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.
19
,
No
.
2
,
A
u
g
u
s
t
20
20
,
p
p
.
9
7
4
~
9
8
2
I
SS
N:
2
5
02
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/i
j
ee
cs.v
1
9
.i
2
.
p
p
974
-
9
8
2
974
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
/
/
ijee
cs.ia
esco
r
e.
co
m
Fused
fas
ter RCN
Ns for e
ff
ici
ent
de
tect
io
n of
t
h
e lice
nse pla
tes
Na
a
m
a
n O
m
a
r
1
,
Adna
n M
o
hs
in A
bd
u
la
ze
ez
2
,
Abdu
l
k
a
d
ir
Seng
ur
3
,
Sa
li
m
G
a
ni
m
Sa
ee
d Al
-
Ali
4
1
De
p
a
rtme
n
t
o
f
In
f
o
rm
a
ti
o
n
T
e
c
h
n
o
l
o
g
y
,
Du
h
o
k
P
o
ly
tec
h
n
ic Un
iv
e
rsit
y
,
Ira
q
2
P
re
sid
e
n
c
y
o
f
Du
h
o
k
P
o
ly
tec
h
n
ic Un
iv
e
rsit
y
,
Du
h
o
k
P
o
ly
tec
h
n
ic Un
iv
e
rsit
y
,
Ira
q
3
De
p
a
rtme
n
t
o
f
El
e
c
tri
c
a
l
-
El
e
c
tro
n
ics
En
g
in
e
e
rin
g
,
F
irat
Un
iv
e
rsity
,
T
u
rk
e
y
4
De
p
a
rtme
n
t
o
f
In
f
o
rm
a
ti
o
n
T
e
c
h
n
o
l
o
g
y
M
a
n
a
g
e
m
e
n
t,
Du
h
o
k
P
o
ly
tec
h
n
ic Un
iv
e
rsity
,
Ira
q
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
an
2
,
2
0
2
0
R
ev
i
s
ed
Feb
2
9
,
2
0
2
0
A
cc
ep
ted
Mar
1
3
,
2
0
2
0
A
u
to
m
a
ti
c
L
i
c
e
n
se
P
late
De
tec
ti
o
n
a
n
d
Re
c
o
g
n
it
i
o
n
(A
L
P
D
-
R)
is
a
n
im
p
o
rtan
t
a
n
d
c
h
a
ll
e
n
g
in
g
a
p
p
li
c
a
ti
o
n
f
o
r
traf
f
i
c
su
rv
e
il
lan
c
e
,
tra
ff
ic
s
a
f
e
t
y
,
se
c
u
rit
y
,
se
rv
i
c
e
s
p
u
rp
o
se
s
a
n
d
p
a
rk
in
g
m
a
n
a
g
e
m
e
n
t.
G
e
n
e
ra
ll
y
,
trad
it
io
n
a
l
im
a
g
e
p
ro
c
e
ss
in
g
ro
u
ti
n
e
s
h
a
v
e
b
e
e
n
u
se
d
in
A
L
P
D
-
R.
A
lt
h
o
u
g
h
th
e
g
e
n
e
ra
l
a
p
p
ro
a
c
h
e
s
p
e
rf
o
rm
we
ll
o
n
AL
P
D
-
R,
n
e
w
a
n
d
e
ff
icie
n
t
a
p
p
r
o
a
c
h
e
s
a
re
n
e
e
d
e
d
t
o
im
p
ro
v
e
th
e
d
e
tec
ti
o
n
a
c
c
u
ra
c
ies
.
T
h
u
s,
i
n
th
is
p
a
p
e
r,
a
n
e
w
a
p
p
ro
a
c
h
,
w
h
ich
is
b
a
se
d
o
n
f
u
sin
g
o
f
m
u
lt
ip
le
F
a
ste
r
Re
g
io
n
s
w
it
h
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
tral
Ne
tw
o
rk
(F
a
ste
r
-
RCNN
)
a
rc
h
it
e
c
tu
re
s,
is
p
ro
p
o
se
d
.
M
o
re
sp
e
c
ially
,
th
e
De
e
p
L
e
a
rn
in
g
(DL
)
is
u
se
d
to
d
e
tec
t
li
c
e
n
s
e
p
late
s
in
g
iv
e
n
i
m
a
g
e
s.
T
h
e
p
ro
p
o
se
d
li
c
e
n
se
p
late
d
e
tec
ti
o
n
m
e
th
o
d
u
se
s
th
r
e
e
F
a
ste
r
-
RCNN
m
o
d
u
les
w
h
e
re
e
a
c
h
f
a
st
e
r
RCNN
m
o
d
u
le
u
se
s
a
p
re
-
tra
in
e
d
CNN
m
o
d
e
l
n
a
m
e
l
y
A
l
e
x
N
e
t,
V
GG
1
6
a
n
d
V
GG
1
9
.
