I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
39
,
No
.
1
,
J
u
l
y
2
0
2
5
,
p
p
.
592
~
6
0
2
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
39
.i
1
.
p
p
5
9
2
-
6
0
2
592
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Rea
l
-
time
driver
dro
wsines
s det
ect
io
n bas
ed on int
e
g
ra
tive
a
ppro
a
ch of deep
lea
rning
and ma
c
hine learning
mo
del
G
o
wrish
a
nk
a
r
Sh
iv
a
Sh
a
nk
a
ra
Cha
ri,
J
y
o
t
hi Ar
co
t
P
r
a
s
ha
nt
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
r
i
n
g
,
M
.
S
.
R
a
ma
i
a
h
U
n
i
v
e
r
s
i
t
y
o
f
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s,
B
e
n
g
a
l
u
r
u
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
an
15
,
2
0
2
5
R
ev
is
ed
Mar
12
,
2
0
2
5
Acc
ep
ted
Mar
26
,
2
0
2
5
Driv
e
r
d
r
o
ws
in
e
ss
is
a
m
a
jo
r
f
a
c
to
r
th
a
t
c
o
n
tri
b
u
ti
n
g
t
o
r
o
a
d
a
c
c
id
e
n
ts.
S
e
v
e
ra
l
re
se
a
rc
h
e
s
a
re
o
n
g
o
i
n
g
t
o
d
e
tec
t
d
ri
v
e
r
d
ro
ws
i
n
e
ss
,
b
u
t
t
h
e
y
s
u
ffe
r
fro
m
th
e
c
o
m
p
lex
i
ty
a
n
d
c
o
st
o
f
th
e
m
o
d
e
ls.
T
h
is
p
a
p
e
r
i
n
tro
d
u
c
e
s
a
h
y
b
ri
d
a
rti
ficia
l
in
telli
g
e
n
c
e
(
AI
)
-
d
riv
e
n
fra
m
e
wo
rk
in
teg
ra
ti
n
g
d
e
e
p
lea
rn
in
g
(DL)
a
n
d
m
a
c
h
i
n
e
lea
rn
i
n
g
(M
L)
m
o
d
e
ls
fo
r
re
a
l
-
ti
m
e
d
r
o
ws
in
e
ss
d
e
te
c
ti
o
n
.
T
h
e
sy
ste
m
u
ti
li
z
e
s
a
ro
b
u
st
DL
m
o
d
e
l
to
c
las
sify
d
riv
e
r
sta
tes
b
a
se
d
o
n
fa
c
ial
ima
g
e
s
a
n
d
su
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
(
S
VM
)
m
o
d
e
l
is
train
e
d
to
d
e
v
e
lo
p
a
c
o
st
-
e
fficie
n
t
y
e
t
ro
b
u
st
fa
c
ial
lan
d
m
a
rk
d
e
tec
to
r
to
e
x
trac
t
k
e
y
fe
a
tu
re
s
su
c
h
a
s
e
y
e
a
sp
e
c
t
ra
ti
o
(EAR)
a
n
d
m
o
u
th
a
sp
e
c
t
ra
ti
o
(M
AR).
We
a
lso
in
tro
d
u
c
e
a
m
u
lt
i
-
sta
g
e
d
e
c
isio
n
fu
si
o
n
m
e
c
h
a
n
ism
th
a
t
c
o
m
b
in
e
s
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
(
CNN
)
p
ro
b
a
b
il
i
ty
sc
o
r
e
s
with
EAR/M
AR
th
re
sh
o
ld
s
t
o
e
n
h
a
n
c
e
d
e
tec
ti
o
n
re
li
a
b
il
it
y
a
n
d
re
d
u
c
e
fa
lse
p
o
si
ti
v
e
s.
Ex
p
e
rime
n
tal
re
su
lt
s
d
e
m
o
n
stra
te
th
a
t
th
e
p
r
o
p
o
se
d
m
o
d
e
l
a
c
h
iev
e
s
9
8
%
a
c
c
u
ra
c
y
a
n
d
F
1
-
sc
o
re
,
sig
n
ifi
c
a
n
t
ly
o
u
tp
e
rfo
rm
in
g
trad
it
io
n
a
l
DL
a
p
p
r
o
a
c
h
e
s.
Ad
d
it
i
o
n
a
ll
y
,
th
e
S
VM
-
b
a
se
d
la
n
d
m
a
rk
p
re
d
icto
r
sh
o
ws
im
p
ro
v
e
d
e
fficie
n
c
y
wit
h
lo
we
r
m
e
a
n
sq
u
a
re
d
e
rro
r
(M
S
E)
with
o
u
t
h
a
v
i
n
g
h
ig
h
e
r
c
o
m
p
u
tatio
n
a
l
re
q
u
irem
e
n
ts.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
Dr
iv
er
d
r
o
wsi
n
ess
d
etec
tio
n
Facial
lan
d
m
ar
k
p
r
ed
ictio
n
Mu
lti
-
s
tag
e
d
ec
is
io
n
f
u
s
io
n
R
o
ad
ac
cid
en
t
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
:
Go
wr
is
h
an
k
ar
Sh
iv
a
Sh
an
k
a
r
a
C
h
a
r
i
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
M
.
S
.
R
a
m
aiah
Un
iv
er
s
ity
of
Ap
p
lied
S
cien
ce
s
B
en
g
alu
r
u
,
I
n
d
ia
E
m
ail:
s
h
a
n
k
a
r
.
i
q
@
g
m
a
il
.
c
o
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
r
o
ad
s
af
ety
r
em
ain
s
a
s
ig
n
if
ican
t c
o
n
ce
r
n
wo
r
ld
wid
e
d
u
e
to
in
cr
ea
s
in
g
tr
af
f
ic
co
n
g
esti
o
n
an
d
th
e
g
r
o
win
g
n
u
m
b
er
o
f
v
eh
icles
o
n
th
e
r
o
ad
s
[
1
]
,
[
2
]
.
On
e
o
f
th
e
m
ajo
r
ca
u
s
es
o
f
r
o
ad
ac
cid
e
n
ts
is
d
r
iv
er
f
atig
u
e
an
d
d
r
o
wsi
n
ess
,
wh
ich
r
ed
u
ce
s
aler
tn
ess
,
s
lo
w
s
d
o
wn
r
ea
ctio
n
tim
e,
an
d
im
p
air
s
d
ec
is
io
n
-
m
ak
in
g
a
b
ilit
y
[
3
]
,
[
4
]
.
Sev
er
al
m
eth
o
d
s
h
av
e
b
ee
n
d
ev
elo
p
e
d
to
d
etec
t
d
r
iv
e
r
d
r
o
wsi
n
ess
u
s
in
g
p
h
y
s
io
lo
g
ical
s
ig
n
als,
b
eh
av
io
r
al
p
atter
n
s
,
an
d
f
ac
ial
ex
p
r
ess
io
n
s
also
b
ased
o
n
f
o
cu
s
ed
o
n
m
o
n
ito
r
in
g
s
teer
in
g
p
atter
n
s
,
lan
e
d
ev
iatio
n
,
an
d
h
ea
d
m
o
v
em
en
ts
[
5
]
,
[
6
]
.
Ho
wev
er
,
th
ese
ap
p
r
o
ac
h
es
o
f
te
n
p
r
o
d
u
ce
f
alse
p
o
s
itiv
es
b
ec
au
s
e
ex
ter
n
al
f
ac
t
o
r
s
lik
e
r
o
ad
c
o
n
d
itio
n
s
,
wea
th
er
ch
an
g
es,
an
d
v
eh
icle
ty
p
es
ca
n
in
f
lu
en
ce
d
r
iv
in
g
b
e
h
av
io
r
.
C
o
m
m
o
n
ap
p
r
o
ac
h
o
f
p
h
y
s
io
lo
g
ical
m
o
n
ito
r
in
g
,
wh
ich
in
clu
d
es e
lectr
o
en
ce
p
h
a
lo
g
r
ap
h
y
(EEG
)
,
h
ea
r
t r
ate
v
ar
iab
ilit
y
(
HR
V)
,
an
d
ey
e
-
tr
ac
k
in
g
s
en
s
o
r
s
a
r
e
h
ig
h
ly
ac
cu
r
ate
in
co
n
tr
o
lled
e
n
v
ir
o
n
m
en
ts
b
u
t
ar
e
ex
p
en
s
iv
e,
in
tr
u
s
iv
e,
a
n
d
im
p
r
ac
tical
f
o
r
r
ea
l
-
wo
r
l
d
ap
p
licatio
n
s
b
ec
au
s
e
th
ey
r
eq
u
ir
e
s
p
ec
ialized
h
ar
d
war
e
[
7
]
.
T
h
e
wo
r
k
ca
r
r
ied
o
u
t
b
y
Vice
n
te
et
a
l.
[
8
]
an
d
Sah
ay
ad
h
as
et
a
l.
[
9
]
ex
p
lo
r
ed
H
R
V
an
d
co
m
b
in
ed
p
h
y
s
io
lo
g
i
ca
l
an
d
b
eh
av
io
r
al
m
ea
s
u
r
es
f
o
r
d
r
o
wsi
n
ess
d
etec
tio
n
,
ac
h
ie
v
in
g
im
p
r
o
v
ed
ac
cu
r
ac
y
i
n
co
n
tr
o
lled
s
ettin
g
s
.
Ho
wev
er
,
r
ea
l
-
tim
e
ap
p
licab
ilit
y
is
lim
ited
d
u
e
to
s
en
s
o
r
p
lace
m
en
t
is
s
u
es
an
d
p
h
y
s
io
lo
g
ical
v
ar
iab
ilit
y
.
S
h
ah
b
ak
h
ti
et
a
l.
[
1
0
]
en
h
an
ce
d
E
E
G
-
b
ased
d
etec
tio
n
b
y
in
teg
r
atin
g
ey
e
b
lin
k
f
ea
tu
r
es
with
b
an
d
p
o
we
r
an
aly
s
i
s
,
b
u
t
E
E
G
s
ig
n
als
r
em
ain
h
ig
h
l
y
s
u
s
ce
p
tib
le
to
n
o
is
e
an
d
m
o
v
em
en
t
ar
tifa
cts.
Similar
ly
,
Fu
jiwar
a
et
a
l.
[
1
1
]
d
ev
elo
p
e
d
a
s
elf
-
atten
tio
n
au
to
en
co
d
er
u
s
in
g
e
lectr
o
ca
r
d
io
g
r
ap
h
y
(
E
C
G
)
R
-
R
in
ter
v
als,
d
em
o
n
s
tr
atin
g
h
ig
h
s
en
s
itiv
ity
,
b
u
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
R
ea
l
-
time
d
r
iver d
r
o
w
s
in
es
s
d
etec
tio
n
b
a
s
ed
o
n
in
teg
r
a
tive
…
(
Go
w
r
is
h
a
n
ka
r
S
h
iva
S
h
a
n
ka
r
a
C
h
a
r
i)
593
r
ea
l
-
wo
r
ld
m
o
n
ito
r
in
g
r
em
ai
n
s
ch
allen
g
in
g
d
u
e
to
m
o
ti
o
n
-
in
d
u
ce
d
d
is
to
r
tio
n
s
.
Ad
v
an
ce
m
en
ts
in
d
ee
p
lear
n
in
g
(
DL
)
a
n
d
m
ac
h
in
e
l
ea
r
n
in
g
(
ML
)
h
av
e
s
ig
n
i
f
ican
tly
im
p
r
o
v
e
d
f
ac
ial
-
b
ased
d
r
o
wsi
n
ess
d
etec
tio
n
.
B
ai
et
a
l.
[
1
2
]
p
r
o
p
o
s
ed
a
s
p
at
ial
-
tem
p
o
r
al
g
r
a
p
h
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN
)
to
ad
d
r
ess
illu
m
in
atio
n
ch
an
g
es
an
d
o
cc
lu
s
io
n
s
,
b
u
t
its
h
ig
h
co
m
p
u
tatio
n
al
co
s
t
lim
its
r
ea
l
-
tim
e
f
ea
s
ib
ilit
y
.
Flo
r
e
z
et
a
l.
[
1
3
]
ev
alu
ated
C
N
N
ar
ch
itectu
r
es
s
u
ch
as
I
n
ce
p
tio
n
V
3
,
VGG1
6
,
an
d
R
esNet5
0
V2
,
ac
h
iev
in
g
9
9
.
7
1
%
ac
cu
r
ac
y
o
n
th
e
NI
T
YM
E
D
d
ataset,
y
et
i
t
s
f
o
cu
s
o
n
n
ig
h
ttime
d
r
iv
in
g
r
estricts
g
en
er
aliza
tio
n
.
Ma
io
r
et
a
l.
[
1
4
]
u
tili
ze
d
ey
e
asp
ec
t
r
atio
(
E
AR
)
with
ML
class
if
ier
s
,
p
er
f
o
r
m
in
g
wel
l
o
n
th
e
DR
OZ
Y
d
ata
s
et,
b
u
t
lack
ed
r
ea
l
-
tim
e
ad
ap
tab
ilit
y
.
Oth
er
ap
p
r
o
ac
h
es,
s
u
ch
as
f
u
zz
y
lo
g
ic
-
b
ase
d
f
ea
tu
r
e
d
etec
tio
n
b
y
AlKis
h
r
i
et
a
l.
[
1
5
]
a
n
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM
)
-
b
ased
d
r
o
wsi
n
ess
d
etec
tio
n
b
y
Sh
u
k
la
et
a
l.
[
1
6
]
,
d
em
o
n
s
tr
ated
im
p
r
o
v
em
e
n
ts
b
u
t
wer
e
o
u
t
p
er
f
o
r
m
ed
b
y
C
NN
-
b
ased
m
o
d
els
o
n
lar
g
er
d
atasets
.
T
o
f
u
r
th
er
e
n
h
an
ce
ef
f
icien
cy
,
B
is
wal
et
a
l.
[
1
7
]
d
ev
elo
p
ed
a
n
I
o
T
-
b
ased
r
ea
l
-
tim
e
f
ac
ial
lan
d
m
ar
k
d
etec
ti
o
n
s
y
s
tem
,
th
o
u
g
h
d
ep
lo
y
m
e
n
t
ch
allen
g
es
r
em
ain
.
L
am
aa
zi
et
a
l.
[
1
8
]
in
tr
o
d
u
ce
d
e
d
g
e
-
b
ased
d
r
o
wsi
n
ess
d
etec
tio
n
,
e
n
s
u
r
in
g
p
r
iv
ac
y
an
d
r
e
d
u
ce
d
laten
cy
,
b
u
t
ed
g
e
co
m
p
u
tin
g
lim
itatio
n
s
co
n
s
tr
ain
DL
m
o
d
el
co
m
p
lex
ity
.
Z
h
an
g
et
a
l.
[
1
9
]
tack
led
th
is
b
y
im
p
lem
e
n
tin
g
f
ed
e
r
ated
t
r
an
s
f
er
lea
r
n
in
g
,
en
a
b
lin
g
d
is
tr
ib
u
ted
tr
ain
in
g
wh
ile
p
r
eser
v
in
g
p
r
iv
ac
y
.
Ah
m
ed
et
a
l.
[
2
0
]
im
p
r
o
v
e
d
ac
cu
r
ac
y
u
s
in
g
a
d
u
al
I
n
ce
p
tio
n
V3
m
o
d
el
with
f
ac
ia
l
s
u
b
s
am
p
lin
g
,
y
et
h
ig
h
co
m
p
u
tatio
n
al
d
em
an
d
s
in
tr
o
d
u
ce
d
in
f
er
e
n
ce
d
elay
s
,
l
im
itin
g
r
ea
l
-
tim
e
u
s
ab
ilit
y
.
J
eb
r
ae
ily
et
a
l.
[
2
1
]
f
u
r
th
er
o
p
tim
ize
d
C
NN
m
o
d
els
u
s
in
g
g
en
etic
alg
o
r
it
h
m
s
,
en
h
a
n
cin
g
p
e
r
f
o
r
m
an
c
e
b
u
t
in
cr
ea
s
in
g
co
m
p
u
tatio
n
al
o
v
er
h
ea
d
.
