I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
p
u
t
er
Science
Vo
l.
1
2
,
No
.
2
,
N
o
v
e
m
b
er
201
8
,
p
p
.
4
5
5
~
4
6
0
I
SS
N:
2502
-
4752
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ee
cs
.
v
1
2
.i
2
.
p
p
455
-
4
6
0
455
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
co
r
e.
co
m/jo
u
r
n
a
ls
/in
d
ex
.
p
h
p
/
ijeec
s
Deep Lea
rni
ng
f
o
r Ro
m
a
n
H
a
ndw
ritt
en
Chara
cter
R
ecog
nition
M
uh
a
a
f
idz
M
d Sa
ufi,
M
o
hd
Af
iq
Z
a
m
a
nh
uri,
No
ra
s
ia
h M
o
ha
m
m
a
d a
nd
Z
a
ida
h I
br
a
hi
m
F
a
c
u
lt
y
o
f
Co
m
p
u
ter an
d
M
a
th
e
m
a
ti
c
a
l
S
c
ien
c
e
s
,
Un
iv
e
rsiti
T
e
k
n
o
lo
g
i
M
A
R
A
,
S
h
a
h
A
la
m
,
S
e
lan
g
o
r
,
M
a
lay
sia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
y
9
,
2
0
1
8
R
ev
i
s
ed
J
u
l
1
2
,
2
0
1
8
A
cc
ep
ted
Ju
l
30
,
2
0
1
8
T
h
e
a
d
v
a
n
tag
e
o
f
d
e
e
p
lea
rn
in
g
i
s
th
a
t
t
h
e
a
n
a
ly
sis
a
n
d
lea
rn
in
g
o
f
m
a
s
siv
e
a
m
o
u
n
ts
o
f
u
n
su
p
e
rv
ise
d
d
a
ta
m
a
k
e
it
a
b
e
n
e
f
icia
l
to
o
l
f
o
r
Big
Da
t
a
a
n
a
ly
si
s.
Co
n
v
o
l
u
ti
o
n
Ne
u
ra
l
Ne
tw
o
rk
(C
NN
)
is
a
d
e
e
p
lea
rn
in
g
m
e
th
o
d
th
a
t
c
a
n
b
e
u
se
d
to
c
las
sify
i
m
a
g
e
,
c
lu
ste
r
th
e
m
b
y
si
m
il
a
rit
y
,
a
n
d
p
e
rf
o
rm
i
m
a
g
e
re
c
o
g
n
it
io
n
i
n
th
e
sc
e
n
e
.
T
h
is
p
a
p
e
r
c
o
n
d
u
c
ts
a
c
o
m
p
a
ra
ti
v
e
stu
d
y
b
e
t
w
e
e
n
th
re
e
d
e
e
p
lea
rn
in
g
m
o
d
e
ls,
w
h
ich
a
re
si
m
p
le
-
CNN
,
A
l
e
x
Ne
t
a
n
d
G
o
o
g
LeN
e
t
f
o
r
Ro
m
a
n
h
a
n
d
w
ri
tt
e
n
c
h
a
ra
c
ter
re
c
o
g
n
it
io
n
u
si
n
g
Ch
a
rs7
4
K
d
a
tas
e
t.
T
h
e
p
ro
d
u
c
e
d
re
su
lt
s
i
n
d
ica
te
t
h
a
t
G
o
o
leN
e
t
a
c
h
iev
e
s
th
e
b
e
st
a
c
c
u
ra
c
y
b
u
t
it
re
q
u
ires
a
l
o
n
g
e
r
ti
m
e
to
a
c
h
iev
e
su
c
h
re
su
lt
w
h
il
e
A
lex
Ne
t
p
ro
d
u
c
e
s les
s ac
c
u
ra
te res
u
lt
b
u
t
a
t
a
f
a
ste
r
ra
te.
K
ey
w
o
r
d
s
:
A
le
x
Net
C
NN
Dee
p
L
ea
r
n
i
n
g
Go
o
g
L
eNe
t
Han
d
w
r
itte
n
Dig
it R
ec
o
g
n
it
io
n
Co
p
y
rig
h
t
©
2
0
1
8
In
stit
u
te o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
rin
g
a
n
d
S
c
ien
c
e
.
All
rig
h
ts
re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Mu
h
aa
f
id
z
Md
Sau
f
i
,
Facu
lt
y
o
f
C
o
m
p
u
ter
an
d
Ma
t
h
e
m
a
tical
Scie
n
ce
s
,
Un
i
v
er
s
iti T
ek
n
o
lo
g
i M
A
R
A
,
Sh
a
h
A
la
m
,
Selan
g
o
r
,
Ma
la
y
s
i
a
.
E
m
ail:
m
u
h
aa
f
id
z4
ev
er
@
g
m
ai
l.c
o
m
1.
I
NT
RO
D
UCT
I
O
N
On
e
o
f
th
e
m
o
s
t
w
id
el
y
u
s
ed
ap
p
licatio
n
s
o
f
A
r
t
if
icial
Ne
u
r
al
Net
w
o
r
k
is
C
h
ar
ac
ter
R
ec
o
g
n
itio
n
[
1
-
2
]
.
Dee
p
lear
n
in
g
is
o
n
e
th
e
m
o
s
t
p
o
w
er
f
u
l
m
ac
h
in
e
lear
n
i
n
g
m
et
h
o
d
o
lo
g
y
f
o
r
s
o
l
v
i
n
g
cr
it
ical
p
r
o
b
lem
s
o
f
i
m
ag
e
p
r
o
ce
s
s
in
g
,
co
m
p
u
ter
v
is
io
n
,
n
at
u
r
al
la
n
g
u
a
g
e
p
r
o
ce
s
s
i
n
g
a
n
d
s
i
g
n
al
p
r
o
ce
s
s
in
g
.