Eac
h
F
a
ste
r
-
RCNN
m
o
d
u
le
is
train
e
d
in
d
e
p
e
n
d
e
n
tl
y
a
n
d
th
e
ir
re
su
lt
s
a
re
f
u
se
d
in
f
u
sin
g
la
y
e
r.
F
u
sin
g
la
y
e
r
u
se
a
v
e
r
a
g
e
o
p
e
ra
to
r
o
n
th
e
X
a
n
d
Y
c
o
o
rd
i
n
a
tes
o
f
th
e
o
u
tp
u
ts
o
f
th
e
F
a
ste
r
-
RCNN
m
o
d
u
le
s
a
n
d
m
a
x
i
m
u
m
o
p
e
ra
to
r
is
e
m
p
lo
y
e
d
o
n
t
h
e
w
id
th
a
n
d
h
e
ig
h
t
o
u
t
p
u
ts
o
f
th
e
F
a
ste
r
-
RCNN
m
o
d
u
les
.
A
p
u
b
li
c
ly
a
v
a
il
a
b
le
d
a
tas
e
t
is
u
se
d
in
e
x
p
e
ri
m
e
n
ts.
T
h
e
a
c
c
u
ra
c
y
is
u
se
d
a
s
a
p
e
rf
o
rm
a
n
c
e
in
d
ica
to
r
o
f
th
e
p
ro
p
o
se
d
m
e
th
o
d
.
F
o
r
1
0
0
te
stin
g
ima
g
e
s,
th
e
p
ro
p
o
se
d
m
e
th
o
d
d
e
tec
ts
th
e
e
x
a
c
t
lo
c
a
ti
o
n
o
f
li
c
e
n
se
p
l
a
tes
f
o
r
9
7
im
a
g
e
s.
T
h
e
a
c
c
u
ra
c
y
o
f
th
e
p
ro
p
o
se
d
m
e
th
o
d
is
9
7
%
.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
Fas
ter
-
R
C
NN
L
ice
n
s
e
p
late
d
etec
tio
n
Veh
icle
i
m
a
g
es
Co
p
y
rig
h
t
©
2
0
2
0
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
.
Al
l
rig
h
ts
re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Naa
m
a
n
O
m
ar
,
Dep
ar
t
m
en
t o
f
I
n
f
o
r
m
atio
n
T
e
ch
n
o
lo
g
y
,
Du
h
o
k
P
o
l
y
tech
n
ic
U
n
i
v
er
s
it
y
,
6
1
Z
ak
h
o
R
o
ad
,
Ma
zi
Qr
،
1
0
0
6
,
Du
h
o
k
4
2
0
0
1
,
I
r
a
q
.
E
m
ail:
ij
ee
cs.iae
s
@
g
m
a
il.c
o
m
1.
I
NT
RO
D
UCT
I
O
N
Dee
p
L
ea
r
n
i
n
g
(
D
L
)
co
n
ce
r
n
s
w
i
th
p
r
o
ce
s
s
i
n
g
i
n
f
o
r
m
atio
n
u
til
izin
g
d
ee
p
n
et
w
o
r
k
s
.
I
t
is
a
p
ar
t
o
f
m
ac
h
i
n
e
lear
n
i
n
g
ap
p
r
o
ac
h
es.
I
n
i
ts
ea
r
lier
ap
p
ea
r
an
ce
in
1
9
4
3
,
DL
w
a
s
ter
m
ed
b
y
Mc
C
lu
llo
ch
an
d
P
itt
s
a
s
“
c
y
b
er
n
etics”.
DL
is
s
u
p
er
io
r
in
d
r
a
w
in
g
o
u
t
-
co
m
es
f
r
o
m
c
o
m
p
le
x
p
r
o
b
le
m
s
.
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
et
w
o
r
k
(
C
NN)
is
a
n
in
s
tan
ce
o
f
d
ee
p
lear
n
in
g
s
tr
ate
g
y
i
s
m
i
m
ic
k
in
g
b
r
ain
f
u
n
ct
io
n
i
n
p
r
o
ce
s
s
i
n
g
i
n
-
f
o
r
m
atio
n
.
R
ec
u
r
r
en
t
Neu
r
al
Net
w
o
r
k
(
R
NN)
,
Dee
p
Neu
r
al
Net
w
o
r
k
(
DNN)
,
an
d
C
o
n
v
o
l
u
tio
n
al
Ne
u
r
al
Net
w
o
r
k
(
C
NN)
ar
e
ar
ch
itectu
r
e
t
y
p
e
s
o
f
D
L
[1
,
2]
.
Au
to
m
a
tic
L
icen
s
e
P
late
R
ec
o
g
n
itio
n
(
AL
P
R
)
is
a
n
i
m
p
o
r
tan
t
to
p
ic
in
i
n
tell
ig
e
n
t
tr
a
n
s
p
o
r
tatio
n
ap
p
licatio
n
s
.
AL
P
R
ca
n
b
e
u
s
ed
to
id
en
tify
th
e
v
e
h
icles
f
r
o
m
its
licen
s
e
p
late.
