I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
14
, N
o.
2
,
A
pr
il
2025
, pp.
843
~
852
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
2
.pp
843
-
852
843
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
E
n
h
a
n
c
i
n
g c
on
vol
u
t
i
o
n
al
n
e
u
r
al
n
e
t
w
or
k
b
ase
d
m
o
d
e
l
f
or
c
h
e
at
i
n
g at
on
l
i
n
e
e
xam
i
n
at
i
o
n
s d
e
t
e
c
t
i
on
S
ar
a O
u
ah
ab
i
1
, R
ih
ab
A
b
ou
d
ih
a
j
1
, N
aw
al
S
ae
l
1
, K
am
al
E
l
G
u
e
m
m
at
2
1
L
a
bor
a
t
or
y
of
M
ode
l
i
ng a
nd I
nf
or
m
a
t
i
on P
r
oc
e
s
s
i
ng, F
a
c
ul
t
y of
S
c
i
e
nc
e
s
B
e
n
M
'
s
i
k, H
a
s
s
a
n I
I
U
ni
ve
r
s
i
t
y, C
a
s
a
bl
a
nc
a
,
M
or
oc
c
o
2
L
a
bor
a
t
or
y
of
S
i
gna
l
s
, D
i
s
t
r
i
but
e
d S
ys
t
e
m
s
,
a
nd A
r
t
i
f
i
c
i
a
l
I
nt
e
l
l
i
ge
nc
e
, E
N
S
E
T
, H
a
s
s
a
n I
I
U
ni
ve
r
s
i
t
y, M
oha
m
m
e
di
a
,
M
or
oc
c
o
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
M
a
r
18, 2024
R
e
vi
s
e
d
N
ov 6, 2024
A
c
c
e
pt
e
d
N
ov 14, 2024
In
the
last
few
years,
e
-
learning
has
revolutioning
education
,
giving
st
udents
access to
diverse an
d adaptabl
e on
-
line resour
ces, but it ha
s also fac
e a
major
challenge:
cheating
on
online
exams.
Student
s
now
use
variant
c
heating
methods
include
consulting
unauthorize
d
documents,
communicatin
g
with
others
during
the
exam,
searching
for
information
on
the
i
n
ternet.
Combating
these
cheatin
g
practic
es
has
become
crucia
l
to
preser
vi
ng
the
integrity
of
academic
assessments.
In
this
context,
artificia
l
intelligenc
e
(AI)
has
emerged
as
an
essential
tool
for
mitiga
ting
this
fraudulent
be
havior.
Equipped
with
advanced
machine
learning
capabilities,
AI
can
examine
a
wide
range
of
data
to
detect
student
suspicious
behavior.
This
study
develops
an
approach
based
on
a
convolut
ional
neural
network
(
CNN
)
model
designed
to
detect
cheating
by
analyzing
candidates'
head
movements
during
online
exams.
By
exploiting
the
FEI
dataset,
this
model
achieves
an
interesting
accura
cy
of
97.28%.
In
addition,
we
compare
this
model
to
the
well
-
known
transfer
learning
models
used
in
the
lit
erature
n
amely,
ResNet50
,
VGG16,
DenseN
et21,
MobileNe
tV2,
and
EfficientNetB0
demonstrating
the
out
performance
of
our
approach
in
detecting
c
heating
during online exams.
K
e
y
w
o
r
d
s
:
C
he
a
ti
ng de
te
c
ti
on
C
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
D
e
e
p l
e
a
r
ni
ng
H
e
a
d
m
ouve
m
e
nt
a
na
ly
s
is
O
nl
in
e
e
xa
m
s
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
S
a
r
a
O
ua
ha
bi
L
a
bor
a
to
r
y
of
M
ode
li
ng a
nd I
nf
or
m
a
ti
on
P
r
oc
e
s
s
in
g, F
a
c
ul
ty
of
S
c
ie
nc
e
s
B
e
n M
'
s
ik
,
H
a
s
s
a
n I
I
U
ni
ve
r
s
it
y
C
a
s
a
bl
a
n
c
a
, M
or
oc
c
o
E
m
a
il
:
s
3.oua
ha
bi
@
gm
a
il
.c
om
1.
I
N
T
R
O
D
U
C
T
I
O
N
E
-
le
a
r
ni
ng
is
pl
a
yi
ng
a
vi
ta
l
r
ol
e
in
th
e
e
xi
s
ti
ng
e
duc
a
ti
ona
l
s
e
tt
in
g,
a
s
it
c
ha
ng
e
s
th
e
e
nt
ir
e
e
duc
a
ti
on
s
ys
te
m
a
nd
be
c
om
e
s
one
of
th
e
gr
e
a
te
s
t
pr
e
f
e
r
r
e
d
to
pi
c
s
f
or
a
c
a
de
m
ic
s
[
1]
.
T
hi
s
s
hi
f
t
ha
s
be
e
n
dr
iv
e
n
by
th
e
ne
e
d
f
or
a
s
a
f
e
a
nd
e
f
f
ic
ie
nt
a
lt
e
r
na
ti
ve
to
i
n
-
pe
r
s
on
le
a
r
ni
ng.
I
n
f
a
c
t,
e
-
le
a
r
ni
ng
a
ll
ow
pr
ovi
di
ng
e
f
f
e
c
ti
ve
te
a
c
hi
ng
m
e
th
ods
,
c
a
te
r
in
g
to
di
ve
r
s
e
le
a
r
ni
ng
s
ty
le
s
a
nd
of
f
e
r
s
a
c
c
e
s
s
ib
il
it
y
to
a
v
a
s
t
a
r
r
a
y of
e
duc
a
ti
ona
l
r
e
s
our
c
e
s
a
nd i
nt
e
r
a
c
ti
ve
oppor
tu
ni
ti
e
s
[
2]
, [
3]
, pr
om
ot
in
g a
c
ti
ve
e
nga
ge
m
e
nt
a
nd c
r
it
ic
a
l
th
in
ki
ng.
H
ow
e
ve
r
,
e
-
le
a
r
ni
ng
f
a
c
e
s
a
ls
o
m
a
jo
r
c
ha
ll
e
nge
s
s
uc
h
a
s
c
he
a
ti
ng.
E
xa
m
f
r
a
ud
is
w
id
e
s
pr
e
a
d
gl
oba
ll
y
[
4]
–
[
6]
,
w
ha
te
ve
r
th
e
le
ve
l
of
de
ve
lo
pm
e
nt
.
A
s
a
r
e
s
u
lt
,
tr
a
di
ti
ona
l
c
he
a
ti
ng
de
te
c
ti
on
m
e
th
ods
m
a
y
no l
onge
r
be
t
ot
a
ll
y e
f
f
e
c
ti
ve
i
n pr
e
ve
nt
in
g e
xa
m
in
a
ti
on f
r
a
ud.
O
nl
in
e
e
xa
m
s
a
r
e
a
n i
nt
e
gr
a
l
pa
r
t
of
e
-
le
a
r
ni
n
g
s
ol
ut
io
ns
f
or
a
ut
he
nt
ic
a
nd
f
a
ir
a
s
s
e
s
s
m
e
nt
of
s
tu
de
nt
p
e
r
f
or
m
a
nc
e
[
7]
.
T
he
d
e
s
ig
n
a
nd
e
xe
c
ut
io
n
of
onl
in
e
e
xa
m
s
a
r
e
th
e
m
os
t
c
ha
ll
e
ngi
ng
a
s
pe
c
ts
of
e
-
le
a
r
ni
ng.
I
n
pa
r
ti
c
ul
a
r
,
onl
in
e
e
xa
m
s
a
r
e
u
s
ua
ll
y
c
ondu
c
te
d
on
e
-
le
a
r
ni
ng
pl
a
tf
or
m
s
w
it
hout
th
e
phys
ic
a
l
pr
e
s
e
nc
e
of
s
tu
de
nt
s
a
nd
in
s
tr
uc
to
r
s
in
th
e
s
a
m
e
pl
a
c
e
.
T
hi
s
c
r
e
a
te
s
s
e
ve
r
a
l
de
f
ic
ie
nc
i
e
s
in
te
r
m
s
of
th
e
in
te
gr
it
y
a
nd
s
e
c
ur
it
y
of
onl
in
e
e
xa
m
s
.
F
or
e
xa
m
pl
e
,
c
a
ndi
da
te
a
ut
he
nt
ic
it
y
ve
r
if
ic
a
ti
on
is
e
xt
r
e
m
e
ly
pr
obl
e
m
a
ti
c
in
a
n
onl
in
e
e
nvi
r
onm
e
nt
,
pa
r
ti
c
ul
a
r
ly
in
th
e
a
b
s
e
nc
e
of
c
ont
in
uous
m
oni
to
r
in
g.
W
ha
t'
s
m
or
e
,
onl
in
e
e
xa
m
s
a
r
e
hi
g
hl
y
c
onduc
iv
e
to
c
he
a
ti
ng,
a
s
th
ous
a
nds
of
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
2
,
A
pr
il
2025
:
843
-
852
844
in
f
or
m
a
ti
on
r
e
s
our
c
e
s
a
r
e
a
c
c
e
s
s
ib
le
to
s
tu
de
nt
s
w
it
hout
a
ny
c
ont
r
ol
s
.
I
n
th
is
c
ont
e
xt
,
w
h
e
r
e
pr
e
s
e
r
vi
ng
th
e
in
te
gr
it
y
of
onl
in
e
e
xa
m
s
is
c
r
uc
ia
l,
a
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
(
A
I
)
of
f
e
r
s
a
dva
nc
e
d
a
n
a
ly
s
is
a
nd
de
t
e
c
ti
on
s
ki
ll
s
,
m
a
ki
ng i
t
a
n i
nva
lu
a
bl
e
a
s
s
e
t
f
or
gua
r
a
nt
e
e
in
g t
he
r
e
li
a
bi
li
ty
of
onl
in
e
a
s
s
e
s
s
m
e
nt
s
. C
he
a
ti
ng on onli
ne
e
xa
m
s
c
a
n
be
de
t
e
c
te
d
a
nd
c
la
s
s
if
ie
d
in
m
ul
ti
tu
de
f
or
m
s
,
f
r
om
c
ol
lu
s
io
n
be
twe
e
n
s
tu
de
nt
s
to
th
e
u
s
e
of
m
obi
le
de
vi
c
e
s
,
s
u
c
h
a
s
phone
s
,
to
m
or
e
s
ubt
le
in
di
c
a
to
r
s
s
u
c
h
a
s
e
y
e
m
ove
m
e
nt
s
[
8]
,
[
9]
,
m
out
h
m
ove
m
e
nt
s
he
a
d
m
ove
m
e
nt
s
[
10]
, a
nd ma
ny othe
r
una
ut
hor
iz
e
d be
ha
vi
or
s
.
I
n
th
e
f
a
c
e
of
th
e
di
ve
r
s
e
m
e
th
ods
of
c
he
a
ti
ng,
th
i
s
s
tu
dy c
onc
e
nt
r
a
te
s
on
a
na
ly
z
in
g
c
a
ndi
da
te
s
'
he
a
d
m
ove
m
e
nt
s
dur
in
g
e
xa
m
s
s
in
c
e
it
is
c
ons
id
e
r
e
d
a
s
ke
y
e
l
e
m
e
nt
to
de
te
c
t
s
e
ve
r
a
l
ot
he
r
be
ha
vi
or
s
a
nd
s
e
r
ve
a
s
pot
e
nt
ia
ll
y s
ig
ni
f
ic
a
nt
i
ndi
c
a
to
r
s
of
c
he
a
ti
ng. T
he
pr
im
a
r
y goa
l
is
t
o e
m
pl
oy t
he
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
(
C
N
N
)
a
ppr
oa
c
h
on
th
e
F
E
I
da
ta
s
e
t,
s
pe
c
if
ic
a
ll
y
de
s
ig
ne
d
to
de
te
c
t
he
a
d
m
ove
m
e
nt
s
.
I
n
a
ddi
ti
on,
a
c
om
pa
r
is
on
of
th
e
a
ppr
oa
c
h'
s
pe
r
f
or
m
a
nc
e
a
ga
in
s
t
ot
he
r
m
ode
ls
in
c
lu
di
ng
V
G
G
16
,
R
e
s
N
e
t5
0,
D
e
ns
e
N
e
t2
1,
E
f
f
ic
ie
nt
N
e
tB
0,
a
nd
M
obi
le
N
e
tV2
is
pe
r
f
or
m
e
d
to
e
va
lu
a
te
it
s
e
f
f
ic
a
c
y
in
c
he
a
ti
ng
d
e
te
c
ti
on
is
done
.
O
ur
a
ppr
oa
c
h
out
pe
r
f
or
m
s
th
e
ot
he
r
m
ode
ls
a
nd
ha
s
th
e
pot
e
nt
ia
l
to
s
ig
ni
f
ic
a
nt
ly
c
ont
r
ib
ut
e
to
th
e
pr
e
ve
nt
io
n
of
c
he
a
ti
ng i
n onli
ne
e
xa
m
s
by of
f
e
r
in
g a
pr
e
c
is
e
a
nd de
pe
nda
bl
e
s
ol
ut
io
n t
o t
hi
s
i
nt
r
ic
a
te
i
s
s
ue
.
T
he
r
e
m
a
in
de
r
of
t
hi
s
a
r
ti
c
le
i
s
s
tr
uc
tu
r
e
d a
s
f
ol
lo
w
. S
e
c
ti
on 2
p
r
ovi
de
s
a
n i
n
-
de
pt
h r
e
vi
e
w
of
r
e
la
te
d
w
or
k
in
th
e
f
ie
ld
of
onl
in
e
c
he
a
ti
ng
de
te
c
ti
on.
S
e
c
ti
on
3
d
e
ta
il
s
th
e
m
e
th
odol
ogy
de
ve
lo
p
e
d,
w
hi
c
h
is
c
e
nt
e
r
e
d
on
de
e
p
le
a
r
ni
ng
(
D
L
)
f
or
a
na
ly
z
in
g
c
a
ndi
da
te
s
'
he
a
d
m
ove
m
e
nt
s
.
I
n
s
e
c
ti
on
4,
m
or
e
de
ta
il
s
a
bout
th
e
e
xpe
r
im
e
nt
a
nd
th
e
obt
a
in
e
d
r
e
s
ul
ts
a
r
e
pr
e
s
e
nt
e
d,
of
f
e
r
in
g
e
s
s
e
nt
ia
l
in
s
ig
ht
s
in
to
th
e
a
ppr
oa
c
h'
s
e
f
f
e
c
ti
ve
ne
s
s
. F
in
a
ll
y,
s
e
c
ti
on 5 s
e
r
ve
s
a
s
t
he
c
onc
lu
s
io
n of
t
he
s
tu
dy.
2.
R
E
L
A
T
E
D
WO
R
K
S
I
n
th
e
f
ie
ld
of
onl
in
e
a
s
s
e
s
s
m
e
nt
,
w
hi
c
h
is
e
vol
vi
ng
r
a
pi
dl
y,
r
e
s
e
a
r
c
he
r
s
f
a
c
e
s
e
ve
r
a
l
c
ha
ll
e
nge
s
a
nd
e
xpl
or
e
nume
r
ous
r
e
s
e
a
r
c
h pos
s
ib
il
it
ie
s
. M
a
ny s
tu
di
e
s
ha
ve
be
e
n c
onduc
te
d t
o i
m
pr
ove
t
he
i
nt
e
gr
it
y
of
onl
in
e
e
xa
m
s
a
nd
a
ddr
e
s
s
th
e
is
s
ue
of
c
he
a
ti
ng.