A
k
e
y
ad
v
an
ta
g
e
o
f
d
ee
p
lear
n
in
g
is
th
at
th
e
a
n
al
y
s
i
s
an
d
lear
n
i
n
g
o
f
m
as
s
i
v
e
a
m
o
u
n
t
s
o
f
u
n
s
u
p
er
v
is
ed
d
ata
m
ak
e
i
t
a
b
en
ef
icial
to
o
l
f
o
r
B
ig
Data
an
al
y
s
is
[
3
]
.
T
h
u
s
,
d
ee
p
lear
n
in
g
o
f
ten
p
r
o
d
u
ce
s
g
o
o
d
r
esu
lt
s
.
Nev
er
th
ele
s
s
,
w
e
m
u
s
t
s
a
y
th
at
d
ee
p
lear
n
i
n
g
ap
p
r
o
ac
h
es
r
eq
u
ir
e
h
i
g
h
co
m
p
u
tin
g
r
eso
u
r
ce
s
co
m
p
ar
ed
t
o
m
o
r
e
tr
ad
itio
n
al
m
ac
h
in
e
lear
n
i
n
g
ap
p
r
o
ac
h
es.
I
n
d
ee
d
it n
ec
ess
i
tates a
p
o
w
er
f
u
l G
P
U
an
d
a
b
ig
d
atab
ase
[
4
].
Ov
er
t
h
e
p
as
t
f
e
w
y
ea
r
s
,
d
ee
p
lear
n
in
g
o
n
r
e
m
o
te
s
e
n
s
in
g
i
m
ag
er
y
s
h
o
w
s
e
x
ce
p
tio
n
al
l
y
g
o
o
d
r
esu
lt
s
o
n
b
o
th
o
p
tical
(
h
y
p
er
s
p
ec
tr
al
an
d
m
u
lti
s
p
ec
tr
al
i
m
ag
er
y
)
an
d
r
ad
ar
im
a
g
es
a
n
d
i
n
m
ap
p
in
g
la
n
d
ar
ea
s
.
Sce
n
e
class
i
f
icatio
n
o
f
h
i
g
h
r
eso
lu
t
i
o
n
r
e
m
o
te
s
e
n
s
in
g
i
m
a
g
e
p
la
y
s
a
m
aj
o
r
r
o
le
f
o
r
a
v
ar
io
u
s
t
y
p
e
s
o
f
ap
p
licatio
n
,
f
o
r
in
s
ta
n
ce
la
n
d
-
u
s
e
d
eter
m
i
n
atio
n
,
n
atu
r
al
h
az
ar
d
d
etec
tio
n
,
n
et
w
o
r
k
i
n
tr
u
s
io
n
d
etec
ti
o
n
[
5
]
,
g
eo
s
p
atial
o
b
j
ec
t
d
etec
tio
n
,
an
d
u
r
b
an
p
lan
n
in
g
[
6
]
.
R
ec
en
tl
y
,
d
ee
p
lear
n
in
g
m
et
h
o
d
esp
ec
iall
y
co
n
v
o
l
u
tio
n
a
l
n
e
u
r
al
n
et
w
o
r
k
h
av
e
s
h
o
w
n
th
e
ir
m
u
ch
s
tr
o
n
g
er
f
ea
t
u
r
e
r
ep
r
esen
tat
io
n
p
o
w
er
in
co
m
p
u
ter
v
i
s
io
n
.
C
o
n
v
o
lu
tio
n
Ne
u
r
al
Net
w
o
r
k
(
C
NN)
is
a
d
ee
p
ar
tific
ial
n
eu
r
al
n
et
w
o
r
k
t
h
at
is
u
s
ed
m
ai
n
l
y
to
class
i
f
y
i
m
a
g
e,
clu
s
ter
th
e
m
b
y
s
i
m
ilar
it
y
,
a
n
d
p
er
f
o
r
m
i
m
ag
e
r
ec
o
g
n
itio
n
in
t
h
e
s
ce
n
e.
C
NN
h
a
s
s
h
o
w
n
i
m
p
r
ess
i
v
e
p
er
f
o
r
m
a
n
ce
s
i
n
ar
tif
icial
i
n
telli
g
e
n
ce
task
s
s
u
ch
as
o
b
j
ec
t
r
ec
o
g
n
itio
n
an
d
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
[
4
]
.
T
h
e
g
en
er
al
s
tr
u
ctu
r
e
o
f
a
C
NN
co
n
s
is
ts
o
f
la
y
er
s
co
m
p
o
s
ed
o
f
n
e
u
r
o
n
s
.
A
n
eu
r
o
n
ta
k
es
in
p
u
t
v
alu
e
s
,
d
o
es
co
m
p
u
tatio
n
s
a
n
d
p
ass
es
t
h
e
r
es
u
lt
to
th
e
n
e
x
t
la
y
er
.
B
esid
es
i
m
a
g
e
r
ec
o
g
n
i
tio
n
,
C
N
N
also
ca
n
b
e
u
s
ed
in
v
id
eo
an
al
y
s
is
a
n
d
also
n
at
u
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
i
n
g
[
7
]
.
Fo
r
ca
s
es
w
h
er
e
th
e
av
a
ilab
l
e
d
ata
is
n
o
t
m
u
c
h
,
p
r
e
-
tr
ain
ed
C
NN
m
o
d
els
s
u
c
h
as
A
le
x
Net
an
d
Go
o
g
L
eNe
t
ca
n
b
e
u
s
ed
.
B
o
th
attai
n
a
lo
w
er
r
o
r
w
h
en
t
r
ain
ed
o
v
er
t
h
e
m
illi
o
n
o
f
i
m
ag
e
s
co
n
ta
in
ed
i
n
I
m
ag
eNe
t
[
8
]
.