AL
P
R
h
as
p
o
ten
tial
to
b
e
u
s
ed
in
v
ar
iet
y
o
f
ap
p
licatio
n
a
r
ea
s
s
u
c
h
as
tr
af
f
ic
s
u
r
v
eil
lan
ce
,
tr
af
f
ic
s
a
f
et
y
,
s
ec
u
r
it
y
,
s
er
v
ices
p
u
r
p
o
s
es
an
d
p
ar
k
in
g
m
a
n
ag
e
m
e
n
t
[3
-
6]
.
L
i
g
h
tin
g
an
d
w
ea
th
er
co
n
d
i
tio
n
s
ef
f
ec
t
AL
P
R
s
y
s
te
m
s
ig
n
i
f
ica
n
tl
y
[7
,
8]
.
An
AL
P
R
s
y
s
te
m
g
e
n
er
all
y
co
m
p
o
s
ed
o
f
f
o
u
r
m
ai
n
b
lo
ck
s
s
u
c
h
as
i
m
a
g
e
ca
p
tu
r
i
n
g
b
lo
ck
,
licen
s
e
p
late
d
etec
tio
n
b
lo
ck
,
c
h
ar
ac
ter
s
eg
m
e
n
tat
io
n
b
lo
ck
an
d
c
h
a
r
ac
ter
r
ec
o
g
n
itio
n
b
lo
ck
,
r
e
s
p
ec
tiv
el
y
[9
-
12]
.
T
h
e
licen
s
e
p
late
d
etec
tio
n
,
w
h
ic
h
is
in
th
e
s
ec
o
n
d
b
lo
ck
o
f
th
e
AL
P
D,
ca
n
b
e
s
ee
n
as
th
e
m
o
s
t
i
m
p
o
r
tan
t
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
F
u
s
ed
fa
s
ter R
C
N
N
s
fo
r
ef
fici
en
t d
etec
tio
n
o
f t
h
e
licen
s
e
p
la
tes
(
N
a
a
ma
n
Oma
r
)
975
p
ar
t.
B
ec
au
s
e,
it
ef
f
ec
ts
t
h
e
r
e
co
g
n
itio
n
p
er
f
o
r
m
a
n
ce
o
f
s
u
b
s
eq
u
en
t
p
ar
ts
o
f
t
h
e
AL
P
D
[
1
3
-
16]
.
A
L
P
s
y
s
te
m
s
ar
e
in
th
e
d
e
m
a
n
d
an
d
i
n
th
e
l
ast
t
w
o
d
ec
ad
es,
s
o
m
an
y
w
o
r
k
s
h
av
e
b
ee
n
ca
r
r
ied
o
u
t
o
n
it.
I
n
th
i
s
w
o
r
k
,
as
we
co
n
ce
n
tr
ated
o
n
t
h
e
lice
n
s
e
p
late
d
etec
tio
n
,
th
e
liter
at
u
r
e
t
h
at
ar
e
r
elate
d
w
it
h
t
h
e
l
icen
s
e
p
late
d
etec
tio
n
is
r
ev
ie
w
ed
.
B
ec
au
s
e
o
f
i
n
cr
ea
s
i
n
g
t
h
e
v
eh
icles
n
u
m
b
er
,
t
h
e
tr
a
f
f
ic
v
io
lat
io
n
a
n
d
tr
af
f
ic
co
n
tr
o
l
b
ec
o
m
e
a
s
er
io
u
s
an
d
ch
alle
n
g
i
n
g
p
r
o
b
le
m
i
n
all
o
v
er
th
e
w
o
r
d
.
Mo
r
eo
v
er
e,
th
e
v
e
h
icle
d
etec
tin
a
n
d
r
ec
o
g
n
itio
n
b
ec
o
m
e
d
if
f
ic
u
lt
f
r
o
m
v
e
h
icle
L
P
.
Va
r
io
u
s
d
ataset
u
s
ed
f
o
r
tr
ain
i
n
g
ev
a
lu
at
io
n
al
g
o
r
ith
m
s
[
1
7
,
1
8]
.
I
n
last
d
ec
ate,
d
if
f
er
e
n
t
al
g
o
r
ith
m
s
a
n
d
m
et
h
o
d
s
h
a
v
e
b
ee
n
p
r
p
p
o
s
ed
f
o
r
d
etec
tio
n
an
d
r
ec
o
g
n
itio
n
v
eh
ic
le
L
P
to
d
ate,
s
u
c
h
as VG
G1
6
,
VGG1
9
,
C
N
N
[
1
9
]
.
Ho
et
al.
[
2
0
]
p
r
o
p
o
s
ed
a
li
ce
n
s
e
p
late
d
etec
tio
n
m
et
h
o
d
f
o
r
o
n
lin
e
ap
p
licatio
n
s
.
T
h
e
p
r
o
p
o
s
e
d
m
et
h
o
d
w
a
s
co
m
p
o
s
ed
o
f
t
w
o
s
tag
e
s
.
