g
ativ
e
is
an
im
ag
e
f
r
o
m
a
d
if
f
er
en
t
s
u
b
ject.
T
h
is
en
s
u
r
es
th
at
th
e
lear
n
ed
em
b
e
d
d
in
g
s
m
in
im
ize
in
t
r
a
-
class
d
is
tan
ce
an
d
m
ax
im
ize
in
te
r
-
cla
s
s
d
is
tan
ce
f
o
r
b
o
th
b
io
m
etr
ic
tr
aits
.
T
h
is
ar
ch
itectu
r
e
is
im
p
lem
en
ted
u
s
in
g
Ker
as
an
d
T
en
s
o
r
Flo
w
lib
r
ar
ies,
en
s
u
r
in
g
m
o
d
u
lar
ity
an
d
s
ca
lab
ilit
y
f
o
r
i
n
teg
r
atio
n
with
o
th
er
s
y
s
tem
s
.
T
h
e
tr
ain
in
g
co
n
f
ig
u
r
atio
n
f
o
r
b
o
t
h
V
GG
-
v
er
y
d
ee
p
1
6
a
n
d
C
NN5
m
o
d
els
is
d
etai
led
in
T
ab
le
5
,
wh
ich
s
p
ec
if
ies
th
e
h
y
p
er
p
ar
am
eter
s
u
s
ed
d
u
r
in
g
m
o
d
el
o
p
tim
izatio
n
.
T
h
ese
p
ar
am
eter
s
wer
e
s
elec
ted
th
r
o
u
g
h
em
p
ir
ical
v
a
lid
atio
n
to
en
s
u
r
e
o
p
tim
al
co
n
v
er
g
en
ce
an
d
g
en
er
aliza
tio
n
.
T
h
e
s
h
ar
e
d
n
e
two
r
k
is
tr
ain
ed
with
a
co
m
b
i
n
atio
n
o
f
E
C
G
an
d
ea
r
b
io
m
e
tr
ic
d
ata,
allo
win
g
th
e
m
o
d
el
to
lear
n
c
o
m
p
lem
en
tar
y
f
ea
tu
r
es
f
o
r
r
o
b
u
s
t
m
u
ltimo
d
al
au
t
h
en
ticatio
n
.
T
h
e
ar
ch
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
d
u
al
m
o
d
el
is
s
h
o
wn
in
Fig
u
r
e
3
.
T
ab
le
5
.
T
r
ai
n
in
g
h
y
p
e
r
p
ar
am
eter
s
f
o
r
th
e
p
r
o
p
o
s
ed
d
u
al
-
m
o
d
al
s
y
s
tem
H
y
p
e
r
p
a
r
a
me
t
e
r
V
a
l
u
e
D
e
scri
p
t
i
o
n
O
p
t
i
mi
z
e
r
A
d
a
m
A
d
a
p
t
i
v
e
m
o
me
n
t
e
s
t
i
m
a
t
i
o
n
Le
a
r
n
i
n
g
r
a
t
e
0
.
0
0
1
I
n
i
t
i
a
l
l
e
a
r
n
i
n
g
r
a
t
e
B
a
t
c
h
si
z
e
32
N
u
mb
e
r
o
f
sam
p
l
e
s
p
e
r
b
a
t
c
h
Ep
o
c
h
s
50
To
t
a
l
n
u
m
b
e
r
o
f
t
r
a
i
n
i
n
g
e
p
o
c
h
s
D
r
o
p
o
u
t
r
a
t
e
0
.
3
R
e
g
u
l
a
r
i
z
a
t
i
o
n
t
o
r
e
d
u
c
e
o
v
e
r
f
i
t
t
i
n
g
Tr
i
p
l
e
t
l
o
ss
marg
i
n
(
α)
0
.
5
M
a
r
g
i
n
p
a
r
a
m
e
t
e
r
u
se
d
i
n
t
r
i
p
l
e
t
l
o
ss
W
e
i
g
h
t
i
n
i
t
i
a
l
i
z
a
t
i
o
n
H
e
n
o
r
ma
l
I
n
i
t
i
a
l
i
z
a
t
i
o
n
sc
h
e
m
e
s
u
i
t
a
b
l
e
f
o
r
R
e
L
U
A
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
R
e
LU
(
h
i
d
d
e
n
)
,
s
o
f
t
m
a
x
(
o
u
t
p
u
t
)
N
o
n
-
l
i
n
e
a
r
t
r
a
n
sf
o
r
m
a
t
i
o
n
s
Lo
ss f
u
n
c
t
i
o
n
Tr
i
p
l
e
t
l
o
ss
+
c
r
o
ss
-
e
n
t
r
o
p
y
Jo
i
n
t
e
m
b
e
d
d
i
n
g
l
e
a
r
n
i
n
g
a
n
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
Ea
r
l
y
st
o
p
p
i
n
g
p
a
t
i
e
n
c
e
1
0
e
p
o
c
h
s
Tr
a
i
n
i
n
g
st
o
p
s i
f
v
a
l
i
d
a
t
i
o
n
l
o
ss
d
o
e
s
n
o
t
i
mp
r
o
v
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
5
2
-
3
2
6
8
3258
Fig
u
r
e
3
.
