I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
14
, N
o.
4
,
A
ugus
t
20
25
, pp.
3172
~
3181
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
4
.pp
3172
-
3181
3172
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
Im
p
r
ovi
n
g t
h
e
t
r
an
sf
e
r
l
e
ar
n
i
n
g f
or
b
at
i
k
b
e
su
r
e
k
t
e
xt
i
l
e
m
ot
i
f
c
l
ass
i
f
i
c
at
i
on
M
ar
is
s
a U
t
a
m
i
1
,
3
, E
r
m
at
it
a E
r
m
at
it
a
1,
2
, A
b
d
ia
n
s
ah
A
b
d
ia
n
s
ah
1,
2
1
D
oc
t
or
a
l
P
r
ogr
a
m
i
n E
ng
i
ne
e
r
i
ng, F
a
c
ul
t
y of
E
ngi
ne
e
r
i
ng, S
r
i
w
i
j
a
ya
U
ni
ve
r
s
i
t
y
,
P
a
l
e
m
ba
ng, I
ndone
s
i
a
3
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
S
c
i
e
nc
e
, F
a
c
ul
t
y of
C
om
put
e
r
S
c
i
e
nc
e
,
S
r
i
w
i
j
a
ya
U
ni
ve
r
s
i
t
y
, P
a
l
e
m
ba
ng, I
ndone
s
i
a
3
D
e
pa
r
t
m
e
nt
of
I
nf
or
m
a
t
i
on
S
ys
t
e
m
, F
a
c
ul
t
y of
E
ngi
ne
e
r
i
ng, U
ni
ve
r
s
i
t
a
s
M
uha
m
m
a
di
ya
h B
e
ngkul
u, B
e
ngkul
u, I
ndone
s
i
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
S
e
p
7
,
2024
R
e
vi
s
e
d
M
a
y
19
,
2025
A
c
c
e
pt
e
d
J
un
8
,
2025
This
proposed
research
discussion
is
a
new
combination
model
for
classify
ing
batik
besurek
fabric
from
the
implementation
transfer
learning
with
mixed
contrast
enhancement,
activation
function,
and
opt
imizer
method.
The
size
of
the
batik
besurek
fabric
motif
image
as
an
input
image
is
250
×
250
with
three
channels
consisting
of
red,
green,
and
blue
totaling
five
classes,
namely
kaligra
fi
,
rafflesia,
burung
kuau
,
relung
pak
u
and
rembula
n
.
All
images
in
the
dataset
will
be
divided
into
trai
n
data
(1540
images),
validate
data
(380
images)
,
and
test
data
(480
image
s)
that
are
taken
directly
from
the
batik
store
in
Bengkulu.
The
division
method
used
is
stratified
random
sampling
to
take
all
the
data,
shuffles
i
t,
and
divides
the
data
sets
for
each
class.
Based
on
the
experiment
r
esults,
ResNet50
obtained
the
best
perfor
mance
compare
d
to
Mobile
NetV2,
InceptionV3
,
and
VGG16,
with
a
training
accuracy
of
99.60%,
a
vali
dation
accuracy
of
97.44%
,
and
a
testing
accuracy
of
98.12%.
In
the
impro
vement
experiment
phase, t
he ResNet
50 model
with A
dam opt
i
mizer,
rectified
linear
unit
(
ReLU
)
activati
on
function
and
contrast
limit
ed
adaptive
hist
ogram
equalizati
on
(
CLAHE
)
as
the
contrast
enhancement
method
obtain
ed
the
highest
test
accuracy
(98.75%),
showing
that
CLAH
E
was
very
effec
tive
in
improving per
formance
on batik
b
esurek dat
a.
K
e
y
w
o
r
d
s
:
A
c
ti
va
ti
on f
unc
ti
on
B
a
ti
k be
s
ur
e
k
C
ont
r
a
s
t
e
nha
nc
e
m
e
nt
O
pt
im
iz
e
r
T
r
a
ns
f
e
r
l
e
a
r
ni
ng
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
:
E
r
m
a
ti
ta
E
r
m
a
ti
ta
D
oc
to
r
a
l
P
r
ogr
a
m
i
n E
ngi
ne
e
r
in
g,
F
a
c
ul
ty
of
E
ngi
ne
e
r
in
g,
S
r
iw
ij
a
ya
U
ni
ve
r
s
it
y
P
a
le
m
ba
ng, I
ndone
s
ia
E
m
a
il
:
e
r
m
a
ti
ta
@
uns
r
i.
a
c
.i
d
1.
I
N
T
R
O
D
U
C
T
I
O
N
I
ndone
s
ia
n
ba
ti
k
w
a
s
in
a
ugur
a
te
d
a
s
a
he
r
it
a
ge
of
hum
a
ni
ty
a
nd
in
ta
ngi
bl
e
c
ul
tu
r
e
a
nd
ha
s
be
e
n
r
e
c
ogni
z
e
d
by
t
he
U
ni
te
d
N
a
ti
ons
E
duc
a
ti
ona
l,
S
c
ie
nt
if
ic
,
a
nd
C
ul
tu
r
a
l
O
r
ga
ni
z
a
ti
on
(
U
N
E
S
C
O
)
a
s
th
e
in
te
ll
e
c
tu
a
l
r
ig
ht
of
th
e
I
ndone
s
ia
n
na
ti
on
on
O
c
to
be
r
2,
2009.
T
he
de
f
in
it
io
n
of
b
a
ti
k
is
a
n
il
lu
s
tr
a
te
d
c
lo
th
th
a
t
is
e
xpl
ic
it
ly
m
a
de
by
w
r
it
in
g
or
e
xpl
a
in
in
g
th
e
ni
ght
on
th
e
c
lo
th
;
a
f
te
r
th
a
t,
it
is
e
xpl
ic
it
ly
m
a
de
by
w
r
it
in
g
or
e
xpl
a
in
in
g
th
e
ni
ght
on
th
e
c
lo
th
,
a
nd
th
e
p
r
oc
e
s
s
in
g
goe
s
th
r
ough
a
s
pe
c
if
ic
pr
oc
e
s
s
[
1]
,
[
2]
.
I
ndone
s
ia
i
s
a
c
ount
r
y t
ha
t
c
ons
is
ts
of
di
f
f
e
r
e
nt
s
e
lf
-
e
vi
de
nc
e
a
nd ha
s
e
xt
r
a
or
di
na
r
y s
oc
ia
l
di
ve
r
s
it
y;
c
ul
tu
r
e
i
s
th
e
r
e
s
ul
t
of
m
in
d a
nd e
ne
r
gy i
n t
he
f
or
m
of
c
r
e
a
ti
on, c
ha
r
it
y, a
nd t
a
s
te
t
ha
t
ha
s
hum
a
n t
e
nde
nc
ie
s
[
3]
, [
4]
.
O
ne
of
th
e
ba
ti
k
ha
ndi
c
r
a
f
t
in
dus
tr
y
pr
oduc
in
g
a
r
e
a
s
w
it
h
it
s
c
ha
r
a
c
te
r
is
ti
c
s
i
s
ba
ti
k
c
r
a
f
ts
f
ound
in
B
e
ngkulu C
it
y, w
hi
c
h i
s
f
a
m
ous
f
o
r
ba
ti
k
be
s
ur
e
k
.
B
a
ti
k be
s
ur
e
k
is
a
t
r
a
di
ti
ona
l
c
r
a
f
t
th
a
t
ha
s
l
ong de
ve
lo
pe
d
a
nd
is
a
le
ga
c
y
of
th
e
a
nc
e
s
to
r
s
of
th
e
B
e
ngkulu
pe
opl
e
f
or
ge
ne
r
a
ti
ons
.
T
hi
s
ba
ti
k
be
s
ur
e
k
c
ont
a
in
s
th
e
m
e
a
ni
ng of
a
l
e
tt
e
r
or
w
r
it
in
g
[
5]
–
[
7]
.
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
I
m
pr
ov
in
g t
he
t
r
ans
fe
r
l
e
ar
ni
ng f
o
r
bat
ik
be
s
u
r
e
k
t
e
x
ti
le
m
ot
if
c
la
s
s
if
ic
at
io
n
(
M
ar
is
s
a U
ta
m
i
)
3173
B
e
s
ur
e
k
c
lo
th
w
a
s
onc
e
onl
y
u
s
e
d
in
r
e
li
gi
ous
r
it
ua
l
c
e
r
e
m
oni
e
s
in
th
e
B
e
ngkulu
r
e
gi
on;
a
lo
ng
w
it
h
th
e
ti
m
e
s
,
th
e
u
s
e
a
nd
d
e
s
ig
n
of
ba
ti
k
be
s
ur
e
k
m
ot
if
s
unde
r
w
e
n
t
m
ode
r
ni
z
a
ti
on.
B
a
ti
k
be
s
ur
e
k
m
ot
if
s
to
ta
li
ng
f
iv
e
f
a
m
ous
m
ot
if
s
,
na
m
e
ly
k
al
ig
r
af
i
,
r
a
f
f
le
s
ia
,
bur
ung
k
uau
,
r
e
lu
ng
pak
u
,
a
nd
r
e
m
bul
an
.
B
a
ti
k
be
s
ur
e
k
is
a
ls
o
a
n
a
r
t,
s
o
one
b
e
s
ur
e
k
c
lo
th
m
ot
if
c
a
n
be
c
r
e
a
te
d,
of
c
our
s
e
,
by
a
r
ti
s
a
ns
w
ho
unde
r
s
ta
nd
th
e
m
ot
if
.
T
hu
s
,
one
ba
ti
k c
lo
th
m
ot
if
w
il
l
not
onl
y ha
ve
one
s
ha
pe
but
w
il
l
ha
ve
m
a
ny s
im
il
a
r
s
ha
pe
s
[
8]
–
[
10]
.
D
ig
it
a
l
im
a
ge
pr
oc
e
s
s
in
g
is
c
a
r
r
ie
d
out
on
im
a
ge
s
to
obt
a
in
c
e
r
ta
in
de
s
ir
e
d
r
e
s
ul
ts
.
U
s
in
g
di
gi
ta
l
im
a
ge
pr
oc
e
s
s
in
g, w
e
c
a
n c
la
s
s
if
y
s
e
ve
r
a
l
s
im
il
a
r
ba
ti
k i
m
a
ge
s
[
11]
. T
hi
s
m
e
th
od c
a
n be
one
w
a
y t
o
s
ol
ve
t
he
pr
obl
e
m
of
th
e
in
tr
oduc
ti
on
of
ba
ti
k
be
s
ur
e
k
m
ot
if
s
[
12]
.
P
r
e
vi
ous
r
e
s
e
a
r
c
he
r
s
ha
v
e
r
e
s
e
a
r
c
he
d
di
gi
ta
l
im
a
g
e
pr
oc
e
s
s
in
g
to
in
tr
oduc
e
ba
ti
k
f
a
br
ic
m
ot
if
s
by
c
om
bi
ni
ng s
e
ve
r
a
l
di
gi
ta
l
im
a
ge
pr
oc
e
s
s
in
g
m
e
th
ods
. R
e
s
e
a
r
c
h
c
onduc
te
d
by
K
us
a
nt
i
a
nd
S
upr
a
pt
o
[
13]
is
r
e
la
te
d
to
th
e
a
na
ly
s
is
of
s
e
ve
n
c
la
s
s
e
s
of
S
ur
a
ka
r
ta
ba
ti
k,
na
m
e
ly
ka
w
ung
m
ot
if
,
s
id
o
m
ukt
i
m
ot
if
,
tr
unt
um
m
ot
if
,
s
a
w
a
t
m
ot
if
,
s
a
tr
io
m
a
na
h
m
ot
if
,
pa
r
a
ng
m
ot
if
,
a
nd
s
e
m
e
n
r
a
nt
e
m
ot
if
.
T
he
da
ta
us
e
d
i
s
100
im
a
ge
s
di
vi
de
d
in
to
70
tr
a
in
in
g
da
ta
a
nd
30
te
s
t
da
ta
.
T
he
r
e
s
ul
ts
s
how
e
d
th
a
t
th
e
a
c
c
ur
a
c
y r
a
te
of
O
ts
u a
nd
C
a
nny wa
s
93%
.
A
ndr
ia
n
e
t
al
.
[
1
4]
c
on
du
c
t
e
d
r
e
s
e
a
r
c
h
r
e
l
a
t
e
d t
o
t
he
c
l
a
s
s
if
i
c
a
ti
o
n
of
L
a
m
p
un
g
b
a
ti
k
m
o
ti
f
s
c
on
s
i
s
ti
n
g
of
j
ung
a
gu
ng,
s
ig
e
r
ke
m
b
a
ng
c
e
n
gk
e
h
,
s
i
ge
r
r
a
tu
a
gun
g,
a
nd
s
e
m
ba
gi
.
T
o
r
e
c
o
gni
z
e
t
he
L
a
m
pu
ng
b
a
ti
k
m
ot
if
,
th
e
gr
a
y l
e
v
e
l
c
o
-
oc
c
ur
r
e
n
c
e
m
a
tr
i
x
(
G
L
C
M
)
f
e
a
tu
r
e
w
a
s
e
xt
r
a
c
t
e
d,
a
n
d k
-
n
e
a
r
e
s
t
n
e
i
ghb
or
(
k
-
N
N
)
to
obt
a
i
n t
h
e
be
s
t
a
c
c
ur
a
c
y
a
c
hi
e
v
e
d
a
t
a
le
ve
l
of
9
7.9
6%
.
R
e
s
e
a
r
c
h
c
on
duc
te
d
by
G
ir
s
a
ng
a
n
d
M
uh
a
t
hi
r
[
1
5]
i
s
th
e
c
la
s
s
if
ic
a
ti
on
of
ba
ti
k
m
ot
if
s
be
c
a
us
e
it
is
c
ha
ll
e
ngi
ng
to
id
e
nt
if
y
ba
ti
k
m
ot
i
f
s
in
I
ndone
s
ia
.
S
o,
it
ta
ke
s
c
la
s
s
if
ic
a
ti
on
w
it
h
pr
e
c
is
e
a
c
c
ur
a
c
y
to
m
a
ke
it
e
a
s
ie
r
to
r
e
c
og
ni
z
e
ba
ti
k
pa
tt
e
r
ns
e
a
s
il
y.
T
hi
s
s
tu
dy
us
e
s
th
e
hi
s
to
gr
a
m
of
t
he
or
ie
nt
e
d g
r
a
di
e
nt
(
H
O
G
)
a
s
a
c
ha
r
a
c
te
r
is
ti
c
e
xt
r
a
c
ti
on pr
oc
e
s
s
t
o obta
in
t
he
c
ha
r
a
c
te
r
is
ti
c
s
of
ba
ti
k
m
ot
if
de
ns
it
y
a
nd
m
ul
ti
la
ye
r
pe
r
c
e
pt
r
on
a
s
th
e
c
la
s
s
if
ic
a
ti
on
m
e
th
od.
T
he
a
c
c
ur
a
c
y
r
a
te
obt
a
in
e
d
in
th
e
s
tu
dy w
a
s
83.4%
[
15]
.
R
e
s
e
a
r
c
h
c
onduc
te
d
by
R
i
s
ki
e
t
al
.
[
16]
is
r
e
la
te
d
to
th
e
i
nt
r
oduc
ti
on
of
M
a
dur
a
ba
ti
k
m
ot
if
s
.