Au
t
h
o
r
s
u
s
ed
A
d
ab
o
o
s
t
m
et
h
o
d
,
s
u
p
p
o
r
t
v
ec
t
o
r
m
ac
h
i
n
e
(
S
VM
)
cla
s
s
i
f
ier
a
n
d
s
ca
le
i
n
v
ar
ian
t
f
ea
tu
r
e
tr
a
n
s
f
o
r
m
(
SIFT
)
d
escr
ip
to
r
s
f
o
r
AL
P
D.
I
n
th
e
f
ir
s
t
s
tag
e
t
h
e
A
d
ab
o
o
s
t
m
et
h
o
d
w
a
s
u
s
ed
to
lo
ca
lize
th
e
lice
n
s
e
p
late
r
eg
io
n
.
T
h
e
SIFT
f
ea
tu
r
es
w
er
e
e
x
tr
ac
ted
f
r
o
m
t
h
e
ch
ar
ac
ter
s
o
f
th
e
lice
n
s
e
p
late
r
eg
io
n
a
n
d
S
VM
class
i
f
ier
w
a
s
u
s
ed
to
r
ec
o
g
n
ize
t
h
e
lice
n
s
e
p
late.
Au
t
h
o
r
s
u
s
ed
a
d
ataset
in
v
o
l
v
es
8
0
0
i
m
a
g
es
i
n
t
h
eir
ex
p
er
i
m
e
n
ts
.
T
h
e
r
ep
o
r
ted
ac
cu
r
ac
y
w
a
s
8
8
%.
Kata
ta
et
al.
p
r
o
p
o
s
ed
a
m
et
h
o
d
to
d
etec
t
licen
s
e
p
late
o
f
a
g
iv
en
i
m
ag
e
b
ased
o
n
Gab
o
r
f
ilte
r
s
an
d
n
e
u
r
al
n
et
w
o
r
k
s
(
N
Ns)
[
2
1
]
.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
w
a
s
co
m
p
o
s
ed
o
f
th
r
ee
s
tep
s
.
T
h
e
f
ir
s
t
s
tep
co
v
er
s
th
e
g
e
n
er
atio
n
o
f
t
h
e
f
ea
tu
r
e
v
ec
to
r
s
f
o
r
b
o
th
tr
ain
i
n
g
a
n
d
test
in
g
i
m
a
g
es.
Au
th
o
r
s
u
s
ed
co
n
tr
ast
l
i
m
i
ted
ad
ap
tiv
e
h
i
s
to
g
r
a
m
eq
u
al
iza
tio
n
(
C
L
A
HE
)
b
e
f
o
r
e
Featu
r
e
E
x
tr
ac
tio
n
(
FE)
f
o
r
i
m
p
r
o
v
i
n
g
t
h
e
q
u
alit
y
o
f
t
h
e
in
p
u
t
i
m
a
g
e
s
.
Gab
o
r
f
ilter
s
w
er
e
u
s
ed
f
ea
t
u
r
e
ex
tr
ac
tio
n
a
n
d
NN
s
w
a
s
u
s
ed
f
o
r
clas
s
i
f
icatio
n
.
5
8
T
u
n
is
ia
n
v
eh
ic
les
i
m
a
g
es
w
er
e
u
s
ed
in
e
x
p
er
i
m
e
n
ts
an
d
ac
ce
p
tab
le
r
esu
lts
w
er
e
r
ep
o
r
ted
b
y
th
e
a
u
t
h
o
r
s
.
Ki
m
et
al.
p
r
esen
ted
a
m
et
h
o
d
f
o
r
licen
s
e
p
late
d
etec
tio
n
th
r
o
u
g
h
t
w
o
s
ta
g
es
[
1
6
]
.
I
n
th
e
f
ir
s
t
s
ta
g
e,
th
e
r
eg
io
n
o
f
v
e
h
icle
w
a
s
lo
ca
ted
in
w
h
o
le
i
m
a
g
e
b
y
u
s
i
n
g
C
NN
alg
o
r
ith
m
to
s
p
ec
if
y
t
h
e
r
eg
io
n
o
f
in
ter
est
(
R
OI
)
ea
s
il
y
.
I
n
th
e
s
ec
o
n
d
s
tag
e,
th
e
d
etec
tio
n
o
f
t
h
e
licen
s
e
p
late
ca
n
d
id
ates
f
r
o
m
v
eh
ic
le
r
eg
io
n
w
as
ac
co
m
p
li
s
h
ed
b
y
u
s
i
n
g
t
h
e
h
ier
ar
ch
ical
s
a
m
p
li
n
g
m
et
h
o
d
.
T
h
e
f
alse
p
o
s
iti
v
e
licen
s
e
p
late
r
eg
io
n
s
w
er
e
eli
m
i
n
ated
b
y
u
s
in
g
d
ee
p
C
NN.
T
h
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
m
e
th
o
d
w
a
s
ev
al
u
ated
o
n
C
altec
h
d
ataset.
T
h
e
o
b
tain
ed
p
r
ec
is
io
n
an
d
r
ec
all
s
co
r
es
w
er
e
9
8
.