B
a
w
a
r
it
h
e
t
al
.
[
11]
pr
opos
e
s
a
m
e
th
odol
ogy
ba
s
e
d
on
c
ont
in
uous
a
ut
he
nt
ic
a
ti
on,
e
ye
tr
a
c
ki
ng,
a
nd
f
in
ge
r
pr
in
t
s
c
a
nni
ng,
w
hi
c
h
w
a
s
a
ppl
ie
d
to
a
pr
iv
a
te
da
t
a
s
e
t.
T
he
r
e
s
ul
ts
de
m
ons
tr
a
te
d
a
s
e
n
s
it
iv
it
y
of
100%
,
a
s
pe
c
if
ic
it
y
of
95.56%
,
a
n
a
c
c
ur
a
c
y
of
95.74%
,
a
n
ove
r
a
ll
a
c
c
ur
a
c
y
of
97.78%
,
a
nd
a
n
F
-
m
e
a
s
ur
e
of
97.83.
F
ur
th
e
r
m
o
r
e
,
J
a
la
li
a
nd
N
oor
be
hba
ha
ni
[
12]
p
r
e
s
e
nt
s
two
di
s
ti
nc
t
m
e
th
ods
f
or
de
te
c
ti
ng
c
he
a
ti
ng
in
onl
in
e
e
xa
m
s
,
bot
h
of
th
e
m
w
e
r
e
de
ve
lo
pe
d
a
nd
te
s
te
d
u
s
in
g
a
pr
iv
a
te
da
ta
s
e
t.
T
he
f
ir
s
t
m
e
th
od
i
s
ba
s
e
d
on
im
a
ge
pr
oc
e
s
s
in
g
u
s
in
g
M
A
T
L
A
B
,
c
a
lc
ul
a
ti
ng
th
e
di
f
f
e
r
e
nc
e
be
twe
e
n
pi
xe
ls
in
th
e
im
a
ge
s
.
W
he
n
th
e
th
r
e
s
hol
d
is
s
e
t
a
t
9,
th
is
m
e
th
od
boa
s
ts
hi
gh
a
c
c
ur
a
c
y,
p
a
r
ti
c
ul
a
r
ly
f
or
de
te
c
ti
ng
e
m
pt
y
s
e
a
ts
,
w
it
h
a
n
a
c
c
ur
a
c
y
of
100%
,
how
e
ve
r
,
it
is
s
e
ns
it
iv
e
to
va
r
ia
ti
ons
in
th
e
c
ol
or
of
w
a
ll
s
,
s
tu
de
nt
s
'
c
lo
th
in
g
a
nd
obj
e
c
t
s
s
uc
h
a
s
s
he
e
ts
of
pa
pe
r
.
T
he
s
e
c
ond
m
e
th
od
is
ba
s
e
d
on
c
lu
s
te
r
in
g
r
e
f
e
r
e
nc
e
im
a
ge
s
us
in
g
th
e
k
-
m
e
doi
ds
a
lg
or
it
hm
.
T
hi
s
m
e
th
od
r
e
duc
e
s
c
o
m
pl
e
xi
ty
c
om
pa
r
e
d
to
th
e
f
ir
s
t
m
e
th
od,
but
it
s
a
c
c
ur
a
c
y
is
s
li
ght
ly
lo
w
e
r
,
w
it
h
a
n
a
ve
r
a
ge
a
c
c
ur
a
c
y
of
68
%
.
I
t
e
xc
e
ls
in
de
te
c
ti
ng
e
m
pt
y
s
e
a
ts
,
a
l
s
o
a
c
hi
e
vi
ng
100%
a
c
c
ur
a
c
y.
G
e
e
th
a
e
t
al
.
[
13]
us
e
d
th
e
E
ig
e
nf
a
c
e
m
e
th
od
f
or
e
xt
r
a
c
ti
ng
f
a
c
i
a
l
f
e
a
tu
r
e
s
f
r
om
f
a
c
ia
l
ve
c
to
r
s
a
nd
th
e
s
uppor
t
ve
c
to
r
m
a
c
hi
ne
(
S
V
M
)
m
ode
l
to
im
pr
ove
de
te
c
ti
on
a
c
c
ur
a
c
y.
T
hi
s
a
ppr
oa
c
h
w
a
s
a
ppl
ie
d
to
a
pr
iv
a
te
da
ta
s
e
t,
a
nd
th
e
obt
a
in
e
d
m
a
t
c
hi
ng
a
c
c
ur
a
c
y
w
a
s
a
ppr
oxi
m
a
te
ly
61%
w
it
h
50
r
e
a
l
-
ti
m
e
i
m
a
ge
s
i
n t
he
da
ta
s
e
t.
T
hi
s
a
c
c
ur
a
c
y
c
a
n be
i
m
pr
ove
d by i
nc
r
e
a
s
in
g t
he
numbe
r
of
i
m
a
ge
s
i
n t
he
da
ta
s
e
t.
T
o
a
ddr
e
s
s
e
s
th
e
i
s
s
ue
of
c
he
a
ti
ng
in
e
xa
m
s
,
w
he
th
e
r
in
pa
pe
r
or
e
le
c
tr
oni
c
e
xa
m
c
opi
e
s
,
R
he
in
e
t
al
.
[
14]
in
tr
oduc
e
F
L
E
X
a
ut
h,
a
n
a
ppl
ic
a
ti
on
th
a
t
ut
i
li
z
e
s
AI
te
c
hni
que
s
f
or
a
ut
hor
ve
r
if
ic
a
ti
on
in
e
le
c
tr
oni
c
pr
ogr
a
m
m
in
g
e
xa
m
s
us
in
g
a
pr
iv
a
te
da
ta
s
e
t.
T
h
e
id
e
a
is
ba
s
e
d
on
th
e
pr
in
c
ip
le
th
a
t
e
a
c
h
s
tu
de
nt
de
ve
lo
ps
a
n
in
di
vi
dua
l
s
ty
le
f
or
a
ns
w
e
r
in
g
c
e
r
ta
in
ty
pe
s
of
e
x
e
r
c
is
e
s
,
w
hi
c
h
c
a
n
be
e
xt
r
a
c
te
d
us
in
g
A
I
to
ol
s
a
nd
th
e
n
c
om
pa
r
e
d
to
r
e
f
e
r
e
nc
e
m
a
te
r
ia
l
w
it
h
ve
r
if
ie
d
a
ut
hor
s
.
T
he
c
onc
e
pt
is
a
ppl
ie
d
to
J
a
v
a
pr
ogr
a
m
m
in
g
e
xa
m
s
,
but
th
e
goa
l
is
to
e
xt
e
nd
s
uppor
t
to
ot
he
r
pr
ogr
a
m
m
in
g
la
ngua
ge
s
a
nd
ty
pe
s
of
a
s
s
ig
nm
e
nt
s
in
th
e
f
ut
ur
e
.
T
he
r
e
s
ul
ts
s
how
th
a
t
th
e
r
a
ndom
f
or
e
s
t
(
R
F
)
m
e
th
od
a
c
hi
e
ve
d
th
e
be
s
t
p
e
r
f
or
m
a
nc
e
w
it
h
a
n
a
c
c
ur
a
c
y
of
up
to
67.15%
f
or
c
la
s
s
if
yi
ng
th
e
to
p
th
r
e
e
opt
io
ns
.
M
or
e
ov
e
r
,
th
e
G
oogl
e
C
ode
J
a
m
2017
da
ta
s
e
t
w
a
s
us
e
d
f
or
t
e
s
ti
ng, whe
r
e
a
n a
c
c
ur
a
c
y of
93.80%
w
a
s
a
c
hi
e
ve
d.
I
n a
ddi
ti
on t
o
m
a
c
hi
ne
l
e
a
r
ni
n
g
(
M
L
)
a
nd
a
ut
h
e
nt
if
ic
a
ti
on b
a
s
e
d
c
he
a
t
in
g de
te
c
t
io
n t
e
c
hni
q
ue
s
, ot
h
e
r
s
pa
pe
r
s
u
s
e
d
DL
te
c
hni
q
ue
s
a
nd
tr
a
ns
f
e
r
le
a
r
ni
ng.
I
n
f
a
c
t
,
O
z
da
m
li
e
t
al
.
[
1
5]
e
m
p
lo
ye
d
c
om
put
e
r
vi
s
i
on
a
lg
or
it
hm
s
f
r
o
m
th
e
O
pe
nC
V
li
br
a
r
y
f
or
im
a
ge
a
c
q
ui
s
it
io
n,
pr
e
pr
o
c
e
s
s
i
ng,
f
e
a
t
ur
e
e
xt
r
a
c
ti
o
n,
de
te
c
ti
on/
s
e
gm
e
nt
a
ti
on
,
hi
g
h
-
le
v
e
l
pr
o
c
e
s
s
i
ng,
a
nd
d
e
c
i
s
io
n
m
a
ki
ng.
C
N
N
s
,
pa
r
ti
c
ul
a
r
ly
th
e
m
in
i
X
c
e
pt
io
n
m
ode
l,
w
e
r
e
u
s
e
d
to
pr
e
di
c
t
f
a
c
ia
l
e
m
ot
io
ns
.
V
a
r
io
u
s
d
a
ta
s
e
t
s
w
e
r
e
u
s
e
d
f
or
tr
a
in
in
g
a
nd
te
s
ti
n
g,
in
c
lu
di
n
g
im
a
ge
s
f
or
f
a
c
i
a
l
v
e
r
if
ic
a
ti
on,
e
m
ot
i
on
r
e
c
ogn
it
io
n,
ga
z
e
tr
a
c
k
in
g,
a
n
d
he
a
d
m
ove
m
e
nt
s
.
T
h
e
s
ys
t
e
m
a
c
hi
e
ve
d
a
n a
c
c
ur
a
c
y of
a
ppr
o
xi
m
a
t
e
ly
99
.38%
on
t
he
L
F
W
da
t
a
s
e
t.
F
or
i
n
-
c
la
s
s
e
m
ot
i
on t
r
a
c
ki
ng
, t
he
a
ve
r
a
ge
a
c
c
ur
a
c
y
w
a
s
a
ppr
oxi
m
a
te
l
y
66%
on
th
e
F
E
R
da
ta
s
e
t.
F
in
a
ll
y,
f
or
m
oni
to
r
in
g
be
ha
vi
or
s
dur
in
g
o
nl
in
e
e
xa
m
s
,
th
e
s
ys
te
m
di
s
pl
a
ye
d a
n
e
y
e
-
tr
a
c
ki
ng
a
c
c
ur
a
c
y of
a
ppr
oxi
m
a
te
l
y 96.
95%
on t
h
e
gi
4E
d
a
ta
s
e
t
a
nd
a
h
e
a
d
m
ove
m
e
nt
tr
a
c
ki
n
g
a
c
c
ur
a
c
y
of
a
ppr
oxi
m
a
te
ly
96
.24%
on
th
e
F
E
I
d
a
ta
s
e
t.
Y
ul
it
a
e
t
al
.
[
16]
us
e
M
obi
l
e
N
e
tV2
a
r
c
hi
t
e
c
tu
r
e
f
or
r
e
c
og
ni
z
in
g
a
c
ti
vi
t
ie
s
dur
in
g
onl
in
e
e
x
a
m
s
,
w
hi
c
h
w
a
s
a
ppl
ie
d
to
th
e
O
E
P
d
a
ta
s
e
t
.
O
pt
im
a
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
nhanc
in
g c
onv
ol
ut
io
nal
ne
ur
al
ne
tw
or
k
ba
s
e
d m
ode
l
fo
r
c
he
at
in
g at
onl
in
e
…
(
Sar
a O
uahabi
)
845
hype
r
p
a
r
a
m
e
te
r
s
w
e
r
e
f
ound,
r
e
s
ul
ti
ng
i
n
a
n
F
1
s
c
or
e
of
84
.52
%
.
A
w
e
b
a
ppl
i
c
a
ti
on
w
a
s
d
e
ve
lo
pe
d
to
a
s
s
is
t
te
a
c
he
r
s
i
n d
e
te
c
ti
ng
a
c
a
d
e
m
ic
f
r
a
ud.
T
h
is
r
e
pr
e
s
e
nt
s
a
s
ig
ni
f
i
c
a
nt
a
dv
a
nc
e
m
e
nt
i
n
e
du
c
a
ti
o
na
l
te
c
hnol
og
y.
I
n
va
r
io
us
s
tu
di
e
s
[
17]
–
[
20]
,
r
e
s
e
a
r
c
he
r
s
ha
ve
e
xpl
or
e
d
e
m
ot
io
n
r
e
c
ogni
ti
on
f
r
om
f
a
c
ia
l
e
xpr
e
s
s
io
ns
f
r
om
di
f
f
e
r
e
nt
pe
r
s
pe
c
ti
ve
s
.
E
l
H
a
m
m
oum
i
e
t
al
.
[
17]
f
oc
us
e
d
on
de
ve
lo
pi
ng
a
f
a
c
ia
l
e
xpr
e
s
s
io
n
r
e
c
ogni
ti
on
s
ys
te
m
ba
s
e
d
on
C
N
N
s
,
w
it
h
a
c
le
a
r
goa
l
of
in
te
gr
a
ti
on
i
nt
o
a
n
e
-
l
e
a
r
ni
ng
s
ys
te
m
.
T
hi
s
m
e
th
odol
ogy
in
vol
ve
d
im
a
ge
pr
e
pr
oc
e
s
s
in
g,
f
e
a
tu
r
e
e
xt
r
a
c
ti
on,
a
nd
im
a
ge
c
la
s
s
if
ic
a
ti
on,
r
e
s
ul
ti
ng
in
a
n
im
pr
e
s
s
iv
e
te
s
t
a
c
c
ur
a
c
y
of
97.53%
on
C
K
+
a
nd
97.18%
on
J
a
pa
ne
s
e
f
e
m
a
le
f
a
c
ia
l
e
xpr
e
s
s
io
n
(
J
A
F
F
E
)
.
O
z
de
m
ir
e
t
al
.
[
18
]
a
dopt
e
d
a
s
im
il
a
r
a
ppr
oa
c
h
us
in
g
DL
a
nd
C
N
N
s
to
c
r
e
a
te
a
n e
m
ot
io
na
l
c
la
s
s
if
ic
a
ti
on
m
ode
l,
m
e
r
gi
ng
va
r
io
us
da
ta
s
e
ts
.
T
h
e
ir
m
ode
l
ba
s
e
d
on
th
e
L
e
N
e
t
a
r
c
hi
te
c
tu
r
e
a
c
hi
e
ve
d
a
tr
a
in
in
g
a
c
c
ur
a
c
y
of
96.43
%
a
nd
a
va
li
da
ti
on
a
c
c
ur
a
c
y
of
91.81%
.
F
ur
th
e
r
m
or
e
,
P
r
a
na
v
e
t
al
.
[
19]
f
oc
us
e
d
on
de
ve
lo
pi
ng
a
de
e
p
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
(
D
C
N
N
)
m
ode
l
to
c
la
s
s
if
y
f
iv
e
di
f
f
e
r
e
nt
hum
a
n
f
a
c
ia
l
e
m
ot
io
ns
,
a
c
hi
e
vi
ng
a
n
a
c
c
ur
a
c
y
of
78.04%
.