E
f
f
o
r
ts
to
o
p
tim
ize
h
a
r
d
war
e
e
f
f
ici
en
cy
h
a
v
e
also
b
ee
n
ex
p
lo
r
ed
.
Ng
u
y
en
et
a
l.
[
2
2
]
d
ev
elo
p
e
d
a
m
in
iatu
r
ized
E
E
G
-
b
ased
s
y
s
tem
with
tin
y
n
eu
r
al
n
etwo
r
k
s
,
r
ed
u
cin
g
p
r
o
ce
s
s
in
g
d
elay
s
b
u
t
s
till
lim
i
ted
b
y
E
E
G
s
ig
n
al
r
eliab
i
lity
in
r
ea
l
-
wo
r
ld
d
r
iv
i
n
g
.
Mo
u
s
av
ik
ia
et
a
l.
[
2
3
]
ac
ce
ler
ate
d
DL
i
n
f
er
en
ce
o
n
FP
GA,
ac
h
iev
in
g
f
aster
co
m
p
u
tatio
n
an
d
lo
wer
p
o
wer
c
o
n
s
u
m
p
tio
n
,
b
u
t
FP
GA
-
b
ased
im
p
lem
en
tatio
n
s
r
eq
u
ir
e
co
s
tly
h
a
r
d
war
e
m
o
d
if
icatio
n
s
.
L
astl
y
,
Ma
d
n
i
et
a
l.
[
2
4
]
lev
e
r
ag
ed
tr
an
s
f
er
lear
n
in
g
with
e
y
e
m
o
v
em
en
t
b
eh
a
v
io
r
a
n
aly
s
is
,
co
m
b
in
i
n
g
v
is
u
al
g
eo
m
etr
y
g
r
o
u
p
(
VGG
)
-
1
6
with
a
lig
h
t
g
r
ad
ie
n
t
-
b
o
o
s
tin
g
class
if
ier
,
y
et
its
r
ea
l
-
wo
r
ld
ef
f
ec
tiv
en
ess
r
eq
u
ir
es f
u
r
th
er
v
a
lid
atio
n
.
Desp
ite
ad
v
an
ce
m
en
ts
in
p
h
y
s
io
lo
g
ical,
b
e
h
av
io
r
al,
AI
-
b
ased
d
r
o
wsi
n
ess
d
etec
tio
n
,
ex
is
tin
g
ap
p
r
o
ac
h
es
f
ac
e
s
ev
er
al
lim
itatio
n
s
th
at
r
ed
u
ce
th
eir
ef
f
ec
t
iv
en
ess
f
o
r
r
ea
l
-
tim
e
d
r
iv
er
m
o
n
ito
r
in
g
in
r
ea
l
-
wo
r
ld
co
n
d
itio
n
s
.
On
e
m
ajo
r
ch
allen
g
e
is
th
at
m
an
y
DL
-
b
a
s
ed
m
o
d
els
ar
e
tr
ain
ed
o
n
lim
i
ted
d
atasets
,
th
at
d
o
n
o
t
ad
eq
u
ately
r
ep
r
esen
t
v
ar
i
atio
n
s
in
lig
h
tin
g
,
f
ac
ial
o
r
ien
tatio
n
s
,
o
r
eth
n
ic
d
iv
er
s
ity
,
o
r
en
v
ir
o
n
m
e
n
tal
f
ac
to
r
s
.
As
a
r
esu
lt,
th
ese
m
o
d
els
s
tr
u
g
g
le
with
o
cc
lu
s
io
n
s
(
e.
g
.
,
g
lass
es,
m
ask
s
,
p
o
o
r
lig
h
tin
g
)
an
d
h
ea
d
m
o
v
em
en
ts
.
Ad
d
itio
n
ally
,
m
a
n
y
s
tu
d
ies
u
s
e
eith
er
DL
o
r
f
ea
tu
r
e
-
b
ased
ap
p
r
o
ac
h
es
in
d
ep
en
d
en
tly
,
with
o
u
t
an
in
teg
r
ated
f
u
s
io
n
m
ec
h
a
n
is
m
th
at
co
m
b
in
es
b
o
th
m
eth
o
d
o
l
o
g
ies
f
o
r
e
n
h
an
ce
d
ac
cu
r
ac
y
.
L
astl
y
,
th
e
ex
is
tin
g
m
o
d
els
o
f
ten
p
r
i
o
r
itize
ac
cu
r
ac
y
o
v
er
c
o
m
p
u
tatio
n
al
ef
f
icien
cy
,
wh
ich
ar
e
n
o
t
s
u
itab
le
f
o
r
r
ea
l
-
tim
e
d
ep
lo
y
m
e
n
t in
au
t
o
m
o
tiv
e
a
p
p
licatio
n
s
wh
er
e
r
eso
u
r
ce
-
e
f
f
ici
en
t p
r
o
ce
s
s
in
g
is
ess
en
tial.
T
h
is
p
ap
er
in
tr
o
d
u
ce
s
a
h
y
b
r
i
d
AI
-
d
r
iv
en
f
r
am
ewo
r
k
th
at
i
n
teg
r
ates
DL
-
b
ased
im
ag
e
cla
s
s
if
icatio
n
with
a
lig
h
tweig
h
t
y
et
r
o
b
u
s
t
SVM
-
b
ased
f
ac
ial
lan
d
m
ar
k
d
etec
to
r
to
en
h
a
n
ce
ac
cu
r
ac
y
wh
ile
m
ain
tain
in
g
co
m
p
u
tatio
n
al
ef
f
icien
cy
f
o
r
r
ea
l
-
tim
e
d
r
iv
er
m
o
n
ito
r
in
g
.
T
h
e
ar
ticu
latio
n
o
f
r
esear
ch
g
a
p
is
as
f
o
llo
ws:
T
h
e
d
etec
tio
n
o
f
d
r
iv
e
r
d
r
o
wsi
n
ess
h
as
d
r
awn
a
lo
t
o
f
in
ter
est
lately
,
b
u
t
m
a
n
y
o
f
th
e
c
u
r
r
e
n
t
tech
n
iq
u
es
h
av
e
d
r
awb
ac
k
s
,
in
clu
d
in
g
lar
g
e
f
alse
p
o
s
itiv
e
r
ates,
in
s
en
s
itiv
ity
to
o
cc
lu
s
io
n
s
,
an
d
p
o
o
r
ac
cu
r
ac
y
in
p
r
ac
tical
s
ettin
g
s
.
E
y
e
clo
s
u
r
e
(
E
AR
)
a
n
d
y
awn
in
g
(
MA
R
)
,
two
in
d
e
p
en
d
en
t
f
ac
ial
m
etr
ics
th
at
ar
e
s
en
s
itiv
e
to
f
leetin
g
f
ac
ial
alter
atio
n
s
an
d
am
b
ien
t
co
n
d
itio
n
s
(
s
u
ch
as
lig
h
tin
g
an
d
o
cc
lu
s
io
n
s
)
,
ar
e
co
m
m
o
n
ly
u
s
ed
in
s
im
p
le
th
r
esh
o
ld
-
b
ased
d
r
o
wsi
n
ess
d
etec
tio
n
s
y
s
tem
s
.
Fu
r
th
er
m
o
r
e,
a
lo
t
o
f
DL
-
b
ased
s
y
s
tem
s
d
o
n
'
t
in
teg
r
ate
m
u
lti
-
m
o
d
al
d
ata
s
o
u
r
ce
s
o
r
i
n
c
o
r
p
o
r
ate
d
if
f
er
e
n
t f
ac
e
s
ig
n
als to
m
ak
e
b
etter
d
ec
is
io
n
s
.
I
n
o
r
d
er
to
o
v
er
c
o
m
e
th
ese
d
if
f
icu
lties
,
th
is
s
tu
d
y
s
u
g
g
ests
a
h
y
b
r
id
AI
-
d
r
iv
e
n
ar
ch
itectu
r
e
th
at
in
co
r
p
o
r
ates
a
m
u
lti
-
s
tag
e
d
ec
is
io
n
f
u
s
io
n
m
ec
h
an
is
m
,
ML
-
d
r
iv
en
f
ac
ial
lan
d
m
ar
k
tr
ac
k
in
g
,
an
d
DL
-
b
ased
f
ac
ial
s
tate
p
r
ed
ictio
n
.
T
h
e
m
a
in
co
n
tr
ib
u
tio
n
s
ar
e:
-
Mu
lti
-
s
tag
e
f
u
s
io
n
ap
p
r
o
ac
h
:
t
o
r
ed
u
ce
f
alse
alar
m
s
b
r
o
u
g
h
t
o
n
b
y
f
leetin
g
f
ac
ial
ch
an
g
es
,
we
in
co
r
p
o
r
ate
E
AR
,
MA
R
,
an
d
th
e
C
N
N
-
b
ased
p
r
o
b
ab
ilit
y
s
co
r
e
in
to
a
co
u
n
ter
-
b
ased
d
ec
is
io
n
f
u
s
io
n
s
y
s
tem
.
T
h
is
m
eth
o
d
g
u
ar
an
tees
th
at
an
aler
t
will
o
n
ly
b
e
tr
ig
g
er
ed
b
y
p
er
s
is
ten
t
s
ig
n
s
o
f
d
r
o
wsi
n
ess
,
s
u
ch
as
ex
ten
d
ed
ey
e
clo
s
in
g
,
f
r
eq
u
e
n
t y
awn
in
g
,
o
r
p
o
o
r
c
o
n
f
i
d
en
ce
in
t
h
e
C
NN
f
o
r
ec
ast.
-
Ad
ap
tab
ilit
y
to
en
v
ir
o
n
m
en
ta
l
v
ar
iab
ilit
y
:
b
y
co
m
b
in
in
g
s
ev
er
al
in
d
icati
o
n
s
th
at
ca
n
id
en
tify
tire
d
n
ess
th
r
o
u
g
h
v
a
r
io
u
s
asp
ec
ts
o
f
f
a
cial
b
eh
av
io
r
,
th
e
m
o
d
el
s
u
cc
ess
f
u
lly
m
an
ag
es
o
cc
l
u
s
io
n
s
(
s
u
ch
as
m
ask
s
o
r
s
u
n
g
lass
es)
an
d
ch
an
g
i
n
g
illu
m
in
atio
n
co
n
d
itio
n
s
.
-
R
ea
l
-
tim
e
p
er
f
o
r
m
an
ce
:
t
h
e
s
y
s
t
em
is
ap
p
r
o
p
r
iate
f
o
r
p
r
ac
t
ical
im
p
lem
en
tatio
n
in
em
b
ed
d
ed
s
y
s
tem
s
f
o
r
d
r
iv
er
m
o
n
ito
r
in
g
s
in
ce
it d
eli
v
er
s
r
ea
l
-
tim
e
ac
cu
r
ac
y
(
9
8
%)
an
d
r
o
b
u
s
tn
ess
with
lo
w
co
m
p
u
tatio
n
al
co
s
t.
-
I
n
cr
ea
s
ed
ac
cu
r
ac
y
an
d
r
o
b
u
s
tn
ess
:
th
e
s
u
g
g
ested
f
u
s
io
n
m
eth
o
d
o
lo
g
y
g
u
ar
an
tees
s
u
s
tain
ed
d
etec
tio
n
,
in
co
n
tr
ast
to
o
th
er
tech
n
iq
u
es
th
at
d
ep
en
d
ed
o
n
a
s
in
g
le
m
etr
ic
o
r
th
r
esh
o
ld
,
m
ak
in
g
th
e
s
y
s
tem
f
o
r
r
ea
l
-
tim
e
s
leep
in
ess
d
etec
tio
n
m
o
r
e
d
ep
en
d
ab
le
an
d
ac
cu
r
ate.
T
h
e
n
o
v
elty
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
tem
is
th
e
in
teg
r
atio
n
o
f
a
lig
h
t
-
weig
h
t
f
ac
ial
lan
d
m
ar
k
d
etec
tio
n
m
o
d
u
le
with
a
r
o
b
u
s
t
DL
-
b
ased
p
r
ed
icti
o
n
m
o
d
el,
wh
ich
en
ab
les ac
cu
r
ate
d
r
o
wsi
n
ess
d
etec
tio
n
u
n
d
e
r
co
m
p
lex
r
ea
l
-
w
o
r
ld
co
n
d
itio
n
s
u
s
in
g
f
ac
ial
f
e
atu
r
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
39
,
No
.
1
,
J
u
ly
20
25
:
592
-
6
0
2
594
2.
M
E
T
H
O
D
T
h
is
s
ec
tio
n
p
r
esen
ts
th
e
s
y
s
te
m
d
esig
n
a
n
d
im
p
lem
en
tatio
n
p
r
o
ce
d
u
r
es
f
o
r
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
A
I
-
d
r
iv
en
f
r
am
ewo
r
k
in
s
p
ir
ed
f
r
o
m
th
e
f
in
d
i
n
g
s
o
f
o
u
r
p
r
io
r
w
o
r
k
[
2
5
]
.
T
h
e
s
y
s
tem
in
teg
r
ates
DL
-
b
ased
im
ag
e
class
if
icatio
n
with
a
lig
h
twei
g
h
t
SVM
-
b
ased
f
ac
ial
lan
d
m
ar
k
d
etec
to
r
f
o
r
ac
cu
r
ate
an
d
r
ea
l
-
tim
e
d
r
i
v
er
d
r
o
wsi
n
ess
d
etec
tio
n
.
I
n
o
r
d
e
r
to
en
s
u
r
e
r
ea
l
-
tim
e
d
etec
tio
n
,
t
h
e
f
r
am
ewo
r
k
co
n
s
id
er
s
b
o
th
DL
p
r
ed
ictio
n
s
an
d
f
ac
ial
la
n
d
m
ar
k
tr
ac
k
i
n
g
o
p
er
ate
s
im
u
ltan
eo
u
s
ly
.
T
h
e
wo
r
k
f
l
o
w
o
f
th
e
p
r
o
p
o
s
ed
i
n
teg
r
ativ
e
s
y
s
tem
is
s
h
o
wn
in
Fig
u
r
e
1
.
As
s
h
o
wn
in
Fig
u
r
e
1
,
th
e
in
p
u
t
to
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
k
i
s
th
e
r
ea
l
-
tim
e
v
id
eo
f
ee
d
s
ap
t
u
r
ed
f
r
o
m
a
d
r
iv
er
-
f
ac
in
g
ca
m
er
a.
E
ac
h
v
id
eo
f
r
am
e
is
ex
tr
ac
ted
s
eq
u
e
n
tially
,
with
ev
er
y
th
ir
d
f
r
am
e
s
am
p
led
to
r
ed
u
ce
co
m
p
u
tatio
n
al
lo
a
d
wh
ile
m
a
in
tain
in
g
d
etec
tio
n
ef
f
icien
cy
.
T
h
e
ex
tr
ac
ted
f
r
a
m
es
ar
e
r
e
s
ized
to
m
atch
th
e
in
p
u
t
s
ize
o
f
th
e
DL
m
o
d
el
an
d
th
e
n
p
r
o
ce
s
s
ed
in
p
ar
a
llel
th
r
ea
d
s
i.e
.
,
o
n
e
f
o
r
DL
-
b
ased
d
r
o
wsi
n
ess
class
if
icatio
n
an
d
th
e
o
t
h
er
f
o
r
SVM
-
b
ased
f
ac
ial
lan
d
m
ar
k
d
etec
tio
n
.
T
h
e
DL
m
o
d
u
le
im
p
lem
en
ts
ef
f
icien
tNet
C
NN
m
o
d
el
class
if
ies
ea
ch
f
r
am
e
as
d
r
o
wsy
o
r
aler
t
b
y
an
aly
zin
g
f
ac
ial
f
ea
tu
r
es.
Simu
ltan
eo
u
s
ly
,
th
e
SVM
-
b
a
s
ed
lan
d
m
ar
k
d
etec
t
o
r
f
ac
ial
lan
d
m
ar
k
s
f
o
r
a
g
iv
en
f
r
am
e
.