T
he
c
l
a
s
s
of
M
a
dur
e
s
e
ba
ti
k
m
ot
if
s
c
on
s
is
ts
of
s
a
to
m
pok
f
lo
w
e
r
,
m
a
nuk
pot
e
r
,
br
oke
n
b
e
li
ng,
s
e
a
w
e
e
d,
a
nd
s
e
ka
r
ja
ga
t
.
T
h
e
G
L
C
M
m
e
th
od
i
s
u
s
e
d
to
e
xt
r
a
c
t
im
a
g
e
f
e
a
tu
r
e
s
,
a
nd
th
e
b
a
c
kpr
opa
ga
ti
on
a
lg
or
it
hm
is
us
e
d
f
or
c
la
s
s
if
ic
a
ti
on.
U
s
in
g
th
e
G
L
C
M
m
e
th
od,
th
e
a
c
c
ur
a
c
y
of
th
e
e
xp
e
r
im
e
nt
r
e
a
c
he
d
98%
in
th
e
te
s
ti
ng
pr
oc
e
s
s
.
R
e
s
e
a
r
c
h
c
onduc
te
d
by
S
e
na
r
a
th
na
a
nd
R
a
ja
k
a
r
una
[
17]
us
e
s
l
oc
a
l
bi
na
r
y
pa
tt
e
r
n
(
L
B
P
)
a
s
a
ve
c
to
r
of
te
xt
ur
e
f
e
a
tu
r
e
s
,
H
u
m
om
e
nt
in
va
r
ia
nt
s
(
H
I
M
)
f
or
th
e
e
xt
r
a
c
ti
on
of
s
ha
p
e
f
e
a
tu
r
e
s
,
a
nd
G
L
C
M
f
or
th
e
e
xt
r
a
c
ti
on of
t
e
xt
ur
e
f
e
a
tu
r
e
s
. T
he
r
e
s
e
a
r
c
h da
ta
s
e
t
u
s
e
d i
n t
hi
s
s
tu
dy c
ons
is
te
d of
300 im
a
ge
s
w
it
h 50 c
la
s
s
e
s
.
T
he
da
ta
a
ugm
e
nt
a
ti
on
m
e
th
od
is
a
ppl
ie
d
to
th
e
pr
im
a
r
y
da
ta
s
e
t
a
nd
ge
ne
r
a
te
s
1200
ne
w
im
a
ge
s
w
it
h
th
e
s
a
m
e
num
be
r
of
c
la
s
s
e
s
.
T
e
s
t
s
c
e
n
a
r
io
s
c
om
pa
r
e
th
e
a
c
c
ur
a
c
y
be
twe
e
n
th
e
or
ig
in
a
l
a
nd
a
ddi
ti
ona
l
da
ta
a
t
a
n
80:
20
r
a
ti
o
f
or
tr
a
in
in
g
a
nd
te
s
ti
ng
da
ta
.
T
hi
s
s
tu
dy
c
la
s
s
if
ie
s
ba
ti
k
im
a
ge
s
by
a
ppl
yi
ng
de
e
p
le
a
r
ni
ng
us
in
g
th
e
R
e
s
N
e
t
m
e
th
od w
it
h a
n a
c
c
ur
a
c
y p
e
r
f
or
m
a
nc
e
of
96%
.
R
e
s
e
a
r
c
h
w
a
s
c
onduc
t
e
d
to
in
tr
oduc
e
s
ix
ba
ti
k
m
ot
if
s
f
r
om
va
r
io
us
r
e
gi
ons
in
I
ndone
s
i
a
.
T
he
ba
ti
k
m
ot
if
s
s
tu
di
e
d
in
c
lu
de
ba
nj
i
m
ot
i
f
s
, c
e
pl
ok mot
if
s
, ka
w
ung
m
ot
if
s
, m
e
ga
m
e
ndung mot
if
s
, pa
r
a
ng
m
ot
if
s
, a
nd
s
e
ka
r
ja
ga
d
m
ot
if
s
.
T
he
r
e
s
e
a
r
c
h
da
ta
s
e
t
c
on
s
is
te
d
of
994
im
a
ge
s
di
vi
de
d
in
to
s
ix
c
la
s
s
e
s
.
T
h
e
r
a
ti
o
of
th
e
di
vi
s
io
n
of
th
e
tr
a
in
in
g
da
ta
s
e
t
a
nd
th
e
te
s
t
da
ta
s
e
t
us
e
d
is
8:
2.
T
he
r
e
s
ul
ts
of
e
xpe
r
im
e
nt
s
on
th
e
te
s
t
da
ta
s
how
e
d
th
a
t
th
e
a
lg
or
it
hm
pr
oduc
e
d
e
xc
e
ll
e
nt
p
e
r
f
or
m
a
nc
e
,
w
hi
c
h
w
a
s
de
m
on
s
tr
a
te
d
w
it
h
94%
a
c
c
ur
a
c
y
us
in
g
th
e
D
e
n
s
e
N
e
t
a
r
c
hi
te
c
tu
r
e
.
I
n
th
is
s
tu
dy,
th
e
da
ta
a
u
gm
e
nt
a
ti
on
m
e
th
od
w
a
s
a
ppl
ie
d
to
pr
ovi
de
va
r
ia
ti
ons
in
tr
a
in
in
g
da
ta
a
nd
pr
e
ve
nt
ove
r
f
it
ti
ng
[
18]
.
B
a
s
e
d
on
th
e
r
e
s
e
a
r
c
h,
m
os
t
pr
e
vi
ous
r
e
s
e
a
r
c
h
u
s
in
g
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
m
e
th
ods
s
uc
h
a
s
H
O
G
,
L
B
P
,
s
c
a
le
-
in
va
r
ia
nt
f
e
a
tu
r
e
tr
a
ns
f
or
m
(
S
I
F
T
)
,
m
om
e
nt
in
va
r
ia
nt
s
(
M
I
)
,
G
L
C
M
f
or
im
pr
ovi
ng
c
la
s
s
if
ic
a
ti
on
pe
r
f
or
m
a
nc
e
.
H
ow
e
ve
r
,
th
is
r
e
s
e
a
r
c
h
a
tt
e
m
pt
e
d
to
im
pr
ove
th
e
c
la
s
s
if
ic
a
ti
on
m
ode
l
by
e
x
a
m
in
in
g
s
e
ve
r
a
l
m
e
th
od
s
,
in
c
lu
di
n
g
c
ont
r
a
s
t
e
nha
nc
e
m
e
nt
,
a
c
ti
va
ti
on
f
unc
ti
on
,
a
nd opti
m
iz
e
r
.
T
he
a
ppr
opr
ia
te
opt
im
iz
e
r
te
c
hni
que
is
in
s
tr
um
e
nt
a
l
in
tr
a
n
s
f
e
r
le
a
r
ni
ng
be
c
a
us
e
of
it
s
a
bi
li
ty
to
a
dj
us
t
th
e
le
a
r
ni
ng
r
a
te
a
d
a
pt
iv
e
ly
,
f
a
s
te
r
c
onv
e
r
ge
nc
e
,
a
nd
be
tt
e
r
gr
a
di
e
nt
m
a
na
ge
m
e
nt
[
19]
–
[
21]
.
T
hi
s
r
e
s
e
a
r
c
h
w
il
l
a
l
s
o
de
te
r
m
in
e
th
e
a
ppr
opr
ia
te
a
c
ti
va
ti
on
f
unc
t
io
n
s
o
th
a
t
th
e
tr
a
n
s
f
e
r
le
a
r
ni
ng
m
ode
l
c
a
n
pr
ovi
de
th
e
ne
c
e
s
s
a
r
y
non
-
li
ne
a
r
it
y,
s
ol
ve
th
e
pr
obl
e
m
of
di
s
a
ppe
a
r
in
g
gr
a
di
e
nt
s
,
im
pr
ove
c
om
put
in
g
e
f
f
ic
ie
nc
y,
a
nd
a
c
c
e
le
r
a
te
tr
a
in
in
g
c
onve
r
ge
n
c
e
[
22]
–
[
24]
.
I
n
a
ddi
ti
on,
c
ont
r
a
s
t
e
nha
nc
e
m
e
nt
in
ba
ti
k
be
s
ur
e
k
da
ta
is
a
ls
o
di
s
c
u
s
s
e
d
b
e
c
a
us
e
m
a
ny
c
a
s
e
s
of
da
ta
s
e
ts
in
th
e
f
ie
ld
ha
ve
poor
c
ont
r
a
s
t
be
c
a
us
e
th
e
im
pl
e
m
e
nt
a
ti
on w
il
l
be
l
a
te
r
on i
m
a
ge
s
t
a
ke
n f
r
om
i
ndoor
s
[
25]
–
[
27]
.
T
hi
s
s
tu
dy
is
di
vi
de
d
in
to
th
e
tr
a
ns
f
e
r
le
a
r
ni
ng
e
xpe
r
im
e
nt
pha
s
e
a
nd
th
e
tr
a
ns
f
e
r
le
a
r
ni
ng
im
pr
ove
m
e
nt
e
xpe
r
im
e
nt
pha
s
e
.
T
hi
s
r
e
s
e
a
r
c
h
c
a
n
be
a
r
e
f
e
r
e
nc
e
f
or
th
e
be
s
t
c
om
bi
na
ti
on
m
ode
l
b
a
s
e
d
on
c
ont
r
a
s
t
e
nha
nc
e
m
e
nt
,
a
c
ti
va
ti
on
f
unc
ti
on
a
nd
opt
im
iz
e
r
f
or
c
la
s
s
if
yi
ng
th
e
ba
ti
k
te
xt
il
e
m
ot
if
.
T
he
e
xpe
r
im
e
nt
w
a
s
c
a
r
r
ie
d
out
f
our
t
im
e
s
us
in
g
di
f
f
e
r
e
nt
tr
a
ns
f
e
r
le
a
r
ni
ng
m
ode
ls
,
na
m
e
ly
M
obi
le
N
e
tV2,
R
e
s
N
e
t5
0,
I
nc
e
pt
io
nV
3
a
nd
V
G
G
16.
T
he
a
r
c
hi
te
c
tu
r
e
of
e
a
c
h
m
ode
l
f
ol
lo
w
s
th
e
m
ode
l
a
r
c
hi
te
c
tu
r
e
in
th
e
pr
e
vi
ous
s
tu
dy w
hi
c
h us
e
d t
h
e
s
a
m
e
m
od
e
l
to
c
la
s
s
if
y t
he
ba
ti
k da
ta
s
e
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
4
,
A
ugus
t
20
25
:
3172
-
3181
3174
2.
M
E
T
H
O
D
B
a
ti
k
be
s
ur
e
k
is
a
ty
pe
of
ba
ti
k
m
ot
if
w
it
h
a
di
s
ti
nc
ti
ve
pa
tt
e
r
n
w
it
h
A
r
a
bi
c
a
c
c
e
nt
s
f
or
a
ni
m
a
l
s
a
nd
pl
a
nt
s
li
vi
ng
in
B
e
ngkulu.
T
hi
s
im
a
g
e
c
l
a
s
s
if
ic
a
ti
on
us
e
s
a
tr
a
ns
f
e
r
le
a
r
ni
ng
m
ode
l
in
th
is
s
tu
dy
th
a
t
ut
il
iz
e
s
tr
a
ns
f
e
r
l
e
a
r
ni
ng mode
ls
t
r
a
in
e
d on la
r
ge
da
ta
s
e
ts
s
u
c
h a
s
I
m
a
g
e
N
e
t.
T
he
a
dv
a
nt
a
ge
of
t
hi
s
t
r
a
ns
f
e
r
l
e
a
r
ni
ng i
s
th
a
t
it
a
ll
ow
s
th
e
m
ode
l
to
ut
il
iz
e
th
e
knowle
dg
e
it
ha
s
le
a
r
ne
d
to
r
e
c
ogni
z
e
c
om
m
on
vi
s
u
a
l
f
e
a
tu
r
e
s
.
I
n
th
is
s
tu
dy,
th
e
da
ta
s
e
t
us
e
d
w
a
s
ba
ti
k
be
s
ur
e
k
. T
hi
s
ba
ti
k
im
a
ge
is
c
ol
le
c
te
d
di
r
e
c
tl
y
us
in
g
m
obi
le
c
a
m
e
r
a
s
f
r
om
va
r
io
us
lo
c
a
ti
ons
,
in
c
lu
di
ng
ba
ti
k
be
s
ur
e
k
S
a
r
i
R
a
s
a
S
to
r
e
on
S
oe
ka
r
no
H
a
tt
a
S
tr
e
e
t
,
of
f
e
r
in
g
a
va
r
ie
ty
of
a
ut
he
nt
ic
m
ot
if
s
;
S
a
ngga
r
B
a
ti
k
B
e
s
ur
e
k
F
a
br
ic
G
r
ya
T
ie
n
C
ol
le
c
ti
on
on
C
il
iwung
S
tr
e
e
t
,
known
f
or
it
s
tr
a
di
ti
ona
l
a
nd
m
ode
r
n
pa
tt
e
r
ns
;
G
a
l
le
r
y
of
B
a
ti
k
B
e
s
ur
e
k
S
w
a
r
na
bum
e
i
on
F
a
tm
a
w
a
ti
S
tr
e
e
t
,
w
hi
c
h
s
how
c
a
s
e
s
in
tr
ic
a
te
b
e
s
ur
e
k
de
s
ig
n
s
;
L
a
-
M
e
nt
iq
ue
B
a
ti
k
B
e
s
ur
e
k
on
S
.
P
a
r
m
a
n
S
tr
e
e
t
,
s
pe
c
ia
li
z
in
g
in
c
ont
e
m
por
a
r
y
s
ty
le
s
;
a
nd
B
a
ti
k
A
ti
k
B
e
s
ur
e
k
on
S
oe
ka
r
no
H
a
tt
a
A
nggut
A
ta
s
S
tr
e
e
t
,
f
e
a
tu
r
in
g
c
la
s
s
ic
B
e
ngkulu
-
in
s
pi
r
e
d ba
ti
k pi
e
c
e
s
.
T
h
e
d
a
ta
s
e
t
th
a
t
h
a
s
be
e
n
c
ol
l
e
c
t
e
d
i
s
b
a
t
ik
b
e
s
ur
e
k
im
a
g
e
w
it
h
th
e
e
xt
e
n
s
i
on
*.
jp
g
w
hi
c
h
c
on
s
i
s
t
s
of
f
our
c
la
s
s
e
s
.
T
he
d
a
t
a
s
e
t
of
im
a
g
e
s
o
f
ba
ti
k
m
o
ti
f
s
th
a
t
h
a
v
e
b
e
e
n
c
o
ll
e
c
te
d
is
gr
ou
pe
d
by
c
a
t
e
g
or
y
or
c
la
s
s
.
D
a
t
a
s
e
t
s
m
us
t
b
e
la
be
le
d
c
or
r
e
c
tl
y
a
c
c
or
d
in
g
to
th
e
e
x
is
ti
n
g
c
a
te
gor
i
e
s
of
b
a
ti
k
m
ot
if
s
,
a
s
de
pi
c
t
e
d
in
F
ig
ur
e
1
:
F
ig
ur
e
1
(
a
)
k
al
ig
r
af
i
,
F
ig
ur
e
1
(
b)
r
a
f
f
le
s
ia
,
F
ig
ur
e
1
(
c
)
b
ur
ung
k
uau
,
F
ig
ur
e
1
(
d)
r
e
lu
ng
pak
u
,
a
nd
F
ig
ur
e
1
(
e
)
r
e
m
bul
an
.