F
in
a
ll
y,
Z
ba
id
a
e
t
al
.
[
20]
e
xa
m
in
e
d
th
e
im
por
ta
nc
e
of
e
m
ot
io
n
r
e
c
ogni
ti
on
in
hum
a
n
-
m
a
c
hi
ne
in
te
r
a
c
ti
on,
e
s
pe
c
i
a
ll
y
in
e
-
l
e
a
r
ni
ng
pl
a
tf
or
m
s
,
by
e
va
lu
a
ti
ng
va
r
io
us
a
ppr
oa
c
he
s
s
u
c
h
a
s
V
G
G
16,
V
G
G
19,
R
e
s
N
e
t5
0V
2,
E
f
f
ic
ie
nt
N
e
tB
0,
a
nd
E
f
f
ic
ie
nt
N
e
tB
7
to
de
ve
lo
p
a
s
pe
c
ia
li
z
e
d
n
e
ur
a
l
ne
twor
k
f
or
a
c
c
ur
a
te
e
m
ot
io
n i
de
nt
if
ic
a
ti
on.
T
he
a
n
a
ly
s
is
of
th
e
c
ur
r
e
nt
s
t
a
te
of
th
e
a
r
t,
r
e
ve
a
ls
t
ha
t
C
N
N
s
h
a
ve
be
c
om
e
im
por
ta
nt
f
or
th
e
ir
e
xc
e
ll
e
nt
pe
r
f
or
m
a
nc
e
in
v
a
r
io
u
s
onl
in
e
a
s
s
e
s
s
m
e
nt
a
r
e
a
s
. M
a
n
y
di
f
f
e
r
e
nt
t
ype
s
of
C
N
N
s
h
a
ve
b
e
e
n
s
t
udi
e
d
in
di
f
f
e
r
e
nt
s
it
ua
ti
ons
a
nd
w
it
h
di
f
f
e
r
e
nt
da
ta
,
s
how
i
ng
f
l
e
xi
bi
li
ty
a
nd
pot
e
nt
ia
l
.
T
h
e
r
e
la
t
e
d
w
or
k
s
s
tu
di
e
s
ha
v
e
ut
il
iz
e
d
C
N
N
s
f
or
v
a
r
io
u
s
ta
s
k
s
s
u
c
h
a
s
r
e
c
ogni
z
in
g
e
m
ot
io
ns
,
a
na
ly
z
in
g
f
a
c
ia
l
e
xpr
e
s
s
io
n
s
,
a
nd
m
oni
t
or
in
g
be
ha
v
io
r
dur
in
g
onl
in
e
e
x
a
m
s
.
H
ow
e
v
e
r
,
w
h
e
n
e
x
pl
or
in
g
r
e
s
e
a
r
c
h,
w
e
c
a
n
c
o
nc
lu
d
e
th
a
t
gr
e
a
t
e
f
f
or
t
s
a
r
e
s
ti
ll
ne
e
d
e
d
to
e
n
ha
n
c
e
t
he
p
e
r
f
or
m
a
n
c
e
of
C
N
N
s
,
e
s
p
e
c
i
a
ll
y
i
n
de
te
c
ti
ng c
he
a
ti
ng
. T
he
f
o
c
u
s
of
t
hi
s
r
e
s
e
a
r
c
h
i
s
on
im
pr
ovi
ng
C
N
N
s
us
in
g t
h
e
F
E
I
da
t
a
s
e
t
d
e
s
i
gne
d s
pe
c
if
ic
a
l
ly
f
or
de
te
c
ti
ng
he
a
d m
o
ve
m
e
nt
s
.
T
he
c
hoi
c
e
to
f
oc
us
on
th
e
a
na
ly
s
i
s
of
he
a
d
m
ove
m
e
nt
s
w
a
s
m
a
de
due
to
it
s
s
ig
ni
f
ic
a
nc
e
in
c
om
pr
e
he
ndi
ng
a
s
tu
de
nt
'
s
be
ha
vi
or
dur
in
g
a
n
e
xa
m
in
a
ti
on.
D
e
te
c
ti
ng
he
a
d
m
ove
m
e
nt
s
is
a
pi
vot
a
l
a
ppr
oa
c
h
f
or
r
e
c
ogni
z
in
g
pot
e
nt
ia
l
s
tu
de
nt
c
he
a
ti
ng,
a
nd
it
c
ons
ti
tu
te
s
a
f
unda
m
e
nt
a
l
a
s
pe
c
t
of
th
e
r
e
s
e
a
r
c
h.
A
n
ove
r
vi
e
w
of
th
e
va
r
io
us
s
tu
di
e
s
c
ove
r
e
d
in
th
e
s
ta
te
of
th
e
a
r
t
is
pr
ovi
de
d.
T
hi
s
s
um
m
a
r
y
a
im
s
to
pr
ovi
de
a
c
onc
is
e
pe
r
s
p
e
c
ti
ve
on
th
e
obj
e
c
ti
ve
s
,
da
t
a
s
e
t
s
,
m
e
th
odol
ogi
e
s
us
e
d
a
nd
r
e
s
ul
ts
obt
a
in
e
d
by
e
a
c
h
a
r
ti
c
le
in
c
lu
de
d
in
th
is
li
te
r
a
tu
r
e
r
e
vi
e
w
.
F
ur
th
e
r
m
or
e
,
in
th
e
r
e
s
ul
ts
c
ol
um
n,
th
e
a
bbr
e
vi
a
ti
on
"
A
c
c
"
s
t
a
nds
f
or
"
A
c
c
ur
a
c
y"
, r
e
pr
e
s
e
nt
in
g t
he
a
c
c
ur
a
c
y v
a
lu
e
obt
a
in
e
d by e
a
c
h
m
e
th
od a
dopt
e
d, a
s
de
ta
il
e
d i
n
T
a
bl
e
1.
T
a
bl
e
1. S
um
m
a
r
y
ta
bl
e
of
r
e
vi
e
w
e
d a
r
ti
c
le
s
R
e
f
O
bj
e
c
t
i
ve
D
a
t
a
s
e
t
na
m
e
D
a
t
a
s
e
t
s
i
z
e
M
e
t
hods
R
e
s
ul
t
s
(
%
)
[
11]
E
ye
m
ove
m
e
nt
de
t
e
c
t
i
on
P
r
i
va
t
e
30 s
t
ude
nt
s
F
i
nge
r
pr
i
nt
, E
ye
t
r
a
c
ki
ng,
E
-
pr
oc
t
or
A
c
c
=97.78
[
12]
B
e
ha
vi
or
de
t
e
c
t
i
on
P
r
i
va
t
e
50 i
m
a
ge
s
of
s
t
ude
nt
s
M
A
T
L
A
B
, K
-
M
e
doi
ds
A
c
c
=78
[
13]
F
a
c
i
a
l
r
e
c
ogni
t
i
on
P
r
i
va
t
e
50 i
m
a
ge
s
of
s
t
ude
nt
s
F
a
c
e
de
t
e
c
t
or
ba
s
e
d on
c
a
f
f
e
,
E
i
ge
nf
a
c
e
, S
V
M
A
c
c
=61
[
14]
A
na
l
yz
i
ng s
t
ude
nt
s
r
e
s
pons
e
s
P
r
i
va
t
e
2 a
s
s
i
gnm
e
nt
s
f
r
om
13 s
t
ude
nt
s
DNN
A
c
c
=37.78
G
oogl
e
C
ode
J
a
m
2017
40 pa
r
t
i
c
i
pa
nt
s
RF
A
c
c
=67.15
S
V
M
A
c
c
=57.69
[
15]
E
m
ot
i
on r
e
c
ogni
t
i
on
F
E
R
35600 i
m
a
ge
s
C
N
N
,
M
i
ni
X
c
e
pt
i
on
A
c
c
=66
F
a
c
i
a
l
r
e
c
ogni
t
i
on
L
F
W
13250 i
m
a
ge
s
of
5,794 i
ndi
vi
dua
l
s
D
e
e
p m
e
t
r
i
c
l
e
a
r
ni
ng, R
e
s
N
e
t
34
A
c
c
=99.38
E
ye
t
r
a
c
ki
ng
G
i
4E
1380 i
m
a
ge
s
T
r
a
i
ne
d m
ode
l
ba
s
e
d on e
ye
t
r
a
c
ki
ng
A
c
c
=96.95
H
e
a
d m
ove
m
e
nt
s
de
t
e
c
t
i
on
F
E
I
2800 i
m
a
ge
s
of
200 i
ndi
vi
dua
l
s
H
a
a
r
c
a
s
c
a
d
e
A
c
c
=96
.
24
O
bj
e
c
t
de
t
e
c
t
i
on
C
O
C
O
5000
i
m
a
ge
s
Y
O
L
O
(
V
3)
A
c
c
=60
[
16]
B
e
ha
vi
or
de
t
e
c
t
i
on
O
E
P
24 pa
r
t
i
c
i
pa
nt
s
M
obi
l
e
N
e
t
V
2
A
c
c
=86.7
[
17]
E
m
ot
i
on de
t
e
c
t
i
on
C
K
+
327 s
e
que
nc
e
s
of
i
m
a
ge
s
C
N
N
, H
a
a
r
c
a
s
c
a
de
A
c
c
=97.53
K
D
E
F
4900 f
a
c
e
i
m
a
ge
s
f
r
om
70 s
ubj
e
c
t
s
J
A
F
F
E
213 i
m
a
ge
s
of
7 J
a
pa
ne
s
e
w
om
e
n
A
c
c
=97.18
[
18]
E
m
ot
i
on de
t
e
c
t
i
on
P
r
i
va
t
e
140 i
m
a
ge
s
w
i
t
h 7 e
xpr
e
s
s
i
ons
L
e
N
e
t
C
N
N
a
r
c
hi
t
e
c
t
ur
e
,
H
a
a
r
c
a
s
c
a
d
e
A
c
c
=96.43
K
D
E
F
4900 f
a
c
e
i
m
a
ge
s
f
r
om
70 s
ubj
e
c
t
s
J
A
F
F
E
213 i
m
a
ge
s
of
7 J
a
pa
ne
s
e
w
om
e
n
[
19]
E
m
ot
i
on de
t
e
c
t
i
on
P
r
i
va
t
e
2550 i
m
a
ge
s
of
5 e
m
ot
i
ons
D
C
N
N
A
c
c
=78.04
[
20]
E
m
ot
i
on r
e
c
ogni
t
i
on
F
E
R
35600 i
m
a
ge
s
V
G
G
16, V
G
G
19, R
e
s
N
e
t
50,
E
f
f
i
c
i
e
nt
N
e
t
B
0, E
f
f
i
c
i
e
nt
N
e
t
B
7
A
c
c
=98.45
3.
M
E
T
H
O
D
T
he
m
a
in
obj
e
c
ti
ve
of
th
i
s
pa
p
e
r
is
to
im
pl
e
m
e
nt
a
C
N
N
m
ode
l
a
r
c
hi
te
c
tu
r
e
f
or
he
a
d
m
ove
m
e
nt
de
te
c
ti
on, with t
he
a
im
of
i
de
nt
if
yi
ng c
he
a
ti
ng be
ha
vi
or
i
n onli
ne
e
xa
m
s
us
in
g t
he
F
E
I
da
ta
s
e
t.
I
n t
hi
s
s
e
c
ti
on,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
2
,
A
pr
il
2025
:
843
-
852
846
th
e
m
e
th
odol
ogy
us
e
d
is
d
e
ta
il
e
d,
a
nd
th
e
c
r
uc
ia
l
s
te
p
s
f
ol
lo
w
e
d
to
e
n
s
ur
e
th
e
r
ig
or
a
nd
r
e
li
a
bi
li
ty
of
th
e
r
e
s
ul
ts
a
r
e
hi
ghl
ig
ht
e
d.
T
he
a
ppr
oa
c
h
is
s
tr
uc
tu
r
e
d
in
s
e
ve
r
a
l
f
unda
m
e
nt
a
l
s
te
ps
.
F
ir
s
t,
a
d
a
ta
c
ol
le
c
ti
on
w
a
s
c
a
r
r
i
f
e
d
out
.
T
hi
s
in
it
ia
l
pha
s
e
w
a
s
f
ol
lo
w
e
d
by
a
c
r
it
ic
a
l
da
ta
c
le
a
ni
ng
s
te
p
a
im
e
d
a
t
e
li
m
in
a
ti
ng
a
ny
in
c
ons
is
te
nc
ie
s
or
a
nom
a
li
e
s
.
N
e
xt
,
th
e
da
ta
w
a
s
la
b
e
le
d
in
to
two
di
s
ti
nc
t
c
a
te
gor
ie
s
:
c
he
a
ti
ng
a
nd
non
-
c
he
a
ti
ng,
to
f
a
c
il
it
a
te
th
e
id
e
nt
if
ic
a
ti
on
of
be
ha
vi
or
a
l
pa
tt
e
r
ns
.
A
f
te
r
w
a
r
ds
,
a
da
t
a
pr
e
-
pr
oc
e
s
s
in
g
ph
a
s
e
w
a
s
c
a
r
r
ie
d
out
.
T
hi
s
in
c
lu
de
d
da
t
a
a
ugm
e
nt
a
ti
on
to
di
ve
r
s
if
y
a
nd
e
nr
ic
h
th
e
d
a
ta
s
e
t,
a
s
w
e
ll
a
s
im
a
ge
s
r
e
s
iz
in
g
to
e
n
s
ur
e
c
on
s
is
te
nc
y
of
m
ode
l
in
put
s
.
T
o
m
a
in
ta
in
a
ba
la
nc
e
be
tw
e
e
n
c
la
s
s
e
s
,
a
b
a
la
nc
in
g
s
tr
a
te
gy
w
a
s
im
pl
e
m
e
nt
e
d.
F
in
a
ll
y,
DL
m
ode
ls
w
e
r
e
im
pl
e
m
e
nt
e
d a
nd
t
he
ir
pe
r
f
or
m
a
nc
e
w
a
s
e
v
a
lu
a
te
d
in
th
e
c
ont
e
xt
of
t
he
s
tu
dy. F
ig
ur
e
1 pr
e
s
e
nt
s
a
n i
ll
us
tr
a
ti
ve
di
a
gr
a
m
of
t
he
m
e
th
odol
ogy a
dopt
e
d.
F
ig
ur
e
1. T
he
pr
opos
e
d a
ppr
oa
c
h
3.1.
D
at
a c
ol
le
c
t
io
n
I
n
th
is
s
tu
d
y, t
h
e
F
E
I
da
ta
s
e
t
w
a
s
u
s
e
d. T
h
i
s
d
a
t
a
s
e
t
w
a
s
g
a
th
e
r
e
d
b
e
tw
e
e
n J
u
n
e
20
05 a
n
d M
a
r
c
h
2
006
a
t
th
e
F
E
I
AI
L
a
bor
a
to
r
y
in
S
ã
o
B
e
r
na
r
do
do
C
a
m
po,
S
ã
o
P
a
ul
o,
B
r
a
z
il
.
I
t
c
om
pr
is
e
s
14
im
a
g
e
s
p
e
r
in
di
vi
dua
l
f
r
om
a
m
ong
200
di
s
ti
nc
t
in
di
vi
dua
ls
,
to
ta
li
ng
280
0
im
a
ge
s
.