T
h
e
s
y
s
tem
th
en
co
m
p
u
tes
E
AR
an
d
MA
R
f
ea
tu
r
e
to
m
ea
s
u
r
e
e
y
e
clo
s
u
r
e
an
d
y
awn
i
n
g
f
r
eq
u
e
n
cy
,
wh
ich
ar
e
cr
itical
in
d
icato
r
s
o
f
d
r
o
wsi
n
ess
.
I
n
o
r
d
er
to
e
n
s
u
r
e
d
etec
tio
n
r
eliab
i
lity
,
th
e
s
y
s
tem
em
p
lo
y
s
a
m
u
lti
-
s
tag
e
d
ec
is
io
n
f
u
s
io
n
m
ec
h
an
is
m
,
wh
ich
c
o
m
b
in
es
DL
-
b
ased
p
r
o
b
ab
ili
ty
s
co
r
es
with
E
AR
an
d
MA
R
th
r
esh
o
ld
s
.
I
f
d
r
o
wsi
n
ess
in
d
icato
r
s
p
er
s
is
t
f
o
r
co
n
s
ec
u
tiv
e
f
r
am
es,
an
al
er
t
is
g
en
er
ated
,
an
d
p
r
o
m
p
d
r
iv
er
to
ta
k
e
ac
tio
n
.
Ad
d
it
io
n
ally
,
t
h
e
aler
t
d
ata
is
s
to
r
ed
i
n
a
clo
u
d
-
b
ased
s
y
s
tem
,
wh
ich
en
ab
les
p
o
s
t
-
d
r
i
v
e
r
ev
iew
v
ia
a
web
ap
p
licatio
n
.
T
h
is
ap
p
r
o
ac
h
e
n
s
u
r
es
th
at
d
r
iv
er
s
ca
n
an
al
y
ze
th
eir
d
r
o
wsi
n
ess
h
is
to
r
y
a
n
d
m
a
k
e
n
ec
ess
ar
y
ad
ju
s
tm
en
ts
to
th
eir
d
r
i
v
in
g
b
eh
av
io
r
.
T
h
e
p
r
o
p
o
s
ed
f
r
am
e
wo
r
k
is
d
esig
n
ed
t
o
b
e
co
m
p
u
tatio
n
ally
ef
f
icien
t,
r
eso
u
r
ce
-
f
r
ie
n
d
ly
,
an
d
s
ca
lab
le,
th
er
eb
y
m
ak
i
n
g
it
s
u
itab
le
f
o
r
r
ea
l
-
wo
r
ld
d
ep
lo
y
m
en
t
s
ce
n
ar
io
with
v
er
y
m
in
im
al
f
alse p
o
s
itiv
es a
n
d
n
e
g
ativ
es.
Fig
u
r
e
1
.
I
ll
u
s
tr
ates w
o
r
k
f
lo
w
o
f
p
r
o
p
o
s
ed
i
n
teg
r
ativ
e
s
y
s
te
m
2
.
1
.
Da
t
a
s
et
des
cr
iptio
n
T
h
is
s
tu
d
y
cr
ea
ted
a
cu
s
to
m
d
ataset
b
y
s
cr
ap
in
g
6
,
0
0
0
im
ag
es
f
r
o
m
s
o
u
r
ce
s
lik
e
Go
o
g
le
an
d
B
in
g
,
an
d
ad
d
in
g
1
2
,
0
0
0
im
a
g
es
f
r
o
m
Kag
g
le.
T
h
e
im
ag
es
wer
e
m
an
u
ally
r
ev
iewe
d
,
a
n
d
1
0
,
4
4
4
r
elev
an
t
im
ag
es
wer
e
s
elec
ted
,
ex
clu
d
in
g
th
o
s
e
with
y
awn
in
g
t
o
f
o
c
u
s
o
n
f
ac
ial
f
ea
tu
r
es
lik
e
ey
e
clo
s
u
r
e
an
d
h
ea
d
p
o
s
itio
n
.
T
h
e
im
ag
es
wer
e
r
esized
to
2
2
4
×2
2
4
p
ix
els
an
d
s
p
lit
in
to
tr
ain
in
g
,
v
alid
atio
n
,
an
d
test
in
g
s
ets
in
a
8
0
:1
0
:1
0
r
atio
,
with
r
an
d
o
m
s
h
u
f
f
lin
g
t
o
av
o
id
o
r
d
e
r
b
ias.
T
h
e
d
atas
et
co
n
s
is
t
o
f
tr
ain
in
g
s
et
o
f
4
,
2
8
4
an
d
d
r
o
wsy
s
et
co
n
s
is
ts
o
f
4
,
0
7
7
im
ag
es
wh
e
r
ea
s
th
er
e
ar
e
5
3
1
im
ag
es
f
o
r
ac
tiv
e
s
tate
an
d
5
0
9
im
a
g
es
f
o
r
d
r
o
wsy
s
tate
f
o
r
v
alid
atio
n
d
ataset.
Fin
ally
,
test
in
g
im
ag
es
co
n
s
is
ts
o
f
5
3
2
i
m
ag
es
f
o
r
ac
tiv
e
s
tate
an
d
5
1
1
im
ag
es
f
o
r
d
r
o
wsy
s
tate.
2
.
2
.
DL
ba
s
ed
driv
er
dro
wsi
nes
s
pre
dict
io
n
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
em
p
lo
y
s
E
f
f
icien
tNetV2
B
0
,
an
o
p
tim
ized
DL
ar
ch
itectu
r
e
d
esig
n
ed
f
o
r
ef
f
icien
t
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
class
if
icatio
n
.
E
f
f
icien
tNet
V2
B
0
is
a
lig
h
tweig
h
t
C
NN
th
at
ap
p
l
ies
m
o
b
ile
b
o
ttlen
ec
k
co
n
v
o
lu
tio
n
s
(
MBC
o
n
v
)
an
d
i
n
v
er
ted
r
esid
u
al
b
lo
ck
s
to
ac
h
iev
e
o
p
tim
al
ac
cu
r
ac
y
with
m
in
im
al
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
R
ea
l
-
time
d
r
iver d
r
o
w
s
in
es
s
d
etec
tio
n
b
a
s
ed
o
n
in
teg
r
a
tive
…
(
Go
w
r
is
h
a
n
ka
r
S
h
iva
S
h
a
n
ka
r
a
C
h
a
r
i)
595
co
m
p
u
tatio
n
al
c
o
s
t.
I
n
th
e
p
r
o
p
o
s
ed
wo
r
k
,
th
is
lear
n
in
g
m
o
d
el
is
en
h
an
ce
d
with
a
d
d
itio
n
al
cu
s
to
m
lay
er
s
to
im
p
r
o
v
e
d
r
o
wsi
n
ess
d
etec
tio
n
ac
cu
r
a
cy
a
n
d
f
i
n
e
-
tu
n
e
d
f
o
r
b
in
ar
y
class
if
icatio
n
(
d
r
o
wsy
v
s
.
ac
tiv
e)
u
s
in
g
a
two
-
p
h
ase
tr
ain
in
g
s
tr
ateg
y
to
b
alan
ce
m
o
d
el
g
en
er
aliza
tio
n
an
d
task
-
s
p
ec
if
ic
ad
ap
tatio
n
.
T
ab
le
1
s
u
m
m
ar
izes
th
e
cu
s
to
m
DL
m
o
d
el
ad
o
p
ted
f
o
r
d
r
iv
er
d
r
o
wsi
n
ess
p
r
ed
icti
o
n
u
s
in
g
f
ac
ial
i
m
ag
e.
I
n
th
e
p
r
o
p
o
s
ed
s
y
s
tem
,
d
r
i
v
er
d
r
o
wsi
n
ess
p
r
ed
ictio
n
is
f
o
r
m
u
lated
as
a
b
in
ar
y
cla
s
s
if
icatio
n
p
r
o
b
lem
,
wh
er
e
t
h
e
m
o
d
el
d
eter
m
in
es
wh
eth
er
a
d
r
iv
er
i
s
d
r
o
wsy
(
Y=1
)
o
r
aler
t
(
Y=0
)
b
ased
o
n
f
ac
ial
im
ag
es.
Fo
r
a
g
iv
e
n
im
a
g
e
X
,
th
e
m
o
d
el
co
m
p
u
tes
th
e
p
r
o
b
a
b
ilit
y
o
f
d
r
o
wsi
n
ess
u
s
in
g
th
e
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
,
s
u
ch
th
at
(
=
1
|
)
=
σ
(
(
)
+
)
,
wh
e
r
e,
an
d
ar
e
th
e
tr
ain
a
b
le
p
a
r
a
m
eter
s
ca
lled
weig
h
ts
an
d
b
ias,
r
esp
ec
tiv
ely
,
(
)
r
ep
r
esen
ts
th
e
f
ea
tu
r
e
v
ec
to
r
ex
tr
ac
ted
f
r
o
m
th
e
f
in
al
d
e
n
s
e
lay
er
,
wh
er
e
σ
(
)
en
s
u
r
es
th
at
th
e
o
u
tp
u
t
is
a
p
r
o
b
ab
ilit
y
s
co
r
e
in
th
e
r
an
g
e
[
0
,
1
]
.
T
h
e
m
o
d
el
is
tr
ain
ed
u
s
in
g
th
e
b
in
ar
y
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
f
u
n
ctio
n
,
wh
ich
is
d
ef
in
ed
as:
[
ℒ
=
−
1
N
∑
[
y
i
l
og
(
y
i
̂
)
+
(
1
−
y
i
)
l
og
(
1
−
y
i
̂
)
]
]
N
i
=
1
(
1
)
W
h
er
e,
r
ep
r
esen
ts
th
e
g
r
o
u
n
d
tr
u
th
lab
el
(
1
f
o
r
d
r
o
wsy
,
0
f
o
r
aler
t)
,
y
i
̂
is
th
e
p
r
ed
icted
p
r
o
b
a
b
ilit
y
an
d
N
is
th
e
to
tal
n
u
m
b
er
o
f
t
h
e
tr
ain
in
g
s
am
p
les.
T
h
e
tr
ain
in
g
s
tr
ateg
y
in
c
o
r
p
o
r
ates
d
ata
au
g
m
en
tatio
n
an
d
a
two
-
p
h
ase
lear
n
in
g
p
r
o
ce
s
s
to
en
h
an
ce
m
o
d
el
g
e
n
er
aliza
tio
n
an
d
p
er
f
o
r
m
an
ce
.
I
n
th
e
i
n
itial
tr
ain
in
g
p
h
ase
(
f
ea
tu
r
e
ex
tr
ac
tio
n
p
h
ase)
,
th
e
b
ase
lay
er
s
o
f
th
e
DL
m
o
d
el
ar
e
f
r
o
ze
n
to
r
etain
th
e
p
r
e
-
tr
a
in
ed
f
ea
tu
r
es
f
r
o
m
I
m
ag
eNe
t,
wh
ile
o
n
l
y
th
e
c
u
s
to
m
lay
er
s
(
d
e
n
s
e,
d
r
o
p
o
u
t,
an
d
o
u
tp
u
t
lay
er
)
ar
e
tr
ain
e
d
.
T
h
e
m
o
d
el
u
n
d
er
g
o
es
2
0
ep
o
ch
s
o
f
tr
ain
in
g
with
a
n
in
itial
lear
n
in
g
r
ate
o
f
0
.
0
0
1
.
I
n
th
e
f
i
n
e
-
tu
n
i
n
g
p
h
ase
(
task
-
s
p
ec
if
ic
ad
ap
tatio
n
p
h
ase)
,
th
e
last
2
0
lay
e
r
s
o
f
DL
m
o
d
el
(
E
f
f
icien
tNetV2
B
0
)
ar
e
u
n
f
r
o
ze
n
t
o
allo
w
task
-
s
p
ec
if
ic
ad
j
u
s
tm
en
ts
wh
ile
lev
er
ag
in
g
p
r
e
-
tr
ain
e
d
k
n
o
wled
g
e
.
I
n
o
r
d
er
to
o
p
tim
ize
m
o
d
el
p
er
f
o
r
m
a
n
ce
,
th
e
f
o
llo
win
g
h
y
p
er
p
ar
am
eter
s
wer
e
f
i
n
e
-
tu
n
ed
as illu
s
tr
ated
in
T
ab
le
2
.
T
ab
le
1
.
T
h
e
ar
ch
itectu
r
al
d
eta
ils
o
f
th
e
cu
s
to
m
DL
La
y
e
r
t
y
p
e
I
n
p
u
t
s
h
a
p
e
F
i
l
t
e
r
s
i
z
e
N
u
mb
e
r
o
f
l
a
y
e
r
s
O
u
t
p
u
t
s
h
a
p
e
I
n
p
u
t
l
a
y
e
r
(
2
2
4
,
2
2
4
,
3
)
-
1
(
2
2
4
,
2
2
4
,
3
)
S
t
e
m
C
o
n
v
2
D
(
B
a
s
e
)
(
2
2
4
,
2
2
4
,
3
)
3
x
3
1
(
1
1
2
,
1
1
2
,
3
2
)
M
B
C
o
n
v
1
_
3
x
3
(
B
a
s
e
)
(
1
1
2
,
1
1
2
,
3
2
)
3
x
3
1
(
1
1
2
,
1
1
2
,
1
6
)
M
B
C
o
n
v
6
_
3
x
3
(
B
a
s
e
)
(
1
1
2
,
1
1
2
,
1
6
)
3
x
3
2
(
5
6
,
5
6
,
2
4
)
M
B
C
o
n
v
6
_
5
x
5
(
B
a
s
e
)
(
5
6
,
5
6
,
2
4
)
5
x
5
2
(
2
8
,
2
8
,
4
0
)
M
B
C
o
n
v
6
_
3
x
3
(
B
a
s
e
)
(
2
8
,
2
8
,
4
0
)
3
x
3
3
(
1
4
,
1
4
,
8
0
)
M
B
C
o
n
v
6
_
5
x
5
(
B
a
s
e
)
(
1
4
,
1
4
,
8
0
)
5
x
5
3
(
7
,
7
,
1
1
2
)
M
B
C
o
n
v
6
_
3
x
3
(
B
a
s
e
)
(
7
,
7
,
1
1
2
)
3
x
3
4
(
7
,
7
,
1
9
2
)
C
o
n
v
2
D
(
C
u
s
t
o
m)
(
7
,
7
,
1
9
2
)
3
x
3
1
(
7
,
7
,
1
2
8
)
G
l
o
b
a
l
A
v
e
r
a
g
e
P
o
o
l
i
n
g
2
D
(
7
,
7
,
1
2
8
)
-
1
(
1
2
8
)
D
e
n
se
l
a
y
e
r
(
1
2
8
)
-
1
(
1
0
2
4
)
D
r
o
p
o
u
t
(
1
0
2
4
)
-
1
(
1
0
2
4
)
D
e
n
se
l
a
y
e
r
(
1
0
2
4
)
-
1
(
1
2
8
)
O
u
t
p
u
t
l
a
y
e
r
(
D
e
n
se)
(
1
2
8
)
-
1
(
1
)
T
ab
le
2
.
T
h
e
d
etails o
f
h
y
p
er
p
ar
am
eter
o
p
tim
izatio
n
H
y
p
e
r
p
a
r
a
me
t
e
r
V
a
l
u
e
P
u
r
p
o
se
B
a
t
c
h
s
i
z
e
32
En
s
u
r
e
s s
t
a
b
l
e
t
r
a
i
n
i
n
g
a
n
d
p
r
e
v
e
n
t
s
memo
r
y
o
v
e
r
l
o
a
d
O
p
t
i
mi
z
e
r
A
d
a
m
A
d
a
p
t
i
v
e
l
e
a
r
n
i
n
g
r
a
t
e
f
o
r
e
f
f
i
c
i
e
n
t
w
e
i
g
h
t
u
p
d
a
t
e
s
I
n
i
t
i
a
l
l
e
a
r
n
i
n
g
r
a
t
e
0
.
0
0
1
F
a
st
e
r
c
o
n
v
e
r
g
e
n
c
e
i
n
t
h
e
i
n
i
t
i
a
l
t
r
a
i
n
i
n
g
p
h
a
s
e
f
i
n
e
-
t
u
n
i
n
g
l
e
a
r
n
i
n
g
r
a
t
e
0
.