I
n
th
e
pr
e
pr
oc
e
s
s
in
g
s
ta
ge
,
th
e
da
ta
s
e
t
is
a
dj
us
te
d
ba
s
e
d
on
th
e
im
a
ge
s
iz
e
to
th
e
s
iz
e
th
a
t
s
ui
ts
th
e
ne
e
ds
of
t
he
M
obi
le
N
e
tV2, R
e
s
N
e
t5
0, I
nc
e
pt
io
nV
3
,
a
nd V
G
G
1
6 m
ode
ls
.
(
a
)
(
b)
(
c
)
(
d)
(
e
)
F
ig
ur
e
1
.
D
a
ta
c
la
s
s
of
ba
ti
k be
s
ur
e
k
m
ot
if
(
a
)
k
al
ig
r
af
i
, (
b)
r
a
f
f
le
s
ia
, (
c
)
bur
ung k
uau
, (
d)
r
e
lu
ng pak
u
,
a
nd
(
e
)
r
e
m
bul
an
A
ll
im
a
ge
s
in
th
e
d
a
ta
s
e
t
w
il
l
be
di
vi
de
d
in
to
tr
a
in
in
g
da
ta
,
a
n
d
te
s
t
da
ta
.
T
he
di
vi
s
io
n
m
e
th
od
us
e
d
is
s
tr
a
ti
f
ie
d
r
a
ndom
s
a
m
pl
in
g
im
pl
e
m
e
nt
e
d
us
in
g
th
e
s
c
ik
it
-
le
a
r
n
li
br
a
r
y.
T
he
s
tr
a
ti
f
ie
d
r
a
ndom
s
a
m
pl
in
g
m
e
th
od t
a
ke
s
t
he
e
nt
ir
e
da
ta
, s
huf
f
le
s
t
he
da
ta
, a
nd
th
e
n di
vi
de
s
t
he
da
ta
i
nt
o
tr
a
in
in
g a
nd
te
s
ti
ng s
e
ts
f
or
e
a
c
h
c
la
s
s
.
T
he
r
a
ti
o
of
s
ha
r
in
g
tr
a
in
da
ta
a
nd
va
li
da
te
-
te
s
t
da
ta
i
s
7
0:
30.
T
he
e
xp
e
r
im
e
nt
c
ons
is
t
s
of
two
pha
s
e
s
,
na
m
e
ly
th
e
c
om
pa
r
is
on
pha
s
e
of
th
e
tr
a
ns
f
e
r
le
a
r
ni
ng
m
ode
l
a
nd
th
e
pha
s
e
of
im
pr
ovi
ng
th
e
tr
a
ns
f
e
r
le
a
r
ni
ng
m
ode
l.
I
n
th
e
tr
a
ns
f
e
r
le
a
r
ni
ng
m
ode
l
im
pr
ove
m
e
nt
pha
s
e
,
th
e
e
xpe
r
im
e
nt
us
e
d
a
c
om
bi
na
ti
on
of
opt
im
iz
e
r
(
O
P
)
,
a
c
ti
va
ti
on
f
unc
ti
on,
r
e
ti
ne
x,
c
ont
r
a
s
t
li
m
it
e
d
a
da
pt
iv
e
h
is
to
gr
a
m
e
qua
li
z
a
ti
on
(
C
L
A
H
E
)
,
a
nd
g
a
m
m
a
c
or
r
e
c
ti
on
(
G
C
)
m
e
th
ods
. T
he
e
xpe
r
im
e
nt
a
l
s
c
e
na
r
io
a
s
s
e
e
n i
n
F
ig
ur
e
2.
F
ig
ur
e
2
. E
xpe
r
im
e
nt
s
c
e
na
r
io
T
he
ne
ur
a
l
ne
twor
k
a
r
c
hi
te
c
tu
r
e
us
e
d
in
th
is
s
tu
dy
i
s
th
e
R
e
s
N
e
t5
0
a
r
c
hi
te
c
tu
r
e
by
us
in
g
a
c
ti
va
ti
on
r
e
c
ti
f
ie
d
li
ne
a
r
uni
t
(
R
e
L
U
)
.
T
he
r
e
a
s
on
of
R
e
L
U
im
pl
e
m
e
n
ta
ti
on
in
th
e
R
e
s
N
e
t5
0
m
ode
l
a
ddr
e
s
s
e
s
th
e
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
I
m
pr
ov
in
g t
he
t
r
ans
fe
r
l
e
ar
ni
ng f
o
r
bat
ik
be
s
u
r
e
k
t
e
x
ti
le
m
ot
if
c
la
s
s
if
ic
at
io
n
(
M
ar
is
s
a U
ta
m
i
)
3175
va
ni
s
hi
ng
gr
a
di
e
nt
pr
obl
e
m
a
nd
e
nha
nc
e
s
c
om
put
a
ti
ona
l
e
f
f
ic
ie
nc
y
by
pr
oduc
in
g
z
e
r
o
out
put
s
f
or
ne
ga
ti
ve
va
lu
e
s
.
A
ddi
ti
ona
ll
y,
us
in
g
R
e
L
U
a
f
te
r
ba
tc
h
nor
m
a
li
z
a
ti
on
a
nd
c
onvolut
io
n
a
ll
ow
s
f
or
be
tt
e
r
le
a
r
ni
ng
o
f
r
e
s
id
ua
l
r
e
pr
e
s
e
nt
a
ti
ons
,
im
pr
ovi
ng
th
e
m
ode
l'
s
a
c
c
ur
a
c
y.
I
n
te
s
ti
ng,
th
e
num
be
r
of
ba
tc
he
s
w
a
s
s
e
t
to
6
a
nd
th
e
num
be
r
of
e
poc
hs
or
f
ir
s
t
ge
ne
r
a
ti
on
w
a
s
50.
I
f
a
f
te
r
50
e
poc
hs
th
e
r
e
ha
s
not
b
e
e
n
c
onve
r
ge
nc
e
,
tr
a
in
in
g
w
il
l
be
s
to
ppe
d a
nd t
r
a
in
in
g
r
e
s
ul
ts
w
il
l
be
t
a
ke
n a
t
th
e
50
th
e
po
c
h. T
r
a
in
in
g i
s
c
a
ll
e
d c
onve
r
ge
nt
i
f
t
he
r
e
is
no
c
ha
nge
in
th
e
tr
a
in
in
g
a
c
c
ur
a
c
y
va
lu
e
(
pl
a
te
a
u)
or
if
th
e
tr
a
in
in
g
a
c
c
ur
a
c
y
va
lu
e
h
a
s
r
e
a
c
he
d
100%
.
T
he
opt
im
iz
e
r
to
be
us
e
d
is
th
e
A
da
m
opt
im
iz
e
r
w
it
h
le
a
r
ni
ng
r
a
te
pa
r
a
m
e
te
r
s
of
0.001.
T
he
a
r
c
hi
te
c
tu
r
a
l
pa
r
a
m
e
te
r
s
t
ha
t
w
il
l
be
us
e
d i
n R
e
s
N
e
t5
0 i
n pha
s
e
2 c
a
n be
s
e
e
n i
n T
a
bl
e
1.
T
a
bl
e
1
. P
a
r
a
m
e
te
r
of
m
ode
l
R
e
s
N
e
t5
0
P
a
r
a
m
e
t
e
r
V
a
l
ue
N
um
be
r
of
l
a
ye
r
s
16
B
a
t
c
h s
i
z
e
6
N
um
be
r
of
e
poc
hs
50
O
pt
i
m
i
z
e
r
A
da
m
(
l
e
a
r
ni
ng r
a
t
e
0.001, be
t
a
_1=0.9, be
t
a
_2=0.999, e
ps
i
l
on=1e
-
07)
A
c
t
i
va
t
i
on f
unc
t
i
on
R
e
L
U
3.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
he
e
xpe
r
im
e
nt
in
th
e
f
ir
s
t
pha
s
e
w
a
s
c
a
r
r
ie
d
out
to
c
om
pa
r
e
th
e
pe
r
f
or
m
a
nc
e
of
th
e
tr
a
ns
f
e
r
le
a
r
ni
ng a
lg
or
it
hm
a
t
t
he
t
r
a
in
in
g, va
li
da
ti
on a
nd
t
e
s
ti
ng s
ta
ge
s
.
T
he
f
ir
s
t
e
xpe
r
im
e
nt
i
n
pha
s
e
one
w
a
s
c
a
r
r
ie
d
out
us
in
g
th
e
M
obi
le
N
e
tV2
m
ode
l.
T
hi
s
m
ode
l
i
s
c
hos
e
n
be
c
a
us
e
M
obi
le
N
e
tV2
is
a
li
ght
w
e
ig
ht
a
r
c
hi
te
c
tu
r
e
de
s
ig
ne
d
f
or
m
obi
le
de
vi
c
e
s
w
it
h
good
c
om
put
in
g
e
f
f
ic
ie
nc
y
.
I
n
th
is
M
obi
le
N
e
tV2
m
ode
l,
s
e
ve
r
a
l
de
n
s
e
la
ye
r
s
a
r
e
a
dde
d
f
or
c
la
s
s
if
ic
a
ti
on
on
to
p
of
th
e
f
e
a
tu
r
e
out
put
of
M
obi
le
N
e
tV2.
B
a
s
e
d
on
th
e
e
xpe
r
im
e
nt
a
l
r
e
s
ul
ts
, t
he
t
r
a
in
in
g pe
r
f
or
m
a
nc
e
f
or
t
he
M
obi
le
N
e
tV2 mode
l
s
how
s
a
ve
r
y hi
gh t
r
a
in
in
g a
c
c
ur
a
c
y of
98.14
%
,
w
hi
c
h
s
how
s
th
a
t
th
is
m
ode
l
c
a
n
le
a
r
n
th
e
f
e
a
tu
r
e
s
f
r
om
th
e
tr
a
in
in
g
da
ta
ve
r
y
w
e
ll
.
T
he
pe
r
f
or
m
a
nc
e
f
o
r
th
e
M
obi
le
N
e
tV2
m
ode
l
a
t
th
e
va
li
da
ti
on
s
ta
ge
w
a
s
s
li
ght
ly
lo
w
e
r
th
a
n
th
a
t
of
th
e
tr
a
in
in
g
a
t
94.60%
,
in
di
c
a
ti
ng
th
e
pos
s
ib
il
it
y
of
s
li
ght
ove
r
f
it
ti
ng.
H
ow
e
ve
r
,
th
is
di
f
f
e
r
e
nc
e
is
m
a
na
ge
a
bl
e
,
s
ugge
s
ti
ng
th
a
t
th
e
m
ode
l
c
a
n
ge
ne
r
a
li
z
e
w
e
ll
.
P
e
r
f
or
m
a
nc
e
a
na
ly
s
is
a
t
th
e
te
s
ti
ng
s
ta
ge
s
ho
w
e
d
th
a
t
th
e
M
obi
le
N
e
tV2
m
ode
l
obt
a
in
e
d
a
n
a
c
c
ur
a
c
y
of
96.46%
,
w
hi
c
h
m
e
a
n
s
th
a
t
M
obi
le
N
e
tV2
a
ls
o
p
e
r
f
or
m
e
d
w
e
ll
on
pr
e
vi
ous
ly
uns
e
e
n
d
a
ta
de
s
pi
t
e
a
de
c
r
e
a
s
e
i
n t
r
a
in
in
g a
c
c
ur
a
c
y.
T
he
s
e
c
ond
e
xpe
r
im
e
nt
in
pha
s
e
one
w
a
s
c
a
r
r
ie
d
out
us
in
g
th
e
R
e
s
N
e
t5
0
m
ode
l.
T
hi
s
m
ode
l
ha
s
a
r
e
s
id
ua
l
a
r
c
hi
te
c
tu
r
e
a
ll
ow
s
in
te
ns
e
ne
twor
k
tr
a
in
in
g
w
it
h
s
ki
p
c
onne
c
ti
ons
.
I
n
th
e
R
e
s
N
e
t5
0
m
ode
l,
a
c
la
s
s
if
ic
a
ti
on
la
ye
r
is
a
dde
d
to
th
e
f
e
a
tu
r
e
s
of
R
e
s
N
e
t5
0
a
nd
f
in
e
-
tu
ne
d
to
th
e
la
s
t
f
e
w
la
ye
r
s
.
B
a
s
e
d
on
th
e
e
xpe
r
im
e
nt
r
e
s
ul
ts
,
R
e
s
N
e
t5
0
ha
s
th
e
hi
ghe
s
t
tr
a
in
in
g
a
c
c
ur
a
c
y
of
99.60%
,
w
hi
c
h
s
how
s
th
a
t
th
is
m
ode
l
c
a
n
le
a
r
n
tr
a
in
in
g
da
ta
.
A
va
li
da
ti
on
a
c
c
ur
a
c
y
of
97.44%
s
how
s
th
a
t
th
is
m
ode
l
ha
s
e
xc
e
ll
e
nt
ge
ne
r
a
li
z
a
ti
on
a
bi
li
ty
,
e
ve
n
w
it
h a
s
li
ght
de
c
r
e
a
s
e
in
tr
a
in
in
g
a
c
c
ur
a
c
y. T
he
te
s
t
a
c
c
ur
a
c
y
of
98.12%
is
th
e
hi
ghe
s
t
a
m
ong
a
ll
m
ode
ls
, i
ndi
c
a
ti
ng t
he
be
s
t
pe
r
f
or
m
a
nc
e
i
n t
he
ne
w
b
a
ti
k be
s
ur
e
k da
ta
.
T
he
th
ir
d
e
xpe
r
im
e
nt
in
pha
s
e
on
e
w
a
s
c
a
r
r
ie
d
out
u
s
in
g
th
e
I
nc
e
pt
io
nV
3
m
ode
l.
T
hi
s
m
ode
l
h
a
s
a
va
r
ie
ty
of
f
il
te
r
s
iz
e
s
in
a
s
in
gl
e
la
ye
r
,
a
ll
ow
in
g
f
or
th
e
e
xt
r
a
c
ti
on
of
c
om
pl
e
x
f
e
a
tu
r
e
s
.
T
he
I
nc
e
pt
io
nV
3
m
ode
l
ha
s
a
dde
d
a
c
la
s
s
if
ic
a
ti
on
la
ye
r
on
to
p
of
th
e
f
e
a
tu
r
e
s
of
I
nc
e
pt
io
nV
3
a
nd
f
in
e
-
tu
ne
d
th
e
ba
ti
k
be
s
ur
e
k
da
ta
s
e
t.
B
a
s
e
d
on
th
e
e
xpe
r
im
e
nt
r
e
s
ul
ts
,
I
nc
e
pt
io
nV
3
s
how
e
d
a
lo
w
e
r
tr
a
in
in
g
a
c
c
ur
a
c
y
th
a
n
ot
he
r
m
ode
ls
,
w
hi
c
h
w
a
s
96.35%
.
T
he
v
a
li
da
ti
on
a
c
c
ur
a
c
y
r
e
a
c
he
d
96.59%
or
a
lm
os
t
e
qui
va
le
nt
to
th
e
tr
a
in
in
g
a
c
c
ur
a
c
y,
s
how
in
g
th
a
t
th
is
m
ode
l
w
a
s
not
to
o
ove
r
f
it
on
th
e
tr
a
in
in
g
da
ta
.