Ar
c
h
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
d
ee
p
d
u
al
-
m
o
d
al
b
io
m
etr
ic
Fig
u
r
e
4
illu
s
tr
ates
p
r
o
p
o
s
ed
h
y
b
r
id
f
u
s
io
n
ar
c
h
itectu
r
e.
T
h
is
f
ig
u
r
e
illu
s
tr
atin
g
d
y
n
a
m
ic
m
o
d
el
s
elec
tio
n
b
etwe
en
VGG
-
v
e
r
y
d
ee
p
1
6
an
d
C
NN5
b
ased
o
n
d
ataset
s
ize,
f
o
llo
wed
b
y
f
ea
t
u
r
e
e
x
tr
ac
tio
n
an
d
f
u
s
io
n
.
T
h
e
s
y
s
tem
d
y
n
am
ica
lly
s
elec
ts
b
etwe
en
VGG
-
v
er
y
d
ee
p
1
6
an
d
C
NN5
b
ased
o
n
d
ataset
s
ize
an
d
co
m
p
u
tatio
n
al
co
n
s
tr
ain
ts
.
F
o
r
lar
g
e
d
atasets
,
C
NN5
is
em
p
lo
y
ed
f
o
r
f
aster
f
ea
t
u
r
e
ex
tr
ac
tio
n
,
wh
ile
VGG
-
v
er
y
d
ee
p
1
6
is
u
s
ed
f
o
r
s
m
aller
d
atasets
r
eq
u
ir
in
g
h
i
g
h
er
ac
cu
r
ac
y
.
Featu
r
es
f
r
o
m
b
o
th
E
C
G
an
d
ea
r
m
o
d
alities
ar
e
ex
tr
ac
ted
v
ia
th
e
s
elec
ted
m
o
d
el,
th
en
f
u
s
ed
u
s
in
g
a
co
m
b
in
atio
n
o
f
weig
h
ted
s
u
m
an
d
m
in
-
m
ax
f
u
s
io
n
tech
n
iq
u
es.
T
h
e
f
u
s
ed
f
ea
tu
r
e
v
ec
to
r
is
p
ass
ed
to
th
e
f
in
al
class
if
icatio
n
lay
er
f
o
r
au
th
en
ticatio
n
d
ec
is
io
n
.
Fig
u
r
e
4
.
Hy
b
r
id
f
u
s
io
n
ar
c
h
itectu
r
e
o
f
t
h
e
p
r
o
p
o
s
ed
s
y
s
tem
3
.
3
.
F
e
a
t
ure
f
us
io
n t
ec
hn
iqu
e
I
n
m
u
ltimo
d
al
b
io
m
etr
ic
s
y
s
tem
s
,
co
m
b
in
in
g
in
f
o
r
m
atio
n
f
r
o
m
d
if
f
er
en
t
s
o
u
r
ce
s
,
s
u
c
h
as
E
C
G
s
ig
n
als
an
d
ea
r
im
ag
es
im
p
r
o
v
es
p
r
ed
ictio
n
ac
c
u
r
ac
y
an
d
s
y
s
tem
r
eliab
ilit
y
.
Fu
s
io
n
te
ch
n
iq
u
es
in
teg
r
ate
co
m
p
lem
en
tar
y
in
f
o
r
m
atio
n
f
r
o
m
b
o
t
h
m
o
d
alities
to
p
r
o
d
u
ce
a
m
o
r
e
r
o
b
u
s
t
r
ep
r
esen
tatio
n
f
o
r
ap
p
licatio
n
s
s
u
ch
as
b
io
m
etr
ic
id
en
tific
atio
n
an
d
h
ea
lth
ca
r
e
an
aly
s
is
.
T
w
o
co
m
m
o
n
f
u
s
io
n
m
eth
o
d
s
a
r
e
u
s
ed
in
t
h
is
wo
r
k
:
weig
h
ted
s
u
m
f
u
s
io
n
a
n
d
m
in
–
m
ax
f
u
s
io
n
.
W
eig
h
ted
s
u
m
f
u
s
io
n
co
m
b
in
es
f
ea
tu
r
es
f
r
o
m
m
u
ltip
le
m
o
d
alities
b
y
ass
ig
n
in
g
a
weig
h
t to
ea
ch
in
p
u
t f
ea
tu
r
e,
allo
win
g
m
o
r
e
r
eliab
le
s
o
u
r
ce
s
to
co
n
tr
ib
u
te
m
o
r
e
s
ig
n
if
ican
tly
to
th
e
f
in
al
r
ep
r
esen
tatio
n
as in
(
6
)
.