Go
o
g
L
eNe
t
a
n
d
A
lex
Ne
t
ar
e
o
f
ten
u
s
ed
in
p
h
o
to
class
i
f
icatio
n
,
as
a
lar
g
e
f
r
ac
tio
n
o
f
ex
a
m
p
le
s
in
I
m
ag
e
Net
t
h
at
ar
e
co
m
p
o
s
ed
o
f
p
h
o
to
s
.
T
h
e
u
s
e
o
f
p
r
e
-
tr
ain
ed
C
NN
m
o
d
el
is
to
tr
an
s
f
er
t
h
e
lear
n
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l
.
1
2
,
No
.
2
,
No
v
e
m
b
er
201
8
:
4
5
5
–
4
6
0
456
p
ar
am
eter
s
.
R
at
h
er
t
h
an
co
n
s
t
r
u
ctin
g
a
n
d
tr
ain
i
n
g
a
n
e
w
n
et
w
o
r
k
,
w
e
ca
n
ta
k
e
a
p
r
etr
ain
e
d
n
et
w
o
r
k
a
n
d
u
s
e
s
it
as
a
s
tar
t
in
g
p
o
in
t
to
lear
n
a
n
e
w
tas
k
a
n
d
d
o
f
in
e
tu
n
i
n
g
.
P
r
e
-
tr
ain
ed
C
N
N
m
o
d
el
s
h
av
e
alr
ea
d
y
lear
n
ed
a
lar
g
e
s
et
o
f
f
ea
t
u
r
es.
T
h
u
s
,
th
e
class
i
f
icatio
n
an
d
r
ec
o
g
n
i
tio
n
p
r
o
ce
s
s
ca
n
b
e
ac
h
ie
v
ed
f
as
t
er
.
I
n
t
h
is
p
ap
er
,
a
co
m
p
ar
is
o
n
b
et
w
ee
n
C
NN
a
n
d
t
w
o
o
f
th
e
p
o
p
u
lar
p
r
e
-
tr
ain
ed
C
NN
m
o
d
el
s
ar
e
co
n
d
u
cted
in
ter
m
s
o
f
ac
cu
r
ac
y
p
er
f
o
r
m
an
ce
.
T
h
ese
t
w
o
m
o
d
els
ar
e
Go
o
g
leNe
t
an
d
A
lex
Net.
T
h
e
o
b
j
ec
tiv
e
o
f
th
i
s
co
m
p
ar
is
o
n
i
s
to
k
n
o
w
w
h
ic
h
m
o
d
els
ar
e
b
est s
u
ited
t
o
r
ec
o
g
n
ize
th
e
d
ata.
2.
T
H
E
P
E
RF
O
RM
ANCE O
VE
RVI
E
W
O
F
CNN
M
O
DE
L
S USI
N
G
H
AND
WRI
T
T
E
N
CAP
I
T
A
L
L
E
T
T
E
R
DA
T
AS
E
T
T
h
e
d
ataset
n
ee
d
s
to
ad
h
er
e
with
t
h
e
co
n
s
tr
ai
n
ts
g
i
v
e
n
b
y
t
h
e
C
NN
m
o
d
els.
O
n
e
o
f
t
h
e
co
n
s
tr
ai
n
t
i
s
r
eg
ar
d
in
g
t
h
e
s
ize
o
f
t
h
e
i
n
d
iv
i
d
u
al
i
m
a
g
e.
A
le
x
Net
n
ee
d
s
to
u
s
e
a
p
r
ec
is
e
i
m
a
g
e
s
ize
f
o
r
tr
ain
i
n
g
.
B
esid
es t
h
e
co
n
s
tr
ain
ts
f
r
o
m
t
h
e
m
o
d
els,
t
h
e
h
ar
d
w
ar
e
u
s
ed
also
p
la
y
s
a
n
i
m
p
o
r
tan
t
p
ar
t
o
n
th
e
p
er
f
o
r
m
an
ce
.
So
,
to
test
th
ese
3
C
NN
m
o
d
els,
a
h
i
g
h
p
er
f
o
r
m
an
ce
g
a
m
i
n
g
lap
to
p
is
u
s
ed
to
co
n
d
u
ct
t
h
e
te
s
t.
T
h
e
l
ap
to
p
n
ee
d
s
to
h
a
v
e
a
d
ed
icate
d
NVI
DI
A
Gr
ap
h
ic
s
P
r
o
ce
s
s
in
g
U
n
it
(
GP
U)
w
it
h
a
m
i
n
i
m
u
m
o
f
8
g
i
g
ab
y
te
o
f
R
a
n
d
o
m
A
cc
e
s
s
Me
m
o
r
y
(
R
A
M)
.
T
h
e
co
m
p
let
e
s
p
ec
if
icatio
n
o
f
t
h
e
lap
to
p
u
s
ed
is
lis
ted
in
T
ab
le
1
.
T
ab
le
1
.
L
ap
to
p
Sp
ec
if
icatio
n
M
o
d
e
l
M
S
I
G
L
6
2
-
6
Q
F
O
p
e
r
a
t
i
n
g
S
y
st
e
m
W
i
n
d
o
w
s 1
0
P
r
o
c
e
sso
r
I
n
t
e
l
C
o
r
e
i
7
-
6
7
0
0
H
Q
R
A
M
8
g
i
g
a
b
y
t
e
s
G
r
a
p
h
i
c
s
C
a
r
d
N
v
i
d
i
a
G
T
X
9
6
0
M
T
h
e
d
ataset
u
s
ed
f
o
r
tr
ain
in
g
o
n
th
e
s
e
t
h
r
ee
m
et
h
o
d
s
is
C
h
ar
s
7
4
K
d
ataset
[
9
]
.