T
he
te
s
t
a
c
c
ur
a
c
y
of
92.50%
is
th
e
lo
w
e
s
t
a
m
ong
a
ll
m
ode
ls
,
in
di
c
a
ti
ng
th
a
t
I
nc
e
pt
io
nV
3
c
oul
d
be
m
or
e
e
f
f
e
c
ti
ve
in
ge
ne
r
a
li
z
in
g
on
pr
e
vi
ous
ly
un
s
e
e
n
ba
ti
k da
ta
t
ha
n ot
he
r
m
ode
ls
.
T
he
f
our
th
e
xpe
r
im
e
nt
in
pha
s
e
one
w
a
s
c
a
r
r
ie
d
out
us
in
g
th
e
V
G
G
16
m
ode
l.
T
hi
s
m
ode
l
us
e
s
a
s
im
pl
e
but
e
f
f
e
c
ti
ve
a
r
c
hi
te
c
tu
r
e
w
it
h
m
a
ny
la
ye
r
s
of
s
m
a
ll
c
onvolut
io
n.
T
he
V
G
G
16
m
ode
l
ha
s
a
dde
d
a
c
la
s
s
if
ic
a
ti
on l
a
ye
r
on t
op of
t
he
V
G
G
16 f
e
a
tu
r
e
out
put
a
nd f
in
e
-
tu
ne
d t
he
ba
ti
k be
s
ur
e
k da
ta
s
e
t.
B
a
s
e
d on the
e
xpe
r
im
e
nt
a
l
r
e
s
ul
ts
,
V
G
G
16
ha
s
a
hi
gh
t
r
a
in
in
g
a
c
c
ur
a
c
y
of
9
9.67%
,
w
hi
c
h
s
how
s
th
a
t
th
e
m
ode
l
le
a
r
ns
th
e
f
e
a
tu
r
e
s
f
r
om
th
e
tr
a
in
in
g
da
ta
ve
r
y
w
e
ll
.
T
he
va
li
da
ti
on
a
c
c
ur
a
c
y
of
96.02%
s
how
s
th
a
t
th
e
V
G
G
16
m
ode
l
c
a
n
s
ti
ll
ge
ne
r
a
li
z
e
w
e
ll
de
s
pi
te
th
e
de
c
r
e
a
s
e
in
tr
a
in
in
g
a
c
c
ur
a
c
y.
A
te
s
t
a
c
c
ur
a
c
y
of
95.00%
a
ls
o
in
di
c
a
te
s
good
pe
r
f
or
m
a
nc
e
but
is
s
li
ght
ly
lo
w
e
r
th
a
n
R
e
s
N
e
t5
0.
T
he
r
e
s
ul
ts
of
th
e
c
om
pa
r
is
on
of
th
e
pe
r
f
or
m
a
nc
e
of
e
a
c
h m
ode
l
in
t
he
e
xpe
r
im
e
nt
i
n pha
s
e
one
c
a
n b
e
s
e
e
n i
n F
ig
ur
e
3.
B
a
s
e
d
on
th
e
pe
r
f
or
m
a
nc
e
c
om
pa
r
is
on
r
e
s
ul
t
s
,
R
e
s
N
e
t5
0
i
s
th
e
be
s
t
-
pe
r
f
or
m
in
g
m
ode
l
ove
r
a
ll
,
w
it
h
th
e
hi
ghe
s
t
a
c
c
ur
a
c
y
on
tr
a
in
in
g,
va
li
da
ti
on,
a
nd
te
s
t
da
ta
. T
hi
s
s
how
s
th
a
t
R
e
s
N
e
t5
0 c
a
n
le
a
r
n
a
nd
ge
n
e
r
a
li
z
e
da
ta
ve
r
y
w
e
ll
.
T
he
V
G
G
16
m
ode
l
a
ls
o
s
how
s
e
x
c
e
ll
e
nt
pe
r
f
or
m
a
nc
e
but
s
li
ght
ly
lo
w
e
r
a
c
c
ur
a
c
y
th
a
n
th
e
R
e
s
N
e
t5
0
m
od
e
l,
e
s
p
e
c
ia
ll
y
in
te
s
t
da
ta
pr
oc
e
s
s
in
g.
M
obi
le
N
e
tV2
a
c
hi
e
ve
s
e
xc
e
ll
e
nt
a
c
c
ur
a
c
y
a
nd
hi
gh
e
r
e
f
f
ic
ie
nc
y c
om
pa
r
e
d t
o V
G
G
16 a
nd R
e
s
N
e
t5
0
,
but
pe
r
f
or
m
s
be
lo
w
t
he
R
e
s
N
e
t5
0 m
ode
l
ove
r
a
ll
. I
nc
e
pt
io
nV
3
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
4
,
A
ugus
t
20
25
:
3172
-
3181
3176
ha
s
r
e
a
s
ona
bl
e
tr
a
in
in
g
a
nd
va
li
da
ti
on
a
c
c
ur
a
c
y
but
s
how
s
lo
w
e
r
te
s
ti
ng
a
c
c
ur
a
c
y
a
nd
pr
obl
e
m
s
,
e
s
p
e
c
ia
ll
y
in
ge
ne
r
a
li
z
a
ti
ons
on t
he
ne
w
b
a
ti
k be
s
ur
e
k da
ta
.
F
ig
ur
e
3
. C
om
pa
r
a
ti
on of
t
r
a
ns
f
e
r
le
a
r
ni
ng pe
r
f
o
r
m
a
nc
e
T
he
e
xp
e
r
im
e
nt
i
n
th
e
s
e
c
ond
p
ha
s
e
w
a
s
c
a
r
r
i
e
d
out
by
i
m
pr
ovi
ng
th
e
b
e
s
t
tr
a
ns
f
e
r
le
a
r
ni
ng
m
o
de
l
in
th
e
s
e
c
ond
ph
a
s
e
of
th
e
e
xpe
r
im
e
nt
.
I
m
pr
o
ve
m
e
nt
s
w
e
r
e
m
a
de
b
y
a
dd
in
g
m
ix
e
d
c
ont
r
a
s
t
e
nha
nc
e
m
e
nt
,
a
c
ti
v
a
ti
on
f
un
c
ti
on,
a
nd
opt
im
i
z
e
r
f
or
th
e
ba
ti
k
b
e
s
ur
e
k
te
xt
i
le
m
ot
if
c
l
a
s
s
if
i
c
a
ti
o
n.
T
h
e
f
ir
s
t
e
xp
e
r
im
e
nt
i
n
th
e
s
e
c
ond
pha
s
e
u
s
e
s
R
e
s
N
e
t
50
w
it
h
A
da
m
opt
im
i
z
e
r
,
R
e
L
U
a
c
ti
va
ti
on
f
un
c
ti
on
a
n
d
r
e
ti
ne
x
a
s
th
e
c
ont
r
a
s
t
e
nha
n
c
e
m
e
nt
m
e
th
o
d.
B
a
s
e
d
on
t
he
e
xp
e
r
im
e
nt
r
e
s
ul
t
s
,
th
e
t
r
a
in
in
g
s
t
a
ge
obt
a
i
ne
d
a
n
a
c
c
ur
a
c
y
s
c
or
e
of
99.87%
,
in
di
c
a
ti
ng
th
a
t
th
i
s
m
od
e
l
c
a
n
l
e
a
r
n
th
e
tr
a
in
in
g
da
ta
m
or
e
e
f
f
e
c
ti
ve
ly
a
f
t
e
r
a
ddi
ng
c
ont
r
a
s
t
e
nha
n
c
e
m
e
nt
t
e
c
hni
que
s
s
uc
h a
s
r
e
ti
ne
x
.
T
h
e
va
li
da
ti
o
n
a
c
c
ur
a
c
y
in
th
is
e
xpe
r
im
e
nt
of
99.
15%
s
ho
w
s
t
ha
t
th
e
m
ode
l
is
a
l
s
o
ve
r
y
good
a
t
ge
n
e
r
a
li
z
in
g
to
th
e
va
li
da
ti
o
n
d
a
ta
.
A
te
s
t
a
c
c
ur
a
c
y
of
96.
25%
in
di
c
a
te
s
s
ig
ni
f
i
c
a
n
t
ove
r
f
it
ti
ng
of
t
r
a
i
ni
ng
a
nd v
a
li
d
a
ti
on
da
t
a
, r
e
s
ul
t
in
g i
n
pe
r
f
or
m
a
nc
e
d
e
gr
a
da
ti
o
n i
n t
he
ne
w
te
s
t
d
a
ta
.
T
he
s
e
c
ond
ph
a
s
e
e
xpe
r
im
e
nt
us
e
s
R
e
s
N
e
t5
0
w
it
h
A
d
a
m
opt
i
m
iz
e
r
,
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
a
nd
r
e
ti
ne
x
a
nd
GC
a
s
c
ont
r
a
s
t
e
nha
nc
e
m
e
nt
m
e
th
ods
.
A
t
th
e
tr
a
in
i
ng
s
ta
ge
,
th
is
m
ode
l
obt
a
in
e
d
100%
a
c
c
ur
a
c
y,
s
how
in
g
th
a
t
th
e
m
ode
l
ha
s
f
ul
ly
m
e
m
or
iz
e
d
th
e
e
xi
s
ti
ng
ba
ti
k
be
s
ur
e
k
da
ta
.
T
he
r
e
w
a
s
a
d
e
c
r
e
a
s
e
in
tr
a
in
in
g
a
c
c
ur
a
c
y
a
t
th
e
va
li
da
ti
on
s
ta
ge
,
w
it
h
a
va
li
da
ti
on
a
c
c
ur
a
c
y
of
98.29%
,
w
hi
c
h
s
how
e
d
th
a
t
th
e
m
ode
l
c
oul
d
s
ti
ll
ge
ne
r
a
li
z
e
th
e
ba
ti
k
be
s
ur
e
k
da
ta
w
e
ll
.
A
t
th
e
te
s
ti
ng
s
ta
ge
,
a
n
a
c
c
ur
a
c
y
of
97.29%
w
a
s
obt
a
in
e
d,
w
hi
c
h
s
how
e
d
e
xc
e
ll
e
nt
pe
r
f
or
m
a
nc
e
in
th
e
ba
ti
k
be
s
ur
e
k
da
ta
.
T
he
GC
te
c
hni
que
he
lp
s
to
im
pr
ove
th
e
qua
li
ty
o
f
te
s
t
r
e
s
ul
ts
c
om
pa
r
e
d t
o u
s
in
g
r
e
ti
ne
x only.
T
he
th
ir
d
e
xpe
r
im
e
nt
in
th
e
s
e
c
ond
pha
s
e
us
e
s
R
e
s
N
e
t5
0
w
it
h
A
da
m
opt
im
iz
e
r
,
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
,
a
nd
C
L
A
H
E
a
s
th
e
c
ont
r
a
s
t
e
nh
a
nc
e
m
e
nt
m
e
th
od.
A
t
th
e
tr
a
in
in
g
s
ta
g
e
,
th
e
m
ode
l
ga
in
e
d
100%
a
c
c
ur
a
c
y,
in
di
c
a
ti
ng
th
a
t
th
e
m
od
e
l
c
a
n
m
e
m
or
iz
e
th
e
b
a
ti
k
da
ta
th
or
oughly.
T
he
a
c
c
ur
a
c
y
of
th
e
va
li
da
ti
on
da
ta
of
95.17%
de
c
r
e
a
s
e
d
c
om
pa
r
e
d
to
ot
he
r
m
ode
ls
,
in
di
c
a
ti
ng
th
a
t
th
e
m
ode
l
m
a
y
be
s
li
ght
ly
ove
r
f
it
te
d
w
it
h
tr
a
in
in
g
da
ta
.
T
he
a
c
c
ur
a
c
y
in
th
e
te
s
t
r
e
a
c
h
e
d
th
e
hi
ghe
s
t
va
lu
e
of
98.75%
,
w
hi
c
h
s
how
s
th
a
t
C
L
A
H
E
e
f
f
e
c
ti
ve
ly
i
m
pr
ove
s
t
he
pe
r
f
or
m
a
nc
e
of
t
he
R
e
s
N
e
t5
0 m
ode
l
o
n ba
ti
k be
s
ur
e
k da
ta
.
T
he
f
our
th
e
xpe
r
im
e
nt
in
th
e
s
e
c
ond
pha
s
e
us
e
s
R
e
s
N
e
t5
0
w
it
h
A
da
m
opt
im
iz
e
r
,
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
,
a
nd
C
L
A
H
E
a
nd
G
C
a
s
c
ont
r
a
s
t
e
nha
n
c
e
m
e
nt
m
e
th
o
ds
.
I
n
th
e
tr
a
in
in
g s
ta
ge
,
th
e
m
ode
l
obt
a
in
e
d
a
n
a
c
c
ur
a
c
y
s
c
or
e
of
99.60%
,
w
hi
c
h
in
d
ic
a
te
s
good
tr
a
in
in
g
pe
r
f
or
m
a
nc
e
w
it
hout
s
ig
ni
f
ic
a
nt
ove
r
f
it
ti
ng.
T
he
va
li
da
ti
on
a
c
c
ur
a
c
y
got
a
s
c
or
e
of
97.73%
,
w
hi
c
h
s
how
s
th
a
t
th
e
m
ode
l
c
a
n
ge
n
e
r
a
li
z
e
th
e
va
li
da
ti
on
da
t
a
w
e
ll
.
A
t
th
e
te
s
t
s
ta
ge
,
th
e
m
od
e
l
a
c
c
ur
a
c
y
of
onl
y
95.00%
r
e
pr
e
s
e
nt
s
a
de
c
r
e
a
s
e
c
om
pa
r
e
d
to
th
e
s
in
gl
e
C
L
A
H
E
c
onf
ig
ur
a
ti
on.
T
hi
s
m
a
y
be
du
e
to
th
e
in
te
r
a
c
ti
on
b
e
twe
e
n
C
L
A
H
E
a
nd
G
C
,
w
hi
c
h
i
s
le
s
s
th
a
n
opt
im
a
l
f
or
t
he
t
e
s
t
da
ta
. T
he
r
e
s
ul
ts
of
t
he
c
om
pa
r
is
on of
t
he
p
e
r
f
or
m
a
nc
e
of
e
a
c
h m
ode
l
in
t
he
e
xpe
r
im
e
nt
i
n
th
e
s
e
c
ond pha
s
e
c
a
n b
e
s
e
e
n i
n T
a
bl
e
2.