0
0
0
1
P
r
e
v
e
n
t
s
su
d
d
e
n
w
e
i
g
h
t
c
h
a
n
g
e
s
d
u
r
i
n
g
f
i
n
e
-
t
u
n
i
n
g
D
r
o
p
o
u
t
r
a
t
e
0
.
5
R
e
d
u
c
e
s
o
v
e
r
f
i
t
t
i
n
g
i
n
d
e
n
se
l
a
y
e
r
s
To
t
a
l
e
p
o
c
h
s
50
2
0
f
o
r
i
n
i
t
i
a
l
t
r
a
i
n
i
n
g
+
3
0
f
o
r
f
i
n
e
-
t
u
n
i
n
g
2
.
3
.
M
L
ba
s
ed
f
a
cia
l la
nd
m
a
rk
predict
io
n m
o
del
Facial
lan
d
m
ar
k
d
etec
tio
n
p
la
y
s
a
cr
u
cial
r
o
le
i
n
id
e
n
tify
in
g
k
ey
f
ac
ial
f
ea
tu
r
es
s
u
ch
as
t
h
e
ey
es
an
d
m
o
u
th
,
wh
ich
s
er
v
e
as
in
d
icato
r
s
o
f
d
r
iv
e
r
v
ig
ilan
ce
.
B
y
ac
c
u
r
ately
d
etec
tin
g
th
ese
la
n
d
m
a
r
k
s
,
th
e
s
y
s
tem
ca
n
co
m
p
u
te
ess
en
tial
m
etr
ics
s
u
ch
as
th
e
E
AR
an
d
MA
R
,
wh
ich
ar
e
f
u
n
d
am
e
n
tal
in
ass
ess
i
n
g
d
r
o
wsi
n
ess
an
d
y
awn
in
g
f
r
eq
u
e
n
cy
.
Alth
o
u
g
h
v
ar
io
u
s
lan
d
m
ar
k
d
etec
tio
n
m
eth
o
d
s
ex
is
t,
b
u
t
d
lib
-
6
8
[
2
6
]
,
[
2
7
]
,
a
p
r
e
-
tr
ain
ed
ML
m
o
d
el,
h
as b
ee
n
wi
d
ely
a
d
o
p
ted
d
u
e
to
its
lig
h
tweig
h
t n
atu
r
e
an
d
r
ea
s
o
n
ab
le
ac
cu
r
ac
y
ac
r
o
s
s
d
iv
er
s
e
f
ac
e
o
r
ien
tatio
n
s
.
Ho
wev
er
,
d
lib
-
6
8
is
h
ig
h
ly
s
en
s
itiv
e
to
o
cc
lu
s
io
n
s
an
d
lack
s
ad
ap
tab
ilit
y
to
d
y
n
am
ic
f
ac
ial
ch
an
g
es,
wh
ich
af
f
ec
ts
its
r
o
b
u
s
tn
ess
in
r
ea
l
-
wo
r
ld
d
r
iv
in
g
s
ce
n
ar
io
s
.
T
h
er
ef
o
r
e,
th
e
p
r
o
p
o
s
ed
s
tu
d
y
d
ev
elo
p
s
a
lig
h
tweig
h
t
y
et
r
o
b
u
s
t
f
ac
ial
lan
d
m
ar
k
d
etec
to
r
u
tili
zin
g
a
p
r
e
-
tr
ain
ed
C
NN
f
o
r
f
ea
tu
r
e
e
x
tr
ac
tio
n
an
d
SVM
f
o
r
lan
d
m
ar
k
p
r
ed
ictio
n
.
T
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
in
teg
r
ate
s
p
u
b
licly
av
ailab
le
iB
u
g
3
0
0
-
W
f
ac
e
lan
d
m
ar
k
d
ataset
[
2
6
]
,
an
d
c
u
s
to
m
d
ata
s
et
f
r
o
m
th
e
p
r
ev
io
u
s
m
o
d
u
le
u
s
ed
f
o
r
d
r
o
wsi
n
ess
p
r
ed
ictio
n
.
Sin
ce
p
u
b
licly
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
39
,
No
.
1
,
J
u
ly
20
25
:
592
-
6
0
2
596
av
ailab
le
d
atasets
o
f
ten
lack
v
ar
iatio
n
s
in
d
y
n
am
ic
co
n
d
itio
n
s
,
th
e
cu
s
to
m
d
ataset
is
en
r
ich
ed
with
d
iv
er
s
e
f
ac
ial
ex
p
r
ess
io
n
s
,
o
cc
lu
s
io
n
s
(
e.
g
.
,
m
as
k
s
,
s
u
n
g
lass
es),
an
d
d
if
f
er
e
n
t
lig
h
tin
g
co
n
d
itio
n
s
.
Her
e,
Dlib
is
em
p
lo
y
ed
to
g
e
n
er
ate
g
r
o
u
n
d
-
tr
u
th
f
ac
ial
lan
d
m
ar
k
s
f
o
r
ea
c
h
im
ag
e
in
t
h
e
o
u
r
cu
s
to
m
d
at
aset,
an
d
th
en
b
o
th
d
atasets
(
iB
u
g
3
0
0
-
W
an
d
o
u
r
cu
s
to
m
im
ag
e
d
ataset)
ar
e
co
m
b
in
ed
i
n
to
a
s
in
g
le
co
m
p
r
eh
en
s
iv
e
d
ataset,
co
n
s
is
tin
g
two
class
es
n
am
el
y
im
ag
es
an
d
la
b
els
(
co
n
tain
in
g
f
ac
ial
co
o
r
d
in
ates)
.
T
h
en
f
o
r
tr
ain
i
n
g
SVM
m
o
d
el,
th
e
d
ataset
is
s
p
lit
in
to
tr
ain
in
g
(
8
0
%),
v
alid
atio
n
(
1
0
%)
an
d
test
in
g
(
1
0
%)
s
u
b
s
ets.
Af
ter
war
d
s
,
th
e
s
tu
d
y
th
en
p
er
f
o
r
m
s
f
ea
tu
r
e
ex
tr
ac
tio
n
o
p
e
r
atio
n
u
s
in
g
a
lig
h
tweig
h
t
R
esNet
-
1
8
m
o
d
el
with
p
r
e
-
tr
ain
ed
I
m
ag
eNe
t
weig
h
ts
.
Sin
ce
R
esNet
-
1
8
is
n
o
t
r
ea
d
ily
av
aila
b
le
in
T
en
s
o
r
Flo
w,
we
cu
s
to
m
ize
R
esNet
-
5
0
b
y
r
ed
u
cin
g
th
e
n
u
m
b
e
r
o
f
lay
er
s
an
d
r
em
o
v
i
n
g
u
n
n
ec
ess
ar
y
b
o
ttlen
ec
k
lay
er
s
b
y
r
e
p
l
ac
in
g
th
r
ee
co
n
v
o
lu
tio
n
al
lay
er
s
with
two
to
m
ain
tain
co
m
p
u
tatio
n
al
e
f
f
icien
cy
wh
i
le
p
r
eser
v
in
g
f
ea
tu
r
e
ex
tr
ac
ti
o
n
ca
p
ab
ilit
y
.
T
h
e
ex
tr
ac
ted
d
ee
p
f
ea
tu
r
es f
r
o
m
R
esNet
-
1
8
f
u
r
th
er
s
er
v
e
as o
p
tim
ized
in
p
u
t r
ep
r
esen
tatio
n
s
f
o
r
tr
ain
in
g
th
e
SVM
m
o
d
el
to
p
r
ed
ict
lan
d
m
a
r
k
c
o
o
r
d
in
ates.
Facial
lan
d
m
ar
k
d
etec
tio
n
is
f
o
r
m
u
lated
as
a
r
eg
r
ess
io
n
p
r
o
b
le
m
,
wh
er
e
th
e
g
o
al
is
to
p
r
ed
ict
t
h
e
an
d
co
o
r
d
in
ates
o
f
f
ac
ial
k
ey
p
o
i
n
ts
b
ased
o
n
e
x
tr
ac
ted
f
ea
tu
r
es.
I
n
t
h
is
r
eg
ar
d
,
f
o
r
a
g
iv
en
i
n
p
u
t
im
a
g
e
I
th
e
ex
tr
ac
ted
f
ea
tu
r
e
v
ec
to
r
F
∈
ℝ
d
f
r
o
m
t
h
e
C
NN
m
o
d
el
s
er
v
es
as
th
e
in
p
u
t
to
th
e
SVM
m
o
d
el
th
at
lear
n
s
a
f
u
n
ctio
n
:
ℝ
→
ℝ
2
th
at
m
a
p
s
ea
ch
in
p
u
t
f
ea
tu
r
e
v
ec
to
r
to
a
p
r
ed
icted
lan
d
m
ar
k
co
o
r
d
in
ate
s
u
ch
th
at:
(
̂
=
(
F
)
=
+
)
,
wh
er
e,
̂
=
(
̂
,
̂
)
r
ep
r
esen
ts
th
e
p
r
ed
icted
f
ac
ial
lan
d
m
ar
k
co
o
r
d
in
ates
an
d
W
an
d
b
ar
e
t
h
e
weig
h
ts
an
d
b
ias
p
ar
am
eter
s
.
Sin
ce
f
ac
ial
l
an
d
m
ar
k
p
r
e
d
ictio
n
in
v
o
lv
es
lear
n
in
g
a
n
o
n
-
lin
ea
r
m
ap
p
in
g
,
th
e
tr
ain
in
g
o
f
SV
M
is
ca
r
r
ied
o
u
t
co
n
s
id
er
in
g
R
B
F
k
er
n
el,
wh
ich
h
elp
s
to
tr
an
s
f
o
r
m
th
e
f
ea
tu
r
e
s
p
ac
e
in
to
a
h
i
g
h
er
-
di
m
e
n
s
io
n
al
r
ep
r
esen
tatio
n
,
s
u
ch
th
at:
(
,
)
=
e
xp
(
−
|
|
−
|
|
2
)
,
wh
er
e,
(
,
)
r
ep
r
esen
ts
th
e
s
im
il
ar
ity
b
etwe
en
two
f
ea
tu
r
e
v
ec
to
r
s
,
is
a
tu
n
ab
le
h
y
p
er
p
ar
am
eter
th
at
co
n
tr
o
ls
th
e
wid
th
o
f
th
e
Gau
s
s
ian
k
er
n
el
an
d
|
|
−
|
|
2
is
th
e
s
q
u
ar
ed
E
u
clid
ea
n
d
is
tan
ce
b
etwe
en
f
ea
tu
r
e
v
ec
to
r
s
.
An
o
th
er
h
y
p
er
p
a
r
am
eter
co
n
s
id
er
ed
is
th
e
r
eg
u
lar
izatio
n
p
ar
am
eter
(
c)
th
at
co
n
tr
o
ls
th
e
tr
ad
e
-
o
f
f
b
etwe
e
n
ac
h
iev
in
g
a
lo
w
er
r
o
r
o
n
t
h
e
tr
ain
in
g
s
et
an
d
m
in
im
izin
g
m
o
d
el
co
m
p
lex
it
y
to
p
r
ev
en
t
o
v
er
f
itti
n
g
.
On
th
e
o
th
er
h
a
n
d
,
a
to
ler
an
ce
f
ac
to
r
is
co
n
s
id
er
ed
to
s
p
ec
if
ies
th
e
m
ar
g
in
o
f
to
le
r
an
c
e
f
o
r
p
r
ed
ictio
n
e
r
r
o
r
s
in
r
eg
r
ess
io
n
p
r
o
b
lem
s
o
f
f
ac
ial
co
o
r
d
in
at
es
p
r
ed
ictio
n
.
T
h
er
ef
o
r
e,
th
e
o
b
jectiv
e
f
u
n
ctio
n
f
o
r
SVM
b
ased
lan
d
m
ar
k
p
r
ed
ictio
n
m
o
d
el
is
f
o
r
m
u
lated
is
a
s
f
o
llo
ws:
[
min
,
1
2
|
|
|
|
2
+
∑
ma
x
(
0
,
|
−
̂
|
−
)
=
1
]
(
2
)
W
h
er
e,
an
d
̂
ar
e
t
h
e
g
r
o
u
n
d
-
tr
u
th
an
d
p
r
e
d
icted
lan
d
m
a
r
k
co
o
r
d
in
ates,
N
is
th
e
n
u
m
b
er
o
f
tr
ai
n
in
g
s
am
p
les
an
d
th
e
ter
m
ma
x
(
0
,
|
−
̂
|
−
)
,
en
f
o
r
c
es
th
e
to
ler
an
ce
m
ar
g
in
,
th
er
e
b
y
en
s
u
r
in
g
th
at
p
r
ed
ictio
n
s
with
in
ar
e
n
o
t
p
en
alize
d
.
On
ce
SVM
m
o
d
el
is
tr
ain
ed
th
en
its
i
s
ev
alu
ated
u
s
in
g
th
e
m
ea
n
s
q
u
ar
ed
er
r
o
r
(
MSE
)
o
n
test
d
ataset.
T
h
e
p
r
e
d
icted
lan
d
m
ar
k
s
ar
e
th
en
u
s
e
d
to
co
m
p
u
te
E
AR
an
d
MA
R
as f
o
llo
ws:
=
|
|
2
−
6
|
|
+
|
|
3
−
5
|
|
2
×
|
|
1
−
4
|
|
(
3
)
W
h
er
e,
1
,
4
d
e
n
o
tes
h
o
r
izo
n
tal
e
y
e
lan
d
m
ar
k
s
,
2
,
6
an
d
3
,
5
ar
e
th
e
v
er
tical
ey
e
lan
d
m
a
r
k
,
|
|
-
|
|
is
th
e
E
u
clid
ea
n
d
is
tan
ce
b
etwe
e
n
p
o
in
ts
an
d
.
Sin
ce
th
e
E
AR
r
em
ain
s
n
ea
r
ly
co
n
s
tan
t
wh
en
ey
es
ar
e
o
p
en
an
d
r
ap
i
d
ly
d
ec
r
ea
s
es d
u
r
in
g
a
b
lin
k
,
it ser
v
es a
s
an
ef
f
ec
tiv
e
in
d
icato
r
f
o
r
ey
e
clo
s
u
r
e
d
ete
ctio
n
.
MAR
=
|
|
2
−
8
|
|
+
|
|
3
−
7
|
|
+
|
|
4
−
6
|
|
2
×
|
|
1
−
5
|
|
(
4
)
W
h
er
e,
P1
,
P5
→
h
o
r
izo
n
tal
m
o
u
th
co
r
n
er
p
o
i
n
ts
,
P2
,
P8
→
v
er
tical
u
p
p
er
an
d
lo
wer
lip
p
o
in
ts
(
o
u
ter
)
,
P3
,
P7
→
v
er
tical
u
p
p
e
r
an
d
lo
we
r
lip
p
o
i
n
ts
(
m
id
d
le)
,
P4
,
P6
→
v
er
tic
al
u
p
p
er
an
d
lo
wer
li
p
p
o
in
ts
(
in
n
er
)
,
an
d
∣
Pa−Pb
∣
r
ep
r
esen
ts
th
e
E
u
clid
ea
n
d
is
tan
ce
b
etwe
en
two
p
o
in
ts
.
2
.
4
.
M
ulti
-
s
t
a
g
e
decisi
o
n f
us
io
n m
ec
ha
nis
m
T
o
en
s
u
r
e
ac
cu
r
ate
an
d
r
eliab
l
e
d
r
o
wsi
n
ess
d
etec
tio
n
,
th
e
p
r
o
p
o
s
ed
s
y
s
tem
im
p
lem
en
ts
a
m
u
lti
-
s
tag
e
d
ec
is
io
n
f
u
s
io
n
m
ec
h
an
is
m
th
at
co
m
b
in
es
DL
-
b
ased
p
r
ed
ict
io
n
s
with
g
eo
m
etr
ic
th
r
esh
o
ld
i
n
g
tech
n
iq
u
es.