T
h
e
d
ataset
co
n
s
is
ts
o
f
th
e
i
m
ag
e
s
o
f
h
a
n
d
w
r
it
ten
ca
p
ital
letter
s
,
s
m
all
letter
s
,
n
u
m
b
er
s
an
d
s
y
m
b
o
ls
.
E
ac
h
o
f
t
h
e
l
etter
co
n
s
is
ts
o
f
5
5
v
ar
iatio
n
s
o
f
id
en
tica
l
i
m
a
g
es
.
Sin
ce
th
e
o
v
er
all
tr
ai
n
i
n
g
t
ak
es
a
l
ong
ti
m
e
an
d
th
e
h
ar
d
w
ar
e
u
s
ed
f
o
r
th
e
tr
ain
i
n
g
is
s
ti
ll
n
o
t
p
o
w
er
f
u
l
en
o
u
g
h
,
th
e
tr
ai
n
in
g
o
n
l
y
co
v
er
s
a
h
an
d
w
r
itte
n
ca
p
ita
l
lette
r
s
o
n
l
y
,
w
h
ic
h
ar
e
letter
s
A
to
Z
,
t
h
u
s
,
t
h
is
m
a
k
es
th
e
to
tal
n
u
m
b
er
o
f
i
m
a
g
es
u
s
ed
i
n
t
h
e
e
x
p
er
i
m
e
n
t
i
s
1
4
3
0
.
T
h
e
s
ize
o
f
t
h
e
i
m
a
g
es i
n
th
e
d
atase
t
is
1
2
0
0
x
9
0
0
p
ix
els.
R
ed
u
ci
n
g
th
e
r
es
o
lu
tio
n
m
a
y
d
ec
r
ea
s
e
t
h
e
v
al
id
atio
n
ac
cu
r
ac
y
.
2
.
1
.
Si
m
ple
-
CN
N
A
C
NN
co
n
s
is
ts
o
f
o
n
e
o
r
m
o
r
e
co
n
v
o
lu
t
io
n
al
la
y
er
s
,
u
s
u
all
y
alo
n
g
w
it
h
a
s
u
b
s
a
m
p
li
n
g
la
y
er
,
w
h
ic
h
ar
e
f
o
llo
w
ed
b
y
o
n
e
o
r
m
o
r
e
f
u
ll
y
co
n
n
ec
ted
la
y
er
s
s
i
m
il
ar
to
a
s
tan
d
ar
d
n
eu
r
al
n
et
wo
r
k
.
T
h
ese
co
m
p
le
x
ar
ch
itect
u
r
es
ar
e
b
u
il
t
f
o
r
cla
s
s
if
ica
tio
n
p
r
o
b
le
m
s
b
y
s
tack
i
n
g
m
u
l
tip
le
an
d
d
i
f
f
er
en
t
la
y
er
s
.
Gen
er
all
y
,
t
h
er
e
ar
e
f
o
u
r
t
y
p
e
s
o
f
la
y
er
s
,
w
h
i
ch
ar
e
co
n
v
o
lu
t
io
n
la
y
er
s
,
p
o
o
li
n
g
/
s
u
b
s
a
m
p
li
n
g
la
y
er
s
,
n
o
n
-
lin
ea
r
la
y
er
s
,
a
n
d
f
u
ll
y
co
n
n
ec
ted
la
y
er
s
[
1
0
]
.
T
h
e
f
ir
s
t
la
y
er
,
co
n
v
o
lu
tio
n
al
la
y
er
s
ar
e
ab
le
to
ex
tr
ac
t
th
e
lo
ca
l
f
ea
tu
r
es
b
ec
au
s
e
th
e
y
r
estrict
th
e
r
ec
ep
tiv
e
f
iel
d
s
o
f
t
h
e
h
id
d
en
la
y
er
s
to
b
e
l
o
ca
l.
T
h
en
,
th
e
p
o
o
lin
g
/
s
u
b
s
a
m
p
li
n
g
la
y
er
r
ed
u
ce
s
th
e
r
eso
l
u
tio
n
o
f
t
h
e
f
ea
t
u
r
es.
T
h
is
p
r
o
ce
s
s
m
a
k
es
t
h
e
f
ea
t
u
r
es
r
o
b
u
s
t
a
g
ai
n
s
t
n
o
i
s
e
a
n
d
d
is
to
r
tio
n
.
Ne
u
r
al
n
et
w
o
r
k
s
g
e
n
er
all
y
an
d
C
NN
s
p
ec
if
icall
y
r
el
y
o
n
a
t
h
ir
d
la
y
er
,
n
o
n
-
lin
ea
r
“
tr
i
g
g
er
”
f
u
n
c
ti
o
n
to
s
i
g
n
al
d
i
s
ti
n
ct
id
en
ti
f
icatio
n
o
f
i
d
en
t
icall
y
f
e
atu
r
es
o
n
ea
ch
h
id
d
en
la
y
er
.
C
NN
m
a
y
u
s
e
a
v
ar
iet
y
o
f
s
p
ec
if
ic
f
u
n
ctio
n
s
s
u
c
h
as
r
ec
tif
ied
lin
ea
r
u
n
its
(
R
e
L
Us)
an
d
co
n
tin
u
o
u
s
tr
ig
g
er
(
n
o
n
-
li
n
ea
r
)
f
u
n
ctio
n
s
to
ef
f
icie
n
tl
y
i
m
p
le
m
en
t
th
i
s
n
o
n
-
li
n
ea
r
tr
i
g
g
er
i
n
g
.
T
h
e
f
in
al
la
y
er
,
Fu
ll
y
co
n
n
ec
ted
la
y
e
r
s
,
to
tal
u
p
a
w
ei
g
h
ti
n
g
o
f
t
h
e
p
r
ev
io
u
s
la
y
er
o
f
f
ea
t
u
r
es,
i
n
d
icatin
g
th
e
p
r
ec
is
e
m
i
x
o
f
“
i
n
g
r
ed
ie
n
ts
”
to
d
et
er
m
in
e
a
s
p
ec
i
f
ic
tar
g
et
o
u
tp
u
t
r
es
u
lt.