T
a
bl
e
2
. C
om
pa
r
a
ti
on of
R
e
s
N
e
t5
0 i
m
pr
ove
m
e
nt
pe
r
f
or
m
a
nc
e
E
xpe
r
i
m
e
nt
T
r
a
i
n
(%)
I
m
pr
ove
m
e
nt
(%)
V
a
l
i
da
t
e
(%)
I
m
pr
ove
m
e
nt
(%)
T
e
s
t
(%)
I
m
pr
ove
m
e
nt
(%)
R
e
s
N
e
t
50
99.60
-
97.44
-
98.12
-
R
e
s
N
e
t
50+A
d
a
m
+R
e
L
U
+
r
e
t
i
ne
x
99.87
(
+)
0.27
99.15
(
+)
1.71
96.25
(
-
)
1.87
R
e
s
N
e
t
50+A
d
a
m
+R
e
L
U
+
r
e
t
i
ne
x+G
C
100
(
+)
0.40
98.29
(
+)
0.85
97.29
(
-
)
0.83
R
e
s
N
e
t
50+A
d
a
m
+R
e
L
U
+C
L
A
H
E
100
(
+)
0.40
95.17
(
-
)
2.27
98.75
(
+)
0.63
R
e
s
N
e
t
50+A
d
a
m
+R
e
L
U
+C
L
A
H
E
+G
C
99.60
(
+)
0.00
97.73
(
+)
0.29
95.00
(
-
)
3.12
98.14%
99.60%
96.35%
99.67%
94.60%
97.44%
96.59%
96.02%
96.46%
98.12%
92.50%
95.00%
92.00%
92.95%
93.90%
94.85%
95.80%
96.75%
97.70%
98.65%
99.60%
M
obi
l
e
N
e
t
V
2
R
e
s
N
e
t
50
I
nc
e
pt
i
onV
3
V
G
G
16
A
c
c
u
r
ac
y
T
r
an
s
f
e
r
l
e
ar
n
i
n
g m
od
e
l
s
C
om
p
ar
at
ion
of
m
od
e
l
ac
c
u
r
ac
y
T
r
a
in
V
a
l
i
da
t
i
on
T
e
s
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
I
m
pr
ov
in
g t
he
t
r
ans
fe
r
l
e
ar
ni
ng f
o
r
bat
ik
be
s
u
r
e
k
t
e
x
ti
le
m
ot
if
c
la
s
s
if
ic
at
io
n
(
M
ar
is
s
a U
ta
m
i
)
3177
T
he
R
e
s
N
e
t5
0
m
ode
l
w
it
h
A
da
m
opt
im
iz
e
r
,
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
a
nd
C
L
A
H
E
a
s
th
e
c
ont
r
a
s
t
e
nha
nc
e
m
e
nt
m
e
th
od
obt
a
in
e
d
th
e
hi
ghe
s
t
te
s
t
a
c
c
ur
a
c
y
(
98.75%
)
,
s
how
in
g
th
a
t
C
L
A
H
E
is
ve
r
y
e
f
f
e
c
ti
ve
in
im
pr
ovi
ng
pe
r
f
o
r
m
a
nc
e
on
ba
ti
k
be
s
ur
e
k
da
ta
.
T
he
R
e
s
N
e
t5
0
m
ode
l
w
it
h
A
da
m
opt
im
iz
e
r
,
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
,
a
nd
r
e
ti
ne
x a
nd G
C
a
s
c
ont
r
a
s
t
e
nha
nc
e
m
e
nt
m
e
th
ods
s
how
e
d e
xc
e
ll
e
nt
t
e
s
t
a
c
c
ur
a
c
y (
97.29%
)
w
it
h
hi
gh
va
li
da
ti
on
a
c
c
ur
a
c
y,
s
ig
na
li
ng
a
good
b
a
la
nc
e
be
twe
e
n
tr
a
in
in
g
a
nd
ge
ne
r
a
li
z
a
ti
on.
S
om
e
m
ode
ls
s
how
100%
tr
a
in
in
g
a
c
c
ur
a
c
y,
w
hi
c
h
m
a
y
s
ig
na
l
ove
r
f
it
ti
ng,
e
s
pe
c
ia
ll
y
in
m
ode
ls
th
a
t
s
how
a
de
c
r
e
a
s
e
in
a
c
c
ur
a
c
y
in
th
e
te
s
t
da
ta
.
B
a
s
e
d
on
c
om
pa
r
in
g
a
c
c
ur
a
c
y
va
lu
e
s
,
R
e
s
N
e
t5
0
w
it
h
A
da
m
opt
im
iz
e
r
,
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
,
a
nd
C
L
A
H
E
a
s
th
e
c
ont
r
a
s
t
e
nha
n
c
e
m
e
nt
m
e
th
od
is
th
e
be
s
t
c
om
bi
na
ti
on.
S
e
ve
r
a
l
e
v
a
lu
a
ti
on
m
a
tr
ic
e
s
s
uc
h
a
s
a
c
c
ur
a
c
y,
lo
s
s
,
pr
e
c
is
io
n,
r
e
c
a
ll
,
F
1
-
s
c
or
e
,
a
n
d
c
onf
us
io
n
m
a
tr
ix
w
e
r
e
us
e
d
to
a
na
ly
z
e
th
e
r
e
s
ul
ts
of
th
e
R
e
s
N
e
t5
0
m
ode
l
w
it
h
A
da
m
opt
im
iz
e
r
,
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
,
a
nd
C
L
A
H
E
a
s
c
ont
r
a
s
t
e
nha
nc
e
m
e
nt
m
e
th
ods
in
m
or
e
de
ta
il
.
T
h
e
f
ir
s
t
e
v
a
lu
a
ti
on
is
ba
s
e
d
on
th
e
lo
s
s
a
nd
a
c
c
ur
a
c
y
gr
a
ph
of
th
e
tr
a
in
in
g da
ta
a
nd va
li
da
ti
on da
ta
s
e
e
n i
n F
ig
ur
e
4.
F
ig
ur
e
4
. A
c
c
ur
a
c
y a
nd l
os
s
v
a
lu
e
of
R
e
s
N
e
t5
0 w
it
h A
da
m
, R
e
L
U
a
nd C
L
A
H
E
B
a
s
e
d
on
th
e
a
c
c
ur
a
c
y a
nd
lo
s
s
gr
a
ph
s
,
th
e
R
e
s
N
e
t5
0
m
ode
l
w
i
th
A
da
m
opt
im
iz
e
r
,
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
,
a
nd
C
L
A
H
E
s
how
s
hi
gh
t
r
a
in
in
g
a
c
c
ur
a
c
y
(
a
bout
9
5
-
100%
)
a
nd
lo
w
tr
a
in
in
g
lo
s
s
a
nd
ve
r
y
lo
w
va
li
da
ti
on
a
c
c
ur
a
c
y
(
a
bout
27
-
20%
)
a
t
th
e
be
gi
nni
ng
of
th
e
e
po
c
h.
T
hi
s
in
di
c
a
te
s
th
a
t
th
e
m
ode
l
i
s
ove
r
f
it
ti
ng
th
e
tr
a
in
in
g
da
ta
a
nd
c
a
nnot
ge
n
e
r
a
li
z
e
th
e
v
a
li
da
ti
on
da
ta
w
e
ll
.
F
r
om
th
e
21
st
to
th
e
30
th
e
poc
h,
th
e
m
ode
l
s
how
e
d a
s
ta
bl
e
va
li
da
ti
on a
c
c
ur
a
c
y of
a
r
ound 94
-
97%
. T
he
t
r
a
in
in
g a
c
c
ur
a
c
y i
s
ve
r
y hi
gh, a
nd t
he
m
ode
l
c
a
n
pe
r
f
or
m
w
e
ll
on
th
e
va
li
da
ti
on
da
ta
.
D
e
s
pi
te
s
om
e
f
lu
c
tu
a
ti
on
s
,
th
e
m
ode
l
r
e
m
a
in
s
in
good
pe
r
f
or
m
a
nc
e
on
va
li
da
ti
on da
ta
. T
he
bl
oc
k di
a
gr
a
m
of
t
he
pr
opos
e
d
a
ppr
oa
c
h c
a
n be
s
e
e
n i
n F
ig
ur
e
5.
F
ig
ur
e
5. B
lo
c
k di
a
gr
a
m
of
pr
opos
e
d R
e
s
N
e
t5
0 m
ode
l
w
it
h m
o
di
f
ic
a
ti
on
V
a
li
da
ti
on
a
c
c
ur
a
c
y
pe
a
ke
d
a
t
a
r
ound
96%
in
th
e
26
th
e
po
c
h,
a
nd
s
ubs
e
que
nt
e
poc
h
s
s
li
ght
ly
de
c
li
ne
d
to
a
r
ound
95%
.
T
h
e
tr
a
in
in
g
a
c
c
ur
a
c
y
r
e
m
a
in
s
s
ta
bl
e
,
in
di
c
a
ti
ng
th
a
t
th
e
m
ode
l
ha
s
a
c
hi
e
v
e
d
s
ta
bi
li
ty
in
tr
a
in
in
g
a
nd
va
li
da
ti
on
da
ta
.
T
he
m
ode
l
s
how
e
d
s
i
gni
f
ic
a
nt
pe
r
f
or
m
a
nc
e
im
pr
ove
m
e
nt
s
f
r
om
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
4
,
A
ugus
t
20
25
:
3172
-
3181
3178
s
ta
r
t
of
tr
a
in
in
g,
w
i
th
va
li
da
ti
on
a
c
c
ur
a
c
y
in
c
r
e
a
s
in
g
c
ons
is
te
nt
ly
a
f
te
r
a
f
e
w
in
it
ia
l
e
poc
hs
.
B
y
th
e
e
nd
of
th
e
tr
a
in
in
g,
th
e
m
ode
l
a
c
hi
e
ve
d
e
xc
e
ll
e
nt
va
li
da
ti
on
pe
r
f
or
m
a
nc
e
,
in
di
c
a
ti
ng
th
a
t
th
e
m
ode
l
ha
d
s
uc
c
e
s
s
f
ul
ly
a
voi
de
d
ove
r
f
it
ti
ng
a
nd
c
oul
d
ge
ne
r
a
li
z
e
w
e
ll
on
th
e
v
a
li
da
ti
on da
ta
.
T
he
s
e
c
ond e
va
lu
a
ti
on
c
a
n
b
e
s
e
e
n
f
r
om
th
e
pr
e
c
is
io
n, r
e
c
a
ll
a
nd F
1
-
s
c
or
e
va
lu
e
s
i
n T
a
bl
e
3.
T
a
bl
e
3
. E
va
lu
a
ti
on ma
tr
ix
of
R
e
s
N
e
t5
0 w
it
h A
da
m
,
R
e
L
U
,
a
n
d C
L
A
H
E
C
l
a
s
s
P
r
e
c
i
s
i
on
R
e
c
a
l
l
F1
-
s
c
or
e
B
ur
ung
k
uau
0.96
1.00
0.98
K
al
i
gr
af
i
1.00
0.98
0.99
R
a
f
l
e
s
i
a
0.98
1.00
0.99
R
e
l
ung
p
ak
u
1.00
1.00
1.00
R
e
m
bul
an
1.00
0.96
0.98
B
a
s
e
d
on t
h
e
r
e
s
ul
t
s
of
t
h
e
c
l
a
s
s
if
ic
a
ti
on
m
ode
l
e
v
a
lu
a
ti
on m
e
tr
i
c
s
i
n
t
he
f
or
m
of
pr
e
c
is
i
on, r
e
c
a
l
l,
a
n
d
F1
-
s
c
or
e
f
or
e
a
c
h
c
l
a
s
s
i
n
th
e
b
a
ti
k
be
s
ur
e
k
d
a
ta
s
e
t,
th
e
bur
ung
k
uau
c
la
s
s
g
ot
a
pr
e
c
i
s
io
n
va
l
ue
of
0
.96,
w
hi
c
h
s
how
s
t
h
a
t
96%
of
t
h
e
m
ode
l'
s
pr
e
di
c
ti
on
s
on t
he
bur
ung k
uau
d
a
ta
a
r
e
c
or
r
e
c
t.
A
r
e
c
a
ll
va
l
ue
of
1.
00 i
ndi
c
a
t
e
s
th
a
t
th
e
m
ode
l
s
uc
c
e
s
s
f
ul
ly
d
e
te
c
te
d
a
ll
in
s
ta
n
c
e
s
of
bu
r
ung
k
u
au
.
A
n
F
1
-
s
c
or
e
of
0.98
f
or
th
e
bi
r
d
kua
u
da
t
a
c
la
s
s
s
ho
w
s
e
x
c
e
ll
e
nt
m
ode
l
pe
r
f
or
m
a
nc
e
w
it
h
a
good
ba
l
a
nc
e
b
e
twe
e
n pr
e
c
i
s
io
n
a
nd
r
e
c
a
ll
.
T
he
k
al
ig
r
af
i
c
la
s
s
ge
t
s
a
pr
e
c
is
io
n
of
1.00,
in
di
c
a
ti
ng
th
a
t
a
ll
k
al
ig
r
af
i
pr
e
di
c
ti
ons
a
r
e
c
or
r
e
c
t.
A
r
e
c
a
ll
va
lu
e
of
0.98
f
or
th
e
k
al
ig
r
af
i
c
la
s
s
in
di
c
a
te
s
th
a
t
98
%
of
k
al
ig
r
af
i
da
ta
a
r
e
s
uc
c
e
s
s
f
ul
ly
de
te
c
te
d.
A
n
F
1
-
s
c
or
e
of
0.99
f
or
th
e
k
al
ig
r
af
i
c
la
s
s
in
di
c
a
te
s
e
xc
e
ll
e
nt
m
ode
l
pe
r
f
or
m
a
nc
e
w
it
h
a
good
ba
la
nc
e
be
twe
e
n
pr
e
c
is
io
n
a
nd
r
e
c
a
ll
.
T
h
e
n,
th
e
r
a
f
f
le
s
ia
c
la
s
s
g
e
ts
a
pr
e
c
is
io
n
of
0.98,
in
di
c
a
ti
ng
th
a
t
98%
of
th
e
r
a
f
f
le
s
ia
da
ta
pr
e
di
c
ti
ons
a
r
e
c
or
r
e
c
t.
T
he
r
e
c
a
ll
va
lu
e
f
or
th
e
r
a
f
f
le
s
ia
da
ta
c
la
s
s
is
1.00,
in
di
c
a
ti
ng
th
a
t
th
e
m
ode
l
de
te
c
te
d
a
ll
r
a
f
f
le
s
ia
in
s
ta
nc
e
s
s
uc
c
e
s
s
f
ul
ly
.
A
n
F
1
-
s
c
or
e
of
0.99
f
or
th
e
r
a
f
f
le
s
ia
da
ta
c
la
s
s
s
how
s
e
xc
e
ll
e
nt
pe
r
f
or
m
a
nc
e
w
it
h a
good ba
la
nc
e
be
twe
e
n pr
e
c
i
s
io
n a
nd r
e
c
a
ll
.
T
he
r
e
lu
ng
pak
u
c
la
s
s
ge
t
s
a
pr
e
c
is
io
n
of
1.00,
in
di
c
a
ti
ng
t
ha
t
a
ll
r
e
lu
ng
pak
u
pr
e
di
c
ti
ons
a
r
e
c
or
r
e
c
t.
A
r
e
c
a
ll
va
lu
e
1.00
f
or
th
e
r
e
lu
ng
pak
u
da
ta
c
la
s
s
i
ndi
c
a
te
s
th
a
t
a
ll
r
e
lu
ng
pak
u
in
s
t
a
nc
e
s
w
e
r
e
s
uc
c
e
s
s
f
ul
ly
de
te
c
te
d.
A
n
F
1
-
s
c
or
e
of
1.00
f
or
th
e
r
e
lu
ng
pak
u
da
ta
c
la
s
s
s
how
s
pe
r
f
e
c
t
pe
r
f
or
m
a
nc
e
w
it
h
a
n
opt
im
a
l
ba
la
nc
e
be
twe
e
n
pr
e
c
is
io
n
a
nd
r
e
c
a
ll
.
T
he
r
e
m
bul
an
c
l
a
s
s
ge
ts
a
pr
e
c
is
io
n
of
1.00,
in
di
c
a
ti
ng
th
a
t
a
ll
m
oon
pr
e
di
c
ti
ons
a
r
e
c
or
r
e
c
t.
A
r
e
c
a
ll
v
a
lu
e
of
0.96
f
or
th
e
r
e
m
bul
an
da
ta
c
la
s
s
in
di
c
a
t
e
s
th
a
t
th
e
m
ode
l
s
uc
c
e
s
s
f
ul
ly
de
te
c
te
d
96%
of
th
e
lu
na
r
in
s
ta
n
c
e
s
.