E
a
c
h
of
th
e
s
e
im
a
ge
s
is
in
c
ol
or
,
c
a
pt
ur
e
d
in
a
s
tr
a
ig
ht
f
r
ont
a
l
pos
it
io
n
on
a
uni
f
or
m
w
hi
te
ba
c
kgr
ound,
w
it
h
pr
of
il
e
r
ot
a
ti
ons
of
up
to
a
ppr
oxi
m
a
te
ly
180 de
gr
e
e
s
. S
e
ve
r
a
l
f
e
a
tu
r
e
s
of
t
hi
s
da
ta
s
e
t
m
a
k
e
i
t
pa
r
ti
c
ul
a
r
ly
s
ui
ta
bl
e
f
or
t
he
t
a
s
k. F
ir
s
tl
y, i
t
pr
ovi
de
s
a
v
a
r
ie
ty
of
pos
e
s
,
c
ov
e
r
in
g
a
n
im
pr
e
s
s
iv
e
r
a
ng
e
of
pr
of
il
e
r
ot
a
ti
ons
,
in
c
lu
di
ng
la
te
r
a
l
h
e
a
d
m
ove
m
e
nt
s
.
T
hi
s
di
ve
r
s
it
y
is
c
r
uc
ia
l
f
or
th
e
pr
oj
e
c
t
a
s
it
r
e
f
le
c
ts
th
e
di
f
f
e
r
e
nt
he
a
d
pos
it
io
ns
a
im
e
d
to
de
te
c
t
in
c
a
s
e
s
of
c
he
a
ti
ng. Additi
ona
ll
y, t
he
c
ons
is
te
nt
s
iz
e
of
i
m
a
ge
s
i
n t
he
F
E
I
da
ta
s
e
t
is
e
s
s
e
nt
ia
l
f
or
e
ns
ur
in
g t
he
c
ons
is
te
nc
y
of
vi
s
ua
l
f
e
a
tu
r
e
s
th
a
t
DL
m
ode
ls
c
a
n
le
a
r
n.
T
hi
s
c
ons
is
te
nc
y
s
im
pl
if
ie
s
th
e
tr
a
in
in
g
of
m
ode
ls
.
F
ur
th
e
r
m
or
e
,
it
is
ge
nde
r
-
ba
la
nc
e
d,
w
it
h
a
n
e
qua
l
num
be
r
of
m
a
le
a
nd
f
e
m
a
le
s
ubj
e
c
ts
(
100
of
e
a
c
h)
.
T
hi
s
ge
nde
r
ba
la
nc
e
i
s
e
s
s
e
nt
ia
l
to
a
voi
d a
ny ge
nde
r
bi
a
s
i
n t
he
de
te
c
ti
on mode
l.
3.2.
D
at
a c
le
an
in
g
T
he
F
E
I
da
ta
s
e
t,
w
hi
le
r
ic
h
in
he
a
d
m
ove
m
e
nt
in
f
or
m
a
ti
on,
c
ont
a
in
s
bot
h
im
a
ge
s
of
hum
a
n
f
a
c
e
s
a
nd
e
m
be
dde
d
im
a
g
e
s
w
it
hout
f
a
c
e
s
.
T
o
e
n
s
ur
e
th
e
qua
li
ty
of
our
tr
a
in
in
g
da
ta
s
e
t,
w
e
und
e
r
to
ok
a
c
le
a
ni
ng
pr
oc
e
s
s
. I
n t
hi
s
pha
s
e
, w
e
c
a
r
e
f
ul
ly
e
xa
m
in
e
d e
ve
r
y i
m
a
ge
i
n t
h
e
F
E
I
da
ta
s
e
t
a
nd e
li
m
in
a
te
d t
hos
e
t
ha
t
di
d not
c
ont
a
in
hum
a
n
f
a
c
e
s
.
T
hi
s
w
a
s
e
s
s
e
nt
ia
l
to
a
voi
d
a
ny
pot
e
n
ti
a
l
c
onf
us
io
n
of
our
m
ode
l
dur
in
g
tr
a
in
in
g,
e
ns
ur
in
g
th
a
t
onl
y
im
a
ge
s
r
e
le
va
nt
to
he
a
d
m
ove
m
e
nt
de
te
c
ti
on
w
e
r
e
r
e
ta
in
e
d.
T
he
c
le
a
ni
ng
pr
oc
e
s
s
w
a
s
c
a
r
r
ie
d out wit
h gr
e
a
t
a
tt
e
nt
io
n t
o de
ta
il
, a
nd e
a
c
h i
m
a
ge
w
a
s
a
s
s
e
s
s
e
d f
or
i
ts
s
ui
ta
bi
li
ty
t
o our
t
a
s
k.
3.3.
D
at
a l
ab
e
li
n
g
T
he
da
ta
s
e
t
u
s
e
d
in
th
e
s
tu
dy
w
a
s
la
b
e
le
d
in
to
two
di
s
ti
nc
t
c
la
s
s
e
s
:
th
e
"
non
-
c
he
a
ti
ng"
c
la
s
s
a
nd
th
e
"
c
he
a
ti
ng"
c
la
s
s
.
T
he
"
non
-
c
he
a
ti
ng"
c
la
s
s
in
c
lu
de
s
im
a
ge
s
of
in
di
vi
dua
ls
c
onc
e
nt
r
a
ti
ng
on
th
e
s
c
r
e
e
n
dur
in
g
a
n
onl
in
e
e
xa
m
. O
n
th
e
ot
he
r
ha
nd,
th
e
"
c
he
a
ti
ng"
c
l
a
s
s
in
c
lu
de
s
im
a
ge
s
of
in
di
vi
dua
ls
pos
it
io
ne
d
out
s
id
e
th
e
s
c
r
e
e
n
f
ie
ld
,
tu
r
ni
ng
th
e
ir
he
a
ds
to
th
e
r
ig
ht
or
le
f
t,
o
r
a
do
pt
in
g
be
ha
vi
or
s
s
ugge
s
ti
ve
of
c
he
a
ti
ng.
T
hi
s
la
be
li
ng s
te
p e
na
bl
e
d t
he
r
e
s
e
a
r
c
he
r
s
t
o s
pe
c
if
ic
a
ll
y t
a
r
ge
t
th
e
c
he
a
ti
ng be
ha
vi
or
s
t
he
y w
e
r
e
s
e
e
ki
ng t
o de
te
c
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
nhanc
in
g c
onv
ol
ut
io
nal
ne
ur
al
ne
tw
or
k
ba
s
e
d m
ode
l
fo
r
c
he
at
in
g at
onl
in
e
…
(
Sar
a O
uahabi
)
847
3.4.
D
at
a au
gm
e
n
t
at
io
n
T
he
da
ta
s
e
t
us
e
d
in
th
e
s
tu
dy
w
a
s
r
e
la
ti
ve
ly
s
m
a
ll
,
s
o
da
ta
a
ug
m
e
nt
a
ti
on
w
a
s
a
ppl
ie
d
to
in
c
r
e
a
s
e
it
s
di
ve
r
s
it
y
a
nd
s
i
z
e
.
N
e
w
im
a
ge
s
w
e
r
e
ge
ne
r
a
te
d
f
r
om
e
xi
s
ti
n
g
one
s
by
a
ppl
yi
ng
va
r
io
us
tr
a
ns
f
or
m
a
ti
ons
,
in
c
lu
di
ng
ve
r
ti
c
a
l
a
nd
hor
iz
ont
a
l
s
hi
f
t
in
g,
hor
iz
ont
a
l
a
nd
ve
r
ti
c
a
l
f
li
ppi
ng,
r
ot
a
ti
on,
a
nd
z
oom
in
g.
T
he
s
e
te
c
hni
que
s
w
e
r
e
a
ppl
ie
d
in
a
c
ont
r
ol
le
d
m
a
nne
r
to
c
r
e
a
te
v
a
r
ia
ti
ons
of
th
e
or
ig
in
a
l
im
a
ge
s
,
in
c
r
e
a
s
in
g
th
e
a
m
ount
of
da
ta
w
hi
le
m
a
in
ta
in
in
g
r
e
le
va
nc
e
to
th
e
de
te
c
ti
on
ta
s
k.
D
a
ta
a
ugm
e
nt
a
ti
on
is
a
n
e
s
s
e
nt
ia
l
to
ol
f
or
r
e
in
f
or
c
in
g t
he
r
obus
tn
e
s
s
of
t
he
m
ode
l
by e
xpos
in
g i
t
to
a
w
id
e
r
va
r
ie
ty
of
pos
s
ib
le
s
it
ua
ti
ons
w
hi
le
a
voi
di
ng
ove
r
f
it
ti
ng.
T
o
pr
ov
id
e
a
c
le
a
r
pe
r
s
pe
c
ti
ve
on
th
e
e
f
f
e
c
t
of
da
t
a
a
ugm
e
nt
a
ti
on,
T
a
bl
e
2
s
how
s
th
e
num
be
r
of
im
a
ge
s
in
th
e
da
ta
s
e
t
be
f
or
e
a
nd
a
f
te
r
a
ugm
e
nt
a
ti
on,
r
e
s
pe
c
ti
v
e
ly
.
T
hi
s
ta
bl
e
hi
ghl
ig
ht
s
th
e
e
xt
e
nt
to
w
hi
c
h
th
e
da
ta
s
e
t
ha
s
be
e
n e
nr
ic
h
e
d by thi
s
t
e
c
hni
que
.
T
a
bl
e
2
. N
um
be
r
of
i
m
a
ge
s
be
f
or
e
a
nd
a
f
te
r
a
ppl
yi
ng da
ta
a
ug
m
e
nt
a
ti
on
C
l
a
s
s
N
um
be
r
of
i
m
a
ge
s
be
f
or
e
a
ppl
yi
ng da
t
a
a
ugm
e
nt
a
t
i
on
N
um
be
r
of
i
m
a
ge
s
a
f
t
e
r
a
ppl
yi
ng da
t
a
a
ugm
e
nt
a
t
i
on
C
he
a
t
1479
4437
No
-
c
he
a
t
1186
3558
3.5.
I
m
age
s
r
e
s
iz
in
g
A
not
he
r
e
s
s
e
nt
ia
l
s
te
p
in
th
e
da
ta
pr
e
pa
r
a
ti
on
pr
oc
e
s
s
in
vol
ve
d
im
a
ge
r
e
s
iz
in
g.
T
he
or
ig
in
a
l
da
ta
s
e
t
in
c
lu
de
d
im
a
ge
s
w
it
h
a
r
e
s
ol
ut
io
n
of
640
×
480
pi
xe
ls
,
w
hi
c
h
w
a
s
uns
ui
ta
bl
e
f
or
th
e
m
ode
l.
T
he
r
e
f
or
e
,
a
ll
im
a
ge
s
w
e
r
e
r
e
s
iz
e
d
to
a
uni
f
or
m
r
e
s
ol
ut
io
n
of
224
×
224
pi
xe
l
s
.
T
hi
s
r
e
s
iz
in
g
w
a
s
don
e
in
s
uc
h
a
w
a
y
a
s
to
r
e
ta
in
th
e
e
s
s
e
nt
ia
l
vi
s
ua
l
in
f
or
m
a
ti
on
w
hi
le
r
e
duc
in
g
th
e
c
o
m
pl
e
xi
ty
of
th
e
da
ta
,
w
hi
c
h
is
b
e
ne
f
ic
ia
l
f
or
tr
a
in
in
g
DL
m
ode
ls
.
T
hi
s
s
te
p
e
n
s
ur
e
s
th
a
t
th
e
in
put
da
ta
is
un
if
or
m
a
nd
r
e
a
dy
to
be
pr
oc
e
s
s
e
d
by
th
e
c
he
a
t
de
te
c
ti
on a
lg
or
it
hm
s
w
hi
le
r
e
duc
in
g t
he
c
om
put
a
ti
ona
l
lo
a
d r
e
q
ui
r
e
d f
or
i
m
a
ge
pr
oc
e
s
s
in
g.
3.6.
C
la
s
s
b
al
an
c
in
g
T
o
a
ddr
e
s
s
th
e
m
a
r
ke
d
im
ba
la
nc
e
be
twe
e
n
th
e
c
he
a
ti
ng
a
nd
non
-
c
he
a
ti
ng
c
la
s
s
e
s
,
th
e
ove
r
s
a
m
pl
in
g
te
c
hni
que
w
a
s
im
pl
e
m
e
nt
e
d
to
m
a
n
a
ge
c
la
s
s
im
ba
la
nc
e
.
T
hi
s
m
e
th
od
c
ons
is
t
s
of
in
c
r
e
a
s
in
g
th
e
num
be
r
of
e
xa
m
pl
e
s
of
th
e
m
in
or
it
y
c
la
s
s
by
dupl
ic
a
ti
ng
or
ge
ne
r
a
ti
ng
ne
w
in
s
ta
nc
e
s
.
I
t
c
a
n
be
u
s
e
f
ul
w
he
n
th
e
r
e
a
r
e
a
s
m
a
ll
num
be
r
of
e
xa
m
pl
e
s
of
th
e
m
in
or
it
y
c
la
s
s
a
nd
th
e
goa
l
is
to
ba
la
nc
e
th
e
c
la
s
s
e
s
.
T
he
r
e
la
te
d
w
or
ks
pr
ovi
de
d s
uppor
t
th
e
us
e
of
ove
r
s
a
m
pl
in
g a
s
a
t
e
c
hni
que
t
o a
dd
r
e
s
s
c
la
s
s
i
m
ba
la
nc
e
i
n
ML
, pa
r
ti
c
ul
a
r
ly
i
n
t
he
c
ont
e
xt
of
c
he
a
ti
ng
de
te
c
ti
on
in
la
r
ge
-
s
c
a
l
e
a
s
s
e
s
s
m
e
nt
s
.
T
hi
s
c
la
s
s
im
ba
l
a
nc
e
m
a
n
a
ge
m
e
nt
is
a
n
im
por
ta
nt
s
te
p
in
e
ns
ur
in
g
th
e
a
c
c
ur
a
c
y
a
nd
r
e
li
a
bi
li
ty
of
th
e
c
h
e
a
ti
ng
de
te
c
ti
on
m
ode
l
by
m
in
im
iz
in
g
th
e
r
is
k
of
f
a
ls
e
pos
it
iv
e
s
or
f
a
ls
e
ne
ga
ti
ve
s
l
in
ke
d t
o t
he
i
ni
ti
a
l
im
ba
la
nc
e
of
t
he
da
ta
.
3.7.
M
od
e
li
n
g
I
n
th
is
s
tu
dy,
th
e
c
he
a
ti
ng
de
te
c
ti
on
m
ode
l
i
s
ba
s
e
d
on
C
N
N
a
r
c
hi
te
c
tu
r
e
.
H
ow
e
ve
r
,
in
or
de
r
to
e
va
lu
a
te
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
th
e
a
ppr
oa
c
h,
s
e
ve
r
a
l
ot
he
r
r
e
f
e
r
e
nc
e
m
ode
ls
w
e
r
e
a
ls
o
in
c
lu
de
d
in
th
e
a
na
ly
s
is
.
T
he
s
e
r
e
f
e
r
e
nc
e
m
od
e
ls
a
r
e
id
e
nt
if
ie
d
f
r
om
th
e
li
te
r
a
tu
r
e
r
e
vi
e
w
a
nd
de
r
iv
e
d
f
r
om
di
f
f
e
r
e
nt
de
e
p
ne
ur
a
l
ne
twor
k
a
nd
s
e
r
ve
a
s
poi
nt
s
of
c
om
pa
r
is
on
to
a
s
s
e
s
s
th
e
c
a
pa
bi
li
ty
of
th
e
a
ppl
ie
d
C
N
N
m
ode
l.