Fi
g
u
r
e
1
illu
s
tr
ates t
h
e
s
i
m
p
le
-
C
NN
ar
c
h
itect
u
r
e.
Fo
r
th
is
C
NN
m
o
d
el,
th
e
i
m
a
g
es
ar
e
r
esized
to
a
s
m
aller
s
ize,
9
3
x
7
0
p
ix
els
to
b
e
p
r
ec
is
e.
T
h
is
is
b
ec
au
s
e
b
y
u
s
i
n
g
th
e
o
r
ig
i
n
a
l
r
eso
lu
tio
n
is
i
m
p
o
s
s
ib
le
d
u
e
to
th
e
lim
i
tatio
n
o
f
Gr
ap
h
i
cs
P
r
o
ce
s
s
in
g
Un
it
(
GP
U)
m
e
m
o
r
y
o
f
th
e
d
ev
ice
u
s
ed
f
o
r
th
is
tr
ain
in
g
.
T
h
e
tr
ain
i
n
g
is
s
e
t
to
1
0
e
p
o
ch
w
i
t
h
1
0
iter
atio
n
s
f
o
r
ev
er
y
e
p
o
ch
.
B
ased
o
n
t
h
e
r
esu
lt,
t
h
e
tr
ai
n
i
n
g
h
a
s
r
ea
c
h
th
e
f
i
n
al
iter
atio
n
at
E
p
o
ch
1
0
w
it
h
0.
8
0
7
7
o
f
v
alid
atio
n
ac
cu
r
ac
y
.
T
h
e
ac
cu
r
ac
y
alr
ea
d
y
n
ea
r
th
e
m
ax
i
m
u
m
d
u
r
in
g
th
e
E
p
o
ch
9
.
Fig
u
r
e
2
s
h
o
w
s
th
e
d
etail
r
esu
lt
s
o
f
t
h
e
tr
ai
n
i
n
g
an
d
v
al
id
atio
n
p
r
o
ce
s
s
.
Fi
g
u
r
e
3
lis
t
ed
s
o
m
e
s
a
m
p
le
i
m
a
g
es
t
h
at
h
av
e
b
ee
n
co
r
r
ec
tl
y
class
i
f
ied
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
Dee
p
Lea
r
n
in
g
fo
r
R
o
ma
n
Ha
n
d
w
r
itten
C
h
a
r
a
cter R
ec
o
g
n
itio
n
(
Mu
h
a
a
fid
z
Md
S
a
u
fi
)
457
Fig
u
r
e
1
.
Si
m
p
le
-
C
N
N
ar
ch
ite
ctu
r
e
[
1
0
]
Fig
u
r
e
2.
Si
m
p
le
-
C
N
N
t
r
ain
i
n
g
an
d
v
al
id
atio
n
d
etails
Fig
u
r
e
3.
Si
m
p
le
-
C
N
N
t
est r
es
u
lts
2
.
2
.
Alex
Net
A
le
x
Net
is
m
ad
e
u
p
o
f
5
c
o
n
v
o
l
u
tio
n
al
la
y
er
s
,
m
a
x
-
p
o
o
lin
g
la
y
er
s
,
d
r
o
p
o
u
t
la
y
er
s
a
n
d
3
f
u
ll
y
co
n
n
ec
ted
la
y
er
s
.
R
el
u
is
ap
p
lied
af
ter
e
v
er
y
co
n
v
o
lu
t
io
n
al
an
d
f
u
ll
y
co
n
n
ec
ted
la
y
er
.
Dr
o
p
o
u
t
is
ap
p
lied
b
ef
o
r
e
th
e
f
ir
s
t
an
d
th
e
s
ec
o
n
d
f
u
l
l
y
co
n
n
ec
ted
la
y
er
[
1
1
]
.
Fi
g
u
r
e
4
ill
u
s
tr
at
es
t
h
e
ar
ch
itect
u
r
e
o
f
A
le
x
Net.
Fo
r
th
is
tr
ai
n
i
n
g
,
t
h
e
i
m
a
g
e
s
s
ize
a
r
e
b
ac
k
to
its
d
ef
au
lt
s
ize.
T
h
e
tr
ain
in
g
is
s
e
t
to
5
ep
o
ch
w
i
th
1
0
1
iter
atio
n
s
p
er
ep
o
ch
.
T
h
e
r
esu
lts
s
h
o
w
th
a
t
th
e
tr
ain
i
n
g
h
as
r
ea
ch
ed
th
e
f
i
n
al
iter
atio
n
w
it
h
0.
94
23
v
alid
atio
n
ac
cu
r
ac
y
.
As
th
e
iter
atio
n
p
r
o
ce
s
s
o
n
A
le
x
Net
is
ten
f
o
ld
t
h
an
Si
m
p
le
-
C
NN,
th
e
tr
ain
i
n
g
ac
c
u
r
ac
y
is
a
lr
ea
d
y
h
it
0.
90
in
t
h
e
m
id
d
le
o
f
t
h
e
s
ec
o
n
d
ep
o
ch
.
T
h
e
r
esu
lts
o
f
t
h
e
tr
ai
n
i
n
g
ca
n
b
e
s
ee
n
i
n
Fi
g
u
r
e
5
w
h
i
le
F
ig
u
r
e
6
l
is
ted
s
o
m
e
s
a
m
p
le
i
m
a
g
es t
h
at
h
a
v
e
b
ee
n
co
r
r
ec
tly
cla
s
s
i
f
ied
b
y
A
le
x
Ne
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l
.
1
2
,
No
.