A
n
F
1
-
s
c
or
e
of
0.98
f
or
th
e
r
e
m
bul
an
da
ta
c
la
s
s
in
di
c
a
te
s
e
xc
e
ll
e
nt
pe
r
f
or
m
a
nc
e
w
it
h a
good ba
la
nc
e
be
twe
e
n pr
e
c
i
s
io
n a
nd r
e
c
a
ll
.
T
he
be
s
t
-
in
-
c
la
s
s
pe
r
f
or
m
a
nc
e
us
e
s
th
e
R
e
s
N
e
t5
0
m
ode
l
w
it
h
A
da
m
opt
im
iz
e
r
,
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
,
a
nd
C
L
A
H
E
,
w
hi
c
h
is
th
e
r
e
lu
ng
pak
u
c
la
s
s
.
T
he
c
la
s
s
ha
s
th
e
hi
ghe
s
t
F
1
-
s
c
or
e
of
1.00,
de
m
ons
tr
a
ti
ng e
xc
e
ll
e
nt
pe
r
f
or
m
a
nc
e
a
nd a
pe
r
f
e
c
t
ba
la
nc
e
be
t
w
e
e
n pr
e
c
is
io
n a
nd r
e
c
a
ll
. T
he
t
hi
r
d e
va
lu
a
ti
on
c
a
n be
s
e
e
n f
r
om
t
he
c
onf
us
io
n m
a
tr
ix
va
lu
e
of
t
he
e
xpe
r
im
e
nt
r
e
s
ul
ts
w
it
h t
he
t
e
s
t
da
ta
s
e
e
n i
n F
ig
ur
e
6
.
F
ig
ur
e
6
. C
onf
us
io
n m
a
tr
ix
of
R
e
s
N
e
t5
0 w
it
h A
da
m
, R
e
L
U
a
nd
C
L
A
H
E
T
h
e
c
onf
u
s
io
n
m
a
tr
i
x
e
v
a
l
ua
ti
o
n
w
a
s
s
e
e
n
b
a
s
e
d
on
th
e
va
lu
e
s
of
tr
ue
p
o
s
it
i
ve
s
(
T
P
)
,
f
a
l
s
e
p
o
s
it
i
ve
s
(
F
P
)
,
f
a
l
s
e
n
e
g
a
ti
ve
s
(
F
N
)
a
n
d
a
c
c
ur
a
c
y
(
A
C
C
)
f
or
e
a
c
h
c
la
s
s
,
na
m
e
l
y
b
ur
un
g
k
ua
u
,
k
a
li
g
r
af
i
,
r
a
f
le
s
ia
,
r
e
lu
n
g
pak
u
,
a
nd
r
e
m
bu
la
n
. T
he
d
a
t
a
c
l
a
s
s
bu
r
u
ng k
u
au
g
e
t
s
a
T
P
va
l
u
e
of
96
,
m
e
a
ni
ng
th
e
a
m
ou
nt
of
da
ta
i
s
c
o
r
r
e
c
tl
y
c
l
a
s
s
if
i
e
d
a
s
bu
r
ung
k
u
au
.
A
n
F
P
v
a
l
ue
of
0
m
e
a
n
s
th
a
t
n
o
d
a
t
a
f
r
om
o
th
e
r
c
la
s
s
e
s
a
r
e
i
nc
or
r
e
c
t
ly
c
l
a
s
s
if
i
e
d
a
s
bu
r
un
g
k
ua
u
.
F
N
v
a
l
ue
of
0
m
e
a
ns
no
bu
r
u
ng
k
u
au
da
ta
in
c
or
r
e
c
tl
y
c
la
s
s
if
ie
d
to
ot
he
r
c
l
a
s
s
e
s
.
A
n
a
c
c
ur
a
c
y
of
100%
m
e
a
ns
th
a
t
a
ll
da
t
a
on
bu
r
ung
k
ua
u
a
r
e
c
l
a
s
s
if
i
e
d
c
or
r
e
c
tl
y
. T
h
e
k
al
ig
r
a
fi
da
t
a
c
l
a
s
s
g
e
t
s
a
T
P
v
a
l
ue
of
94
da
t
a
c
or
r
e
c
t
ly
c
la
s
s
if
i
e
d
a
s
k
a
li
g
r
a
fi
.
A
n
F
P
va
lu
e
of
0
in
d
ic
a
t
e
s
th
a
t
no
d
a
t
a
f
r
om
o
th
e
r
c
l
a
s
s
e
s
i
s
in
c
or
r
e
c
tl
y
c
l
a
s
s
if
i
e
d
a
s
k
al
i
gr
af
i
. A
n
F
N
va
lu
e
of
2 i
nd
ic
a
t
e
s
t
h
a
t
th
e
a
m
ou
nt
of
da
ta
f
r
o
m
k
a
li
g
r
af
i
i
s
in
c
or
r
e
c
tl
y c
l
a
s
s
if
i
e
d
in
to
a
no
th
e
r
c
l
a
s
s
.
T
h
e
c
l
a
s
s
a
c
c
u
r
a
c
y
of
97
.9
2%
o
f
t
he
k
al
i
gr
af
i
da
t
a
i
s
c
or
r
e
c
tl
y
c
la
s
s
if
ie
d.
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
I
m
pr
ov
in
g t
he
t
r
ans
fe
r
l
e
ar
ni
ng f
o
r
bat
ik
be
s
u
r
e
k
t
e
x
ti
le
m
ot
if
c
la
s
s
if
ic
at
io
n
(
M
ar
is
s
a U
ta
m
i
)
3179
T
he
r
a
f
f
le
s
ia
da
ta
c
la
s
s
ge
ts
a
T
P
va
l
ue
o
f
9
6
da
ta
,
m
e
a
ni
ng
t
he
a
m
ou
nt
of
da
ta
is
c
o
r
r
e
c
tl
y
c
la
s
s
i
f
ie
d
a
s
r
a
f
f
le
s
ia
.
A
n
F
P
va
lu
e
o
f
0
m
e
a
ns
t
ha
t
n
o
da
ta
f
r
o
m
o
th
e
r
c
la
s
s
e
s
is
in
c
or
r
e
c
tl
y
c
la
s
s
i
f
ie
d
a
s
r
a
f
f
le
s
i
a
.
A
n
F
N
va
lu
e
o
f
0
m
e
a
ns
no
r
a
f
f
le
s
ia
da
ta
is
m
is
c
la
s
s
if
ie
d
t
o
a
n
ot
he
r
c
la
s
s
.
T
he
a
c
c
u
r
a
c
y
o
f
th
e
c
la
s
s
i
s
1
00
%
w
h
ic
h
m
e
a
ns
t
ha
t
a
ll
th
e
r
a
f
f
le
s
i
a
da
t
a
is
c
o
r
r
e
c
t
ly
c
la
s
s
i
f
ie
d.
T
he
r
e
lu
n
g
pa
k
u
da
ta
c
l
a
s
s
ge
ts
a
T
P
va
lu
e
o
f
9
6
da
ta
,
m
e
a
n
in
g
th
e
a
m
ou
nt
o
f
d
a
ta
is
c
o
r
r
e
c
t
l
y
c
l
a
s
s
i
f
ie
d
a
s
a
r
e
lu
ng
pa
k
u
.
A
n
F
P
va
lu
e
o
f
0
m
e
a
ns
t
ha
t
n
o
da
ta
f
r
om
o
th
e
r
c
l
a
s
s
e
s
is
in
c
or
r
e
c
tl
y
c
la
s
s
if
i
e
d
a
s
a
r
e
l
ung
p
ak
u
.
A
n
F
N
v
a
l
ue
of
0
w
hi
c
h
m
e
a
ns
no
da
ta
r
e
lu
n
g
p
ak
u
a
r
e
m
is
c
l
a
s
s
i
f
ie
d
to
ot
he
r
c
l
a
s
s
e
s
.
T
he
a
c
c
u
r
a
c
y
o
f
th
e
c
la
s
s
is
1
00
%
w
hi
c
h
m
e
a
ns
t
ha
t
a
l
l
r
e
lu
ng
p
ak
u
d
a
ta
is
c
o
r
r
e
c
t
ly
c
l
a
s
s
i
f
ie
d.
T
he
r
e
m
bu
la
n
da
t
a
c
la
s
s
ge
ts
a
T
P
va
lu
e
of
92,
m
e
a
n
in
g
th
e
a
m
o
un
t
o
f
da
ta
is
c
o
r
r
e
c
tl
y
c
la
s
s
if
ie
d
a
s
r
e
m
b
ul
an
.
A
n
F
P
va
lu
e
of
0
da
t
a
in
di
c
a
te
s
th
a
t
th
e
num
be
r
f
r
o
m
ot
h
e
r
c
la
s
s
e
s
is
in
c
o
r
r
e
c
t
ly
c
l
a
s
s
i
f
ie
d
a
s
r
e
m
bu
l
an
.
A
n
F
N
va
lu
e
o
f
4
in
di
c
a
te
s
t
he
a
m
oun
t
o
f
da
ta
f
r
om
th
e
r
e
m
bul
an
c
la
s
s
i
f
i
e
d
i
nt
o
a
no
th
e
r
c
l
a
s
s
.
T
he
a
c
c
ur
a
c
y
o
f
th
e
c
la
s
s
w
a
s
9
5.8
3
%
,
o
f
th
e
r
e
m
bu
la
n
da
ta
w
a
s
c
o
r
r
e
c
t
ly
c
la
s
s
if
ie
d.
U
s
in
g
R
e
s
N
e
t5
0
w
it
h
th
e
A
da
m
opt
im
iz
e
r
,
R
e
L
U
a
c
ti
va
ti
on
f
u
nc
ti
on,
a
nd
C
L
A
H
E
pr
e
pr
oc
e
s
s
in
g
is
hi
ghl
y
e
f
f
e
c
ti
ve
f
or
ba
ti
k
m
ot
i
f
c
la
s
s
if
ic
a
ti
on.
R
e
s
N
e
t5
0,
a
de
e
p
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k,
le
ve
r
a
ge
s
r
e
s
id
ua
l
c
onne
c
ti
ons
to
a
voi
d
va
ni
s
hi
ng
gr
a
di
e
nt
is
s
ue
s
,
e
na
bl
i
ng
it
to
le
a
r
n
c
om
pl
e
x
pa
tt
e
r
ns
w
it
hi
n
in
tr
ic
a
te
ba
ti
k
m
ot
if
s
.
T
he
A
da
m
opt
im
iz
e
r
,
known
f
or
it
s
a
da
pt
a
b
il
it
y
a
nd
e
f
f
ic
ie
nc
y,
e
nha
n
c
e
s
th
e
m
ode
l’
s
c
onve
r
ge
nc
e
,
m
a
ki
ng
it
w
e
ll
-
s
ui
te
d
f
or
ha
ndl
in
g
h
ig
h
-
va
r
ia
ti
on
da
ta
li
ke
ba
ti
k
pa
tt
e
r
ns
.
R
e
L
U
f
ur
th
e
r
a
id
s
by
a
ddi
ng
non
-
li
ne
a
r
it
y
a
nd
s
pa
r
s
it
y,
h
e
lp
in
g
th
e
m
ode
l
f
oc
u
s
on
e
s
s
e
nt
ia
l
f
e
a
tu
r
e
s
.
C
L
A
H
E
pr
e
pr
oc
e
s
s
in
g
e
nha
nc
e
s
c
ont
r
a
s
t
in
ba
ti
k
im
a
ge
s
,
m
a
ki
ng
s
ubt
le
de
ta
il
s
m
or
e
pr
onounc
e
d
a
nd
boos
ti
ng
f
e
a
tu
r
e
e
xt
r
a
c
ti
on,
ul
ti
m
a
te
ly
i
m
pr
ovi
ng c
la
s
s
if
ic
a
ti
on a
c
c
ur
a
c
y f
or
i
nt
r
ic
a
te
ba
ti
k m
ot
if
s
.
F
ut
ur
e
r
e
s
e
a
r
c
h
c
oul
d
e
xpl
or
e
e
nh
a
nc
in
g
ba
ti
k
m
ot
if
c
la
s
s
if
ic
a
ti
on
by
c
om
bi
ni
ng
R
e
s
N
e
t5
0
w
it
h
a
dva
nc
e
d
te
c
hni
que
s
s
u
c
h
a
s
a
tt
e
nt
io
n
m
e
c
ha
ni
s
m
s
or
f
e
a
tu
r
e
f
us
io
n
m
e
th
ods
to
f
ur
th
e
r
r
e
f
in
e
in
tr
ic
a
te
pa
tt
e
r
n
r
e
c
ogni
ti
on.
E
xpe
r
im
e
nt
in
g
w
it
h
ot
he
r
opt
im
iz
e
r
s
li
ke
R
M
S
pr
op
or
gr
a
di
e
nt
c
li
ppi
ng
m
a
y
a
l
s
o
s
ta
bi
li
z
e
tr
a
in
in
g
a
nd
im
pr
ove
p
e
r
f
or
m
a
nc
e
on
c
om
pl
e
x,
hi
gh
-
va
r
ia
nc
e
ba
ti
k
pa
tt
e
r
ns
.
A
ddi
ti
ona
ll
y,
us
in
g
tr
a
ns
f
e
r
le
a
r
ni
ng
f
r
om
ot
he
r
dom
a
in
s
or
e
m
pl
oyi
ng
hybr
id
m
ode
ls
th
a
t
in
te
gr
a
te
C
N
N
s
w
it
h
tr
a
ns
f
or
m
e
r
s
c
oul
d
yi
e
ld
m
or
e
r
obus
t
f
e
a
tu
r
e
r
e
pr
e
s
e
nt
a
ti
ons
.
E
xpa
ndi
ng
th
e
ba
ti
k
da
ta
s
e
t
a
nd
t
e
s
ti
ng
th
e
m
ode
l
on
v
a
r
io
us
m
ot
if
s
ty
le
s
,
c
ol
or
s
,
a
nd
f
a
br
ic
te
xt
ur
e
s
c
oul
d
pr
ovi
de
in
s
ig
ht
s
in
to
th
e
a
da
pt
a
bi
li
ty
of
R
e
s
N
e
t5
0,
f
ur
th
e
r
a
dva
nc
in
g a
ut
om
a
te
d ba
ti
k c
la
s
s
if
ic
a
ti
on t
e
c
hni
qu
e
s
.
4.
C
O
N
C
L
U
S
I
O
N
D
ig
it
a
l
im
a
ge
pr
oc
e
s
s
in
g
is
done
on
im
a
ge
s
to
obt
a
in
s
pe
c
if
ic
r
e
s
ul
ts
a
c
c
or
di
ng
to
ne
e
ds
.
U
s
in
g
di
gi
ta
l
im
a
ge
pr
oc
e
s
s
in
g,
w
e
c
a
n
c
la
s
s
if
y
s
e
ve
r
a
l
s
im
il
a
r
ba
ti
k
be
s
ur
e
k
im
a
ge
s
.
T
hi
s
m
e
th
od
c
a
n
be
one
w
a
y
to
s
ol
ve
th
e
pr
obl
e
m
of
th
e
in
tr
oduc
ti
on
of
ba
ti
k
be
s
ur
e
k
m
ot
if
s
.