I
n
th
e
f
ol
lo
w
in
g s
e
c
ti
on, a
c
lo
s
e
r
l
ook is
t
a
ke
n
a
t
th
e
s
e
r
e
f
e
r
e
nc
e
m
od
e
ls
.
3.7.1. Convol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
C
N
N
is
a
ne
twor
k
m
ode
l
pr
opos
e
d
by
L
e
c
un
[
21]
a
r
e
a
c
la
s
s
of
ne
ur
a
l
ne
twor
ks
th
a
t
ha
v
e
pr
ove
d
hi
ghl
y
e
f
f
e
c
ti
ve
in
f
ie
ld
s
s
uc
h
a
s
im
a
ge
r
e
c
ogni
ti
on
a
nd
c
la
s
s
if
ic
a
ti
on.
C
N
N
s
a
r
e
a
ty
pe
of
f
or
w
a
r
d
-
pr
opa
ga
ti
ng
ne
ur
a
l
ne
twor
k
c
om
pos
e
d
of
s
e
ve
r
a
l
la
ye
r
s
.
T
he
y
c
ons
is
t
of
f
il
te
r
s
,
ke
r
ne
ls
or
ne
u
r
ons
w
it
h
a
dj
us
ta
bl
e
w
e
ig
ht
s
or
p
a
r
a
m
e
te
r
s
a
nd
bi
a
s
e
s
.
E
a
c
h
f
il
te
r
ta
k
e
s
i
nput
s
,
pe
r
f
or
m
s
a
c
onvolut
io
n,
a
nd
e
ve
nt
ua
ll
y
a
ppl
ie
s
a
non
-
li
ne
a
r
it
y.
T
he
s
tr
uc
tu
r
e
of
a
C
N
N
in
c
lu
de
s
c
onv
ol
ut
io
n,
pool
in
g,
r
e
c
ti
f
ie
d
li
ne
a
r
uni
t
(
R
e
L
U
)
a
nd
f
ul
ly
c
onne
c
te
d
la
ye
r
s
[
22]
.
T
o
e
va
lu
a
te
th
e
out
pe
r
f
or
m
a
nc
e
of
our
pr
opos
e
d
C
N
N
m
ode
l,
w
e
ha
ve
e
xpl
or
e
d t
he
m
os
t
pe
r
f
or
m
a
nt
D
L
m
ode
ls
us
e
d i
n t
he
l
it
e
r
a
tu
r
e
r
e
vi
e
w
. T
he
m
ode
ls
a
dopt
e
d a
r
e
a
s
f
ol
lo
w
s
.
3.7.2. Re
s
id
u
al
n
e
t
w
or
k
50
I
n
pr
in
c
ip
le
,
a
ddi
ng
e
xt
r
a
la
ye
r
s
to
a
n
e
ur
a
l
ne
twor
k
s
houl
d
im
pr
ove
m
ode
l
qua
li
ty
pr
ovi
de
d
th
e
ove
r
f
it
ti
ng
pr
obl
e
m
is
a
ddr
e
s
s
e
d.
H
ow
e
ve
r
,
m
uc
h
a
r
c
hi
te
c
tu
r
e
f
a
c
e
th
e
c
ha
ll
e
nge
of
"
va
ni
s
hi
ng
gr
a
di
e
nt
s
"
.
T
he
R
e
s
N
e
t
a
r
c
hi
te
c
tu
r
e
ha
s
be
e
n
de
s
ig
ne
d
to
s
ol
ve
th
is
pr
obl
e
m
by
in
tr
oduc
in
g
s
hor
tc
ut
c
onne
c
ti
ons
.
T
h
e
s
e
c
onne
c
ti
ons
e
ns
ur
e
th
a
t
a
ddi
ng
la
ye
r
s
doe
s
not
r
e
qui
r
e
le
a
r
ni
ng
a
n
id
e
nt
ic
a
l
tr
a
ns
f
or
m
a
ti
on
to
m
a
in
ta
in
or
s
ur
pa
s
s
t
he
pe
r
f
or
m
a
nc
e
of
a
r
c
hi
te
c
tu
r
e
w
it
h f
e
w
e
r
l
a
ye
r
s
. T
hi
s
is
m
a
de
pos
s
ib
le
by t
he
i
m
m
e
di
a
te
a
ddi
ti
on of
a
di
r
e
c
t
c
onne
c
ti
on be
twe
e
n t
h
e
out
put
of
e
a
c
h l
a
ye
r
a
nd t
he
i
np
ut
of
t
he
ne
xt
l
a
ye
r
[
23]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
2
,
A
pr
il
2025
:
843
-
852
848
3.7.3. Vis
u
al
ge
om
e
t
r
y gr
ou
p
-
16
I
s
a
C
N
N
a
r
c
hi
te
c
tu
r
e
e
xt
e
n
s
iv
e
ly
e
m
pl
oye
d
in
di
ve
r
s
e
c
o
m
put
e
r
vi
s
io
n
ta
s
ks
,
pa
r
ti
c
ul
a
r
ly
in
e
m
ot
io
n
r
e
c
ogni
ti
on
ba
s
e
d
on
f
a
c
ia
l
e
xpr
e
s
s
io
ns
.
D
e
ve
lo
pe
d
by
th
e
V
G
G
a
t
th
e
U
ni
ve
r
s
it
y
of
O
x
f
or
d,
th
e
V
G
G
16
m
ode
l
s
ta
nds
out
f
or
it
s
de
e
p
s
tr
uc
tu
r
e
,
c
om
pr
is
in
g
16
la
ye
r
s
,
in
c
lu
di
ng
13
c
onvolut
io
na
l
la
ye
r
s
a
nd
3
f
ul
ly
c
onne
c
te
d
la
ye
r
s
.
N
ot
a
bl
e
f
or
it
s
s
im
pl
ic
it
y
a
nd
c
ons
is
te
nc
y,
th
e
V
G
G
16
a
r
c
hi
te
c
tu
r
e
f
e
a
tu
r
e
s
c
onvolut
io
na
l
la
ye
r
s
w
it
h
a
s
m
a
ll
r
e
c
e
pt
iv
e
f
ie
ld
(
3
×
3)
,
s
ta
c
ke
d
s
e
que
nt
ia
ll
y.
T
hi
s
a
r
c
hi
te
c
tu
r
a
l
a
ppr
oa
c
h
e
na
bl
e
s
th
e
m
ode
l
to
a
c
qui
r
e
in
tr
ic
a
te
f
e
a
tu
r
e
s
by
c
a
pt
ur
in
g
lo
c
a
l
pa
tt
e
r
ns
in
th
e
e
a
r
ly
la
y
e
r
s
,
pr
ogr
e
s
s
iv
e
ly
m
ovi
ng on to m
or
e
c
om
pl
e
x pa
tt
e
r
ns
a
s
t
he
de
pt
h i
nc
r
e
a
s
e
s
[
24
]
.
3.7.4. De
n
s
e
ly
c
on
n
e
c
t
e
d
c
on
vol
u
t
io
n
al
n
e
t
w
or
k
s
21
R
e
pr
e
s
e
nt
s
a
c
ont
e
m
por
a
r
y
C
N
N
a
r
c
hi
te
c
tu
r
e
de
s
ig
n
e
d
f
or
vi
s
ua
l
obj
e
c
t
r
e
c
ogni
ti
on,
a
c
hi
e
vi
ng
s
ta
te
-
of
-
th
e
-
a
r
t
pe
r
f
or
m
a
nc
e
w
it
h
a
r
e
duc
e
d
num
be
r
of
pa
r
a
m
e
te
r
s
.
W
hi
le
s
h
a
r
in
g
s
om
e
f
unda
m
e
nt
a
l
s
im
il
a
r
it
ie
s
w
it
h
R
e
s
N
e
t,
D
e
ns
e
N
e
t
in
tr
oduc
e
s
not
a
bl
e
m
odi
f
ic
a
ti
ons
.
U
nl
ik
e
R
e
s
N
e
t'
s
a
ddi
ti
ve
a
tt
r
ib
ut
e
(
+
)
th
a
t
m
e
r
ge
s
pr
e
vi
ous
a
nd
f
ut
ur
e
la
ye
r
s
,
D
e
ns
e
N
e
t
e
m
pl
oys
c
onc
a
te
na
ti
on
(
.)
to
c
om
bi
ne
th
e
out
put
of
th
e
pr
e
vi
ous
la
ye
r
w
it
h
th
a
t
of
th
e
s
ub
s
e
que
nt
la
y
e
r
.
T
hi
s
a
r
c
hi
te
c
tu
r
a
l
di
s
ti
nc
ti
on
a
ddr
e
s
s
e
s
th
e
c
onne
c
ti
vi
ty
c
ha
ll
e
nge
by de
ns
e
ly
i
nt
e
r
c
onne
c
ti
ng a
ll
l
a
ye
r
s
[
25]
.
3.7.5. M
ob
il
e
N
e
t
V
2
I
s
a
c
om
pa
c
t,
s
w
if
t,
a
nd
pr
e
c
is
e
D
C
N
N
ta
il
or
e
d
f
or
c
la
s
s
if
ic
a
ti
on
a
nd
de
te
c
ti
on
a
s
s
ig
nm
e
nt
s
.
E
ngi
ne
e
r
e
d
to
e
xc
e
l
in
te
r
m
s
of
bot
h
s
pe
e
d
a
nd
s
iz
e
e
f
f
ic
ie
nc
y
.
T
hi
s
ne
twor
k
e
ns
ur
e
s
not
e
w
or
th
y
a
c
c
ur
a
c
y
in
va
r
io
us
c
om
put
e
r
vi
s
io
n t
a
s
ks
s
uc
h a
s
obj
e
c
t
c
la
s
s
if
ic
a
ti
on a
nd
de
te
c
ti
on
[
26]
.
3.7.6. E
f
f
ic
ie
n
t
N
e
t
S
ta
nds
out
a
s
a
C
N
N
a
r
c
hi
te
c
tu
r
e
a
c
c
la
im
e
d
f
or
it
s
e
f
f
ic
ie
nc
y
a
nd
out
s
ta
ndi
ng
p
e
r
f
or
m
a
nc
e
a
c
r
os
s
di
ve
r
s
e
c
om
put
e
r
vi
s
io
n t
a
s
ks
, i
nc
lu
di
ng t
he
di
s
c
e
r
nm
e
nt
of
e
m
ot
io
ns
ba
s
e
d on f
a
c
ia
l
e
xpr
e
s
s
io
n
s
. E
m
pl
oyi
ng
a
c
om
pound
s
c
a
li
ng
a
ppr
oa
c
h,
th
e
E
f
f
ic
ie
nt
N
e
tB
0
a
r
c
hi
te
c
tu
r
e
uni
f
or
m
ly
a
dj
us
ts
th
e
ne
twor
k'
s
de
pt
h,
w
id
th
,
a
nd
r
e
s
ol
ut
io
n.
T
hi
s
s
c
a
li
ng
s
tr
a
te
gy
e
n
a
bl
e
s
th
e
m
ode
l
to
s
tr
ik
e
a
c
om
m
e
nda
bl
e
ba
la
nc
e
be
tw
e
e
n
c
a
pa
c
it
y
a
nd c
om
put
a
ti
ona
l
e
f
f
ic
ie
nc
y
[
27]
.
4.
E
X
P
E
R
I
M
E
N
T
S
A
N
D
R
E
S
U
L
T
S
I
n
th
is
s
e
c
ti
on,
w
e
pr
e
s
e
nt
th
e
e
xp
e
r
im
e
nt
s
c
onduc
t
e
d
a
nd
th
e
c
or
r
e
s
ponding
r
e
s
ul
ts
obt
a
in
e
d.
T
he
pr
im
a
r
y
a
im
of
th
e
s
e
e
xpe
r
im
e
nt
s
w
a
s
to
a
s
s
e
s
s
th
e
e
f
f
e
c
ti
ve
n
e
s
s
of
th
e
pr
opos
e
d
C
N
N
m
ode
l
a
r
c
hi
te
c
tu
r
e
in
de
te
c
ti
ng
he
a
d
m
ove
m
e
nt
s
dur
in
g
onl
in
e
e
xa
m
s
.
W
e
a
ls
o
c
o
m
pa
r
e
it
s
pe
r
f
or
m
a
nc
e
w
it
h
ot
he
r
m
ode
ls
to
e
va
lu
a
te
i
ts
c
a
p
a
bi
li
ty
.
4
.1.
P
r
op
os
e
d
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
T
he
C
N
N
m
ode
l
f
ol
lo
w
s
a
m
ul
ti
-
s
ta
ge
a
r
c
hi
te
c
tu
r
e
.
F
ir
s
tl
y,
th
e
f
ir
s
t
pha
s
e
of
th
e
m
ode
l
c
om
pr
is
e
s
th
r
e
e
c
onvolut
io
n
la
ye
r
s
,
w
it
h
a
R
e
L
U
a
c
ti
v
a
ti
on
f
unc
ti
on,
w
hi
c
h
ta
ke
a
s
in
put
im
a
ge
s
of
s
iz
e
244
×
244
×
3.
A
f
te
r
e
a
c
h
c
onvolut
io
n,
a
m
a
x
-
pool
in
g
la
ye
r
w
it
h
a
pool
w
in
do
w
of
2×
2
is
a
ppl
ie
d
to
r
e
duc
e
th
e
di
m
e
ns
io
ns
.
A
d
r
opout
la
ye
r
is
th
e
n
in
s
e
r
te
d
a
t
a
r
a
te
of
0.5
to
r
e
duc
e
ove
r
f
it
ti
ng.
T
he
s
e
c
ond
pha
s
e
be
gi
n
s
w
it
h
a
f
la
tt
e
n
la
ye
r
t
ha
t
tr
a
ns
f
or
m
s
t
he
t
w
o
-
di
m
e
ns
io
na
l
da
ta
i
nt
o a
one
-
di
m
e
ns
io
na
l
a
r
r
a
y. N
e
xt
, t
w
o f
ul
ly
c
onne
c
te
d l
a
ye
r
s
(
d
e
ns
e
)
w
it
h
a
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
a
r
e
a
dde
d
to
le
a
r
n
m
o
r
e
a
bs
tr
a
c
t
f
e
a
tu
r
e
s
in
th
e
da
ta
.
F
in
a
ll
y,
f
or
th
e
out
put
of
th
e
m
ode
l,
a
d
e
ns
e
la
ye
r
w
it
h
a
S
ig
m
oi
d
a
c
ti
va
ti
on
f
unc
ti
on
is
in
c
lu
de
d,
in
di
c
a
ti
ng
a
bi
na
r
y
c
la
s
s
if
ic
a
ti
on
(
c
he
a
t
or
non
-
c
he
a
t)
.
T
he
m
ode
l
is
c
om
pi
le
d
w
it
h
th
e
A
da
m
opt
im
iz
e
r
,
a
bi
na
r
y_c
r
os
s
e
nt
r
opy
lo
s
s
f
unc
ti
on
a
da
pt
e
d
to
bi
na
r
y
c
la
s
s
if
ic
a
ti
on.
F
ig
ur
e
2
s
how
s
a
s
um
m
a
r
y
of
th
e
C
N
N
a
r
c
hi
te
c
tu
r
e
a
dopt
e
d.