3
9
%
an
d
9
6
.
8
3
%,
r
esp
ec
tiv
el
y
.
Yu
a
n
et
al.
p
r
o
p
o
s
ed
a
m
et
h
o
d
to
d
etec
t
v
eh
icle
licen
s
e
p
late
[
4
]
.
I
n
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
lin
e
d
en
s
it
y
f
ilter
w
a
s
u
s
ed
to
f
in
d
th
e
ca
n
d
id
ate
r
eg
io
n
s
.
T
h
en
,
t
h
e
p
o
s
iti
v
e
(
tr
u
e)
lice
n
s
e
p
late
w
a
s
id
en
t
if
ied
b
ased
o
n
li
n
ea
r
SVMs.
Fo
r
p
er
f
o
r
m
a
n
ce
e
v
al
u
atio
n
,
t
h
e
C
a
ltech
licen
s
e
p
late
d
atas
et
w
a
s
u
s
ed
.
Au
th
o
r
s
a
ls
o
u
s
ed
an
o
th
er
d
atase
t
th
at
co
n
tai
n
s
3
8
2
8
im
a
g
es.
T
h
e
au
th
o
r
s
r
ep
o
r
ted
9
6
.
6
2
% a
v
er
ag
e
ac
cu
r
ac
y
s
co
r
e.
Z
h
ao
et
al.
p
r
o
p
o
s
ed
a
m
et
h
o
d
w
h
ic
h
w
a
s
co
m
p
o
s
ed
o
f
t
h
e
Haa
r
-
li
k
e
ca
s
ca
d
e
cla
s
s
if
ier
an
d
A
d
ab
o
o
s
t
f
o
r
v
eh
ic
le
licen
s
e
p
late
d
etec
tio
n
[
1
3
]
.
T
h
e
d
ataset,
w
h
ich
w
a
s
u
s
ed
i
n
ex
p
er
i
m
en
ts
,
w
as
co
llected
f
r
o
m
d
i
f
f
er
en
t
en
v
ir
o
n
m
e
n
ts
in
C
h
in
a.
T
h
e
ac
cu
r
ac
y
o
f
t
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
w
as
8
9
.
5
%.
Ma
s
o
o
d
et
al.
p
r
o
p
o
s
ed
an
alg
o
r
ith
m
f
o
r
L
P
DR
b
ased
o
n
d
ee
p
C
o
n
v
o
lu
tio
n
al
Ne
u
r
al
Net
w
o
r
k
s
(
C
NN
s
)
[
2
2
]
.
T
h
e
alg
o
r
ith
m
w
a
s
ap
p
lied
u
n
d
er
v
ar
io
u
s
wea
th
er
co
n
d
itio
n
s
a
n
d
licen
s
e
p
late
s
h
ap
es.
T
h
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
p
er
f
o
r
m
ed
d
etec
tio
n
,
ch
ar
ac
ter
s
eg
m
en
ta
tio
n
a
n
d
r
ec
o
g
n
itio
n
.
T
w
o
d
if
f
er
en
t
d
ataset
s
h
av
e
b
ee
n
u
s
ed
to
ev
alu
ate
th
e
p
er
f
o
r
m
a
n
ce
,
3
2
8
i
m
ag
e
s
f
r
o
m
US
A
a
n
d
5
5
0
i
m
ag
e
s
f
r
o
m
E
u
r
o
p
ea
n
.
T
h
e
p
er
f
o
r
m
an
c
e
w
a
s
9
9
.
0
9
%
an
d
9
9
.
6
4
% f
o
r
USA
an
d
E
u
r
o
p
ea
n
d
atasets
,
r
esp
ec
ti
v
el
y
.
A
za
m
et
al.
p
r
o
p
o
s
ed
an
alg
o
r
ith
m
to
d
et
ec
ted
v
eh
icle
lice
n
s
e
p
late
r
e
g
io
n
i
n
d
i
f
f
er
e
n
t
h
az
ar
d
o
u
s
i
m
a
g
e
co
n
d
itio
n
s
[
1
0
]
.
Var
io
u
s
p
r
e
-
p
r
o
ce
s
s
i
n
g
s
tep
s
w
er
e
ap
p
lied
o
n
in
p
u
t
i
m
a
g
e
f
o
r
n
o
is
e
r
e
m
o
v
a
l
an
d
co
n
tr
ast
e
n
h
a
n
ce
m
e
n
t.
Au
t
h
o
r
s
[
1
0
]
u
s
ed
R
ad
o
n
tr
an
s
f
o
r
m
a
n
d
tilt
co
r
r
ec
tio
n
f
o
r
d
ete
ctio
n
o
f
th
e
lice
n
s
e
p
late.
A
d
ataset
t
h
at
co
n
tai
n
s
8
5
0
v
eh
icle
i
m
ag
e
s
f
o
r
d
if
f
er
en
t
co
n
d
itio
n
s
w
er
e
u
s
ed
i
n
e
x
p
er
i
m
e
n
ts
an
d
th
e
o
b
tain
ed
ac
c
u
r
ac
y
s
co
r
e
w
a
s
9
4
%.