2
,
No
v
e
m
b
er
201
8
:
4
5
5
–
4
6
0
458
Fig
u
r
e
4
.
A
le
x
Net
ar
ch
i
tectu
r
e
[
1
1
]
Fig
u
r
e
5.
A
le
x
Net
t
r
ai
n
in
g
a
n
d
v
alid
atio
n
d
etails
Fig
u
r
e
6.
Sa
m
p
le
A
le
x
Net
t
est
r
esu
lts
2
.
3
.
G
o
o
g
L
eNe
t
Go
o
g
L
eNe
t
i
s
a
p
r
e
-
tr
ain
ed
C
NN
th
a
t
h
as
b
ee
n
tr
ai
n
ed
o
n
o
v
er
a
m
illi
o
n
i
m
a
g
es
an
d
ca
n
class
i
f
y
i
m
a
g
es
in
to
ca
teg
o
r
ies.
T
h
e
n
et
w
o
r
k
ta
k
es
an
i
m
ag
e
as
i
n
p
u
t
a
n
d
o
u
tp
u
t
s
a
lab
el
f
o
r
th
e
o
b
j
ec
t
in
th
e
i
m
ag
e
to
g
eth
er
w
it
h
th
e
p
r
o
b
ab
ilit
ies
f
o
r
ea
ch
o
f
th
e
o
b
j
ec
t
ca
teg
o
r
ies.
I
t
h
as
2
2
lay
er
s
.
T
h
is
ar
ch
itectu
r
e
(
Fig
u
r
e
7
)
is
o
u
t
o
f
th
e
n
o
r
m
,
it
v
ee
r
ed
o
f
f
f
r
o
m
t
h
e
g
en
er
al
ap
p
r
o
ac
h
o
f
s
i
m
p
l
y
s
tac
k
i
n
g
co
n
v
an
d
p
o
o
lin
g
la
y
er
s
o
n
to
p
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
Dee
p
Lea
r
n
in
g
fo
r
R
o
ma
n
Ha
n
d
w
r
itten
C
h
a
r
a
cter R
ec
o
g
n
itio
n
(
Mu
h
a
a
fid
z
Md
S
a
u
fi
)
459
o
f
ea
ch
o
t
h
er
in
s
eq
u
e
n
ce
[
1
2
]
.
Fo
r
Go
o
g
L
eNe
t
C
NN
m
o
d
el,
th
e
i
m
a
g
es
s
ize
is
also
u
s
i
n
g
t
h
eir
d
ef
a
u
lt
s
ize.
Go
o
g
L
eNe
t
tr
ai
n
i
n
g
an
d
v
al
id
atio
n
d
etails
s
h
i
w
n
i
n
Fi
g
u
r
e
8
.
T
h
e
r
esu
lts
o
f
Go
o
g
L
eNe
t
ar
e
s
h
o
w
n
i
n
Fi
g
u
r
e
9
alo
n
g
w
i
th
th
e
tr
ain
i
n
g
p
er
ce
n
tag
e
s
o
f
ea
ch
letter
.
T
h
e
tr
ain
in
g
i
s
s
e
t
to
6
ep
o
ch
s
w
i
t
h
1
0
1
iter
atio
n
s
p
er
ep
o
ch
.
T
h
e
r
esu
lts
s
h
o
w
t
h
at
t
h
e
tr
ain
i
n
g
h
a
s
r
ea
ch
ed
th
e
f
i
n
al
iter
atio
n
w
it
h
0
.
9
4
4
7
o
f
v
ali
d
atio
n
ac
cu
r
ac
y
.
Fig
u
r
e
7
.
Go
o
g
L
eNe
t a
r
ch
itect
u
r
e
[
1
2
]
Fig
u
r
e
8
.
Go
o
g
L
eNe
t
tr
ai
n
i
n
g
an
d
v
alid
atio
n
d
etail
s
Fig
u
r
e
9
.
Sa
m
p
le
Go
o
g
L
eNe
t
test
r
esu
lts
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l
.
1
2
,
No
.
2
,
No
v
e
m
b
er
201
8
:
4
5
5
–
4
6
0
460
3.
CO
NCLU
SI
O
N
T
ab
le
2
s
h
o
w
s
a
n
o
v
er
v
ie
w
o
f
t
h
e
ac
c
u
r
ac
y
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
t
h
r
ee
m
o
d
els
b
ased
o
n
C
h
ar
s
7
4
K
d
ataset
.
B
y
lo
o
k
in
g
at
T
ab
le
2
,
w
e
ca
n
s
ee
t
h
at
Go
o
g
L
eN
et
is
b
etter
th
a
n
s
i
m
p
le
-
C
NN
an
d
A
le
x
Net
b
u
t
i
t
to
o
k
a
lo
n
g
er
ti
m
e
to
ac
h
ie
v
e
t
h
is
r
es
u
lt.
T
ab
le
2
.
T
h
e
P
e
r
f
o
r
m
a
n
ce
O
v
er
v
ie
w
o
n
C
NN
Mo
d
els Usi
n
g
C
ap
ital L
et
ter
s
f
r
o
m
C
h
ar
s
7
4
K
Data
s
et.
C
N
N
M
o
d
e
l
S
i
mp
l
e
-
C
N
N
G
o
o
g
L
e
N
e
t
A
l
e
x
N
e
t
V
a
l
i
d
a
t
i
o
n
A
c
c
u
r
a
c
y
0
.
8
0
7
7
0
.
9
4
4
7
0
.