T
o
e
nha
n
c
e
th
e
pe
r
f
or
m
a
nc
e
,
s
pe
e
d
a
nd
r
e
le
va
n
c
e
of
our
pr
opos
e
d
DL
m
ode
ls
,
w
e
h
a
ve
a
dopt
e
d
th
e
f
ol
lo
w
in
g
hype
r
pa
r
a
m
e
te
r
s
,
a
s
pr
e
s
e
nt
e
d
in
T
a
bl
e
3
.
T
he
s
e
hype
r
pa
r
a
m
e
te
r
s
w
e
r
e
c
a
r
e
f
ul
ly
s
e
le
c
te
d
to
opt
im
iz
e
t
r
a
in
in
g e
f
f
ic
ie
nc
y a
nd e
ns
ur
e
t
ha
t
th
e
m
ode
l
c
onve
r
ge
s
a
ppr
opr
ia
te
ly
.
4.2.
R
e
s
u
lt
s
of
t
h
e
p
r
op
os
e
d
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
m
od
e
l
A
f
te
r
s
uc
c
e
s
s
f
ul
ly
tr
a
in
in
g
th
e
C
N
N
m
ode
l,
a
n
a
c
c
ur
a
c
y
of
97.28%
w
a
s
a
c
hi
e
ve
d.
F
or
de
ta
il
e
d
r
e
s
ul
ts
,
in
c
lu
di
ng
th
e
e
vol
ut
io
n
of
a
c
c
ur
a
c
y
a
nd
lo
s
s
ove
r
th
e
20
tr
a
in
in
g
e
poc
hs
,
r
e
f
e
r
to
F
ig
ur
e
s
3
a
nd
4
.
T
he
s
e
gr
a
phs
pr
ovi
de
a
c
le
a
r
r
e
pr
e
s
e
nt
a
ti
on
of
th
e
m
ode
l'
s
pe
r
f
or
m
a
nc
e
ove
r
t
im
e
.
B
a
s
e
d
on
F
ig
ur
e
3,
th
e
e
vol
ut
io
n
of
th
e
C
N
N
m
ode
l
ove
r
20
tr
a
in
in
g
e
poc
hs
in
di
c
a
te
s
a
c
ont
in
uous
im
pr
ove
m
e
nt
in
it
s
pe
r
f
or
m
a
nc
e
.
T
he
a
c
c
ur
a
c
y
on
th
e
v
a
li
da
ti
on
s
e
t
gr
a
dua
ll
y
in
c
r
e
a
s
e
s
,
s
how
c
a
s
in
g
th
e
m
ode
l'
s
e
nha
nc
e
d
pr
e
di
c
ti
ve
c
a
pa
bi
li
ti
e
s
. T
hi
s
c
on
s
is
te
nt
upw
a
r
d
tr
a
je
c
to
r
y
hi
ghl
ig
ht
s
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
th
e
tr
a
in
in
g
pr
oc
e
s
s
a
nd
r
e
f
le
c
ts
th
e
m
ode
l'
s
gr
ow
in
g
a
bi
li
ty
to
ge
ne
r
a
li
z
e
to
uns
e
e
n
da
ta
.
I
n
c
o
nt
r
a
s
t,
F
ig
ur
e
4
il
lu
s
tr
a
te
s
th
e
e
vol
ut
io
n
of
lo
s
s
ove
r
th
e
s
a
m
e
20
tr
a
in
in
g
e
po
c
hs
.
R
e
m
a
r
ka
bl
y,
th
e
lo
s
s
de
c
r
e
a
s
e
s
s
ig
ni
f
ic
a
nt
ly
,
r
e
f
le
c
ti
ng
th
e
m
od
e
l'
s
a
bi
li
ty
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
nhanc
in
g c
onv
ol
ut
io
nal
ne
ur
al
ne
tw
or
k
ba
s
e
d m
ode
l
fo
r
c
he
at
in
g at
onl
in
e
…
(
Sar
a O
uahabi
)
849
to
opt
im
iz
e
i
ts
pr
e
di
c
ti
ons
. T
he
s
e
f
in
di
ngs
c
ol
le
c
ti
ve
ly
de
m
ons
t
r
a
te
t
he
pos
it
iv
e
pr
ogr
e
s
s
io
n a
nd e
f
f
e
c
ti
ve
ne
s
s
of
th
e
C
N
N
m
ode
l
th
r
oughout
th
e
tr
a
in
in
g
pr
oc
e
s
s
.
F
ur
th
e
r
m
or
e
,
th
e
r
e
duc
ti
on
in
lo
s
s
c
or
r
obor
a
te
s
th
e
m
ode
l'
s
c
a
pa
bi
li
ty
t
o m
in
im
iz
e
e
r
r
or
s
, t
he
r
e
by e
nha
nc
in
g i
ts
ov
e
r
a
ll
pe
r
f
or
m
a
nc
e
.
F
ig
ur
e
2. C
N
N
m
ode
l
a
r
c
hi
te
c
tu
r
e
us
e
d
T
a
bl
e
3
.
H
ype
r
pa
r
a
m
e
te
r
s
C
a
t
e
gor
y
H
ype
r
pa
r
a
m
e
t
e
r
s
V
a
l
ue
or
c
onf
i
gur
a
t
i
on
L
a
ye
r
hype
r
pa
r
a
m
e
t
e
r
s
D
r
opout
50%
K
e
r
ne
l
s
i
z
e
(
3x3)
A
c
t
i
va
t
i
on f
unc
t
i
on of
t
he
f
i
na
l
l
a
ye
r
S
i
gm
oi
d
A
c
t
i
va
t
i
on f
unc
t
i
on of
hi
dde
n l
a
ye
r
s
R
e
L
U
C
om
pi
l
a
t
i
on hype
r
pa
r
a
m
e
t
e
r
s
O
pt
i
m
i
z
a
t
i
on f
unc
t
i
on
A
da
m
E
r
r
or
f
unc
t
i
on
bi
na
r
y_c
r
os
s
e
nt
r
opy
L
e
a
r
ni
ng r
a
t
e
0.001
E
xe
c
ut
i
on hype
r
pa
r
a
m
e
t
e
r
s
B
a
t
c
h s
i
z
e
32
N
um
be
r
of
e
poc
hs
20
F
ig
ur
e
3. E
vol
ut
io
n o
f
t
r
a
in
in
g a
nd va
li
da
ti
on a
c
c
ur
a
c
y
F
ig
ur
e
4. E
vol
ut
io
n o
f
of
tr
a
in
in
g a
nd va
li
da
ti
on l
os
s
A
ddi
ti
ona
ll
y,
F
ig
ur
e
5
pr
e
s
e
nt
s
th
e
c
onf
us
io
n
m
a
tr
ix
,
a
ll
ow
in
g
f
or
a
c
om
pr
e
he
ns
iv
e
a
s
s
e
s
s
m
e
nt
of
th
e
m
ode
l'
s
c
la
s
s
if
ic
a
ti
on
c
a
pa
bi
li
ty
.
T
hi
s
m
a
tr
ix
pr
ovi
de
s
va
lu
a
bl
e
in
s
ig
ht
s
in
to
th
e
m
ode
l'
s
s
tr
e
ngt
hs
a
nd
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
2
,
A
pr
il
2025
:
843
-
852
850
w
e
a
kne
s
s
e
s
,
hi
ghl
ig
ht
in
g
a
r
e
a
s
f
or
pot
e
nt
ia
l
im
pr
ove
m
e
nt
.
T
h
e
c
onf
us
io
n
m
a
tr
ix
obt
a
in
e
d
a
f
te
r
tr
a
in
in
g
th
e
m
ode
l
to
c
la
s
s
if
y
c
he
a
ti
ng
a
nd
non
-
c
he
a
ti
ng
c
a
s
e
s
r
e
ve
a
ls
im
p
r
e
s
s
iv
e
pe
r
f
or
m
a
nc
e
,
m
e
a
ni
ng
th
a
t
th
e
m
ode
l
c
or
r
e
c
tl
y i
de
nt
if
ie
d c
he
a
ti
ng a
nd non
-
c
he
a
ti
ng c
a
s
e
s
i
n t
he
m
a
jo
r
it
y of
s
it
ua
ti
ons
.
A
f
te
r
e
xa
m
in
in
g
th
e
obt
a
in
e
d
m
e
tr
ic
s
,
it
is
c
le
a
r
th
a
t
th
e
C
N
N
m
ode
l
ha
s
a
c
hi
e
ve
d
out
s
ta
ndi
ng
pe
r
f
or
m
a
nc
e
.
W
it
h
a
n
a
c
c
ur
a
c
y
of
97.28
%
,
th
e
m
ode
l
ha
s
de
m
ons
tr
a
te
d
a
s
tr
ong
a
bi
li
ty
to
c
or
r
e
c
tl
y
c
la
s
s
if
y
in
put
da
ta
.
F
ur
th
e
r
m
o
r
e
,
a
pr
e
c
is
io
n
s
c
or
e
of
98.00%
a
tt
e
s
ts
to
it
s
c
a
pa
bi
li
ty
in
m
in
im
iz
in
g
f
a
ls
e
pos
it
iv
e
s
.
T
he
s
e
r
e
s
ul
ts
r
e
ve
a
l
th
e
e
f
f
ic
ie
nc
y
a
nd
r
e
li
a
bi
li
ty
of
th
e
m
ode
l,
pos
it
io
ni
ng
it
a
s
a
pr
om
is
in
g
s
ol
ut
io
n
f
or
th
e
gi
ve
n
ta
s
k.
T
he
y
a
l
s
o
c
onf
ir
m
th
a
t
th
e
m
ode
l
is
w
e
ll
-
ta
il
or
e
d
to
th
e
da
ta
a
nd
pr
ovi
de
s
a
s
ui
ta
bl
e
ba
la
nc
e
be
twe
e
n
id
e
nt
if
yi
ng
tr
ue
pos
it
iv
e
s
a
nd
m
in
im
iz
in
g
c
la
s
s
if
ic
a
ti
on
e
r
r
or
s
.
T
he
s
e
r
e
s
ul
ts
a
r
e
e
nc
our
a
gi
ng
f
or
th
e
f
ut
ur
e
of
t
he
pr
oj
e
c
t
a
nd s
ugge
s
t
th
a
t
th
e
m
ode
l
is
r
e
a
dy f
or
de
p
lo
ym
e
nt
i
n r
e
a
l
-
w
or
ld
s
c
e
na
r
io
s
.
F
ig
ur
e
5
.
C
onf
us
io
n m
a
tr
ix
4.3.
C
om
p
ar
is
on
w
it
h
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
-
b
as
e
d
p
r
e
-
t
r
ai
n
e
d
m
od
e
ls
I
n
th
e
e
va
lu
a
ti
on
pha
s
e
,
th
e
a
im
is
to
a
s
s
e
s
s
th
e
p
e
r
f
or
m
a
nc
e
of
th
e
C
N
N
m
od
e
l
by
c
om
pa
r
in
g
it
w
it
h
s
e
ve
r
a
l
w
id
e
ly
r
e
c
ogni
z
e
d
pr
e
-
tr
a
in
e
d
a
r
c
hi
te
c
tu
r
e
s
,
in
c
lu
di
ng
R
e
s
N
e
t5
0,
V
G
G
16,
D
e
ns
e
N
e
t1
21,
M
obi
le
N
e
tV2, a
nd E
f
f
ic
ie
nt
N
e
t.
O
nc
e
va
r
io
us
m
ode
ls
w
e
r
e
s
uc
c
e
s
s
f
ul
ly
t
r
a
in
e
d, a
de
ta
il
e
d c
om
pa
r
is
on of
t
he
r
e
s
ul
ts
obt
a
in
e
d
w
it
h
th
e
C
N
N
m
ode
l
w
a
s
c
onduc
te
d.
I
n
or
de
r
to
s
um
m
a
r
iz
e
th
e
s
e
r
e
s
ul
t
s
a
nd
f
a
c
il
it
a
te
th
e
c
om
pa
r
is
on,
c
ons
ul
t
T
a
bl
e
4
f
or
a
s
um
m
a
r
y
of
a
ll
th
e
m
e
tr
ic
s
a
nd
pe
r
f
or
m
a
nc
e
s
r
e
c
or
de
d.
T
hi
s
c
om
pa
r
a
ti
ve
a
s
s
e
s
s
m
e
nt
w
il
l
e
n
a
bl
e
a
be
tt
e
r
e
va
lu
a
ti
on of
t
he
C
N
N
m
ode
l
in
us
e
.
A
f
te
r
a
s
e
r
ie
s
of
in
-
de
pt
h
tr
a
in
in
g
s
e
s
s
io
ns
a
nd
e
va
lu
a
ti
on
s
,
w
e
a
r
e
a
bl
e
to
obs
e
r
ve
th
e
pe
r
f
or
m
a
nc
e
of
e
a
c
h
m
ode
l.
T
he
s
e
tr
a
in
in
g
a
nd
e
va
lu
a
ti
on
s
ta
ge
s
w
e
r
e
c
r
uc
ia
l
to
unde
r
s
ta
ndi
ng
how
e
a
c
h
m
ode
l
pe
r
f
or
m
e
d
in
th
e
s
pe
c
if
ic
c
ont
e
xt
of
our
onl
in
e
e
x
a
m
c
h
e
a
ti
ng
de
te
c
ti
on
pr
oj
e
c
t.
T
he
f
in
a
l
r
e
s
ul
ts
r
e
ve
a
l
th
a
t,
a
m
ong
th
e
m
ode
ls
e
va
lu
a
te
d,
our
C
N
N
a
r
c
hi
te
c
tu
r
e
a
c
hi
e
ve
d
th
e
m
os
t
pe
r
f
o
r
m
a
nc
e
,
w
it
h
a
n
a
c
c
ur
a
c
y
of
97.28%
.
C
om
pa
r
e
d
to
ot
he
r
c
om
m
onl
y
us
e
d
a
r
c
hi
te
c
tu
r
e
s
,
s
uc
h
a
s
R
e
s
N
e
t,
M
obi
le
N
e
tV2,
E
f
f
ic
ie
nt
N
e
t,
D
e
ns
e
N
e
t,
a
nd
V
G
G
16,
our
C
N
N
m
ode
l
s
ta
nds
out
s
ig
ni
f
ic
a
nt
ly
.
F
or
e
xa
m
pl
e
,
a
lt
hough
R
e
s
N
e
t
i
s
r
e
c
ogni
z
e
d
a
s
a
le
a
di
ng
a
r
c
hi
te
c
tu
r
e
,
our
C
N
N
m
ode
l
out
pe
r
f
or
m
s
it
w
it
h
a
n
a
c
c
ur
a
c
y
of
97.28%
,
c
om
pa
r
e
d
w
it
h 95.24%
f
or
R
e
s
N
e
t.
S
im
il
a
r
ly
, M
obi
le
N
e
t
a
c
hi
e
ve
d 96.93
%
a
c
c
ur
a
c
y.