T
h
is
ap
p
r
o
ac
h
m
in
im
izes
f
alse
p
o
s
itiv
es
an
d
f
alse
n
eg
ati
v
es
b
y
i
n
teg
r
atin
g
th
e
C
NN
p
r
o
b
ab
ilit
y
s
co
r
e
with
E
AR
an
d
MA
R
.
T
h
e
s
y
s
tem
co
n
tin
u
o
u
s
ly
m
o
n
ito
r
s
E
AR
,
MA
R
,
an
d
th
e
C
NN
p
r
o
b
ab
ilit
y
s
co
r
e.
I
f
an
y
o
f
th
ese
m
etr
ics
ex
ce
ed
th
eir
p
r
e
d
ef
in
e
d
th
r
esh
o
ld
s
f
o
r
a
s
p
ec
if
ied
n
u
m
b
er
o
f
c
o
n
s
ec
u
tiv
e
f
r
a
m
es,
an
aler
t
is
tr
ig
g
er
ed
.
T
h
e
p
r
ed
ef
i
n
ed
th
r
esh
o
l
d
v
alu
es
an
d
co
n
d
itio
n
s
f
o
r
d
ec
is
io
n
f
u
s
io
n
ar
e
s
u
m
m
ar
ized
in
T
a
b
le
3
.
Her
e,
E
AR
is
u
s
ed
to
d
eter
m
in
e
p
r
o
lo
n
g
ed
ey
e
clo
s
u
r
e,
MA
R
is
u
s
ed
to
d
etec
t
y
awn
in
g
,
a
n
d
th
e
C
N
N
p
r
o
b
ab
ilit
y
s
co
r
e
ass
es
s
es
th
e
o
v
er
all
lik
elih
o
o
d
o
f
d
r
o
wsi
n
ess
b
ased
o
n
f
ac
ial
f
ea
tu
r
e
an
aly
s
is
.
T
h
ese
co
m
b
in
ed
in
d
icato
r
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
R
ea
l
-
time
d
r
iver d
r
o
w
s
in
es
s
d
etec
tio
n
b
a
s
ed
o
n
in
teg
r
a
tive
…
(
Go
w
r
is
h
a
n
ka
r
S
h
iva
S
h
a
n
ka
r
a
C
h
a
r
i)
597
en
s
u
r
e
th
at
tem
p
o
r
a
r
y
d
is
tr
ac
tio
n
s
o
r
b
r
ief
b
lin
k
s
d
o
n
o
t
m
is
tak
en
ly
tr
ig
g
e
r
aler
ts
,
th
er
eb
y
im
p
r
o
v
i
n
g
th
e
s
y
s
tem
’
s
r
o
b
u
s
tn
ess
.
I
n
a
d
d
itio
n
,
a
co
u
n
ter
-
b
ased
ap
p
r
o
ac
h
is
im
p
lem
en
ted
to
tr
ac
k
th
e
n
u
m
b
er
o
f
c
o
n
s
ec
u
ti
v
e
f
r
a
m
es
wh
er
e
E
AR
,
MA
R
,
o
r
C
NN
p
r
o
b
ab
ilit
y
ex
ce
ed
th
ei
r
r
esp
ec
tiv
e
th
r
esh
o
ld
s
.
E
ac
h
co
u
n
ter
is
u
p
d
ated
p
er
f
r
am
e,
e
n
s
u
r
in
g
t
h
at
o
n
l
y
s
u
s
tain
ed
in
d
icato
r
s
o
f
d
r
o
wsi
n
e
s
s
lead
to
an
alar
m
.
T
h
is
p
r
e
v
en
ts
f
alse
alar
m
s
ca
u
s
ed
b
y
m
o
m
e
n
t
ar
y
c
h
an
g
e
s
in
f
ac
ial
ex
p
r
ess
io
n
,
lig
h
tin
g
co
n
d
itio
n
s
,
o
r
b
r
ie
f
h
ea
d
m
o
v
em
en
ts
.
I
f
an
y
o
f
th
ese
co
u
n
ter
s
s
u
r
p
ass
th
eir
d
ef
in
ed
lim
it,
th
e
s
y
s
tem
g
en
er
ates
an
aler
t
an
d
lo
g
s
th
e
ev
en
t
f
o
r
f
u
r
th
er
an
aly
s
is
.
T
ab
le
3
.
Mu
lti
-
s
tag
e
d
ec
is
io
n
f
u
s
io
n
th
r
esh
o
l
d
s
M
e
t
r
i
c
P
u
r
p
o
se
Th
r
e
s
h
o
l
d
C
o
n
se
c
u
t
i
v
e
f
r
a
mes
Ev
a
l
u
a
t
i
o
n
c
r
i
t
e
r
i
a
EA
R
D
e
t
e
c
t
s
p
r
o
l
o
n
g
e
d
e
y
e
c
l
o
su
r
e
0
.
2
6
15
Tr
i
g
g
e
r
a
l
e
r
t
i
f
EA
R
<
0
.
2
6
f
o
r
1
5
c
o
n
se
c
u
t
i
v
e
f
r
a
mes
M
A
R
I
d
e
n
t
i
f
i
e
s
y
a
w
n
i
n
g
f
r
e
q
u
e
n
c
y
0
.
0
5
20
R
e
c
o
r
d
y
a
w
n
i
f
M
A
R
>
0
.
0
5
f
o
r
2
0
c
o
n
se
c
u
t
i
v
e
f
r
a
mes
C
N
N
p
r
o
b
a
b
i
l
i
t
y
M
e
a
su
r
e
s
o
v
e
r
a
l
l
l
i
k
e
l
i
h
o
o
d
o
f
d
r
o
w
si
n
e
ss
0
.
3
5
15
Tr
i
g
g
e
r
a
l
e
r
t
i
f
C
N
N
p
r
o
b
a
b
i
l
i
t
y
<
0
.
3
f
o
r
1
5
c
o
n
se
c
u
t
i
v
e
f
r
a
mes
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
d
esig
n
an
d
d
ev
elo
p
m
en
t o
f
th
e
s
y
s
tem
was
ca
r
r
ie
d
o
u
t u
s
in
g
th
e
Py
th
o
n
p
r
o
g
r
am
m
in
g
lan
g
u
ag
e
,
an
d
th
e
e
x
ec
u
tio
n
o
f
th
e
m
o
d
e
l w
as d
o
n
e
o
n
th
e
An
ac
o
n
d
a
d
is
tr
ib
u
tio
n
o
n
a
W
in
d
o
ws 1
0
m
ac
h
in
e.
-
T
h
r
esh
o
ld
s
elec
tio
n
an
d
ju
s
tific
atio
n
:
bot
h
s
tatis
tical
an
aly
s
is
an
d
em
p
ir
ical
ex
p
er
im
en
tati
o
n
wer
e
u
s
ed
t
o
d
eter
m
in
e
th
e
cr
iter
ia
f
o
r
C
N
N
p
r
o
b
ab
ilit
y
,
E
AR
,
an
d
MA
R
.
Sev
er
al
th
r
esh
o
ld
v
alu
es
w
er
e
ex
am
in
ed
in
th
e
ea
r
ly
p
h
ases
o
f
m
o
d
el
b
u
il
d
in
g
to
ass
ess
th
eir
ef
f
ec
ts
o
n
f
alse
p
o
s
itiv
es
an
d
f
alse
n
eg
a
tiv
es.
B
ased
o
n
an
o
p
tim
izatio
n
p
r
o
ce
d
u
r
e
m
e
an
t
to
r
ed
u
ce
d
etec
tio
n
m
is
tak
es,
th
ese
th
r
esh
o
ld
s
wer
e
c
h
o
s
en
.
Fo
r
ea
ch
o
f
th
e
in
d
icato
r
s
(
E
AR
,
MA
R
,
an
d
C
NN
p
r
o
b
ab
ilit
y
)
,
we
p
er
f
o
r
m
e
d
a
r
ec
eiv
er
o
p
e
r
atin
g
ch
ar
ac
te
r
is
tic
(
R
OC
)
cu
r
v
e
an
aly
s
is
in
o
r
d
e
r
to
s
tatis
tically
s
u
p
p
o
r
t
th
e
v
al
u
es.
T
h
r
o
u
g
h
t
h
ese
ev
alu
atio
n
s
,
we
wer
e
ab
le
to
ch
o
o
s
e
th
r
esh
o
ld
s
th
at
o
f
f
er
ed
th
e
b
est
p
o
s
s
ib
le
b
alan
ce
b
etwe
en
tr
u
e
p
o
s
itiv
e
r
ate
(
T
PR
)
an
d
f
alse
p
o
s
itiv
e
r
ate
(
FP
R
)
.
I
n
o
u
r
te
s
t
d
ataset,
th
e
b
est
cr
iter
ia
f
o
r
attain
in
g
lo
w
f
alse
-
p
o
s
itiv
e
r
ates
an
d
g
o
o
d
ac
cu
r
ac
y
wer
e
f
o
u
n
d
to
b
e
E
AR
=0
.
2
6
,
MA
R
=0
.
0
5
,
an
d
C
NN
p
r
o
b
a
b
ilit
y
=0
.
3
5
.
I
n
o
r
d
er
to
m
a
k
e
s
u
r
e
o
u
r
m
o
d
el'
s
p
er
f
o
r
m
an
ce
was
in
lin
e
with
th
e
m
o
s
t
ad
v
an
ce
d
tech
n
iq
u
es
in
th
is
f
ield
,
we
also
ex
am
in
e
d
em
p
ir
ical
r
esear
ch
th
at
h
ad
em
p
lo
y
ed
co
m
p
ar
ab
le
th
r
e
s
h
o
ld
s
f
o
r
f
ac
ial
lan
d
m
a
r
k
s
an
d
d
r
o
wsi
n
ess
d
etec
tio
n
an
d
m
o
d
if
ie
d
th
o
s
e
r
esu
lts
.
-
T
h
r
esh
o
ld
ev
alu
atio
n
:
a
cr
o
s
s
-
v
alid
atio
n
p
r
o
ce
d
u
r
e
was
u
s
ed
to
f
u
r
th
er
r
e
f
in
e
th
e
ass
es
s
m
en
t
o
f
th
ese
th
r
esh
o
ld
s
.
W
e
co
n
f
ir
m
e
d
th
at
th
e
ch
o
s
en
th
r
esh
o
ld
s
p
r
o
d
u
ce
d
t
h
e
b
est
p
o
s
s
ib
le
tr
a
d
e
-
o
f
f
b
etwe
en
s
en
s
itiv
ity
an
d
s
p
ec
if
icity
b
y
ex
am
in
in
g
p
r
ec
is
io
n
-
r
ec
all
c
u
r
v
es.
T
h
e
v
alid
atio
n
p
h
ase'
s
em
p
ir
ical
d
at
a
v
er
if
ied
th
at
th
e
th
r
esh
o
ld
s
s
elec
ted
wer
e
ap
p
r
o
p
r
iate
f
o
r
r
e
al
-
wo
r
ld
s
itu
atio
n
s
wh
er
e
p
r
o
m
p
t
an
d
p
r
ec
is
e
id
en
tific
atio
n
o
f
d
r
iv
e
r
f
at
ig
u
e
is
ess
en
tial.
T
h
e
p
er
f
o
r
m
an
ce
e
v
alu
atio
n
o
f
th
e
s
u
g
g
ested
d
r
o
wsi
n
ess
p
r
ed
ictio
n
m
o
d
el
is
s
h
o
wn
in
Fig
u
r
e
2
.
On
e
o
f
th
e
m
ain
j
u
s
tific
atio
n
s
f
o
r
ch
o
o
s
in
g
E
f
f
icien
tNetV2
B
0
is
its
co
m
p
o
u
n
d
s
ca
lin
g
ap
p
r
o
ac
h
,
wh
ich
ju
d
icio
u
s
ly
m
a
x
im
izes
th
e
m
o
d
el'
s
d
ep
th
,
w
id
th
,
an
d
r
eso
lu
t
io
n
.
T
h
is
allo
ws
f
o
r
h
ig
h
ac
cu
r
ac
y
s
ca
lin
g
o
f
th
e
m
o
d
el
with
o
u
t a
s
ig
n
if
ican
t in
cr
ea
s
e
in
co
m
p
u
tatio
n
al
co
m
p
l
ex
ity
.
Sin
ce
d
r
iv
er
d
r
o
wsi
n
ess
d
etec
tio
n
is
a
r
ea
l
-
tim
e
p
r
o
ce
s
s
wh
er
e
ac
cu
r
ac
y
a
n
d
s
p
ee
d
ar
e
cr
u
cial,
E
f
f
icien
t
NetV2
B
0
o
f
f
er
s
a
g
r
ea
t b
alan
c
e
b
etwe
en
th
e
two
.
T
h
e
co
m
p
a
r
is
o
n
with
o
th
e
r
m
o
d
els ar
e
as f
o
llo
ws:
-
R
esNet:
b
ec
au
s
e
o
f
its
d
ee
p
ar
ch
itectu
r
e
an
d
r
esid
u
al
c
o
n
n
ec
tio
n
s
,
R
esNet
ten
d
s
to
d
em
an
d
g
r
ea
ter
p
r
o
ce
s
s
in
g
r
eso
u
r
ce
s
ev
e
n
if
it
is
p
o
p
u
lar
an
d
h
as
s
h
o
wn
s
u
c
ce
s
s
in
DL
ap
p
licatio
n
s
.
R
esN
et
m
ig
h
t
n
o
t
b
e
as
ef
f
ec
tiv
e
as
E
f
f
icien
tNetV2
B
0
in
r
ea
l
-
tim
e
ap
p
licatio
n
s
s
u
ch
as
s
leep
d
etec
tio
n
in
ter
m
s
o
f
co
m
p
u
te
co
s
t a
n
d
in
f
er
e
n
ce
tim
e,
wh
ich
is
cr
u
cial
f
o
r
g
u
ar
an
teein
g
lo
w
-
laten
cy
an
s
wer
s
.
-
Mo
b
ileNet:
co
m
p
ar
ed
to
m
o
r
e
s
o
p
h
is
tica
ted
m
o
d
els,
Mo
b
ileNet
is
f
a
s
ter
an
d
b
etter
s
u
ited
f
o
r
d
ev
ices w
ith
less
r
eso
u
r
ce
s
,
b
u
t
it
also
l
o
s
es
s
o
m
e
ac
cu
r
ac
y
.
Fo
r
th
i
s
wo
r
k
,
E
f
f
icien
tNetV2
B
0
w
as
ch
o
s
en
o
v
e
r
Mo
b
ileNet
d
u
e
to
th
e
r
eq
u
ir
e
m
en
t
f
o
r
h
ig
h
ac
cu
r
ac
y
in
id
e
n
tify
in
g
s
m
all
in
d
icato
r
s
o
f
d
r
o
wsi
n
ess
,
s
u
ch
as
s
lig
h
t e
y
e
clo
s
es o
r
ch
an
g
es in
h
ea
d
p
o
s
itio
n
.
-
Vis
io
n
t
r
an
s
f
o
r
m
er
s
(
ViT
)
:
p
ar
ticu
lar
ly
wh
en
wo
r
k
in
g
with
b
ig
d
atasets
,
v
is
io
n
tr
an
s
f
o
r
m
er
s
h
av
e
d
em
o
n
s
tr
ated
ex
c
ep
tio
n
al
p
er
f
o
r
m
an
ce
i
n
p
ictu
r
e
class
if
ica
tio
n
task
s
.
T
o
f
u
n
ctio
n
at
th
ei
r
b
est,
t
h
ey
ar
e
s
aid
to
n
ee
d
a
lo
t
o
f
d
ata
a
n
d
a
lo
t
o
f
p
r
o
ce
s
s
in
g
p
o
wer
.
E
f
f
icien
tNetV2
B
0
o
f
f
er
s
a
b
e
tter
o
p
tio
n
th
an
v
is
io
n
tr
an
s
f
o
r
m
e
r
s
b
ec
au
s
e
t
h
e
s
tu
d
y
'
s
d
ataset
is
r
elativ
ely
s
m
all
an
d
its
o
b
jectiv
e
is
to
o
b
tain
ef
f
ec
tiv
e
,
r
ea
l
-
tim
e
p
er
f
o
r
m
an
ce
.