9
4
2
3
T
r
a
i
n
i
n
g
F
i
n
i
s
h
e
d
Y
e
s
Y
e
s
Y
e
s
El
a
p
se
d
T
i
me
(
s)
11
1
2
0
2
7
9
8
Ep
o
c
h
1
0
o
f
1
0
6
o
f
6
5
o
f
5
I
t
e
r
a
t
i
o
n
1
0
0
o
f
1
0
0
6
0
6
o
f
6
0
6
5
0
5
o
f
5
0
5
V
a
l
i
d
a
t
i
o
n
F
r
e
q
u
e
n
c
y
3
0
i
t
e
r
a
t
i
o
n
s
3
i
t
e
r
a
t
i
o
n
s
3
i
t
e
r
a
t
i
o
n
s
Dee
p
lear
n
in
g
i
s
o
n
e
o
f
th
e
m
o
s
t p
o
w
er
f
u
l
m
ac
h
i
n
e
lear
n
i
n
g
m
et
h
o
d
s
f
o
r
s
o
l
v
i
n
g
cr
it
ical
p
r
o
b
lem
s
in
co
m
p
u
ter
v
is
io
n
.
I
n
th
is
p
ap
er
,
th
r
ee
m
o
d
els
o
f
Dee
p
L
ea
r
n
in
g
h
a
v
e
b
ee
n
u
s
ed
to
t
est
t
h
eir
ac
c
u
r
ac
y
p
er
f
o
r
m
a
n
ce
.
B
ased
o
n
t
h
e
e
x
p
er
im
e
n
tal
r
e
s
u
l
ts
,
t
h
e
r
ec
o
m
m
en
d
ed
m
o
d
el
f
o
r
lear
n
i
n
g
al
p
h
ab
ets
is
Alex
Net
as
it
ta
k
es
a
s
h
o
r
ter
ti
m
e
to
le
ar
n
th
e
d
ata
w
h
er
e
i
t
o
n
l
y
t
o
o
k
1
5
0
s
ec
o
n
d
s
p
er
ep
o
ch
w
h
il
e
th
e
n
ea
r
est
r
i
v
al,
w
h
ic
h
is
Go
o
g
L
eNe
t
t
ook
2
0
0
s
ec
o
n
d
s
p
er
ep
o
c
h
.
T
h
is
i
s
d
u
e
to
th
e
ar
ch
itec
u
r
e
o
f
th
e
s
e
m
o
d
els
w
h
er
e
t
h
e
la
y
er
s
i
n
A
le
x
tNet
ar
e
n
o
t
as
d
ee
p
as
in
Go
o
g
L
eNe
t
.
T
h
e
Si
m
p
le
-
C
NN
m
o
d
el
d
o
es
n
o
t
p
r
o
d
u
ce
a
g
o
o
d
r
esu
lt
co
m
p
ar
ed
to
Go
o
g
L
eNe
t
o
r
A
le
x
Net
b
ec
au
s
e
it
r
eq
u
ir
es
t
r
e
m
en
d
o
u
s
a
m
o
u
n
t
o
f
d
ata
f
o
r
tr
ain
i
n
g
p
u
r
p
o
s
es
.
Fu
t
u
r
e
r
esear
c
h
i
s
to
e
n
h
a
n
ce
th
e
s
i
m
p
le
-
C
NN
ar
ch
itect
u
r
e
an
d
i
n
cr
ea
s
e
t
h
e
s
ize
o
f
t
h
e
d
a
taset
to
o
b
tai
n
m
o
r
e
ac
cu
r
ate
an
d
f
a
s
ter
r
esu
l
ts
.
4.
ACK
NO
WL
E
D
G
E
M
E
NT
T
h
e
au
th
o
r
s
w
o
u
ld
li
k
e
to
t
h
an
k
Fac
u
lt
y
o
f
C
o
m
p
u
ter
an
d
Ma
th
e
m
atica
l
Scien
ce
s
,
Un
i
v
er
s
iti
T
ek
n
o
lo
g
i M
A
R
A
,
S
h
a
h
A
la
m
,
Sela
n
g
o
r
,
f
o
r
s
p
o
n
s
o
r
i
n
g
t
h
is
r
esear
ch
.
RE
F
E
R
E
NC
E
S
[1
]
P
a
ti
l
V
.
&
S
h
im
p
i
S
.
(2
0
1
1
).
Ha
n
d
w
rit
ten
En
g
li
sh
c
h
a
ra
c
ter
re
c
o
g
n
it
io
n
u
sin
g
n
e
u
ra
l
n
e
tw
o
rk
El
ix
ir
Co
m
p
.
S
c
i.
&
En
g
g
.
4
1
5
5
8
7
-
5
5
9
1
[2
]
S
M
u
tal
ib
,
R
Ra
m
li
,
S
A
Ra
h
m
a
n
,
M
Yu
so
f
f
,
A
M
o
h
a
m
e
d
.
(2
0
0
8
)
.
T
o
wa
rd
s e
m
o
ti
o
n
a
l
c
o
n
tro
l
re
c
o
g
n
it
io
n
t
h
ro
u
g
h
h
a
n
d
w
rit
in
g
u
si
n
g
f
u
z
z
y
in
f
e
r
e
n
c
e
.
IT
S
im
2
0
0
8
.
[3
]
Nijh
a
w
a
n
,
R.
.
,
S
h
a
rm
a
,
H.,
S
a
h
n
i
,
H.,
&
Ba
tra,
A
.
(2
0
1
8
,
A
p
ril
).
A
d
e
e
p
lea
rn
in
g
h
y
b
rid
CNN
f
ra
m
e
w
o
rk
a
p
p
ro
a
c
h
f
o
r
v
e
g
e
tatio
n
c
o
v
e
r
m
a
p
p
in
g
u
sin
g
d
e
e
p
f
e
a
tu
re
s.
In
S
ig
n
a
l
-
I
m
a
g
e
T
e
c
h
n
o
lo
g
y
&
In
tern
e
t
-
Ba
se
d
S
y
st
e
m
s
(S
I
T
IS
),
2
0
1
7
1
3
t
h
I
n
tern
a
ti
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
.