T
a
bl
e
4
. M
od
e
ls
m
e
tr
ic
s
M
ode
l
s
A
c
c
ur
a
c
y
P
r
e
c
i
s
i
on
R
e
c
a
l
l
F1
-
s
c
or
e
P
r
opos
e
d C
N
N
m
ode
l
97.28
98.00
96.60
97.30
R
e
s
N
e
t
50
95.24
97.00
92.00
95.10
M
obi
l
e
N
e
t
V
2
96.93
95.20
98.00
97.00
D
e
ns
e
N
e
t
21
94.88
96.80
92.80
94.70
V
G
G
16
96.32
94.10
98.70
96.40
E
f
f
i
c
i
e
nt
N
e
t
B
0
96.12
97.00
95.80
96.40
O
n
th
e
ot
he
r
ha
nd,
th
e
97
.28%
a
c
c
ur
a
c
y
r
a
te
a
c
hi
e
ve
d
il
l
us
tr
a
t
e
s
th
e
r
e
m
a
r
ka
bl
e
e
f
f
e
c
ti
v
e
ne
s
s
of
th
e
m
e
th
od
in
c
l
a
s
s
if
yi
ng
h
e
a
d
m
ove
m
e
nt
s
in
th
e
F
E
I
da
ta
s
e
t,
w
hi
c
h
is
di
vi
d
e
d
in
t
o
two
di
s
ti
nc
t
c
l
a
s
s
e
s
:
c
he
a
ti
ng
a
nd
non
-
c
he
a
ti
ng.
C
om
pa
r
i
ng
t
he
s
e
r
e
s
ul
t
s
w
it
h
th
o
s
e
of
pr
e
vi
ous
s
t
udi
e
s
doc
um
e
nt
e
d
in
th
e
li
t
e
r
a
tu
r
e
,
a
l
s
o
c
ondu
c
te
d
on
th
e
s
a
m
e
F
E
I
da
ta
s
e
t,
it
is
i
nt
e
r
e
s
ti
n
g
to
not
e
th
a
t
th
e
s
e
e
a
r
l
ie
r
w
or
k
s
a
c
hi
e
ve
d
a
n
a
c
c
ur
a
c
y
r
a
te
of
96.24%
[
15]
.
T
h
e
r
e
f
or
e
,
th
e
s
e
r
e
s
ul
t
s
s
ig
ni
f
i
c
a
n
tl
y
s
ur
pa
s
s
pr
e
vi
ou
s
ly
r
e
por
te
d
p
e
r
f
or
m
a
n
c
e
,
de
m
o
ns
tr
a
ti
ng
th
a
t
th
e
C
N
N
-
b
a
s
e
d
a
ppr
oa
c
h r
e
pr
e
s
e
n
ts
a
s
ub
s
ta
n
ti
a
l
a
dv
a
nc
e
i
n s
ol
vi
ng t
h
is
pa
r
ti
c
ul
a
r
pr
obl
e
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
nhanc
in
g c
onv
ol
ut
io
nal
ne
ur
al
ne
tw
or
k
ba
s
e
d m
ode
l
fo
r
c
he
at
in
g at
onl
in
e
…
(
Sar
a O
uahabi
)
851
5.
C
O
N
C
L
U
S
I
O
N
I
n
c
onc
lu
s
io
n,
th
is
r
e
s
e
a
r
c
h
e
xpl
or
e
d
th
e
us
e
of
DL
to
de
te
c
t
c
he
a
ti
ng
in
onl
in
e
e
xa
m
s
,
f
oc
u
s
in
g
s
pe
c
if
ic
a
ll
y
on
th
e
de
te
c
ti
on
of
he
a
d
m
ove
m
e
nt
s
.
T
he
pr
opos
e
d
C
N
N
m
ode
l
de
m
ons
tr
a
te
d
e
xc
e
pt
io
na
l
pe
r
f
or
m
a
nc
e
,
out
pe
r
f
or
m
in
g
ot
he
r
e
va
lu
a
te
d
m
ode
ls
s
uc
h
a
s
R
e
s
N
e
t,
V
G
G
16,
D
e
ns
e
N
e
t2
1,
M
obi
le
N
e
tV2
,
a
nd
E
f
f
ic
ie
nt
N
e
tB
0.
T
h
e
pr
opos
e
d
a
ppr
oa
c
h
a
c
hi
e
v
e
d
97.28%
a
c
c
ur
a
c
y,
w
it
h
98.00%
pr
e
c
is
io
n
a
nd
96.60%
r
e
c
a
ll
.
T
he
s
e
r
e
s
ul
t
s
a
r
e
s
ig
ni
f
ic
a
nt
ly
be
tt
e
r
th
a
n
pr
e
vi
ous
w
or
k
on
th
e
s
a
m
e
F
E
I
da
ta
s
e
t,
w
hi
c
h
a
c
hi
e
v
e
d
a
pr
e
c
is
io
n of
96.24%
. T
he
e
f
f
e
c
ti
ve
ne
s
s
of
t
he
C
N
N
m
ode
l
in
de
te
c
ti
ng c
he
a
ti
ng i
n onli
ne
e
xa
m
s
s
ugge
s
ts
t
ha
t
it
c
oul
d
be
s
uc
c
e
s
s
f
ul
ly
de
pl
oye
d
in
r
e
a
l
-
w
or
ld
s
c
e
na
r
io
s
,
he
lp
in
g
to
e
ns
ur
e
th
e
in
te
gr
it
y
of
onl
in
e
a
s
s
e
s
s
m
e
nt
s
.
H
ow
e
ve
r
,
it
is
im
por
ta
nt
to
not
e
th
a
t
th
is
m
ode
l
i
s
not
in
f
a
ll
ib
le
a
nd
m
a
y
not
be
a
s
e
f
f
e
c
ti
ve
in
de
te
c
ti
ng
m
or
e
s
ubt
le
f
or
m
s
of
c
h
e
a
ti
ng,
s
uc
h
a
s
v
e
r
ba
l
c
he
a
ti
ng
or
di
s
c
r
e
e
t
c
ol
la
bor
a
ti
on
be
twe
e
n
s
tu
de
nt
s
.
T
he
s
e
f
or
m
s
of
c
he
a
ti
ng
m
a
y
not
in
vol
ve
vi
s
ib
le
he
a
d
m
ove
m
e
nt
s
,
m
a
ki
ng
th
e
ir
de
te
c
ti
on
m
or
e
d
if
f
ic
ul
t
f
o
r
our
c
ur
r
e
nt
m
ode
l.
6.
P
E
R
S
P
E
C
T
I
V
E
S
I
n our
f
ut
ur
e
w
or
k,
w
e
pl
a
n t
o e
xpl
or
e
ot
he
r
pe
r
s
pe
c
ti
ve
s
a
nd s
t
udy va
r
io
us
f
or
m
s
of
onl
in
e
c
he
a
ti
ng,
in
c
lu
di
ng
th
os
e
th
a
t
a
r
e
m
or
e
c
om
pl
e
x
a
nd
di
f
f
ic
ul
t
to
de
te
c
t.
F
or
e
xa
m
pl
e
,
w
e
pl
a
n
to
s
tu
dy e
ye
m
ove
m
e
nt
s
,
f
a
c
ia
l
e
xpr
e
s
s
io
ns
a
nd
voi
c
e
r
e
c
ogni
ti
on
a
s
pot
e
nt
ia
l
m
e
a
ns
of
de
te
c
ti
ng
onl
in
e
c
he
a
ti
ng.
B
y
e
xpa
ndi
ng
ou
r
c
he
a
ti
ng
de
te
c
ti
on
to
ol
ki
t,
w
e
hope
to
s
tr
e
ngt
he
n
our
m
ode
l
a
n
d
im
pr
ove
it
s
a
bi
li
ty
to
de
te
c
t
a
w
id
e
r
r
a
nge
of
onl
in
e
c
he
a
ti
ng be
ha
vi
or
s
.
R
E
F
E
R
E
N
C
E
S
[
1]
A
.
M
.
M
a
a
t
uk,
E
.
K
.
E
l
be
r
ka
w
i
,
S
.
A
l
j
a
w
a
r
ne
h,
H
.
R
a
s
ha
i
de
h,
a
nd
H
.
A
l
ha
r
bi
,
“
T
he
c
ovi
d
-
19
pa
nde
m
i
c
a
nd
e
-
l
e
a
r
ni
ng:
c
ha
l
l
e
nge
s
a
nd
oppor
t
uni
t
i
e
s
f
r
om
t
he
pe
r
s
pe
c
t
i
ve
of
s
t
ude
nt
s
a
nd
i
ns
t
r
uc
t
or
s
,
”
J
our
nal
of
C
om
put
i
ng
i
n
H
i
ghe
r
E
duc
at
i
on
,
vol
.
34, no. 1, pp. 21
–
38, 2022, doi
:
10.1007/
s
12528
-
021
-
09274
-
2.
[
2]
S
.
O
ua
ha
bi
,
A
.
E
dda
oui
,
E
.
H
.
L
a
br
i
j
i
,
E
.
B
e
nl
a
hm
a
r
,
a
nd
K
.
E
l
G
ue
m
m
a
t
,
“
I
m
pl
e
m
e
nt
a
t
i
on
of
a
nove
l
e
duc
a
t
i
ona
l
m
ode
l
i
ng
a
ppr
oa
c
h
f
or
c
l
oud
c
om
put
i
ng,”
I
nt
e
r
nat
i
onal
J
our
nal
of
E
m
e
r
gi
ng
T
e
c
hnol
ogi
e
s
i
n
L
e
ar
ni
ng
,
vol
.
9,
no.
6,
pp.
49
–
53,
2014,
doi
:
10.3991/
i
j
e
t
.v9i
6.4153.
[
3]
S
.
O
u
a
ha
b
i
,
K
.
E
l
G
u
e
m
m
a
t
,
M
.
A
z
ou
a
z
i
,
a
n
d
S
.
E
l
F
i
l
a
l
i
,
“
A
s
u
r
ve
y
o
f
d
i
s
t
a
nc
e
l
e
a
r
n
i
n
g
i
n
m
o
r
oc
c
o
d
u
r
i
n
g
c
o
vi
d
-
1
9
,”
I
n
d
o
n
e
s
i
an
J
o
u
r
n
a
l
o
f
E
l
e
c
t
r
i
c
a
l
E
n
g
i
ne
e
r
i
n
g
a
n
d
C
o
m
p
u
t
e
r
Sc
i
e
n
c
e
,
v
o
l
.
2
2
,
n
o
.
2,
p
p
.
1
0
8
7
–
10
9
5
,
2
02
1
,
do
i
:
1
0.
1
1
5
91
/
i
j
e
e
c
s
.
v
22
.
i
2
.
pp
1
0
8
7
-
10
9
5
.
[
4]
A
.
A
.
B
a
m
a
ni
k
a
r
,
A
.
P
a
t
i
l
,
D
.
S
i
ngh,
M
.
K
um
bha
r
,
a
nd
V
.
B
a
dhe
,
“
O
nl
i
ne
e
xa
m
por
t
a
l
us
i
ng
m
a
c
hi
ne
l
e
a
r
ni
ng
a
nd
f
a
c
e
de
t
e
c
t
i
on,”
I
nt
e
r
nat
i
onal
J
our
nal
f
o
r
R
e
s
e
a
r
c
h
i
n
A
ppl
i
e
d
Sc
i
e
nc
e
and
E
ngi
n
e
e
r
i
ng
T
e
c
hnol
ogy
,
vol
.
11,
no.
5,
pp.
4624
–
4630,
2023, doi
:
10.22214/
i
j
r
a
s
e
t
.2023.52692.
[
5]
A
.
M
.
S
e
ndur
,
“
A
c
a
de
m
i
c
m
a
l
pr
a
c
t
i
c
e
i
n
t
e
s
t
s
a
nd
e
x
a
m
s
f
r
om
a
n
i
nt
e
r
na
t
i
ona
l
pe
r
s
pe
c
t
i
ve
,”
P
r
z
e
gl
ąd
B
adań
E
duk
ac
y
j
ny
c
h
,
vol
.
36, no. 1, 2022, doi
:
10.12775/
pbe
.2022.007.
[
6]
L
.
C
.
O
.
T
i
ong
a
nd
H
.
J
.
L
e
e
,
“
E
-
c
he
a
t
i
ng
pr
e
ve
nt
i
on
m
e
a
s
ur
e
s
:
de
t
e
c
t
i
on
of
c
he
a
t
i
ng
a
t
onl
i
ne
e
xa
m
i
na
t
i
on
s
us
i
ng
de
e
p
l
e
a
r
ni
ng
a
ppr
oa
c
h
--
a
c
a
s
e
s
t
udy,”
a
r
X
i
v
-
C
om
put
e
r
Sc
i
e
n
c
e
, pp. 1
–
9, 2021.
[
7]
A
.
W
.
M
uz
a
f
f
a
r
,
M
.
T
a
hi
r
,
M
.
W
.
A
nw
a
r
,
Q
.
C
ha
udr
y,
S
.
R
.
M
i
r
,
a
nd
Y
.
R
a
s
he
e
d,
“
A
s
ys
t
e
m
a
t
i
c
r
e
vi
e
w
of
onl
i
ne
e
xa
m
s
s
ol
ut
i
ons
i
n
e
-
l
e
a
r
ni
ng:
t
e
c
hni
que
s
,
t
ool
s
,
a
nd
gl
oba
l
a
dopt
i
on,”
I
E
E
E
A
c
c
e
s
s
,
vol
.
9,
pp.
32689
–
32712,
2021,
doi
:
10.1109/
A
C
C
E
S
S
.2021.3060192.
[
8]
N
.
D
i
l
i
ni
,
A
.
S
e
na
r
a
t
ne
,
T
.
Y
a
s
a
r
a
t
hna
,
N
.
W
a
r
na
j
i
t
h,
a
nd
L
.
S
e
n
e
vi
r
a
t
ne
,
“
C
he
a
t
i
ng
de
t
e
c
t
i
on
i
n
br
ow
s
e
r
-
ba
s
e
d
onl
i
ne
e
xa
m
s
t
hr
ough
e
ye
ga
z
e
t
r
a
c
ki
ng,”
i
n
2021
6t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
I
nf
or
m
at
i
on
T
e
c
hnol
ogy
R
e
s
e
ar
c
h
(
I
C
I
T
R
)
,
I
E
E
E
,
2021,
pp.
1
–
8
, doi
:
10.1109/
I
C
I
T
R
54349.2021.9657277.
[
9]
A
.
J
a
ve
d
a
nd
Z
.
A
s
l
a
m
,
“
A
n
i
nt
e
l
l
i
ge
nt
a
l
a
r
m
ba
s
e
d
vi
s
ua
l
e
y
e
t
r
a
c
ki
ng
a
l
gor
i
t
hm
f
or
c
he
a
t
i
ng
f
r
e
e
e
xa
m
i
na
t
i
on
s
ys
t
e
m
,
”
I
nt
e
r
nat
i
onal
J
our
nal
of
I
nt
e
l
l
i
ge
nt
Sy
s
t
e
m
s
and A
ppl
i
c
at
i
ons
, vol
. 5, no. 10, pp
. 86
–
92, 2013, doi
:
10.5815/
i
j
i
s
a
.2013.10.11.
[
10]
M
.
M
.
M
a
s
ud,
K
.
H
a
ya
w
i
,
S
.
S
.
M
a
t
he
w
,
T
.
M
i
c
ha
e
l
,
a
nd
M
.
E
l
B
a
r
a
c
hi
,
“
S
m
a
r
t
onl
i
ne
e
xa
m
p
r
oc
t
or
i
ng
a
s
s
i
s
t
f
or
c
he
a
t
i
ng
de
t
e
c
t
i
on,”
i
n
A
dv
anc
e
d D
at
a M
i
ni
ng and A
ppl
i
c
at
i
ons
, 2022, pp. 118
–
132
, doi
:
10.1007/
978
-
3
-
030
-
95405
-
5_9.
[
11]
R
.
B
a
w
a
r
i
t
h,
A
.