ℎ
=
(
1
,
2
,
…
,
)
=
1
∗
1
+
2
∗
2
+
…
+
∗
(
6
)
Min
–
m
ax
f
u
s
io
n
ca
p
tu
r
es
th
e
m
in
im
u
m
an
d
m
ax
im
u
m
v
alu
es
b
etwe
en
m
o
d
alities
to
p
r
eser
v
e
f
ea
t
u
r
e
v
ar
iab
ilit
y
an
d
h
ig
h
lig
h
t d
is
tin
ctiv
e
ch
ar
ac
ter
is
tics
as in
(
7
)
.
−
=
(
1
,
2
)
=
[
(
1
,
2
)
,
(
1
,
2
)
]
(
7
)
T
h
ese
f
u
s
io
n
s
tr
ateg
ies
im
p
r
o
v
e
s
y
s
tem
ac
cu
r
ac
y
an
d
r
o
b
u
s
tn
ess
b
y
lev
er
ag
in
g
co
m
p
lem
en
tar
y
in
f
o
r
m
atio
n
f
r
o
m
E
C
G
an
d
ea
r
m
o
d
alities
wh
ile
r
em
ain
in
g
co
m
p
u
t
atio
n
ally
ef
f
icien
t
an
d
ad
a
p
tab
le
to
d
i
f
f
er
en
t
ap
p
licatio
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
B
io
metric a
u
th
en
tica
tio
n
u
s
in
g
d
u
a
l
-
mo
d
a
l
d
ee
p
lea
r
n
in
g
b
a
s
ed
-
o
n
E
C
G
a
n
d
…
(
Mo
h
a
m
ed
S
.
K
h
a
la
f
)
3259
3
.
4
.
F
o
ld
cr
o
s
s
v
a
lid
a
t
io
n a
n
d per
f
o
rm
a
nce
m
e
t
rics
T
o
en
s
u
r
e
r
o
b
u
s
t
e
v
alu
atio
n
a
n
d
r
e
d
u
ce
d
ata
p
ar
titi
o
n
in
g
b
i
as,
a
5
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
s
tr
ateg
y
was
em
p
lo
y
ed
.
T
h
e
co
m
b
in
ed
d
a
taset
(
E
C
G
-
I
D
an
d
AM
I
)
w
as
r
an
d
o
m
ly
d
iv
i
d
ed
in
to
f
i
v
e
s
tr
atif
ied
f
o
ld
s
,
p
r
eser
v
in
g
t
h
e
class
d
is
tr
ib
u
tio
n
in
ea
c
h
f
o
l
d
.
Fo
r
ea
c
h
iter
atio
n
,
f
o
u
r
f
o
ld
s
wer
e
u
s
ed
f
o
r
tr
ain
in
g
an
d
th
e
r
em
ain
in
g
f
o
ld
f
o
r
test
in
g
.
T
h
i
s
p
r
o
ce
s
s
was r
ep
ea
ted
f
iv
e
ti
m
es so
th
at
ea
ch
f
o
ld
s
er
v
ed
o
n
ce
as th
e
test
s
et.
T
h
e
m
o
d
el
p
er
f
o
r
m
an
ce
was
ev
alu
ated
u
s
in
g
s
tan
d
ar
d
cla
s
s
if
icatio
n
m
etr
ics,
in
clu
d
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
-
s
co
r
e
,
wh
ich
m
ea
s
u
r
e
o
v
er
all
c
o
r
r
e
ctn
ess
,
p
r
ed
ictio
n
r
eliab
ilit
y
,
d
etec
tio
n
ca
p
ab
ilit
y
,
an
d
th
e
b
alan
ce
b
etwe
en
p
r
ec
is
io
n
an
d
r
ec
all.
T
h
ese
m
etr
ics
wer
e
ca
lcu
lated
o
n
th
e
test
d
ata
in
ea
ch
f
o
ld
,
an
d
th
e
r
esu
lts
wer
e
o
b
tain
ed
b
y
a
v
er
ag
in
g
th
e
p
er
f
o
r
m
a
n
ce
ac
r
o
s
s
all
f
iv
e
f
o
ld
s
.
T
h
e
a
v
er
ag
e
ev
alu
atio
n
r
esu
lts
ar
e
p
r
esen
ted
in
s
ec
tio
n
4
to
p
r
o
v
id
e
a
r
eliab
le
ass
ess
m
en
t o
f
th
e
p
r
o
p
o
s
ed
d
u
al
-
m
o
d
al
au
th
en
ticatio
n
s
y
s
tem
.
3
.
5
.