IE
EE
[4
]
T
u
a
m
a
,
A
.
,
Co
m
b
y
,
F
.
,
&
Ch
a
u
m
o
n
t,
M
.
(2
0
1
6
,
De
c
e
m
b
e
r).
Ca
m
e
ra
m
o
d
e
l
id
e
n
ti
f
ica
ti
o
n
w
it
h
t
h
e
u
se
o
f
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s.
In
In
f
o
rm
a
ti
o
n
F
o
re
n
sic
s
a
n
d
S
e
c
u
rit
y
(W
IF
S
),
2
0
1
6
I
EE
E
I
n
tern
a
ti
o
n
a
l
W
o
rk
sh
o
p
o
n
(
p
p
.
1
-
6
).
IE
EE
.
[5
]
KA
J
a
li
l,
M
H Ka
m
a
ru
d
in
,
M
N
M
a
sre
k
.
(2
0
1
0
)
.
Co
m
p
a
riso
n
o
f
m
a
c
h
in
e
lea
rn
i
n
g
a
lg
o
rit
h
m
s p
e
rf
o
rm
a
n
c
e
in
d
e
tec
ti
n
g
n
e
tw
o
rk
in
tru
si
o
n
.
Ne
tw
o
rk
in
g
a
n
d
In
f
o
rm
a
ti
o
n
T
e
c
h
n
o
lo
g
y
(ICNIT
)
[6
]
Ch
e
n
g
,
G
.
,
M
a
,
C.
,
Z
h
o
u
,
P
.
,
Ya
o
,
X.,
&
Ha
n
,
J.
(2
0
1
6
,
J
u
ly
).
S
c
e
n
e
c
las
si
f
ica
ti
o
n
o
f
h
ig
h
re
so
lu
ti
o
n
re
m
o
te
se
n
sin
g
im
a
g
e
s
u
sin
g
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s.
In
G
e
o
sc
ien
c
e
a
n
d
Re
m
o
te S
e
n
sin
g
S
y
m
p
o
siu
m
(I
GA
R
S
S
),
2
0
1
6
IEE
E
In
tern
a
ti
o
n
a
l
(
p
p
.
7
6
7
-
7
7
0
).
IEE
E
.
[7
]
Z
Ib
ra
h
im
,
D Isa
,
R
Ra
jk
u
m
a
r,
G
Ke
n
d
a
ll
.
(
2
0
0
9
).
Do
c
u
m
e
n
t
z
o
n
e
c
o
n
ten
t
c
las
sif
ica
ti
o
n
f
o
r
tec
h
n
ica
l
d
o
c
u
m
e
n
t
im
a
g
e
s u
sin
g
a
rti
f
i
c
ial
n
e
u
ra
l
n
e
t
w
o
rk
s an
d
su
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
s.
A
p
p
li
c
a
ti
o
n
s o
f
Dig
it
a
l
In
f
o
rm
a
ti
o
n
a
n
d
W
e
b
T
e
c
h
n
o
lo
g
ies
,
ICA
DI
WT
'
0
9
[8
]
Ba
ll
e
ste
r,
P
.
,
&
d
e
A
ra
ú
jo
,
R.
M
.
(2
0
1
6
,
F
e
b
r
u
a
ry
).
On
th
e
P
e
rf
o
r
m
a
n
c
e
o
f
G
o
o
g
L
e
N
e
t
a
n
d
A
l
e
x
Ne
t
A
p
p
li
e
d
to
S
k
e
tch
e
s.
In
AA
A
I
(p
p
.
1
1
2
4
-
1
1
2
8
).
[9
]
T
h
e
Ch
a
rs7
4
K d
a
tas
e
t.
Re
tri
e
v
e
d
f
ro
m
h
tt
p
:/
/w
ww
.
e
e
.
su
rre
y
.
a
c
.
u
k
/CV
S
S
P
/d
e
m
o
s/c
h
a
rs7
4
k
/
[1
0
]
Hijaz
i,
S
.
,
Ku
m
a
r,
R.
,
&
R
o
w
e
n
,
C.
(2
0
1
5
)
.
Us
in
g
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s
f
o
r
i
m
a
g
e
re
c
o
g
n
it
io
n
.
Ca
d
e
n
c
e
De
sig
n
S
y
ste
m
s In
c
.
:
S
a
n
Jo
se
,
C
A
,
US
A
.
[1
1
]
G
a
o
,
H.
(2
0
1
7
,
A
u
g
u
st
0
7
)
.
A
Walk
-
th
ro
u
g
h
o
f
A
le
x
Ne
t
–
Ha
o
G
a
o
–
M
e
d
iu
m
.
Re
tri
e
v
e
d
f
ro
m
h
tt
p
s:/
/m
e
d
iu
m
.
c
o
m
/@s
m
a
ll
f
i
sh
b
ig
se
a
/a
-
wa
lk
-
th
ro
u
g
h
-
of
-
a
lex
n
e
t
-
6
c
b
d
1
3
7
a
5
6
3
7
[1
2
]
De
sh
p
a
n
d
e
,
A
.
T
h
e
9
De
e
p
L
e
a
r
n
in
g
P
a
p
e
rs
Yo
u
Ne
e
d
T
o
Kn
o
w
A
b
o
u
t
(Un
d
e
rsta
n
d
i
n
g
CNN
s
P
a
rt
3
).
Re
tri
e
v
e
d
f
ro
m
h
tt
p
s://
a
d
e
sh
p
a
n
d
e
3
.
g
it
h
u
b
.
i
o
/T
h
e
-
9
-
De
e
p
-
L
e
a
rn
in
g
-
P
a
p
e
rs
-
Y
o
u
-
Ne
e
d
-
To
-
Kn
o
w
-
A
b
o
u
t.
h
tm
l
Evaluation Warning : The document was created with Spire.PDF for Python.