T
h
e
co
n
f
u
s
io
n
m
at
r
ix
,
d
is
p
lay
ed
in
Fig
u
r
e
2
(
a
)
,
d
em
o
n
s
tr
ate
s
th
e
m
o
d
el'
s
ac
cu
r
ac
y
in
i
d
en
tify
in
g
th
e
d
r
iv
er
'
s
s
tatu
s
as
eith
er
ac
tiv
e
o
r
s
leep
y
.
T
h
e
m
o
d
el
h
as
a
h
i
g
h
tr
u
e
p
o
s
itiv
e
an
d
tr
u
e
n
eg
a
tiv
e
r
ate,
with
f
ew
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
39
,
No
.
1
,
J
u
ly
20
25
:
592
-
6
0
2
598
f
alse
p
o
s
itiv
es
an
d
f
alse
n
e
g
ativ
es,
ac
co
r
d
in
g
to
th
e
c
o
n
f
u
s
io
n
m
atr
ix
.
T
h
e
ac
cu
r
ac
y
o
f
th
e
s
u
g
g
ested
im
p
r
o
v
e
d
E
f
f
icien
tNet
is
co
n
tr
asted
with
th
at
o
f
o
th
er
p
o
p
u
lar
C
NN
ar
ch
itectu
r
es,
s
u
c
h
as
Den
s
eNe
t1
2
1
,
VGG
-
1
6
,
an
d
E
f
f
icien
tNet,
i
n
Fig
u
r
e
2
(
b
)
.
As
d
em
o
n
s
tr
at
ed
,
th
e
s
u
g
g
ested
m
o
d
el
o
u
tp
er
f
o
r
m
s
co
m
p
etin
g
ar
ch
i
tectu
r
es
b
y
a
s
ig
n
i
f
ican
t
m
ar
g
in
a
n
d
attain
s
th
e
m
a
x
im
u
m
ac
cu
r
ac
y
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
r
etain
s
a
h
ig
h
F1
-
s
co
r
e,
g
u
ar
an
teein
g
th
at
b
o
th
p
r
ec
is
io
n
an
d
r
ec
all
ar
e
o
p
tim
ized
an
d
v
er
if
y
in
g
its
r
o
b
u
s
tn
ess
in
d
etec
tin
g
d
r
o
wsi
n
ess
,
as
s
h
o
wn
in
th
e
co
m
p
ar
ativ
e
s
tu
d
y
o
f
th
e
F1
-
s
co
r
e
in
Fig
u
r
e
2
(
c
)
.
B
ec
au
s
e
o
f
its
ef
f
ec
tiv
en
ess
an
d
p
e
r
f
o
r
m
an
ce
b
ala
n
ce
,
E
f
f
icien
tNetV2
B
0
was
s
elec
ted
as
th
e
f
o
u
n
d
atio
n
al
m
o
d
el
f
o
r
d
r
iv
er
d
r
o
wsi
n
ess
d
etec
tio
n
in
th
is
in
v
esti
g
atio
n
.
E
f
f
icien
tNetV2
B
0
is
a
lig
h
tweig
h
t
C
NN
ar
ch
itectu
r
e
th
at
u
s
es
in
v
er
ted
r
esid
u
al
b
lo
c
k
s
an
d
m
o
b
ile
b
o
ttlen
ec
k
co
n
v
o
l
u
tio
n
s
(
MBC
o
n
v
)
to
ac
h
iev
e
g
o
o
d
p
er
f
o
r
m
an
ce
at
lo
w
co
m
p
u
tatio
n
al
co
s
t.
Fo
r
r
ea
l
-
t
im
e
ap
p
licatio
n
s
,
th
is
p
ar
ad
ig
m
is
p
er
f
ec
t,
esp
ec
ially
in
r
e
s
o
u
r
ce
-
co
n
s
tr
ain
e
d
s
ettin
g
s
lik
e
em
b
ed
d
e
d
s
y
s
tem
s
o
r
m
o
b
ile
d
ev
ices,
w
h
ich
ar
e
co
m
m
o
n
ap
p
licatio
n
s
f
o
r
d
r
iv
e
r
m
o
n
ito
r
in
g
s
y
s
tem
s
.
E
f
f
icien
tNetV2
B
0
is
a
g
o
o
d
o
p
tio
n
f
o
r
th
is
ap
p
licat
io
n
s
in
ce
it
p
r
o
v
id
es
th
e
b
est
p
o
s
s
ib
le
b
alan
ce
o
f
s
p
ee
d
,
ac
cu
r
ac
y
,
a
n
d
ef
f
icien
c
y
f
o
r
d
etec
tin
g
d
r
iv
er
s
leep
in
e
s
s
.
T
h
e
an
aly
s
is
in
F
i
g
u
r
e
2
d
e
m
o
n
s
tr
ates
th
at
th
e
p
r
o
p
o
s
ed
DL
m
o
d
el
o
u
t
p
er
f
o
r
m
s
o
th
er
DL
m
o
d
els,
ac
h
iev
in
g
9
8
%
ac
cu
r
ac
y
a
n
d
F1
-
s
co
r
e.
T
h
e
c
o
n
f
u
s
io
n
m
atr
ix
co
n
f
ir
m
s
t
h
at
th
e
m
o
d
el
ef
f
ec
tiv
el
y
d
is
tin
g
u
is
h
es
b
etwe
en
d
r
o
wsy
an
d
ac
tiv
e
s
tates,
with
m
in
im
al
m
is
class
if
icatio
n
s
.
W
h
en
co
m
p
ar
in
g
with
Den
s
eNe
t1
2
1
,
VGG
-
1
6
,
an
d
E
f
f
icien
tN
et,
th
e
f
in
e
-
tu
n
ed
E
f
f
icien
tNet
ex
h
ib
its
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
with
r
ed
u
ce
d
f
als
e
p
o
s
itiv
es,
en
s
u
r
in
g
r
o
b
u
s
t d
r
o
wsi
n
ess
d
etec
tio
n
ac
r
o
s
s
d
iv
er
s
e
f
ac
ial
v
a
r
iatio
n
s
an
d
en
v
ir
o
n
m
en
tal
co
n
d
itio
n
s
.
Fig
u
r
e
3
d
em
o
n
s
tr
ates
p
er
f
o
r
m
an
ce
an
al
y
s
is
o
f
th
e
p
r
o
p
o
s
ed
SVM
b
ased
f
ac
ia
l la
n
d
m
ar
k
d
etec
tio
n
.
(
a)
(
b
)
(
c)
Fig
u
r
e
2
.
Per
f
o
r
m
an
c
e
ev
alu
at
io
n
o
f
t
h
e
p
r
o
p
o
s
ed
d
r
o
wsi
n
ess
p
r
ed
ictio
n
m
o
d
el,
in
clu
d
in
g
:
(
a)
co
n
f
u
s
io
n
m
atr
ix
,
(
b
)
co
m
p
ar
ativ
e
an
al
y
s
is
o
f
ac
cu
r
ac
y
,
an
d
(
c)
co
m
p
ar
ativ
e
an
aly
s
is
o
f
F1
-
s
co
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
R
ea
l
-
time
d
r
iver d
r
o
w
s
in
es
s
d
etec
tio
n
b
a
s
ed
o
n
in
teg
r
a
tive
…
(
Go
w
r
is
h
a
n
ka
r
S
h
iva
S
h
a
n
ka
r
a
C
h
a
r
i)
599
T
h
e
ef
f
ec
tiv
en
ess
o
f
th
e
SVM
-
b
ased
f
ac
ial
lan
d
m
a
r
k
p
r
ed
ic
tio
n
m
o
d
el
is
ass
ess
ed
in
Fig
u
r
e
3
.
T
h
e
m
o
d
el'
s
ab
ilit
y
to
c
o
r
r
ec
tly
an
ticip
ate
im
p
o
r
tan
t
f
ac
ial
f
ea
tu
r
es
in
clu
d
in
g
t
h
e
ey
es,
n
o
s
e,
a
n
d
m
o
u
th
is
s
ee
n
i
n
Fig
u
r
e
3
(
a)
,
w
h
er
e
t
h
e
p
r
ed
ict
ed
lan
d
m
ar
k
s
ar
e
d
is
p
lay
e
d
o
n
a
s
am
p
le
im
ag
e.
T
h
e
m
o
d
el
'
s
ab
ilit
y
to
h
an
d
le
v
ar
iatio
n
s
in
f
ac
e
e
x
p
r
ess
io
n
s
is
d
em
o
n
s
tr
ated
b
y
th
e
g
o
o
d
a
lig
n
m
en
t
b
etwe
en
t
h
e
p
r
e
d
icted
an
d
ac
t
u
al
f
ac
ial
lan
d
m
ar
k
s
.
T
h
e
MSE
an
aly
s
is
f
o
r
th
e
test
an
d
v
alid
atio
n
d
a
tasets
is
s
h
o
wn
in
Fig
u
r
e
3
(
b
)
.
Hig
h
ac
cu
r
ac
y
i
n
f
ac
e
lan
d
m
ar
k
p
r
ed
ictio
n
is
in
d
icate
d
b
y
th
e
lo
w
MSE
v
alu
es
f
o
r
b
o
th
d
atasets
,
wh
ich
ar
e
0
.
0
0
3
9
8
f
o
r
v
alid
atio
n
an
d
0
.
0
0
3
9
7
f
o
r
t
esti
n
g
.
L
ast
b
u
t
n
o
t
leas
t,
Fi
g
u
r
e
3
(
c
)
co
n
tr
asts
th
e
p
r
o
ce
s
s
in
g
tim
es
o
f
th
e
s
u
g
g
ested
SVM
m
o
d
el
a
n
d
Dlib
.
I
t
d
e
m
o
n
s
tr
ates
th
at,
d
esp
ite
a
m
in
o
r
co
m
p
u
tatio
n
al
tim
e
in
cr
ea
s
e
(
0
.
3
2
0
s
v
s
.
0
.
2
8
0
s
)
f
o
r
th
e
SVM
-
b
as
ed
m
o
d
el,
th
is
s
m
all
d
elay
is
co
m
p
en
s
ated
f
o
r
b
y
th
e
en
h
an
ce
d
ac
cu
r
ac
y
a
nd
r
o
b
u
s
tn
ess
u
n
d
e
r
v
ar
io
u
s
f
ac
ia
l c
o
n
d
itio
n
s
.
T
h
e
n
ex
t
Fig
u
r
e
4
s
h
o
ws
an
aly
s
is
with
r
esp
ec
t
to
r
ea
l
-
ti
m
e
test
in
g
o
f
th
e
p
r
o
p
o
s
ed
d
r
o
wsi
n
ess
d
etec
tio
n
m
o
d
el
o
n
a
s
u
b
ject
u
n
d
er
d
if
f
er
e
n
t
co
n
d
itio
n
s
.
T
h
e
f
ir
s
t
im
ag
e
in
d
icate
s
an
al
er
t
s
tate,
wh
er
e
th
e
m
o
d
el
co
r
r
e
ctly
id
en
tifie
s
n
o
s
ig
n
s
o
f
d
r
o
wsi
n
ess
.
T
h
e
s
ec
o
n
d
im
ag
e
s
ig
n
alin
g
a
y
awn
in
g
ev
en
t
an
d
s
y
s
tem
tr
ig
g
er
s
an
aler
t w
ith
o
u
t w
ak
e
u
p
.
On
th
e
o
th
er
h
an
d
,
th
e
th
i
r
d
im
ag
e
class
if
y
in
g
th
e
s
u
b
jec
t a
s
awa
k
e.
W
e
ex
p
an
d
th
e
p
er
f
o
r
m
an
ce
a
n
aly
s
is
to
in
clu
d
e
s
en
s
iti
v
ity
,
s
p
ec
if
icity
,
an
d
AUC
-
R
OC
in
a
d
d
itio
n
to
p
r
ec
is
io
n
an
d
r
ec
all
in
o
r
d
e
r
to
g
iv
e
a
th
o
r
o
u
g
h
ass
ess
m
en
t
o
f
th
e
s
u
g
g
ested
ap
p
r
o
ac
h
.
T
h
ese
m
ea
s
u
r
es
p
r
o
v
id
e
a
m
o
r
e
im
p
ar
tial
ev
alu
atio
n
o
f
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
,
esp
ec
ially
wh
en
it
co
m
es
to
b
in
ar
y
class
if
ic
atio
n
task
s
lik
e
d
etec
tin
g
d
r
iv
er
d
r
o
wsi
n
ess
.
T
h
e
m
ajo
r
p
er
f
o
r
m
an
ce
in
d
icato
r
s
f
o
r
th
e
s
u
g
g
ested
m
o
d
el
ar
e
co
m
p
iled
in
th
e
a
cc
o
m
p
an
y
in
g
ta
b
le
(
T
a
b
le
4
)
,
wh
ich
also
c
o
n
tr
asts
th
em
with
th
o
s
e
o
f
o
th
er
cu
ttin
g
-
ed
g
e
DL
m
o
d
els cu
r
r
e
n
tly
in
u
s
e.
T
h
e
s
u
g
g
ested
m
o
d
el
s
u
cc
ess
f
u
lly
b
alan
ce
s
p
r
ec
is
io
n
(
0
.
9
8
)
,
r
ec
all
(
0
.
9
8
)
,
s
en
s
itiv
ity
(
0
.
9
8
)
,
an
d
s
p
ec
if
icity
(
0
.
9
8
)
,
as
d
e
m
o
n
s
tr
ated
in
T
ab
le
4
,
d
em
o
n
s
tr
atin
g
its
r
esil
ien
ce
in
ac
cu
r
ately
id
e
n
tify
in
g
b
o
t
h
aler
t
an
d
d
r
o
wsy
s
tates.
T
h
e
m
o
d
el's
o
u
ts
tan
d
in
g
ab
ilit
y
to
d
i
s
cr
im
in
ate
b
etwe
en
t
h
e
two
class
es
is
f
u
r
th
er
ev
id
en
ce
d
b
y
its
AUC
-
R
OC
s
co
r
e
o
f
0
.
9
9
.
T
h
e
s
u
g
g
ested
m
o
d
el
p
er
f
o
r
m
s
b
etter
th
an
o
t
h
er
in
d
u
s
tr
y
-
lead
i
n
g
tech
n
iq
u
es
lik
e
Den
s
eNe
t1
2
1
,
VGG
-
1
6
,
a
n
d
ef
f
icien
tN
et
in
e
v
er
y
im
p
o
r
tan
t
p
er
f
o
r
m
an
ce
in
d
icato
r
,
d
em
o
n
s
tr
atin
g
its
ex
ce
p
tio
n
al
ef
f
icac
y
in
d
etec
tin
g
d
r
iv
er
d
r
o
wsi
n
ess
.
Key
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
as
f
o
llo
ws:
a
h
y
b
r
id
AI
-
d
r
iv
en
f
r
am
ewo
r
k
f
o
r
r
ea
l
-
tim
e
d
r
iv
er
d
r
o
wsi
n
ess
d
etec
tio
n
is
p
r
ese
n
ted
in
t
h
is
p
ap
er
,
with
a
n
a
s
to
n
is
h
in
g
9
8
%
ac
c
u
r
ac
y
r
ate
in
d
if
f
er
en
tiatin
g
b
etwe
en
ac
tiv
e
an
d
d
r
o
wsy
s
tates.
A
s
tr
o
n
g
an
d
d
ep
e
n
d
ab
le
s
o
lu
tio
n
is
p
r
o
v
id
ed
b
y
th
e
m
u
lti
-
s
tag
e
d
ec
is
io
n
f
u
s
io
n
tech
n
iq
u
e,
wh
ich
co
m
b
in
es
f
ac
ial
lan
d
m
ar
k
d
etec
ti
o
n
with
DL
-
b
ased
f
ac
ial
an
al
y
s
is
.
T
h
e
ap
p
r
o
ac
h
lo
wer
s
f
alse
alar
m
s
an
d
g
u
ar
a
n
tees
th
at
o
n
l
y
p
e
r
s
is
ten
t
s
leep
in
ess
in
d
icato
r
s
ca
u
s
e
aler
ts
b
y
co
m
b
in
in
g
C
NN
p
r
o
b
a
b
ilit
y
r
atin
g
s
,
E
AR
,
a
n
d
m
o
u
th
asp
ec
t
r
atio
(
MA
R
)
.