Co
m
pu
t
a
t
io
na
l
c
o
m
plex
it
y
a
na
ly
s
is
T
h
e
co
m
p
u
tatio
n
al
co
m
p
lex
it
y
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
els
is
an
aly
ze
d
in
ter
m
s
o
f
p
ar
a
m
eter
co
u
n
ts
an
d
f
lo
atin
g
-
p
o
in
t
o
p
er
atio
n
s
(
FL
OPs
)
.
T
h
e
co
m
p
lex
ity
was
ca
lcu
lated
f
o
r
a
s
in
g
le
f
o
r
war
d
p
ass
with
s
tan
d
ar
d
in
p
u
t
d
im
en
s
io
n
s
(
E
C
G:
5
0
0
0
s
am
p
les,
e
ar
:
2
2
4
×2
2
4
×3
)
.
T
ab
le
6
p
r
esen
ts
th
e
co
m
p
u
tatio
n
al
co
m
p
lex
ity
m
etr
ics
f
o
r
all
ev
alu
ated
ar
c
h
itectu
r
es,
in
clu
d
in
g
p
ar
am
et
er
co
u
n
ts
,
FLOPs,
m
em
o
r
y
r
eq
u
ir
em
e
n
ts
,
an
d
in
f
er
en
ce
tim
es.
As
s
h
o
wn
,
C
NN5
o
f
f
er
s
th
e
m
o
s
t
f
av
o
r
ab
l
e
b
alan
ce
b
etwe
en
co
m
p
lex
it
y
an
d
p
er
f
o
r
m
an
ce
with
o
n
ly
4
.
2
M
p
ar
am
eter
s
c
o
m
p
ar
ed
to
VGG
-
v
er
y
d
ee
p
1
6
'
s
1
3
8
M
p
ar
am
eter
s
.
As
ev
id
en
t
f
r
o
m
T
ab
le
6
,
C
NN5
p
r
o
v
id
es
t
h
e
o
p
tim
al
b
alan
ce
b
etwe
en
co
m
p
u
tatio
n
al
ef
f
icien
cy
an
d
p
e
r
f
o
r
m
an
c
e,
with
o
n
ly
4
.
2
M
p
ar
am
eter
s
an
d
0
.
8
g
ig
a
f
lo
atin
g
-
p
o
in
t
o
p
e
r
atio
n
s
(
GFL
OPs
)
co
m
p
ar
ed
to
VGG
-
v
e
r
y
d
ee
p
1
6
'
s
1
3
8
M
p
ar
am
eter
s
an
d
1
5
.
5
GFLO
Ps
.
T
h
is
m
a
k
es
C
NN5
p
ar
ticu
la
r
ly
s
u
itab
le
f
o
r
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
en
v
ir
o
n
m
en
ts
with
o
u
t sig
n
if
ican
tly
co
m
p
r
o
m
is
in
g
ac
cu
r
ac
y
.
T
ab
le
6
.
C
o
m
p
u
tatio
n
al
co
m
p
l
ex
ity
o
f
e
v
alu
ated
ar
c
h
itectu
r
e
s
M
o
d
e
l
P
a
r
a
me
t
e
r
s
F
LO
P
s
M
e
m
o
r
y
(
M
B
)
I
n
f
e
r
e
n
c
e
t
i
me
(
ms)
VGG
-
v
e
r
y
d
e
e
p
1
6
1
3
8
M
1
5
.
5
G
5
4
8
4
5
.
2
C
N
N
5
4
.
2
M
0
.
8
G
42
8
.
7
C
N
N
6
8
.
1
M
1
.
9
G
78
1
4
.
3
C
N
N
7
1
2
.
3
M
3
.
2
G
1
1
2
2
1
.
8
C
N
N
8
1
8
.
7
M
5
.
1
G
1
6
5
3
2
.
5
*
F
LO
P
s:
i
n
f
e
r
e
n
c
e
t
i
me
me
a
su
r
e
d
o
n
N
V
I
D
I
A
R
TX
3
0
9
0
*
4.
E
XP
E
R
I
M
E
N
T
A
L
RE
SUL
T
S
T
o
ass
ess
th
e
ef
f
ec
tiv
e
n
ess
an
d
r
o
b
u
s
tn
ess
o
f
t
h
e
p
r
o
p
o
s
ed
d
u
al
-
m
o
d
al
b
io
m
etr
ic
ar
ch
itectu
r
e
in
teg
r
atin
g
E
C
G
an
d
ea
r
f
ea
t
u
r
es,
th
e
d
ataset
was
p
ar
titi
o
n
ed
u
s
in
g
an
8
0
:2
0
r
atio
f
o
r
t
r
ain
in
g
an
d
test
in
g
,
r
esp
ec
tiv
ely
.