Nai
m
i
et
a
l
.
p
r
o
p
o
s
ed
an
al
g
o
r
ith
m
f
o
r
lice
n
s
e
p
late
d
etec
tio
n
f
o
r
v
ar
io
u
s
n
atio
n
a
n
d
m
u
lti
cr
ite
r
io
n
s
p
late
in
d
if
f
er
en
t
co
n
d
it
i
o
n
[
1
1
]
.
T
h
e
p
r
o
p
o
s
ed
w
o
r
k
w
a
s
b
ased
o
n
d
ee
p
lear
n
in
g
w
h
er
e
s
el
f
-
ta
u
g
h
t
f
ea
tu
r
es
w
er
e
u
s
ed
f
o
r
d
etec
tio
n
o
f
th
e
licen
s
e
p
late.
T
h
e
au
t
h
o
r
s
co
m
b
in
ed
r
eg
io
n
p
r
o
p
o
s
al
n
et
w
o
r
k
w
it
h
C
NN
to
im
p
r
o
v
e
t
h
e
p
er
f
o
r
m
an
ce
.
Firstl
y
,
f
ea
tu
r
e
s
w
er
e
e
x
tr
ac
ted
b
y
C
N
N
an
d
p
r
esen
ted
th
e
f
ea
t
u
r
e
m
ap
to
t
h
e
R
P
N.
Seco
n
d
l
y
,
s
o
f
t
-
m
a
x
class
i
f
ier
w
as
u
s
ed
to
d
etec
t
th
e
L
P
r
e
g
io
n
.
5
0
0
0
co
lo
r
im
a
g
es
w
er
e
u
s
ed
in
e
x
p
er
i
m
en
ts
an
d
9
9
% a
cc
u
r
ac
y
s
c
o
r
e
w
a
s
o
b
tain
ed
.
L
i
et
al.
p
r
o
p
o
s
ed
an
alg
o
r
ith
m
f
o
r
v
e
h
icle
l
icen
s
e
p
late
d
etec
tio
n
an
d
r
ec
o
g
n
itio
n
u
s
in
g
d
ee
p
C
NN
an
d
L
ST
Ms
[
7
]
.
I
n
t
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
,
t
w
o
C
NN
cla
s
s
i
f
ier
s
w
er
e
u
s
ed
,
o
n
e
w
as
u
s
ed
f
o
r
d
etec
tio
n
o
f
t
h
e
ch
ar
ac
ter
s
f
r
o
m
i
m
ag
e
s
,
a
n
d
th
e
s
ec
o
n
d
o
n
e
w
a
s
u
s
ed
to
r
e
m
o
v
e
f
al
s
e
p
o
s
iti
v
e.
C
altec
h
ca
r
s
d
ataset
an
d
A
O
L
P
d
ataset
w
er
e
u
s
ed
f
o
r
d
etec
tio
n
p
er
f
o
r
m
a
n
ce
.
T
h
e
a
ch
iev
ed
p
er
f
o
r
m
a
n
ce
s
w
er
e
9
7
.
5
6
%
p
r
ec
is
io
n
an
d
9
5
.
2
4
%
r
ec
all.
L
ali
m
i
et
a
l.
p
r
esen
ted
an
a
u
to
m
a
tic
al
g
o
r
ith
m
f
o
r
v
e
h
icle
l
icen
s
e
p
late
d
etec
tio
n
[
2
3
]
.
T
h
e
in
p
u
t
i
m
ag
e
s
w
er
e
en
h
a
n
ce
d
b
y
u
s
in
g
ed
g
e
d
e
n
s
it
y
an
d
i
n
ten
s
it
y
v
ar
ian
ce
.
Ver
ti
ca
l
ed
g
e
d
etec
tio
n
,
m
o
r
p
h
o
lo
g
ical
f
ilter
in
g
an
d
g
eo
m
e
tr
ical
f
ea
t
u
r
es
w
er
e
u
s
e
d
to
o
b
tain
th
e
co
r
r
ec
t
lo
ca
ti
o
n
o
f
licen
s
e
p
late
.
T
h
e
d
ataset
u
s
ed
in
th
i
s
p
ap
er
co
n
s
is
t
s
o
f
4
2
5
co
llected
im
a
g
es
.
T
h
e
ac
ce
s
s
r
ate
ac
h
iev
ed
f
o
r
VL
P
D
in
th
i
s
alg
o
r
ith
m
is
9
4
.
1
2
%.
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.
19
,
No
.
2
,
A
u
g
u
s
t
20
2
0
:
9
7
4
-
982
976
I
n
th
i
s
p
ap
er
,
a
n
o
v
el
ap
p
r
o
ac
h
,
w
h
ic
h
i
s
b
ased
o
n
DL
,
i
s
p
r
o
p
o
s
ed
f
o
r
d
etec
tio
n
o
f
licen
s
e
p
lates
in
g
iv
e
n
i
m
a
g
es.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
d
o
es
n
o
t
u
s
e
an
y
p
r
e
-
p
r
o
ce
s
s
in
g
f
o
r
i
m
p
r
o
v
in
g
t
h
e
q
u
alit
y
o
f
t
h
e
i
n
p
u
t
i
m
a
g
es.