T
h
e
s
u
g
g
ested
a
p
p
r
o
ac
h
is
n
o
w
at
th
e
f
o
r
ef
r
o
n
t
o
f
d
r
iv
er
s
af
ety
tec
h
n
o
lo
g
y
th
a
n
k
s
to
th
ese
f
in
d
in
g
s
.
W
h
en
co
m
p
ar
e
d
to
ea
r
lier
r
es
ea
r
ch
,
o
u
r
m
eth
o
d
o
v
er
co
m
es
th
e
s
h
o
r
tco
m
in
g
s
o
f
ea
r
lier
t
ec
h
n
iq
u
es
b
y
r
ec
o
g
n
izin
g
e
n
v
ir
o
n
m
en
ta
l
ch
an
g
es
an
d
f
ac
ial
o
cc
lu
s
io
n
s
,
wh
ich
f
r
eq
u
en
tly
m
a
k
e
it
m
o
r
e
d
if
f
icu
lt
to
d
etec
t
d
r
o
wsi
n
ess
in
r
ea
l
-
wo
r
ld
s
itu
atio
n
s
.
Ou
r
m
o
d
el
u
s
es
a
m
o
r
e
d
y
n
a
m
ic
f
u
s
io
n
ap
p
r
o
ac
h
,
in
c
lu
d
in
g
b
o
th
f
ac
ial
g
eo
m
etr
y
i
n
f
o
r
m
atio
n
an
d
DL
p
r
e
d
ictio
n
s
,
in
co
n
tr
ast
to
co
n
v
en
tio
n
al
s
y
s
tem
s
th
at
ju
s
t
u
s
e
b
asic
th
r
esh
o
ld
s
o
r
ce
r
tain
f
ac
ial
tr
aits
.
Un
d
er
a
v
ar
iety
o
f
d
r
iv
i
n
g
cir
cu
m
s
tan
ce
s
,
th
is
h
y
b
r
id
m
eth
o
d
im
p
r
o
v
es
ac
cu
r
ac
y
an
d
r
esil
ien
ce
.
I
n
s
u
m
m
a
r
y
,
th
is
s
tu
d
y
s
h
o
ws
h
o
w
AI
-
p
o
wer
e
d
s
y
s
tem
s
m
ig
h
t
e
n
h
an
ce
d
r
iv
er
s
af
ety
b
y
p
r
ec
is
ely
id
e
n
tify
in
g
tire
d
n
ess
in
r
ea
l
tim
e.
As
tech
n
o
lo
g
y
d
ev
elo
p
s
,
it
m
ay
b
e
a
b
le
to
s
ig
n
if
ican
tly
r
ed
u
ce
d
r
iv
er
f
atig
u
e
-
r
elate
d
co
llis
io
n
s
,
u
ltima
tely
m
ak
in
g
r
o
ad
way
s
s
af
er
ev
er
y
wh
e
r
e.
(
a)
(
b
)
(
c)
Fig
u
r
e
3
.
Per
f
o
r
m
an
c
e
ev
alu
at
io
n
o
f
t
h
e
p
r
o
p
o
s
ed
SVM
b
ase
d
l
an
d
m
ar
k
p
r
ed
ictio
n
,
i
n
clu
d
i
n
g
:
(
a)
p
r
ed
icted
f
ac
ial
lan
d
m
ar
k
,
(
b
)
MSE
an
al
y
s
is
,
an
d
(
c)
p
r
o
ce
s
s
in
g
tim
e
a
n
aly
s
is
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
39
,
No
.
1
,
J
u
ly
20
25
:
592
-
6
0
2
600
Fig
u
r
e
4
.
Dem
o
n
s
tr
ates r
ea
l
-
tim
e
test
in
g
o
f
th
e
p
r
o
p
o
s
ed
d
r
o
wsi
n
ess
d
etec
tio
n
m
o
d
el
T
ab
le
4
.
C
o
m
p
a
r
is
o
n
o
f
p
er
f
o
r
m
an
ce
m
etr
ics f
o
r
d
if
f
er
e
n
t m
o
d
els
M
o
d
e
l
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
S
e
n
s
i
t
i
v
i
t
y
S
p
e
c
i
f
i
c
i
t
y
F1
-
S
c
o
r
e
AUC
-
R
O
C
P
r
o
p
o
se
d
m
o
d
e
l
0
.
9
8
0
.
9
8
0
.
9
8
0
.
9
8
0
.
9
8
0
.
9
9
D
e
n
seN
e
t
1
2
1
0
.
9
4
0
.
9
6
0
.
9
5
0
.
9
4
0
.
9
5
0
.
9
7
VGG
-
16
0
.
9
2
0
.
9
4
0
.
9
3
0
.
9
2
0
.
9
3
0
.
9
6
Ef
f
i
c
i
e
n
t
N
e
t
0
.
9
5
0
.
9
7
0
.
9
6
0
.
9
5
0
.
9
6
0
.
9
8
4.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
s
u
g
g
ested
a
h
y
b
r
i
d
AI
-
d
r
iv
e
n
ar
ch
itectu
r
e
th
at
co
m
b
in
es
a
m
u
lti
-
s
tag
e
d
ec
is
io
n
f
u
s
io
n
m
ec
h
an
is
m
with
DL
-
b
ased
p
r
ed
ictio
n
s
f
o
r
r
ea
l
-
tim
e
d
r
iv
er
s
leep
in
ess
d
etec
tio
n
.
T
h
r
o
u
g
h
th
e
in
teg
r
atio
n
o
f
f
ac
ial
lan
d
m
ar
k
m
o
n
ito
r
in
g
with
a
ML
-
b
ased
f
ac
ial
lan
d
m
ar
k
p
r
ed
ictio
n
m
o
d
el,
th
e
s
u
g
g
ested
s
y
s
tem
ac
h
iev
ed
a
h
ig
h
d
etec
tio
n
ac
c
u
r
ac
y
o
f
9
8
%,
p
r
o
v
id
in
g
d
e
p
e
n
d
ab
le
ca
te
g
o
r
izatio
n
b
etwe
en
d
r
o
wsy
an
d
ac
tiv
e
s
tates.
Ad
d
itio
n
ally
,
th
e
co
u
n
ter
-
b
ased
m
eth
o
d
o
f
tr
ac
k
in
g
s
u
cc
ess
iv
e
f
r
am
es
g
u
ar
an
tees
t
h
at
an
aler
t
is
o
n
l
y
tr
ig
g
er
ed
b
y
p
er
s
is
ten
t
s
ig
n
s
o
f
d
r
o
wsi
n
ess
,
m
in
im
izin
g
f
als
e
alar
m
s
b
r
o
u
g
h
t
o
n
b
y
f
leetin
g
h
ea
d
m
o
v
em
e
n
ts
o
r
ch
an
g
es
in
f
ac
ial
ex
p
r
ess
io
n
.
T
h
e
d
ev
ice
n
o
tifie
s
th
e
d
r
i
v
er
in
r
ea
l
tim
e
wh
en
an
aler
t
is
s
et
o
f
f
,
an
d
all
in
s
tan
ce
s
o
f
d
r
o
wsi
n
ess
ar
e
r
e
co
r
d
ed
f
o
r
s
tu
d
y
af
ter
th
e
d
r
i
v
e.
Fu
tu
r
e
r
esear
ch
m
u
s
t
ad
d
r
ess
a
f
ew
s
h
o
r
tco
m
i
n
g
s
,
ev
en
if
th
e
s
u
g
g
ested
ap
p
r
o
ac
h
h
as
g
r
ea
t
ac
cu
r
ac
y
an
d
r
esil
ien
ce
in
co
n
tr
o
lled
s
e
ttin
g
s
.
First,
h
ar
s
h
en
v
ir
o
n
m
e
n
tal
co
n
d
itio
n
s
lik
e
in
ten
s
e
g
lar
e,
p
o
o
r
lig
h
tin
g
,
o
r
s
ev
er
e
wea
t
h
er
m
ig
h
t
alter
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
b
y
a
f
f
ec
tin
g
th
e
ac
cu
r
ac
y
o
f
lan
d
m
a
r
k
p
r
ed
ictio
n
a
n
d
th
e
v
is
ib
ilit
y
o
f
f
ac
ial
f
ea
tu
r
es.
Fu
r
th
er
m
o
r
e,
th
e
s
y
s
tem
is
p
r
i
m
ar
ily
d
ep
e
n
d
en
t
o
n
f
ac
ial
la
n
d
m
ar
k
d
etec
tio
n
,
wh
ich
m
ay
h
a
v
e
t
r
o
u
b
le
d
etec
tin
g
o
cc
l
u
s
io
n
s
(
s
u
ch
a
s
lo
n
g
h
air
,
s
u
n
g
lass
es,
o
r
f
ac
e
m
ask
s
)
.
T
o
im
p
r
o
v
e
th
e
m
o
d
el'
s
r
esil
ien
ce
in
th
ese
d
if
f
icu
lt
cir
cu
m
s
tan
ce
s
,
m
o
r
e
i
n
v
esti
g
atio
n
is
r
eq
u
ir
ed
.
Fu
tu
r
e
d
ir
ec
tio
n
s
f
o
r
th
e
s
y
s
tem
co
u
ld
in
clu
d
e
ad
d
in
g
m
o
r
e
s
en
s
o
r
s
,
i
n
clu
d
in
g
ey
e
-
tr
ac
k
in
g
d
ev
ices
o
r
in
-
ca
r
ca
m
er
a
s
,
to
b
etter
id
en
tify
in
d
icato
r
s
o
f
tire
d
n
ess
wh
en
f
ac
e
o
cc
l
u
s
io
n
s
ar
e
p
r
es
en
t.
I
n
o
r
d
er
t
o
f
ac
ilit
ate
r
ea
l
-
tim
e
d
ep
lo
y
m
e
n
t in
au
to
m
o
tiv
e
s
y
s
tem
s
,
th
e
m
o
d
e
l
ca
n
also
b
e
tu
n
ed
f
o
r
q
u
ick
e
r
p
r
o
ce
s
s
in
g
.
T
h
e
cr
ea
tio
n
o
f
ad
ap
tiv
e
r
u
le
-
b
ased
s
y
s
tem
s
th
at
d
y
n
am
ica
lly
m
o
d
if
y
d
etec
tio
n
th
r
esh
o
ld
s
in
r
e
s
p
o
n
s
e
to
cu
r
r
en
t
en
v
ir
o
n
m
en
tal
co
n
d
itio
n
s
,
s
u
ch
lig
h
tin
g
o
r
t
h
e
d
r
iv
er
'
s
p
o
s
itio
n
,
m
a
y
b
e
th
e
s
u
b
ject
o
f
f
u
tu
r
e
s
tu
d
y
.
Fu
r
th
e
r
m
o
r
e,
b
r
o
ad
e
n
in
g
t
h
e
d
ataset
to
en
co
m
p
ass
m
o
r
e
v
ar
ied
an
d
a
u
th
en
tic
d
r
iv
i
n
g
s
ce
n
ar
io
s
,
in
c
lu
d
in
g
n
i
g
h
ttime
o
r
e
x
ten
d
e
d
r
id
es,
m
ay
im
p
r
o
v
e
th
e
m
o
d
el'
s
ca
p
ac
ity
f
o
r
g
en
er
aliza
tio
n
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
e
au
th
o
r
s
s
tate
th
at
n
o
f
u
n
d
i
n
g
was in
v
o
lv
e
d
in
th
is
r
esear
ch
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Go
wr
is
h
an
k
ar
Sh
iv
a
Sh
an
k
ar
a
C
h
ar
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
J
y
o
th
i A
r
co
t Pr
ash
an
t
✓
✓
✓
✓
✓
✓
✓
✓
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
R
ea
l
-
time
d
r
iver d
r
o
w
s
in
es
s
d
etec
tio
n
b
a
s
ed
o
n
in
teg
r
a
tive
…
(
Go
w
r
is
h
a
n
ka
r
S
h
iva
S
h
a
n
ka
r
a
C
h
a
r
i)
601
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
atasets
u
s
ed
an
d
an
aly
z
ed
d
u
r
i
n
g
th
is
s
tu
d
y
ar
e
av
ail
ab
le
f
r
o
m
t
h
e
co
r
r
esp
o
n
d
i
n
g
a
u
th
o
r
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
D
.
A
l
b
a
l
a
t
e
a
n
d
X
.
F
a
g
e
d
a
,
“
C
o
n
g
e
st
i
o
n
,
r
o
a
d
safe
t
y
,
a
n
d
t
h
e
e
f
f
e
c
t
i
v
e
n
e
ss
o
f
p
u
b
l
i
c
p
o
l
i
c
i
e
s
i
n
u
r
b
a
n
a
r
e
a
s,”
S
u
st
a
i
n
a
b
i
l
i
t
y
,
v
o
l
.
1
1
,
n
o
.
1
8
,
p
.
5
0
9
2
,
2
0
1
9
,
d
o
i
:
1
0
.
3
3
9
0
/
s
u
1
1
1
8
5
0
9
2
.
[
2
]
E.
B
a
s
i
l
i
e
t
a
l
.
,
“
T
h
e
i
mp
a
c
t
o
f
r
o
a
d
a
c
c
i
d
e
n
t
s o
n
h
o
s
p
i
t
a
l
a
d
mi
ss
i
o
n
s a
n
d
t
h
e
p
o
t
e
n
t
i
a
l
o
f
A
D
A
S
i
n
c
o
n
t
a
i
n
i
n
g
h
e
a
l
t
h
e
x
p
e
n
d
i
t
u
r
e
:
Ev
i
d
e
n
c
e
f
r
o
m Pi
e
d
mo
n
t
d
a
t
a
,
”
T
ra
n
s
p
.
Re
s.
I
n
t
e
r
d
i
s
c
i
p
.
Pe
rs
p
e
c
t
.
,
v
o
l
.
2
5
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
t
r
i
p
.
2
0
2
4
.
1
0
1
1
2
5
.
[
3
]
A
.
C
h
o
p
r
a
,
N
.
K
u
mar,
a
n
d
R
.
K
.
K
a
u
s
h
a
l
,
“
A
c
o
m
p
r
e
h
e
n
si
v
e
a
n
a
l
y
si
s
o
f
d
r
i
v
e
r
d
r
o
w
si
n
e
ss
d
e
t
e
c
t
i
o
n
t
e
c
h
n
i
q
u
e
s,
”
i
n
Ap
p
l
i
e
d
D
a
t
a
S
c
i
e
n
c
e
a
n
d
S
m
a
r
t
S
y
s
t
e
m
s
,
L
o
n
d
o
n
:
C
R
C
P
r
e
ss,
2
0
2
4
,
p
p
.
1
3
4
–
1
3
9
.
d
o
i
:
1
0
.
1
2
0
1
/
9
7
8
1
0
0
3
4
7
1
0
5
9
-
1
9
.
[
4
]
D
.
Lé
g
e
r
,
E
.
P
e
p
i
n
,
a
n
d
G
.
C
a
e
t
a
n
o
,
“
Th
e
e
c
o
n
o
mi
c
b
u
r
d
e
n
o
f
s
l
e
e
p
y
d
r
i
v
i
n
g
,
”
S
l
e
e
p
M
e
d
.
C
l
i
n
.
,
v
o
l
.
1
4
,
n
o
.
4
,
p
p
.
4
2
3
–
4
2
9
,
2
0
1
9
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
sm
c
.
2
0
1
9
.
0
7
.
0
0
4
.
[
5
]
M
.
K
a
m
b
o
j
e
t
.
a
l
.
,
“
A
d
v
a
n
c
e
d
d
e
t
e
c
t
i
o
n
t
e
c
h
n
i
q
u
e
s
f
o
r
d
r
i
v
e
r
d
r
o
w
si
n
e
s
s:
a
c
o
m
p
r
e
h
e
n
s
i
v
e
r
e
v
i
e
w
o
f
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
,
d
e
e
p
l
e
a
r
n
i
n
g
,
a
n
d
p
h
y
s
i
o
l
o
g
i
c
a
l
a
p
p
r
o
a
c
h
e
s,”
Mu
l
t
i
m
e
d
.