A
f
ix
ed
r
an
d
o
m
s
ee
d
was
a
p
p
lied
d
u
r
i
n
g
th
e
d
ata
s
p
lit
to
m
ain
tain
co
n
s
is
ten
cy
a
n
d
r
ep
r
o
d
u
cib
ilit
y
ac
r
o
s
s
ex
p
er
i
m
en
ts
.
T
h
e
ev
alu
atio
n
f
o
c
u
s
ed
o
n
k
ey
p
er
f
o
r
m
an
ce
in
d
icat
o
r
s
s
u
ch
as
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e,
wh
ich
wer
e
ca
lcu
lated
to
p
r
o
v
id
e
a
co
m
p
r
eh
en
s
iv
e
an
al
y
s
is
o
f
th
e
m
o
d
el’
s
p
r
ed
ictiv
e
ca
p
ab
ilit
ies.
T
h
e
E
C
G
-
I
D
d
ataset
u
s
ed
f
o
r
E
C
G
s
ig
n
als,
wh
ile
th
e
AM
I
d
ataset
p
r
o
v
i
d
ed
ea
r
im
ag
es.
B
o
th
d
atasets
o
f
f
er
ed
d
iv
er
s
e
an
d
h
ig
h
-
q
u
ality
s
am
p
les,
m
ak
in
g
th
em
s
u
itab
le
f
o
r
r
o
b
u
s
t
b
io
m
etr
ic
au
th
en
ticatio
n
s
tu
d
ies.
T
o
en
s
u
r
e
o
p
tim
al
p
er
f
o
r
m
an
ce
ac
r
o
s
s
v
ar
y
in
g
d
ataset
s
izes,
th
is
s
tu
d
y
ad
o
p
ts
an
ad
ap
tiv
e
ar
ch
itectu
r
e
s
tr
ateg
y
th
at
d
y
n
am
ically
s
elec
ts
b
etwe
en
C
NN5
an
d
VGG
-
v
er
y
d
ee
p
1
6
b
ased
o
n
d
ataset
s
ca
le.
E
x
ten
s
iv
e
ex
p
er
im
e
n
ts
d
em
o
n
s
tr
ated
th
at
C
NN5
,
with
it
s
li
g
h
ter
ar
ch
itectu
r
e
a
n
d
lo
wer
c
o
m
p
u
tatio
n
al
lo
a
d
,
s
ig
n
if
ican
tly
r
e
d
u
ce
d
tr
ain
in
g
tim
e
an
d
m
em
o
r
y
u
s
ag
e
o
n
lar
g
e
-
s
ca
le
d
atasets
with
o
u
t
s
u
b
s
tan
tial
lo
s
s
in
ac
cu
r
ac
y
.
C
o
n
v
er
s
ely
,
f
o
r
s
m
a
ller
d
atasets
wh
er
e
tr
ain
i
n
g
s
p
ee
d
is
less
cr
itical,
VGG
-
v
er
y
d
ee
p
1
6
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
ed
o
th
er
m
o
d
els
in
ac
cu
r
ac
y
d
u
e
to
its
d
ee
p
er
la
y
er
s
an
d
en
h
a
n
ce
d
f
ea
tu
r
e
ex
t
r
ac
tio
n
ca
p
ab
ilit
ies.
T
h
is
ad
ap
tiv
e
s
elec
tio
n
en
s
u
r
es
th
e
s
y
s
tem
r
e
m
ain
s
s
ca
lab
le
an
d
p
r
ac
tical
f
o
r
d
ep
lo
y
m
en
t
in
b
o
th
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
en
v
ir
o
n
m
en
ts
an
d
h
ig
h
-
ac
cu
r
ac
y
u
s
e
ca
s
es,
r
ein
f
o
r
cin
g
th
e
s
y
s
t
em
'
s
v
er
s
ati
lity
an
d
r
ea
l
-
wo
r
ld
a
p
p
licab
ilit
y
.
All
ex
p
er
im
en
ts
wer
e
im
p
lem
en
t
ed
in
Py
th
o
n
u
s
in
g
Ker
as
an
d
T
en
s
o
r
Flo
w.
T
h
e
m
o
d
els
wer
e
tr
ain
ed
an
d
ev
al
u
ated
o
n
a
wo
r
k
s
tatio
n
eq
u
i
p
p
ed
with
an
I
n
tel
C
o
r
e
i7
C
PU,
1
6
GB
R
AM
,
an
d
an
NVI
DI
A
R
T
X
3
0
9
0
GPU.
T
h
e
s
am
e
h
ar
d
war
e
co
n
f
ig
u
r
atio
n
was
u
s
ed
f
o
r
all
u
n
im
o
d
al
(
E
C
G
-
o
n
ly
an
d
ear
-
o
n
ly
)
an
d
m
u
ltimo
d
al
(
E
C
G
–
ea
r
f
u
s
io
n
)
ex
p
e
r
im
en
ts
to
en
s
u
r
e
a
f
air
c
o
m
p
ar
is
o
n
.