Mo
r
e
s
p
ec
if
icall
y
,
th
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
u
s
es
Fa
s
t
er
R
C
NN
f
o
r
d
etec
tio
n
o
f
th
e
licen
s
e
p
lates
[
1
5
]
.
T
h
r
ee
Fas
ter
R
C
NN
m
o
d
u
le
s
ar
e
u
s
ed
w
h
er
e
ea
c
h
o
f
t
h
e
m
u
s
e
s
a
p
r
e
-
tr
ai
n
ed
C
NN
m
o
d
el
s
u
ch
as
Alex
Net,
VGG1
6
an
d
VGG1
9
,
r
esp
ec
ti
v
el
y
.
E
ac
h
f
as
t
er
R
C
NN
p
r
o
d
u
ce
s
a
r
ec
tan
g
le
th
at
i
s
d
ef
i
n
e
d
b
y
f
o
u
r
p
ar
a
m
eter
s
s
u
c
h
as
X
a
n
d
Y
co
o
r
d
in
ates
o
f
t
h
e
u
p
p
er
co
r
n
er
o
f
th
e
r
ec
tan
g
le
an
d
w
id
th
(
W
)
a
n
d
h
eig
h
t
(
H)
o
f
t
h
e
r
ec
tan
g
le.
T
h
u
s
,
th
r
ee
r
ec
tan
g
les
ar
e
p
r
o
d
u
ce
d
b
y
t
h
r
ee
f
as
ter
R
C
NN
m
o
d
el
s
.
A
f
u
s
io
n
la
y
er
is
u
s
ed
af
ter
f
aster
R
C
NN
m
o
d
els.
T
h
e
f
u
s
io
n
is
h
a
n
d
led
b
y
u
s
in
g
a
v
er
ag
e
o
p
er
ato
r
o
n
co
lu
m
n
w
i
s
e
f
o
r
X
a
n
d
Y
co
o
r
d
in
ates
f
r
o
m
all
f
a
s
ter
R
C
NNs
an
d
m
a
x
i
m
u
m
o
p
er
ato
r
s
is
u
s
ed
o
n
co
lu
m
n
w
is
e
f
o
r
W
an
d
H
v
alu
es
t
h
a
t
o
b
tain
ed
f
r
o
m
all
f
aster
R
C
N
Ns.
A
d
ataset,
t
h
at
co
n
tai
n
s
5
0
2
im
a
g
es,
is
u
s
ed
in
o
u
r
ex
p
e
r
i
m
en
ts
.
T
h
e
i
m
ag
e
s
w
er
e
co
llected
f
r
o
m
v
ar
io
u
s
en
v
ir
o
n
m
e
n
t
s
an
d
v
ar
io
u
s
co
n
d
itio
n
s
f
o
r
m
ak
in
g
th
e
d
ata
s
et
m
o
r
e
r
ea
lis
tic.
T
h
ese
all
i
m
a
g
es
ar
e
u
s
ed
f
o
r
tr
ain
i
n
g
o
f
th
e
f
a
s
ter
R
C
NN
m
o
d
el
s
a
n
d
t
h
e
f
lip
p
ed
an
d
r
o
ta
ted
v
er
s
io
n
s
o
f
th
e
i
m
a
g
es
ar
e
u
s
ed
i
n
test
i
n
g
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
.
T
h
e
o
b
tain
ed
r
esu
l
ts
s
h
o
w
th
at
t
h
e
p
r
o
p
o
s
ed
m
eth
o
d
is
q
u
ite
s
u
cc
ess
f
u
l i
n
d
etec
tio
n
o
f
th
e
l
icen
s
e
p
lates.
a)
T
h
e
o
r
ig
in
al
co
n
tr
ib
u
tio
n
o
f
t
h
is
w
o
r
k
is
as
f
o
llo
w
i
n
g
.
T
h
e
p
r
e
-
tr
ain
ed
C
NN
m
o
d
els
b
ased
f
aster
R
C
NN
m
o
d
el
s
ar
e
g
en
er
all
y
u
s
ed
in
v
ar
i
o
u
s
o
b
j
ec
t
d
etec
tio
n
ap
p
licatio
n
s
.
Ho
w
e
v
er
,
a
f
u
s
io
n
o
p
er
atio
n
o
v
er
m
u
ltip
le
f
aster
R
C
NN
m
o
d
els
h
a
s
n
o
t
b
ee
n
ap
p
lied
o
n
lice
n
s
e
p
late
d
etec
tio
n
u
n
t
il
s
o
f
ar
.
T
h
e
f
u
s
io
n
o
p
er
atio
n
is
j
u
s
ti
f
ied
b
o
th
m
a
t
h
e
m
a
ticall
y
a
n
d
ex
p
er
i
m
e
n
tall
y
.