B
a
s
e
d
on
th
e
e
xpe
r
im
e
nt
r
e
s
ul
ts
,
R
e
s
N
e
t5
0
pe
r
f
or
m
e
d
a
t
th
e
f
ir
s
t
-
be
s
t
w
it
h a
tr
a
in
in
g
a
c
c
ur
a
c
y
of
99.60%
, a
va
li
da
ti
on
a
c
c
ur
a
c
y
of
97.44%
, a
nd
a
te
s
ti
ng
a
c
c
ur
a
c
y of
98.12%
. T
he
M
obi
le
N
e
tV2 mode
l
obt
a
in
e
d t
he
s
e
c
ond
-
be
s
t
pe
r
f
or
m
a
nc
e
w
it
h a
t
r
a
in
in
g a
c
c
ur
a
c
y
of
98.14%
,
a
va
li
da
ti
on
a
c
c
ur
a
c
y
of
94.60%
a
nd
a
te
s
ti
ng
a
c
c
ur
a
c
y
of
96.46%
. T
he
V
G
G
16
m
ode
l
pe
r
f
or
m
e
d
a
t
th
e
th
ir
d
hi
ghe
s
t,
w
it
h
a
tr
a
in
in
g
a
c
c
ur
a
c
y
of
99.67%
,
a
va
li
da
ti
on
a
c
c
ur
a
c
y
of
96.02%
,
a
nd
a
te
s
ti
ng
a
c
c
ur
a
c
y
of
95.00%
.
T
h
e
I
nc
e
pt
io
nV
3
m
ode
l
pe
r
f
or
m
e
d
a
t
t
he
th
ir
d
hi
ghe
s
t,
w
it
h
a
tr
a
in
in
g
a
c
c
ur
a
c
y
of
96.35%
,
a
va
li
da
ti
on
a
c
c
ur
a
c
y
of
96.59
%
,
a
nd
a
te
s
ti
ng
a
c
c
ur
a
c
y
of
92.50%
.
T
he
R
e
s
N
e
t5
0
m
ode
l
w
it
h
A
da
m
opt
im
iz
e
r
,
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
a
nd
C
L
A
H
E
a
s
th
e
c
ont
r
a
s
t
e
nha
nc
e
m
e
nt
m
e
th
od
obt
a
in
e
d
th
e
hi
ghe
s
t
te
s
t
a
c
c
ur
a
c
y
(
98.75%
)
,
s
how
in
g
th
a
t
C
L
A
H
E
is
ve
r
y
e
f
f
e
c
ti
ve
in
im
pr
ovi
ng
pe
r
f
or
m
a
nc
e
on
ba
ti
k
be
s
ur
e
k
da
ta
.
T
he
n,
th
is
m
ode
l
a
ls
o
s
ho
w
s
hi
gh
tr
a
in
in
g
a
c
c
ur
a
c
y
(
a
bout
95
-
100%
)
a
nd
lo
w
tr
a
in
in
g
lo
s
s
a
nd
ve
r
y
lo
w
va
li
da
ti
on
a
c
c
ur
a
c
y
(
a
bout
27
-
20%
)
a
t
th
e
be
gi
nni
ng
of
th
e
e
poc
h.
T
hi
s
in
di
c
a
te
s
th
a
t
th
e
m
ode
l
is
ove
r
f
it
ti
ng t
he
t
r
a
in
in
g da
ta
a
nd c
a
nnot
ge
ne
r
a
li
z
e
t
he
va
li
da
ti
o
n da
ta
w
e
ll
.
A
C
K
N
O
WL
E
D
G
E
M
E
N
T
S
T
he
a
ut
hor
s
w
oul
d
li
ke
to
e
xpr
e
s
s
th
e
ir
gr
a
ti
tu
de
to
S
r
iwi
ja
ya
U
ni
ve
r
s
it
y
,
U
ni
ve
r
s
it
a
s
M
uha
m
m
a
di
ya
h
B
e
ngkulu,
D
e
pa
r
tm
e
nt
of
E
duc
a
ti
on
a
nd
C
ul
t
ur
e
of
B
e
ngkulu,
a
nd
th
e
M
in
is
tr
y
of
H
ig
he
r
E
duc
a
ti
on, S
c
ie
nc
e
, a
nd T
e
c
hnol
ogy f
or
t
he
ir
s
uppor
t
in
t
hi
s
r
e
s
e
a
r
c
h.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
W
e
a
knowle
dge
to
S
r
iwi
ja
ya
U
ni
ve
r
s
it
y
a
nd
D
ir
e
c
to
r
a
te
o
f
R
e
s
e
a
r
c
h,
T
e
c
hnol
ogy,
a
nd
C
om
m
uni
ty
S
e
r
vi
c
e
;
D
ir
e
c
to
r
a
te
G
e
ne
r
a
l
of
H
ig
he
r
E
duc
a
ti
on,
R
e
s
e
a
r
c
h,
a
nd
T
e
c
hnol
ogy;
M
in
is
tr
y
of
E
duc
a
ti
on,
C
ul
tu
r
e
,
R
e
s
e
a
r
c
h
a
nd
T
e
c
hnol
ogy
of
th
e
R
e
publ
ic
of
I
ndo
ne
s
ia
th
r
ough
r
e
s
e
a
r
c
h
f
undi
ng
w
it
h
num
be
r
090/
E
5/
P
G
.02.00.P
L
/2
024 c
q. 0016.030/UN9/S
B
1.L
P
2M
.P
T
/2
024.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
4
,
A
ugus
t
20
25
:
3172
-
3181
3180
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
di
T
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on.
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
M
a
r
is
s
a
U
t
a
m
i
✓
✓
✓
✓
✓
✓
✓
E
r
m
a
ti
ta
E
r
m
a
ti
ta
✓
✓
✓
✓
✓
✓
A
bdi
a
ns
a
h A
bdi
a
n
s
a
h
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
l
i
z
a
t
i
on
M
:
M
e
t
hodol
ogy
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
da
t
i
on
Fo
:
Fo
r
m
a
l
a
na
l
ys
i
s
I
:
I
nve
s
t
i
ga
t
i
on
R
:
R
e
s
our
c
e
s
D
:
D
a
t
a
C
ur
a
t
i
on
O
:
W
r
i
t
i
ng
-
O
r
i
gi
na
l
D
r
a
f
t
E
:
W
r
i
t
i
ng
-
R
e
vi
e
w
&
E
di
t
i
ng
Vi
:
Vi
s
ua
l
i
z
a
t
i
on
Su
:
Su
pe
r
vi
s
i
on
P
:
P
r
oj
e
c
t
a
dm
i
ni
s
t
r
a
t
i
on
Fu
:
Fu
ndi
ng a
c
qui
s
i
t
i
on
C
O
N
F
L
I
C
T
O
F
I
N
T
E
R
E
S
T
S
T
A
T
E
M
E
N
T
T
he
a
ut
hor
s
de
c
la
r
e
th
a
t
th
e
y
h
a
ve
no
kno
w
n
c
om
pe
ti
ng
f
in
a
nc
ia
l
in
te
r
e
s
ts
or
p
e
r
s
ona
l
r
e
la
ti
ons
hi
p
s
th
a
t
c
oul
d ha
ve
a
ppe
a
r
e
d t
o i
nf
lu
e
nc
e
t
h
e
w
or
k r
e
por
te
d i
n t
hi
s
pa
pe
r
.
A
ut
hor
s
s
ta
te
no c
onf
li
c
t
of
i
nt
e
r
e
s
t.
I
N
F
O
R
M
E
D
C
O
N
S
E
N
T
W
e
ha
ve
obt
a
in
e
d i
nf
or
m
e
d c
ons
e
nt
f
r
om
a
ll
i
ndi
vi
dua
ls
i
nc
lu
de
d i
n t
hi
s
s
tu
dy.
D
A
T
A
A
V
A
I
L
A
B
I
L
I
T
Y
T
he
da
ta
th
a
t
s
uppor
t
th
e
f
in
di
ngs
of
th
is
s
tu
dy
a
r
e
a
va
il
a
bl
e
on
r
e
que
s
t
f
r
om
th
e
a
ut
hor
,
[
M
U
]
.
T
he
da
ta
a
r
e
not
publi
c
ly
a
v
a
il
a
bl
e
due
t
o c
e
r
ta
in
r
e
s
tr
ic
ti
ons
.
R
E
F
E
R
E
N
C
E
S
[
1]
I
. N
ur
ha
i
da
, R
. A
. M
. Z
e
n, V
. A
yum
i
,
a
nd H
. W
e
i
,
“
D
e
t
e
r
m
i
ni
ng t
he
num
be
r
of
ba
t
i
k m
ot
i
f
obj
e
c
t
ba
s
e
d on hi
e
r
a
r
c
hi
c
a
l
s
ym
m
e
t
r
y
de
t
e
c
t
i
on
a
ppr
oa
c
h,”
I
ndone
s
i
an
J
our
nal
of
E
l
e
c
t
r
i
c
al
E
ngi
ne
e
r
i
ng
and
I
nf
or
m
at
i
c
s
,
vol
.
9,
no.
1,
pp.
141
–
152,
2021,
doi
:
10.11591/
i
j
e
e
i
.v9i
1.2369.
[
2]
S
.
I
.
S
.
S
ha
ha
r
uddi
n
e
t
al
.
,
“
A
r
e
vi
e
w
on
t
he
M
a
l
a
ys
i
a
n
a
nd
I
ndone
s
i
a
n
ba
t
i
k
pr
oduc
t
i
on,
c
ha
l
l
e
nge
s
,
a
nd
i
nnova
t
i
ons
i
n
t
he
21s
t
c
e
nt
ur
y,”
Sage
O
pe
n
, vol
. 11, no. 3, pp. 1
–
19, 2021, doi
:
10.1177/
21582440211040128.
[
3]
S
.
A
l
a
m
,
A
.
B
udi
m
a
n,
D
.
H
i
da
ya
t
,
a
nd
S
una
r
t
o,
“
B
a
t
i
k
B
om
ba
:
S
ym
bo
l
i
c
i
nt
e
r
a
c
t
i
on
t
hr
ough
a
r
t
w
or
k,”
i
n
Su
s
t
ai
nabl
e
D
e
v
e
l
opm
e
nt
i
n
C
r
e
at
i
v
e
I
ndus
t
r
i
e
s
:
E
m
br
ac
i
ng
D
i
gi
t
al
C
ul
t
ur
e
f
or
H
um
ani
t
i
e
s
,
L
ondon:
R
out
l
e
dge
,
2023,
pp.
1
–
6
,
doi
:
10.1201/
9781003372486
-
1.
[
4]
D
.
T
.
S
e
t
i
a
w
a
n
a
nd
B
.
W
i
r
j
o
di
r
d
j
o
,
“
T
h
e
de
ve
l
op
m
e
n
t
s
t
r
a
t
e
gy
o
f
b
a
t
i
k
s
m
a
l
l
a
n
d
m
e
d
i
u
m
e
nt
e
r
p
r
i
s
e
s
(
S
M
E
)
i
n
K
a
m
p
un
g
B
a
t
i
k
J
e
t
i
s
S
i
d
oa
r
j
o
,”
I
O
P
C
on
f
e
r
e
nc
e
Se
r
i
e
s
:
E
ar
t
h
a
n
d
E
nv
i
r
on
m
e
nt
a
l
Sc
i
e
nc
e
,
vo
l
.
56
2
,
no
.
1
,
2
0
20
,
d
oi
:
1
0.
1
08
8
/
1
75
5
-
1
3
15
/
56
2/
1
/
0
1
20
23
.
[
5]
F
.
A
.
P
ut
r
a
e
t
al
.
,
“
C
l
a
s
s
i
f
i
c
a
t
i
on
of
ba
t
i
k
a
ut
he
nt
i
c
i
t
y
us
i
ng
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k
a
l
gor
i
t
hm
w
i
t
h
t
r
a
ns
f
e
r
l
e
a
r
ni
ng
m
e
t
hod,”
i
n
2021
Si
x
t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
I
nf
or
m
at
i
c
s
and
C
om
put
i
ng
(
I
C
I
C
)
,
2021,
pp.
1
–
6
,
doi
:
10.1109/
I
C
I
C
54025.2021.9632937.
[
6]
H
.
N
opr
i
s
s
on,
E
.
E
r
m
a
t
i
t
a
,
A
.
A
bdi
a
ns
a
h,
V
.
A
yum
i
,
M
.
P
u
r
ba
,
a
nd
H
.
S
e
t
i
a
w
a
n,
“
F
i
ne
-
t
uni
ng
t
r
a
ns
f
e
r
l
e
a
r
ni
ng
m
ode
l
i
n
w
ove
n
f
a
br
i
c
pa
t
t
e
r
n
c
l
a
s
s
i
f
i
c
a
t
i
on,”
I
nt
e
r
nat
i
onal
J
our
nal
of
I
nnov
at
i
v
e
C
om
put
i
ng,
I
nf
or
m
at
i
on
and
C
ont
r
ol
,
vol
.
18,
no.
6,
pp. 1885
–
1894, 2022, doi
:
10.24507/
i
j
i
c
i
c
.18.06.1885.
[
7]
E
r
m
a
t
i
t
a
,
H
.
N
opr
i
s
s
on,
a
nd
A
bdi
a
ns
a
h,
“
P
a
l
e
m
ba
ng
s
ongke
t
f
a
br
i
c
m
ot
i
f
i
m
a
ge
de
t
e
c
t
i
on
w
i
t
h
da
t
a
a
ugm
e
nt
a
t
i
on
ba
s
e
d
o
n
R
e
s
N
e
t
us
i
ng
dr
opout
,”
B
ul
l
e
t
i
n
of
E
l
e
c
t
r
i
c
al
E
ngi
ne
e
r
i
ng
and
I
nf
or
m
at
i
c
s
,
vol
.
13,
no.
3,
pp.
1991
–
1999,
2024,
doi
:
10.11591/
e
e
i
.v13i
3.6883.
[
8]
D
.
S
i
na
ga
,
C
.
J
a
t
m
oko,
a
nd
N
.
H
e
ndr
i
ya
nt
o,
“
M
ul
t
i
-
l
a
ye
r
c
onvol
ut
i
ona
l
ne
u
r
a
l
ne
t
w
or
ks
f
or
ba
t
i
k
i
m
a
ge
c
l
a
s
s
i
f
i
c
a
t
i
on,”
Sc
i
e
nt
i
f
i
c
J
our
nal
of
I
nf
or
m
at
i
c
s
, vol
. 11, no. 2, pp. 477
–
484, 2024.
[
9]
Z
.
M
.
M
a
ya
s
a
r
i
,
“
A
ppl
i
c
a
t
i
on
of
f
r
a
c
t
a
l
pr
i
nc
i
pl
e
s
i
n
r
e
de
s
i
gni
ng
a
n
a
r
a
bi
c
c
a
l
l
i
gr
a
phy
a
nd
r
a
f
f
l
e
s
i
a
f
l
ow
e
r
m
ot
i
f
i
n
ba
t
i
k
be
s
ur
e
k
,”
J
our
nal
of
P
hy
s
i
c
s
:
C
onf
e
r
e
nc
e
Se
r
i
e
s
, vol
. 1731, no. 1, 2021, doi
:
10
.1088/
1742
-
6596/
1731/
1/
012029.