4
.
1
.
E
a
r
bio
m
et
ric
re
s
ults
T
h
e
tr
ain
in
g
an
d
v
alid
atio
n
a
cc
u
r
ac
y
a
n
d
lo
s
s
cu
r
v
es
f
o
r
t
h
e
ea
r
b
io
m
etr
ic
m
o
d
el,
tr
ai
n
ed
o
n
th
e
AM
I
d
ataset,
ar
e
s
h
o
wn
in
Fig
u
r
e
5
,
wh
er
e
th
e
m
o
d
el
ac
h
i
ev
ed
a
tr
ain
in
g
ac
c
u
r
ac
y
o
f
9
3
%
an
d
a
v
alid
atio
n
ac
cu
r
ac
y
o
f
9
0
%
af
ter
1
0
e
p
o
ch
s
,
d
em
o
n
s
tr
atin
g
s
tr
o
n
g
g
en
er
aliza
tio
n
with
clo
s
ely
co
n
v
e
r
g
in
g
c
u
r
v
es
an
d
n
o
s
ig
n
if
ican
t
o
v
er
f
itti
n
g
,
as
s
h
o
wn
in
Fig
u
r
e
5
(
a)
.
T
h
e
lo
s
s
c
u
r
v
es
ex
h
i
b
ited
a
co
n
s
is
ten
t
d
ec
lin
e,
s
tab
ilizin
g
at
ap
p
r
o
x
im
ately
0
.
2
f
o
r
tr
ain
in
g
an
d
0
.
3
f
o
r
v
alid
atio
n
,
co
n
f
i
r
m
in
g
ef
f
ec
tiv
e
lear
n
in
g
an
d
r
o
b
u
s
tn
ess
,
as
s
h
o
wn
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
5
2
-
3
2
6
8
3260
in
Fig
u
r
e
5
(
b
)
.
T
h
ese
r
esu
lt
s
h
ig
h
lig
h
t
th
e
m
o
d
el'
s
r
elia
b
ilit
y
f
o
r
ea
r
-
b
ased
au
th
e
n
ticatio
n
,
as
th
e
h
ig
h
ac
cu
r
ac
y
a
n
d
lo
w
lo
s
s
v
alu
es
in
d
icate
m
in
im
al
m
is
class
if
icatio
n
r
is
k
,
m
a
k
in
g
it
a
v
iab
le
s
o
lu
tio
n
f
o
r
s
ec
u
r
e
b
io
m
etr
ic
s
y
s
tem
s
,
with
p
o
ten
tial
f
o
r
f
u
r
th
er
e
n
h
an
ce
m
en
t
th
r
o
u
g
h
f
in
e
-
t
u
n
in
g
o
r
d
ata
au
g
m
en
tatio
n
.
Fig
u
r
e
6
p
r
esen
ts
th
e
p
r
ed
ictio
n
s
co
r
es
f
o
r
ea
r
im
ag
e
class
if
icatio
n
,
d
e
m
o
n
s
tr
atin
g
th
e
m
o
d
el's
d
is
cr
im
in
ativ
e
ca
p
ab
ilit
y
b
etwe
en
g
e
n
u
in
e
a
n
d
im
p
o
s
ter
s
am
p
les.
(
a)
(
b
)
Fig
u
r
e
5
.
T
r
ain
in
g
a
n
d
v
alid
atio
n
p
er
f
o
r
m
a
n
ce
f
o
r
th
e
AM
I
e
ar
b
io
m
etr
ic
m
o
d
el:
(
a)
ac
cu
r
a
cy
an
d
(
b
)
l
o
s
s
T
h
e
h
is
to
g
r
am
o
f
p
r
ed
ictio
n
p
r
o
b
a
b
ilit
ies
in
Fig
u
r
e
6
s
h
o
ws
a
clea
r
b
im
o
d
al
d
is
tr
ib
u
tio
n
,
with
g
en
u
in
e
m
atch
es
co
n
s
is
ten
tly
r
ec
eiv
in
g
h
ig
h
co
n
f
id
en
ce
s
co
r
es
(
ty
p
ically
ab
o
v
e
0
.
8
5
)
wh
ile
im
p
o
s
ter
co
m
p
ar
is
o
n
s
y
ield
s
ig
n
if
ican
tl
y
lo
wer
p
r
o
b
a
b
ilit
ies
(
g
en
er
al
ly
u
n
d
er
0
.
1
5
)
.
T
h
is
d
is
tin
ct
s
ep
ar
atio
n
b
etwe
en
th
e
two
d
is
tr
ib
u
tio
n
s
in
d
icate
s
r
o
b
u
s
t
class
d
is
cr
im
in
atio
n
,
with
m
in
im
al
o
v
er
lap
p
in
g
r
e
g
io
n
s
u
g
g
esti
n
g
f
ew
f
alse
ac
ce
p
tan
ce
o
r
r
ejec
tio
n
ca
s
es.