B
a
s
uha
i
l
,
A
.
F
a
t
t
ouh
,
a
nd
S
.
G
a
m
a
l
e
l
-
D
i
n
,
“
E
-
e
xa
m
c
he
a
t
i
ng
de
t
e
c
t
i
on
s
ys
t
e
m
,”
I
nt
e
r
nat
i
onal
J
our
nal
o
f
A
dv
anc
e
d C
om
put
e
r
Sc
i
e
n
c
e
and A
ppl
i
c
at
i
ons
, vol
. 8, no. 4, 2017, doi
:
10.14569/
i
j
a
c
s
a
.2017.080425.
[
12]
K
. J
a
l
a
l
i
a
nd F
.
N
oor
be
hba
ha
ni
, “
A
n
a
ut
om
a
t
i
c
m
e
t
hod f
or
c
he
a
t
i
ng
de
t
e
c
t
i
on i
n onl
i
ne
e
xa
m
s
by
pr
oc
e
s
s
i
ng
t
he
s
t
ude
nt
s
w
e
bc
a
m
i
m
a
ge
s
,”
i
n
3r
d
C
onf
e
r
e
nc
e
on E
l
e
c
t
r
i
c
al
and C
om
put
e
r
E
ngi
ne
e
r
i
ng T
e
c
hnol
o
gy
(
E
-
T
e
c
h 2017)
,
T
e
hr
a
n, I
r
a
n
, 2017, pp. 1
–
6.
[
13]
M
.
G
e
e
t
ha
,
R
.
S
.
L
a
t
ha
,
S
.
K
.
N
i
ve
t
ha
,
S
.
H
a
r
i
pr
a
s
a
t
h,
S
.
G
ow
t
ha
m
,
a
nd
C
.
S
.
D
e
e
pa
k,
“
D
e
s
i
gn
of
f
a
c
e
de
t
e
c
t
i
on
a
nd
r
e
c
ogni
t
i
o
n
s
ys
t
e
m
t
o
m
oni
t
or
s
t
ud
e
nt
s
dur
i
ng
onl
i
ne
e
x
a
m
i
na
t
i
ons
u
s
i
ng
m
a
c
hi
ne
l
e
a
r
ni
ng
a
l
gor
i
t
hm
s
,”
i
n
2021
I
nt
e
r
nat
i
onal
C
onf
e
r
e
n
c
e
o
n
C
om
put
e
r
C
om
m
uni
c
at
i
on and I
nf
or
m
at
i
c
s
(
I
C
C
C
I
)
, I
E
E
E
, 2021, pp. 1
–
4
, doi
:
10.1109/
I
C
C
C
I
50826.2021.9402553.
[
14]
J
.
O
.
-
R
he
i
n,
B
.
K
üppe
r
s
,
a
nd
U
.
S
c
hr
oe
de
r
,
“
A
n
a
ppl
i
c
a
t
i
on
t
o
di
s
c
ove
r
c
he
a
t
i
ng
i
n
di
gi
t
a
l
e
xa
m
s
,”
i
n
P
r
oc
e
e
di
ngs
of
t
he
18t
h
K
ol
i
C
al
l
i
ng I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on C
om
put
i
ng E
duc
at
i
on R
e
s
e
ar
c
h
, 201
8, pp. 1
–
5
, doi
:
10.1145/
3279720.3279740.
[
15]
F
.
O
z
da
m
l
i
,
A
.
A
l
j
a
r
r
a
h,
D
.
K
a
r
a
goz
l
u,
a
nd
M
.
A
ba
bne
h,
“
F
a
c
i
a
l
r
e
c
ogni
t
i
on
s
ys
t
e
m
t
o
de
t
e
c
t
s
t
ude
nt
e
m
ot
i
ons
a
nd
c
he
a
t
i
ng
i
n
di
s
t
a
nc
e
l
e
a
r
ni
ng,”
Sus
t
ai
nabi
l
i
t
y
,
vol
. 14, no. 20, 2022, doi
:
10.3390/
s
u142013
230.
[
16]
I
.
N
.
Y
ul
i
t
a
,
F
.
A
.
H
a
r
i
z
,
I
. S
ur
ya
na
,
a
nd
A
.
S
. P
r
a
buw
ono,
“
E
duc
a
t
i
ona
l
i
nnov
a
t
i
on
f
a
c
e
d
w
i
t
h
c
ovi
d
-
19:
de
e
p
l
e
a
r
ni
ng f
or
on
l
i
ne
e
xa
m
c
he
a
t
i
ng de
t
e
c
t
i
on,”
E
duc
at
i
on Sc
i
e
nc
e
s
, vol
. 13, no. 2, 2023, doi
:
10.339
0/
e
duc
s
c
i
13020194.
[
17]
O
.
E
l
H
a
m
m
ou
m
i
,
F
.
B
e
nm
a
r
r
a
k
c
h
i
,
N
.
O
u
he
r
r
ou
,
J
.
E
l
K
a
f
i
,
a
n
d
A
.
E
l
H
o
r
e
,
“
E
m
ot
i
o
n
r
e
c
ogn
i
t
i
o
n
i
n
e
-
l
e
a
r
ni
ng
s
ys
t
e
m
s
,”
i
n
2
018
6t
h
I
nt
e
r
na
t
i
on
al
C
on
f
e
r
e
nc
e
on
M
ul
t
i
m
e
d
i
a C
om
put
i
n
g
and
Sy
s
t
e
m
s
(
I
C
M
C
S
)
, 2
018
, p
p.
1
–
6
,
do
i
:
1
0.1
10
9/
I
C
M
C
S
.2
01
8.8
52
587
2.
[
18]
M
.
A
.
O
z
de
m
i
r
,
B
.
E
l
a
goz
,
A
.
A
l
a
ybe
yogl
u,
R
.
S
a
di
ghz
a
de
h,
a
nd
A
.
A
ka
n,
“
R
e
a
l
t
i
m
e
e
m
ot
i
on
r
e
c
ogni
t
i
on
f
r
om
f
a
c
i
a
l
e
xpr
e
s
s
i
ons
us
i
ng
c
nn
a
r
c
hi
t
e
c
t
ur
e
,”
i
n
2019
M
e
di
c
al
T
e
c
hnol
ogi
e
s
C
ongr
e
s
s
(
T
I
P
T
E
K
N
O
)
,
I
E
E
E
,
2019,
pp.
1
–
4
,
do
i
:
10.1109/
T
I
P
T
E
K
N
O
.2019.8895215.
[
19]
E
.
P
r
a
na
v,
S
.
K
a
m
a
l
,
C
.
S
.
C
ha
ndr
a
n,
a
nd
M
.
H
.
S
upr
i
ya
,
“
F
a
c
i
a
l
e
m
ot
i
on
r
e
c
o
gni
t
i
on
us
i
ng
de
e
p
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k,
”
i
n
2020
6t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
A
dv
anc
e
d
C
om
put
i
ng
and
C
om
m
uni
c
at
i
on
Sy
s
t
e
m
s
,
I
C
A
C
C
S
2020
,
2020,
pp.
317
–
320
,
doi
:
10.1109/
I
C
A
C
C
S
48705.2020.9074302.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
2
,
A
pr
il
2025
:
843
-
852
852
[
20]
A
.
Z
ba
i
da
,
M
.
K
oda
d,
Y
.
M
oha
m
e
d,
a
nd
N
.
B
e
nm
ous
s
a
,
“
D
e
e
p
l
e
a
r
ni
ng
a
nd
e
m
ot
i
on
r
e
c
ogni
t
i
on
f
o
r
a
n
e
-
l
e
a
r
ni
ng
pl
a
t
f
or
m
,”
R
e
s
e
ar
c
h G
at
e
, pp. 1
–
11, 2023.
[
21]
W
.
L
u,
J
.
L
i
,
J
.
W
a
ng,
a
nd
L
.
Q
i
n,
“
A
C
N
N
-
bi
L
S
T
M
-
A
M
m
e
t
hod
f
or
s
t
oc
k
pr
i
c
e
pr
e
di
c
t
i
on,”
N
e
ur
al
C
om
put
i
ng
and
A
ppl
i
c
at
i
ons
, vol
. 33, no. 10, pp. 4741
–
4753, 2021, doi
:
10.1007/
s
00521
-
020
-
0
5532
-
z.
[
22]
M
. C
os
kun,
A
. U
c
a
r
, O
. Y
i
l
di
r
i
m
, a
nd Y
. D
e
m
i
r
, “
F
a
c
e
r
e
c
ogni
t
i
on ba
s
e
d on c
o
nvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k,”
i
n
2017 I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on M
ode
r
n E
l
e
c
t
r
i
c
al
and E
ne
r
gy
S
y
s
t
e
m
s
(
M
E
E
S)
, N
ov. 2017, pp.
376
–
379
, doi
:
10.1109/
M
E
E
S
.2017.8248937.
[
23]
V
. A
t
l
i
ha
a
nd D
. S
e
s
ok, “
C
om
pa
r
i
s
on
of
VGG
a
nd r
e
s
ne
t
us
e
d
a
s
e
nc
ode
r
s
f
or
i
m
a
ge
c
a
pt
i
oni
ng,”
i
n
2020 I
E
E
E
O
pe
n C
onf
e
r
e
nc
e
of
E
l
e
c
t
r
i
c
al
, E
l
e
c
t
r
oni
c
and I
nf
or
m
at
i
on Sc
i
e
nc
e
s
(
e
St
r
e
am
)
, I
E
E
E
, 2020, pp. 1
–
4
, doi
:
10.1109/
e
S
t
r
e
a
m
50540.2020.9108880.
[
24]
A
. K
. D
ube
y a
nd
V
. J
a
i
n, “
A
ut
om
a
t
i
c
f
a
c
i
a
l
r
e
c
ogni
t
i
on us
i
ng
V
G
G
16 ba
s
e
d t
r
a
ns
f
e
r
l
e
a
r
ni
ng m
ode
l
,”
J
ou
r
nal
of
I
nf
or
m
at
i
on an
d
O
pt
i
m
i
z
at
i
on Sc
i
e
nc
e
s
, vol
. 41, no. 7, pp. 1589
–
1596, 2020, doi
:
10.1080/
0252
2667.2020.1809126.
[
25]
N
. H
a
s
a
n, Y
.
B
a
o,
A
. S
ha
w
on,
a
nd
Y
. H
ua
ng,
“
D
e
n
s
e
N
e
t
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
ks
a
ppl
i
c
a
t
i
on f
or
pr
e
di
c
t
i
ng
c
ovi
d
-
19 us
i
n
g
c
t
i
m
a
ge
,”
SN
C
om
put
e
r
Sc
i
e
nc
e
, vol
. 2, no. 5, 2021, doi
:
10.1007/
s
42979
-
021
-
00782
-
7.
[
26]
H
.
A
bd
o
,
K
.
M
.
A
m
i
n
,
a
n
d
A
.
M
.
H
a
m
a
d
,
“
F
a
l
l
d
e
t
e
c
t
i
on
ba
s
e
d
o
n r
e
t
i
na
n
e
t
a
nd
m
o
b
i
l
e
ne
t
c
o
n
v
o
l
ut
i
o
n
a
l
n
e
u
r
a
l
ne
t
w
o
r
k
s
,”
i
n
2
0
2
0 1
5
t
h
I
n
t
e
r
na
t
i
o
n
a
l
C
on
f
e
r
e
nc
e
o
n
C
o
m
p
ut
e
r
E
n
g
i
ne
e
r
i
n
g
a
n
d
Sy
s
t
e
m
s
(
I
C
C
E
S
)
,
2
02
0
,
pp
.
1
–
7
,
do
i
:
1
0.
1
1
0
9
/
I
C
C
E
S
51
5
6
0
.2
0
2
0
.9
3
3
4
57
0
.
[
27]
Ü
.
A
t
i
l
a
,
M
.
U
ç
a
r
,
K
.
A
kyol
,
a
nd
E
.
U
ç
a
r
,
“
P
l
a
nt
l
e
a
f
di
s
e
a
s
e
c
l
a
s
s
i
f
i
c
a
t
i
on
us
i
ng
e
f
f
i
c
i
e
nt
ne
t
de
e
p
l
e
a
r
ni
ng
m
ode
l
,”
E
c
ol
ogi
c
al
I
nf
or
m
at
i
c
s
, vol
. 61, 2021, doi
:
10.1016/
j
.e
c
oi
nf
.2020.101182.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
S
Sara
Ouahabi
is
a
Habilitated
Professor
(PH)
at
Departm
ent
of
Ma
thematics
and
Computer
Science
of
Hassan
II
University
and
member
of
Computer
Science
and
Information
Proce
ssing
L
aboratory
at
Faculty
of
Scienc
e
Ben
M’sik.
Her
res
earch
includes
artificial
intelligence
,
computer
security,
semantic
web,
E
-
learning,
co
mputer
communications
(
n
etworks)
,
and
the
internet
of
things
(IoT).
She
conducts
cutting
-
edg
e
research
in
these
areas
while
actively
engaging
in
higher
education
to
educate
the
next
generation
of
computer
professionals. She can be contac
ted at email: s3.ouahabi@
gmail.com.
Rihab
Aboudihaj
received
the
B.Sc.
degree
in
mathematical
sciences
and
computer
science
from
the
Faculty
of
Sciences
Ben
M’Sick,
H
assan
II
Universi
ty
of
Casablan
ca,
Morocc
o,
in
2022,
and
the
M.Sc.
degree
in
data
science
a
nd
big
data
from
Hassan
II
University
of
Casablanca,
in
2024.
Her
research
interests
inclu
de
the
exploration
and
analysis
of
data,
deep
learning
and
machine
learning
algorit
hms
,
and
computer
visio
n
applicati
ons.
She
focuses
particular
ly
on
developi
ng
and
improvi
ng
classifi
cation
and
prediction
models,
as
well
as
visualizing
complex
data
to
uncover
use
ful
insig
hts
and
patterns.
She ca
n be c
ontact
ed at
email:
aboudih
ajriha
b@
gmail.c
om.
Nawal
Sael
received
the
engineering
degree
in
software
eng
ineering
from
ENSIAS,
Morocco,
in
2002.
She
has
been
a
teacher
-
researcher
,
sin
ce
2012,
an
Authorized
Profes
sor,
since
2014,
and
a
professor
of
higher
education
wi
th
the
Department
of
Mathematics
and
Computer
Science,
Faculty
of
Sciences
Ben
M’Sick,
Hassan
II
University
of
Casablan
ca,
Casablan
ca,
since
2020.
Her
resea
rch
interes
ts
include
data
mining,
educat
ional
data
mining,
machine
learning,
deep
learning,
and
the
internet
of
thin
gs
.
She
can
be
contacted
at email
: nawal.s
ael@
univh2c.
ma.
Kamal El Guemma
t
is Professor
at ENSET, H
assan
II Unive
rsity
of Casabla
nca
.
He
is
the
coordinator
of
the
engineering
cycle
(computer
engineer
ing,
cyber
security
and
digital
trust
II
-
CCN).
He
has
several
skills
in
s
emantic
indexing
,
search
engines,
natural
language
processing
,
digital
learning,
web
mining,
text
mining,
info
rmation
extractio
n,
data
science,
artificial
intelligence,
networking,
data
mining,
ontologies,
m
achine
learning,
and
big
data
. He can be contac
ted at email: k.elguemmat@
gmail.com.
Evaluation Warning : The document was created with Spire.PDF for Python.