[
10]
Z
. A
bi
di
n, M
. S
upr
i
a
t
na
, T
. H
e
r
m
a
n,
L
. F
a
r
okha
h, a
nd
R
. F
e
br
i
a
ndi
,
“
T
he
ge
o
m
e
t
r
i
c
pa
t
t
e
r
ns
i
n K
a
w
ung
S
ur
a
ka
r
t
a
ba
t
i
k m
ot
i
f
:
a
n
e
t
hnom
a
t
he
m
a
t
i
c
a
l
e
xpl
or
a
t
i
on,”
A
I
P
C
onf
e
r
e
nc
e
P
r
oc
e
e
di
ng
s
, vol
. 2727, 2023, doi
:
10.1063/
5.0141663.
[
11]
H
.
S
a
s
t
ypr
a
t
i
w
i
,
H
.
M
uha
r
di
,
a
nd
Y
.
Y
ul
i
a
nt
i
,
“
B
a
t
i
k
r
e
c
ogni
t
i
on
a
nd
c
l
a
s
s
i
f
i
c
a
t
i
on
us
i
ng
t
r
a
ns
f
e
r
l
e
a
r
ni
ng
a
nd
M
obi
l
e
N
e
t
a
ppr
oa
c
h,”
J
O
I
V
:
I
nt
e
r
nat
i
onal
J
our
nal
on
I
nf
or
m
at
i
c
s
V
i
s
ual
i
z
at
i
on
,
vol
.
8,
no.
4,
pp.
2400
–
2410,
2024,
doi
:
10.62527/
j
oi
v.8.4.2407.
[
12]
A
.
H
.
R
a
ngkut
i
,
A
.
H
a
r
j
oko,
a
nd
A
.
P
ut
r
a
,
“
A
nove
l
r
e
l
i
a
bl
e
a
ppr
oa
c
h
f
or
i
m
a
ge
ba
t
i
k
c
l
a
s
s
i
f
i
c
a
t
i
on
t
h
a
t
i
nva
r
i
a
nt
w
i
t
h
s
c
a
l
e
a
n
d
r
ot
a
t
i
on
us
i
ng
M
U
2E
C
S
-
L
B
P
a
l
gor
i
t
hm
,”
P
r
oc
e
di
a
C
om
put
e
r
Sc
i
e
nc
e
,
vol
.
179,
pp.
863
–
870,
2021,
doi
:
10.1016/
j
.pr
oc
s
.2021.01.075.
[
13]
J
.
K
us
a
nt
i
a
nd
A
.
S
upr
a
pt
o,
“
C
om
bi
na
t
i
on
of
O
t
s
u
a
nd
C
a
nny
m
e
t
hod
t
o
i
de
nt
i
f
y
t
he
c
ha
r
a
c
t
e
r
i
s
t
i
c
s
of
S
ol
o
B
a
t
i
k
a
s
S
ur
a
ka
r
t
a
t
r
a
di
t
i
ona
l
ba
t
i
k,”
i
n
2019
2nd
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
of
C
om
put
e
r
and
I
nf
or
m
at
i
c
s
E
ngi
ne
e
r
i
ng
(
I
C
2I
E
)
,
S
e
p.
2019,
pp. 63
–
68
, doi
:
10.1109/
I
C
2I
E
47452.2019.8940884.
[
14]
R
.
A
ndr
i
a
n,
M
.
A
.
N
a
uf
a
l
,
B
.
H
e
r
m
a
nt
o,
A
.
J
una
i
di
,
a
nd
F
.
R
.
L
um
ba
nr
a
j
a
,
“
k
-
ne
a
r
e
s
t
n
e
i
ghbor
(
k
-
N
N
)
c
l
a
s
s
i
f
i
c
a
t
i
on
f
or
r
e
c
ogni
t
i
on
of
t
he
B
a
t
i
k
L
a
m
pung
m
ot
i
f
s
,”
J
our
nal
of
P
hy
s
i
c
s
:
C
onf
e
r
e
nc
e
Se
r
i
e
s
,
vol
.
1338,
2019,
doi
:
10.1088/
1742
-
6596/
1338/
1/
012061.
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
I
m
pr
ov
in
g t
he
t
r
ans
fe
r
l
e
ar
ni
ng f
o
r
bat
ik
be
s
u
r
e
k
t
e
x
ti
le
m
ot
if
c
la
s
s
if
ic
at
io
n
(
M
ar
is
s
a U
ta
m
i
)
3181
[
15]
N
.
D
.
G
i
r
s
a
ng
a
nd
M
uha
t
hi
r
,
“
C
l
a
s
s
i
f
i
c
a
t
i
on
of
ba
t
i
k
i
m
a
ge
s
us
i
ng
m
ul
t
i
l
a
ye
r
pe
r
c
e
pt
r
on
w
i
t
h
hi
s
t
ogr
a
m
of
or
i
e
nt
e
d
gr
a
di
e
nt
f
e
a
t
ur
e
e
xt
r
a
c
t
i
on,”
i
n
P
r
oc
e
e
di
ng I
nt
e
r
nat
i
ona
l
C
onf
e
r
e
n
c
e
Sc
i
e
nc
e
E
ngi
ne
e
r
i
ng
, 2021, vol
. 4, pp. 197
–
204.
[
16]
A
.
R
i
s
ki
,
E
.
B
.
W
i
na
t
a
,
a
nd
A
.
K
a
m
s
ya
ka
w
uni
,
“
P
a
t
t
e
r
n
r
e
c
ogni
t
i
on
of
B
a
t
i
k
M
a
dur
a
us
i
ng
ba
c
kpr
opa
ga
t
i
on
a
l
gor
i
t
hm
,”
i
n
P
r
oc
e
e
di
ngs
of
t
he
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
M
at
he
m
at
i
c
s
,
G
e
om
e
t
r
y
,
St
at
i
s
t
i
c
s
,
and
C
om
put
at
i
on
(
I
C
-
M
aG
e
St
i
C
2021)
,
2022,
pp. 238
–
243
, doi
:
10.2991/
a
c
s
r
.k.220202.044.
[
17]
B
.
P
.
H
.
K
.
M
.
D
.
S
e
na
r
a
t
hna
a
nd
R
.
M
.
T
.
P
.
R
a
j
a
ka
r
una
,
“
F
e
a
t
ur
e
de
s
c
r
i
pt
or
f
or
S
r
i
L
a
nka
n
B
a
t
i
k
pa
t
t
e
r
ns
us
i
ng
hu
m
om
e
nt
i
nva
r
i
a
nt
s
a
nd
G
L
C
M
,
”
i
n
2021
10t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
I
nf
or
m
at
i
on
and
A
ut
om
at
i
on
f
or
Sus
t
ai
nabi
l
i
t
y
(
I
C
I
A
f
S)
,
2021,
pp. 197
–
202
, doi
:
10.1109/
I
C
I
A
f
S
52090.2021.9606106.
[
18]
K
. M
a
ha
r
a
na
,
S
. M
onda
l
,
a
nd B
.
N
e
m
a
de
,
“
A
r
e
vi
e
w
:
D
a
t
a
pr
e
-
pr
oc
e
s
s
i
ng
a
nd
da
t
a
a
ugm
e
nt
a
t
i
on t
e
c
hni
que
s
,”
G
l
obal
T
r
ans
i
t
i
on
s
P
r
oc
e
e
di
ngs
, vol
. 3, no. 1, pp. 91
–
99, 2022, doi
:
10.1016/
j
.gl
t
p.2022.04.020.
[
19]
K
.
B
i
a
n
a
nd
R
.
P
r
i
ya
da
r
s
hi
,
“
M
a
c
hi
ne
l
e
a
r
ni
ng
opt
i
m
i
z
a
t
i
on
t
e
c
hni
que
s
:
a
s
ur
ve
y,
c
l
a
s
s
i
f
i
c
a
t
i
on,
c
ha
l
l
e
nge
s
,
a
nd
f
ut
ur
e
r
e
s
e
a
r
c
h
i
s
s
ue
s
,
”
A
r
c
hi
v
e
s
of
C
om
put
at
i
onal
M
e
t
hods
i
n E
ngi
ne
e
r
i
ng
, 2024, doi
:
10.100
7/
s
11831
-
024
-
10110
-
w.
[
20]
P
.
S
ha
r
m
a
a
nd
R
.
S
.
A
na
nd,
“
A
c
om
pr
e
he
ns
i
ve
e
va
l
ua
t
i
on
of
de
e
p
m
ode
l
s
a
nd
opt
i
m
i
z
e
r
s
f
or
I
ndi
a
n
s
i
gn
l
a
ngua
ge
r
e
c
ogni
t
i
on,”
G
r
aphi
c
s
and V
i
s
ual
C
om
put
i
ng
, vol
. 5, 2021, doi
:
10.1016/
j
.gvc
.2021.200032.
[
21]
N
.
C
ha
khi
m
,
M
.
L
ouz
a
r
,
A
.
L
a
m
ni
i
,
a
nd
M
.
A
l
a
oui
,
“
I
m
a
ge
r
e
c
ons
t
r
uc
t
i
on
i
n
di
f
f
us
e
opt
i
c
a
l
t
om
ogr
a
phy
us
i
ng
a
da
pt
i
ve
m
om
e
n
t
gr
a
di
e
nt
ba
s
e
d
opt
i
m
i
z
e
r
s
:
A
s
t
a
t
i
s
t
i
c
a
l
s
t
udy,”
A
ppl
i
e
d
Sc
i
e
nc
e
s
,
vol
.
10,
no.
24,
pp.
1
–
18,
2020,
doi
:
10.3390/
a
pp10249117.
[
22]
S
. R
. D
ube
y, S
. K
. S
i
ngh,
a
nd B
.
B
. C
h
a
udhur
i
, “
A
c
t
i
va
t
i
on f
unc
t
i
ons
i
n de
e
p l
e
a
r
ni
ng:
A
c
om
pr
e
he
ns
i
ve
s
ur
ve
y a
nd be
n
c
hm
a
r
k,”
N
e
ur
oc
om
put
i
ng
, vol
. 503, pp. 92
–
108, 2022, doi
:
10.1016/
j
.ne
uc
om
.2022.06.1
11.
[
23]
G
.
H
a
bi
b
a
nd
S
.
Q
ur
e
s
hi
,
“
O
pt
i
m
i
z
a
t
i
on
a
nd
a
c
c
e
l
e
r
a
t
i
on
of
c
onvol
ut
i
ona
l
n
e
ur
a
l
ne
t
w
or
ks
:
A
s
ur
ve
y,”
J
our
nal
of
K
i
ng
Sau
d
U
ni
v
e
r
s
i
t
y
-
C
om
put
e
r
and I
nf
or
m
at
i
on Sc
i
e
nc
e
s
, vol
. 34, no. 7, pp. 4244
–
4268
, 2022, doi
:
10.1016/
j
.j
ks
uc
i
.2020.10.004.
[
24]
H
.
A
bde
l
-
N
a
bi
,
G
.
A
l
-
N
a
ym
a
t
,
M
.
Z
.
A
l
i
,
a
nd
A
.
A
w
a
j
a
n,
“
H
c
L
S
H
:
A
nove
l
non
-
l
i
ne
a
r
m
onot
oni
c
a
c
t
i
va
t
i
on
f
unc
t
i
on
f
or
de
e
p
l
e
a
r
ni
ng m
e
t
hods
,”
I
E
E
E
A
c
c
e
s
s
, vol
. 11, pp. 47794
–
47815, 2023, doi
:
10.1109
/
A
C
C
E
S
S
.2023.3276298.
[
25]
M
.
T
.
R
a
s
he
e
d,
D
.
S
hi
,
a
nd
H
.
K
ha
n,
“
A
c
om
pr
e
he
ns
i
ve
e
xpe
r
i
m
e
nt
-
ba
s
e
d
r
e
vi
e
w
of
l
ow
-
l
i
ght
i
m
a
ge
e
nha
nc
e
m
e
nt
m
e
t
hods
a
nd
be
nc
hm
a
r
ki
ng l
ow
-
l
i
ght
i
m
a
ge
qua
l
i
t
y a
s
s
e
s
s
m
e
nt
,”
Si
gnal
P
r
oc
e
s
s
i
ng
, vol
. 20
4, 2023, doi
:
10.1016/
j
.s
i
gpr
o.2022.108821.
[
26]
W
.
W
a
ng,
X
.
W
u,
X
.
Y
ua
n,
a
nd
Z
.
G
a
o,
“
A
n
e
xpe
r
i
m
e
nt
-
ba
s
e
d
r
e
vi
e
w
of
l
ow
-
l
i
ght
i
m
a
ge
e
nha
nc
e
m
e
nt
m
e
t
hod
s
,”
I
E
E
E
A
c
c
e
s
s
,
vol
. 8, pp. 87884
–
87917, 2020, doi
:
10.1109/
A
C
C
E
S
S
.2020.2992749.
[
27]
J
.
L
i
,
X
.
F
e
ng,
a
nd
Z
.
H
ua
,
“
L
ow
-
l
i
ght
i
m
a
ge
e
nha
nc
e
m
e
nt
vi
a
pr
ogr
e
s
s
i
ve
-
r
e
c
ur
s
i
ve
ne
t
w
or
k,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
C
i
r
c
ui
t
s
and Sy
s
t
e
m
s
f
o
r
V
i
de
o T
e
c
hnol
ogy
, vol
. 31, no. 11, pp. 4227
–
4240, 2021, doi
:
10.1109/
T
C
S
V
T
.2021.3049940.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
S
Marissa
Utami
is
lecturer
of
Information
System
in
Universitas
M
uhammadiyah
Bengkulu,
Indones
ia. She is
curre
ntly studying
at
Doctor
al Progr
am in
Enginee
ring, Fac
ulty of
Engineering,
Sriwija
ya
Univer
sity
,
Palembang,
Indonesia.
She
recei
ved
a
master’s
degrees
from Master o
f Informatics,
Universit
as AMIKOM Yogy
akarta. Her r
esearch interests
are
data
science and
information system
. She can be contacted a
t email: marissautami@
umb.ac.id.
Ermatit
a
Ermatit
a
received
a
mathematics
bachelor
from
Univers
itas
Lampung,
a
master’s
degree
in
computer
science
from
Universitas
Indonesia,
a
nd
a
doctoral
degree
in
Computer Sc
ience
from Un
iversita
s
Gadjah
Mada.
She
is curr
ently wo
rking in th
e Depa
rtmen
t
of
Computer
Science,
Faculty
of
Computer
Science,
Sriwija
ya
Un
iversit
y
,
Indonesia.
Her
researches
include
artificial
intellig
ence,
data
mining,
machine
learning,
and
informatio
n
systems.
Her
most
cited
researc
h
articles
are
related
to
electric
me
thods
in
solving
group
decision
support
system
bioinformatics
on
gene
mutation
detect
ion
simulation.
She
can
be
contacted
at the em
ail: erm
atita@
unsri.ac.
id
.
Abdiansa
h
Abdiansa
h
is
lecturer
of
the
Department
of
Com
puter
Science,
Faculty
of
Computer
Scienc
e,
Sriwija
ya
Univer
sity
,
Indonesia
.
H
e
re
ceived
doctoral
degrees
from
Universitas
Gadjah
Mada.
His
research
interests
are
artific
ial
intelligence,
natural
language
processing
,
and
intelligent
tutoring
system
.
His
most
cited
r
esearch
is
related
to
the
time
complexity
analysis
of
support
vector
machines
(SVM)
in
Li
bSVM
and
a
survey
on
answer
validation
for
the
Indonesian
question
answering
system
(IQA
S).
He
can
be
contacted
at email
: abdian
sah@
unsri.ac.
id
.
Evaluation Warning : The document was created with Spire.PDF for Python.