T
h
e
s
h
ar
p
p
ea
k
n
ea
r
p
r
o
b
ab
ilit
y
o
f
1
.
0
f
o
r
g
en
u
in
e
s
am
p
les
p
ar
ticu
lar
ly
u
n
d
er
s
co
r
es
th
e
m
o
d
el'
s
co
n
f
id
en
ce
in
c
o
r
r
ec
t
i
d
en
tific
atio
n
s
,
wh
ile
th
e
im
p
o
s
ter
d
is
tr
ib
u
t
io
n
'
s
co
n
ce
n
tr
atio
n
n
ea
r
ze
r
o
r
e
f
lects
ef
f
ec
tiv
e
r
ejec
tio
n
o
f
u
n
a
u
th
o
r
ize
d
atte
m
p
ts
.
T
h
ese
r
esu
lts
q
u
a
n
titativ
ely
v
alid
ate
t
h
e
m
o
d
el'
s
s
tr
o
n
g
au
th
en
ticatio
n
p
er
f
o
r
m
a
n
ce
,
with
th
e
clea
r
s
co
r
e
s
ep
ar
atio
n
ex
ce
ed
in
g
th
e
ty
p
ical
d
ec
is
io
n
th
r
esh
o
ld
o
f
0
.
5
b
y
a
c
o
n
s
id
er
ab
le
m
a
r
g
in
.
Su
ch
r
eliab
le
p
r
o
b
ab
ilis
tic
o
u
tp
u
ts
co
n
f
ir
m
t
h
e
s
y
s
tem
'
s
o
p
er
atio
n
al
r
ea
d
in
ess
f
o
r
r
ea
l
-
wo
r
ld
b
io
m
etr
ic
d
ep
lo
y
m
en
t,
wh
er
e
m
ain
tain
in
g
b
o
th
h
ig
h
ac
cu
r
ac
y
a
n
d
s
ec
u
r
ity
is
p
ar
am
o
u
n
t.
Fig
u
r
e
6
.
E
x
am
p
les f
o
r
th
e
r
es
u
lts
af
ter
p
r
ed
ictin
g
th
e
ea
r
im
ag
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
B
io
metric a
u
th
en
tica
tio
n
u
s
in
g
d
u
a
l
-
mo
d
a
l
d
ee
p
lea
r
n
in
g
b
a
s
ed
-
o
n
E
C
G
a
n
d
…
(
Mo
h
a
m
ed
S
.
K
h
a
la
f
)
3261
4
.2
.
E
lect
ro
c
a
rdio
g
ra
m
a
ut
hentic
a
t
io
n
r
esu
lt
s
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
E
C
G
-
b
ased
au
th
en
ticatio
n
m
o
d
el
was
ev
alu
ated
ac
r
o
s
s
1
0
tr
ain
in
g
ep
o
c
h
s
,
as
s
h
o
wn
in
Fig
u
r
e
7
.
Fig
u
r
e
7
(
a)
illu
s
tr
ates
th
e
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
cu
r
v
es,
wh
ile
Fig
u
r
e
7
(
b
)
p
r
esen
ts
th
e
co
r
r
esp
o
n
d
in
g
l
o
s
s
cu
r
v
es.
I
n
Fig
u
r
e
7
(
a)
,
t
h
e
m
o
d
el
ex
h
ib
its
a
co
n
s
is
ten
t
in
cr
ea
s
e
in
b
o
th
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
a
cy
o
v
er
th
e
e
p
o
ch
.
T
r
ain
in
g
ac
cu
r
ac
y
b
e
g
in
s
at
ap
p
r
o
x
im
ately
9
1
%
an
d
s
tead
ily
r
is
es
to
ar
o
u
n
d
9
8
.
2
%
b
y
ep
o
c
h
1
0
,
in
d
icatin
g
th
at
th
e
m
o
d
e
l
ef
f
ec
tiv
ely
lear
n
s
th
e
u
n
d
er
l
y
in
g
p
atter
n
s
in
th
e
E
C
G
d
ata.
Similar
ly
,
th
e
v
al
id
atio
n
ac
cu
r
ac
y
im
p
r
o
v
es
f
r
o
m
r
o
u
g
h
ly
9
0
%
to
9
6
.
3
%,
with
a
s
tab
le
tr
en
d
em
er
g
in
g
af
ter
e
p
o
ch
8
.
T
h
e
n
ar
r
o
win
g
g
ap
b
etwe
en
th
e
tr
ain
in
g
an
d
v
alid
atio
n
cu
r
v
es
s
u
g
g
ests
s
tr
o
n
g
g
en
er
aliza
tio
n
p
er
f
o
r
m
a
n
ce
an
d
m
in
im
al
o
v
er
f
itti
n
g
.