T
o
o
l
s
Ap
p
l
.
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
0
2
4
-
1
9
7
3
8
-
z.
[
6
]
G
.
S
i
k
a
n
d
e
r
a
n
d
S
.
A
n
w
a
r
,
“
D
r
i
v
e
r
f
a
t
i
g
u
e
d
e
t
e
c
t
i
o
n
s
y
s
t
e
ms
:
A
r
e
v
i
e
w
,
”
I
EEE
T
r
a
n
s.
I
n
t
e
l
l
.
T
r
a
n
sp
.
S
y
st
.
,
v
o
l
.
2
0
,
n
o
.
6
,
p
p
.
2
3
3
9
–
2
3
5
2
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
t
i
t
s.
2
0
1
8
.
2
8
6
8
4
9
9
.
[
7
]
A
.
A
.
S
a
l
e
e
m
e
t
.
a
l
.
,
“
A
sy
s
t
e
m
a
t
i
c
r
e
v
i
e
w
o
f
p
h
y
si
o
l
o
g
i
c
a
l
s
i
g
n
a
l
s
-
b
a
se
d
d
r
i
v
e
r
d
r
o
w
si
n
e
ss
d
e
t
e
c
t
i
o
n
s
y
st
e
ms,
”
C
o
g
n
.
N
e
u
r
o
d
y
n
.
,
v
o
l
.
1
7
,
p
p
.
1
2
2
9
–
1
2
5
9
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
5
7
1
-
0
2
2
-
0
9
8
9
8
-
9.
[
8
]
J.
V
i
c
e
n
t
e
,
P
.
La
g
u
n
a
,
A
.
B
a
r
t
r
a
,
a
n
d
R
.
B
a
i
l
ó
n
,
“
D
r
o
w
s
i
n
e
ss d
e
t
e
c
t
i
o
n
u
s
i
n
g
h
e
a
r
t
r
a
t
e
v
a
r
i
a
b
i
l
i
t
y
,
”
Me
d
.
B
i
o
l
.
E
n
g
.
C
o
m
p
u
t
.
,
v
o
l
.
5
4
,
n
o
.
6
,
p
p
.
9
2
7
–
9
3
7
,
2
0
1
6
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
5
1
7
-
0
1
5
-
1
4
4
8
-
7.
[
9
]
A
.
S
a
h
a
y
a
d
h
a
s,
K
.
S
u
n
d
a
r
a
j
,
a
n
d
M
.
M
u
r
u
g
a
p
p
a
n
,
“
D
r
o
w
si
n
e
ss
d
e
t
e
c
t
i
o
n
d
u
r
i
n
g
d
i
f
f
e
r
e
n
t
t
i
m
e
s
o
f
d
a
y
u
s
i
n
g
m
u
l
t
i
p
l
e
f
e
a
t
u
r
e
s,
”
Au
st
ra
l
a
s
.
P
h
y
s
.
E
n
g
.
S
c
i
.
Me
d
.
,
v
o
l
.
3
6
,
n
o
.
2
,
p
p
.
2
4
3
–
2
5
0
,
2
0
1
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
3
2
4
6
-
0
1
3
-
0
2
0
0
-
6.
[
1
0
]
M
.
S
h
a
h
b
a
k
h
t
i
e
t
a
l
.
,
“
S
i
m
u
l
t
a
n
e
o
u
s
e
y
e
b
l
i
n
k
c
h
a
r
a
c
t
e
r
i
z
a
t
i
o
n
a
n
d
e
l
i
mi
n
a
t
i
o
n
f
r
o
m
l
o
w
-
c
h
a
n
n
e
l
p
r
e
f
r
o
n
t
a
l
EEG
s
i
g
n
a
l
s
e
n
h
a
n
c
e
s
d
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
d
e
t
e
c
t
i
o
n
,
”
I
E
EE
J
.
Bi
o
m
e
d
.
H
e
a
l
t
h
I
n
f
o
rm
.
,
v
o
l
.
2
6
,
n
o
.
3
,
p
p
.
1
0
0
1
–
1
0
1
2
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
j
b
h
i
.
2
0
2
1
.
3
0
9
6
9
8
4
.
[
1
1
]
K
.
F
u
j
i
w
a
r
a
,
H
.
I
w
a
mo
t
o
,
K
.
H
o
r
i
,
a
n
d
M
.
K
a
n
o
,
“
D
r
i
v
e
r
d
r
o
w
si
n
e
s
s
d
e
t
e
c
t
i
o
n
u
s
i
n
g
R
-
R
i
n
t
e
r
v
a
l
o
f
e
l
e
c
t
r
o
c
a
r
d
i
o
g
r
a
m
a
n
d
se
l
f
-
a
t
t
e
n
t
i
o
n
a
u
t
o
e
n
c
o
d
e
r
,
”
I
E
EE
T
r
a
n
s.
I
n
t
e
l
l
.
V
e
h
.
,
v
o
l
.
9
,
n
o
.
1
,
p
p
.
2
9
5
6
–
2
9
6
5
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
t
i
v
.
2
0
2
3
.
3
3
0
8
5
7
5
.
[
1
2
]
J.
B
a
i
e
t
a
l
.
,
“
Tw
o
-
s
t
r
e
a
m
s
p
a
t
i
a
l
–
t
e
m
p
o
r
a
l
g
r
a
p
h
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
s
f
o
r
d
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
d
e
t
e
c
t
i
o
n
,
”
I
EEE
T
ra
n
s
.
C
y
b
e
rn
.
,
v
o
l
.
5
2
,
n
o
.
1
2
,
p
p
.
1
3
8
2
1
–
1
3
8
3
3
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
t
c
y
b
.
2
0
2
1
.
3
1
1
0
8
1
3
.
[
1
3
]
R
.
F
l
o
r
e
z
e
t
.
a
l
.
,
“
A
C
N
N
-
b
a
s
e
d
a
p
p
r
o
a
c
h
f
o
r
d
r
i
v
e
r
d
r
o
w
si
n
e
ss
d
e
t
e
c
t
i
o
n
b
y
r
e
a
l
-
t
i
me
e
y
e
st
a
t
e
i
d
e
n
t
i
f
i
c
a
t
i
o
n
,
”
Ap
p
l
.
S
c
i
.
(
B
a
se
l
)
,
v
o
l
.
1
3
,
n
o
.
1
3
,
p
.
7
8
4
9
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
3
1
3
7
8
4
9
.
[
1
4
]
C
.
B
.
S
.
M
a
i
o
r
,
M
.
J.
d
a
s
C
.
M
o
u
r
a
,
J.
M
.
M
.
S
a
n
t
a
n
a
,
a
n
d
I
.
D
.
Li
n
s,
“
R
e
a
l
-
t
i
m
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
f
o
r
a
u
t
o
n
o
mo
u
s
d
r
o
w
si
n
e
ss
d
e
t
e
c
t
i
o
n
u
si
n
g
e
y
e
a
s
p
e
c
t
r
a
t
i
o
,
”
Ex
p
e
rt
S
y
s
t
.
Ap
p
l
.
,
v
o
l
.
1
5
8
,
n
o
.
1
1
3
5
0
5
,
p
.
1
1
3
5
0
5
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
sw
a
.
2
0
2
0
.
1
1
3
5
0
5
.
[
1
5
]
W
.
A
l
K
i
s
h
r
i
,
A
.
A
b
u
a
l
k
i
sh
i
k
,
a
n
d
M
.
A
l
-
B
a
h
r
i
,
“
E
n
h
a
n
c
e
d
i
m
a
g
e
p
r
o
c
e
ssi
n
g
a
n
d
f
u
z
z
y
l
o
g
i
c
a
p
p
r
o
a
c
h
f
o
r
o
p
t
i
m
i
z
i
n
g
d
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
d
e
t
e
c
t
i
o
n
,
”
A
p
p
l
.
C
o
m
p
u
t
.
I
n
t
e
l
l
.
S
o
f
t
C
o
m
p
u
t
.
,
v
o
l
.
2
0
2
2
,
p
p
.
1
–
1
4
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
2
/
9
5
5
1
2
0
3
.
[
1
6
]
R
.
K
.
S
h
u
k
l
a
,
A
.
K
.
T
i
w
a
r
i
,
a
n
d
A
.
K
.
J
h
a
,
“
A
n
e
f
f
i
c
i
e
n
t
a
p
p
r
o
a
c
h
o
f
f
a
c
e
d
e
t
e
c
t
i
o
n
a
n
d
p
r
e
d
i
c
t
i
o
n
o
f
d
r
o
w
si
n
e
ss
u
s
i
n
g
S
V
M
,
”
Ma
t
h
.
Pr
o
b
l
.
E
n
g
.
,
v
o
l
.
2
0
2
3
,
n
o
.
1
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
3
/
2
1
6
8
3
6
1
.
[
1
7
]
A
.
K
.
B
i
sw
a
l
,
D
.
S
i
n
g
h
,
B
.
K
.
P
a
t
t
a
n
a
y
a
k
,
D
.
S
a
ma
n
t
a
,
a
n
d
M
.
-
H
.
Y
a
n
g
,
“
I
o
T‐
b
a
se
d
s
mart
a
l
e
r
t
s
y
s
t
e
m
f
o
r
d
r
o
w
sy
d
r
i
v
e
r
d
e
t
e
c
t
i
o
n
,
”
W
i
re
l
.
C
o
m
m
u
n
.
Mo
b
.
C
o
m
p
u
t
.
,
v
o
l
.
2
0
2
1
,
n
o
.
1
,
p
p
.
1
–
1
3
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
1
/
6
6
2
7
2
1
7
.
[
1
8
]
H
.
La
m
a
a
z
i
,
A
.
A
l
q
a
ss
a
b
,
R
.
A
.
F
a
d
u
l
,
a
n
d
R
.
M
i
z
o
u
n
i
,
“
S
mart
e
d
g
e
-
b
a
se
d
d
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
d
e
t
e
c
t
i
o
n
i
n
m
o
b
i
l
e
c
r
o
w
d
s
o
u
r
c
i
n
g
,
”
I
EEE
A
c
c
e
ss
,
v
o
l
.
1
1
,
p
p
.
2
1
8
6
3
–
2
1
8
7
2
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
a
c
c
e
ss
.
2
0
2
3
.
3
2
5
0
8
3
4
.
[
1
9
]
L.
Zh
a
n
g
,
H
.
S
a
i
t
o
,
L.
Y
a
n
g
,
a
n
d
J.
W
u
,
“
P
r
i
v
a
c
y
-
p
r
e
s
e
r
v
i
n
g
f
e
d
e
r
a
t
e
d
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
f
o
r
d
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
d
e
t
e
c
t
i
o
n
,
”
I
EE
E
Ac
c
e
ss
,
v
o
l
.
1
0
,
p
p
.
8
0
5
6
5
–
8
0
5
7
4
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
a
c
c
e
ss.
2
0
2
2
.
3
1
9
2
4
5
4
.
[
2
0
]
M
.
A
h
me
d
,
S
.
M
a
s
o
o
d
,
M
.
A
h
ma
d
,
a
n
d
A
.
A
.
A
b
d
El
-
L
a
t
i
f
,
“
I
n
t
e
l
l
i
g
e
n
t
d
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
d
e
t
e
c
t
i
o
n
f
o
r
t
r
a
f
f
i
c
safe
t
y
b
a
se
d
o
n
mu
l
t
i
-
C
N
N
d
e
e
p
mo
d
e
l
a
n
d
f
a
c
i
a
l
s
u
b
sam
p
l
i
n
g
,
”
I
EEE
T
ra
n
s.
I
n
t
e
l
l
.
T
r
a
n
sp
.
S
y
st
.
,
v
o
l
.
2
3
,
n
o
.
1
0
,
p
p
.
1
9
7
4
3
–
1
9
7
5
2
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
t
i
t
s
.
2
0
2
1
.
3
1
3
4
2
2
2
.
[
2
1
]
Y
.
Je
b
r
a
e
i
l
y
,
Y
.
S
h
a
r
a
f
i
,
a
n
d
M
.
Te
s
h
n
e
h
l
a
b
,
“
D
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
d
e
t
e
c
t
i
o
n
b
a
se
d
o
n
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
a
r
c
h
i
t
e
c
t
u
r
e
o
p
t
i
m
i
z
a
t
i
o
n
u
s
i
n
g
g
e
n
e
t
i
c
a
l
g
o
r
i
t
h
m,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
1
2
,
p
p
.
4
5
7
0
9
–
4
5
7
2
6
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
a
c
c
e
ss.
2
0
2
4
.
3
3
8
1
9
9
9
.
[
2
2
]
H.
-
T.
N
g
u
y
e
n
,
N
.
-
D
.
M
a
i
,
B
.
G
.
Le
e
,
a
n
d
W
.
-
Y
.
C
h
u
n
g
,
“
B
e
h
i
n
d
-
t
h
e
-
e
a
r
E
EG
-
b
a
se
d
w
e
a
r
a
b
l
e
d
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
d
e
t
e
c
t
i
o
n
s
y
st
e
m
u
si
n
g
e
m
b
e
d
d
e
d
t
i
n
y
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
I
EEE
S
e
n
s
.
J
.
,
v
o
l
.
2
3
,
n
o
.
1
9
,
p
p
.
2
3
8
7
5
–
2
3
8
9
2
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
j
se
n
.
2
0
2
3
.
3
3
0
7
7
6
6
.
[
2
3
]
S
.
K
.
M
o
u
sa
v
i
k
i
a
,
E
.
G
h
o
l
i
z
a
d
e
h
a
z
a
r
i
,
M
.
M
o
u
saz
a
d
e
h
,
a
n
d
S
.
B
.
O
.
Y
a
l
c
i
n
,
“
I
n
st
r
u
c
t
i
o
n
se
t
e
x
t
e
n
s
i
o
n
o
f
a
R
i
sc
V
b
a
se
d
S
o
C
f
o
r
d
r
i
v
e
r
d
r
o
w
s
i
n
e
ss
d
e
t
e
c
t
i
o
n
,
”
I
EEE
A
c
c
e
ss
,
v
o
l
.
1
0
,
p
p
.
5
8
1
5
1
–
5
8
1
6
2
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
a
c
c
e
ss
.
2
0
2
2
.
3
1
7
7
7
4
3
.
[
2
4
]
H
.
A
.
M
a
d
n
i
,
A
.
R
a
z
a
,
R
.
S
e
h
a
r
,
N
.
Th
a
l
j
i
,
a
n
d
L.
A
b
u
a
l
i
g
a
h
,
“
N
o
v
e
l
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
f
o
r
d
r
i
v
e
r
d
r
o
w
si
n
e
ss
d
e
t
e
c
t
i
o
n
u
si
n
g
e
y
e
mo
v
e
me
n
t
b
e
h
a
v
i
o
r
,
”
I
E
EE
Ac
c
e
ss
,
v
o
l
.
1
2
,
p
p
.
6
4
7
6
5
–
6
4
7
7
8
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
a
c
c
e
ss.
2
0
2
4
.
3
3
9
2
6
4
0
.
[
2
5
]
G
.
S
.
S
h
a
n
k
a
r
a
C
h
a
r
i
a
n
d
J.
A
.
P
r
a
sh
a
n
t
,
“
S
c
a
l
i
n
g
e
f
f
e
c
t
i
v
i
t
y
i
n
ma
n
i
f
o
l
d
me
t
h
o
d
o
l
o
g
i
e
s
t
o
d
e
t
e
c
t
d
r
i
v
e
r
’
s
f
a
t
i
g
u
e
n
e
ss
a
n
d
d
r
o
w
s
i
n
e
ss
s
t
a
t
e
,
”
I
A
ES
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
I
J
-
AI
)
,
v
o
l
.
1
3
,
n
o
.
2
,
p
.
1
2
2
7
,
Ju
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
1
3
.
i
2
.
p
p
1
2
27
-
1
2
4
0
.
Evaluation Warning : The document was created with Spire.PDF for Python.