Fig
u
r
e
5
(
b
)
p
r
o
v
id
es
f
u
r
th
er
in
s
ig
h
t
i
n
to
m
o
d
el
lear
n
in
g
b
y
d
ep
ictin
g
th
e
tr
ain
in
g
an
d
v
alid
atio
n
lo
s
s
.
T
h
e
tr
ain
in
g
lo
s
s
d
r
o
p
s
s
h
ar
p
ly
f
r
o
m
ab
o
v
e
0
.
7
to
u
n
d
er
0
.
1
,
d
em
o
n
s
tr
atin
g
e
f
f
icien
t
co
n
v
e
r
g
en
ce
.
T
h
e
v
alid
atio
n
lo
s
s
,
al
th
o
u
g
h
s
lig
h
tly
h
ig
h
er
t
h
an
th
e
tr
ain
in
g
l
o
s
s
,
also
f
o
llo
ws
a
d
o
wn
war
d
tr
en
d
f
r
o
m
0
.
4
8
to
ap
p
r
o
x
im
ately
0
.
2
7
,
co
n
f
ir
m
in
g
th
at
th
e
m
o
d
el
is
n
o
t
o
n
ly
f
itti
n
g
th
e
tr
ain
in
g
d
ata
b
u
t
also
ad
ap
tin
g
well
to
u
n
s
ee
n
s
am
p
les.
T
h
e
co
m
b
in
atio
n
o
f
h
ig
h
v
alid
atio
n
ac
cu
r
ac
y
a
n
d
lo
w
v
alid
atio
n
lo
s
s
u
n
d
e
r
s
co
r
es th
e
r
o
b
u
s
tn
ess
o
f
th
e
E
C
G
-
b
ased
au
th
en
ticatio
n
m
o
d
el.
T
h
ese
r
esu
lts
h
ig
h
lig
h
t th
e
m
o
d
el’
s
ab
ilit
y
to
e
x
tr
ac
t
d
is
cr
im
in
ativ
e
f
ea
tu
r
es
f
r
o
m
p
h
y
s
io
lo
g
ical
s
ig
n
als
an
d
s
u
g
g
e
s
t
its
s
u
itab
ili
ty
f
o
r
r
ea
l
-
wo
r
ld
b
i
o
m
etr
ic
s
y
s
tem
s
wh
er
e
r
eliab
ilit
y
an
d
g
en
e
r
aliza
b
ilit
y
ar
e
cr
itical.
(
a)
(
b
)
Fig
u
r
e
7
.
Per
f
o
r
m
an
c
e
cu
r
v
es o
f
th
e
E
C
G
-
b
ased
u
n
im
o
d
al
a
u
th
en
ticatio
n
m
o
d
el:
(
a)
ac
c
u
r
ac
y
an
d
(
b
)
l
o
s
s
T
o
f
u
r
th
er
v
alid
ate
th
e
E
C
G
au
th
en
ticatio
n
p
r
o
ce
s
s
,
r
aw
E
C
G
s
ig
n
als
wer
e
an
aly
ze
d
f
r
o
m
th
e
E
C
G
-
I
D
d
ataset
an
d
m
ar
k
ed
th
e
v
en
tr
icu
la
r
d
ep
o
lar
izatio
n
(
QR
S)
p
o
in
ts
in
ea
ch
p
er
io
d
,
as sh
o
wn
in
Fig
u
r
e
8
.
T
h
e
d
ee
p
m
o
d
el
alg
o
r
ith
m
s
u
cc
ess
f
u
lly
id
en
tifie
d
th
e
co
r
r
e
ct
p
ea
k
s
,
wh
ich
ar
e
cr
itical
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
h
e
d
is
s
im
ilar
ity
s
co
r
es
co
m
p
u
ted
d
u
r
in
g
au
th
e
n
ticatio
n
r
an
g
e
d
b
etwe
en
0
.
3
7
an
d
1
.
2
8
,
co
n
f
ir
m
in
g
s
u
cc
ess
f
u
l
au
th
en
ticatio
n
f
o
r
a
ll
test
ca
s
e
s
.
T
h
ese
s
co
r
es
in
d
icate
a
h
ig
h
lev
el
o
f
co
n
f
id
en
c
e
in
th
e
E
C
G
-
b
ased
au
th
en
ticatio
n
s
y
s
tem
,
as
lo
wer
d
is
s
im
ilar
ity
s
co
r
es
co
r
r
esp
o
n
d
t
o
h
ig
h
er
s
im
ilar
ity
b
et
wee
n
th
e
in
p
u
t
a
n
d
en
r
o
lled
tem
p
lates.
Fig
u
r
e
8
.
C
o
r
r
ec
te
d
QR
S p
ea
k
d
etec
tio
n
Evaluation Warning : The document was created with Spire.PDF for Python.