W
h
ile
s
i
n
g
le
f
aster
R
C
N
N
m
o
d
els
p
r
o
d
u
ce
lo
w
d
etec
tio
n
r
ates,
th
e
f
u
s
ed
m
o
d
el
h
i
g
h
l
y
i
m
p
r
o
v
es t
h
e
ex
p
er
i
m
en
tal
r
es
u
lts
.
T
h
e
r
est
o
f
th
e
p
ap
er
is
o
r
g
a
n
ized
as
f
o
llo
w
.
Ne
x
t
s
ec
tio
n
b
r
ief
l
y
in
tr
o
d
u
ce
s
th
e
th
eo
r
y
o
f
th
e
f
aster
R
C
N
N.
Sectio
n
3
d
escr
ib
es
t
h
e
p
r
o
p
o
s
ed
m
eth
o
d
in
d
etai
l.
E
x
p
er
i
m
en
tal
w
o
r
k
s
a
n
d
r
esu
lt
s
ar
e
g
i
v
en
i
n
Sectio
n
4
.
T
h
e
p
ap
er
is
co
n
clu
d
ed
in
Sectio
n
5
.
2.
P
RO
P
O
SE
D
M
E
T
H
O
D
T
h
e
p
r
o
p
o
s
ed
licen
s
e
p
late
d
etec
tio
n
s
y
s
te
m
u
s
es
m
u
ltip
le
f
aster
r
e
g
io
n
al
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
et
w
o
r
k
s
(
Fas
ter
R
C
NN)
f
o
r
ef
f
icie
n
t
d
etec
tio
n
o
f
th
e
li
ce
n
s
e
p
late
r
eg
io
n
o
n
a
g
i
v
e
n
i
m
a
g
e.
Fi
g
u
r
e
1
s
h
o
w
s
t
h
e
ill
u
s
tr
atio
n
o
f
t
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
.
Fig
u
r
e
1
.
T
h
e
illu
s
tr
atio
n
o
f
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
d
o
es
n
o
t
u
s
e
an
y
p
r
e
-
p
r
o
ce
s
s
i
n
g
s
ta
g
e
f
o
r
i
m
p
r
o
v
in
g
t
h
e
q
u
alit
y
o
f
th
e
i
n
p
u
t
i
m
a
g
e.
B
ec
au
s
e,
t
h
e
f
aster
R
C
NN
m
o
d
els
ar
e
q
u
ite
r
o
b
u
s
t
a
g
ain
s
t
t
h
e
n
o
is
e
a
n
d
o
th
er
ef
f
e
cts.
T
h
is
also
s
a
v
e
s
th
e
co
m
p
u
ta
tio
n
ti
m
e
o
f
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
.
T
h
r
ee
Fas
ter
R
C
NN
m
o
d
u
le
s
ar
e
u
s
ed
w
h
er
e
ea
ch
o
f
th
e
m
u
s
e
s
a
p
r
e
-
tr
ain
ed
C
NN
m
o
d
e
l.
A
s
s
ee
n
in
Fi
g
u
r
e
1
,
th
e
f
u
s
io
n
la
y
er
is
in
t
h
e
h
ea
r
t
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
.
I
t
f
u
s
e
s
t
h
e
r
es
u
lt
s
t
h
at
ar
e
f
ed
f
r
o
m
a
ll
t
h
e
f
a
s
ter
R
C
NN
m
o
d
u
le
s
.
E
ac
h
Fa
s
ter
R
C
N
N
m
o
d
u
le
p
r
o
d
u
c
e
s
a
r
ec
tan
g
u
lar
r
eg
io
n
(
X,
Y
co
o
r
d
in
ates
an
d
t
h
e
w
id
th
a
n
d
h
ei
g
h
t
o
f
th
e
r
ec
ta
n
g
le
r
eg
io
n
)
as
th
e
lo
ca
tio
n
o
f
t
h
e
licen
s
e
p
late.
T
h
is
r
eg
io
n
m
i
g
h
t
n
o
t
b
e
lo
ca
ted
o
n
th
e
ex
a
ct
lo
ca
tio
n
o
f
t
h
e
licen
s
e
p
lat
e
s
o
f
u
s
io
n
la
y
er
is
u
s
ed
to
f
i
x
th
is
p
r
o
b
le
m
.
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
F
u
s
ed
fa
s
ter R
C
N
N
s
fo
r
ef
fici
en
t d
etec
tio
n
o
f t
h
e
licen
s
e
p
la
tes
(
N
a
a
ma
n
Oma
r
)
977
Fig
u
r
e
2
ill
u
s
tr
ates
th
e
VGG
1
9
,
VGG1
6
an
d
A
le
x
Net
ar
c
h
itect
u
r
e.
T
h
e
f
ir
s
t
Fas
ter
R
C
NN
m
o
d
el
u
s
e
s
th
e
w
e
ll
-
k
n
o
w
n
Alex
Net
m
o
d
el.
A
le
x
Net
m
o
d
el
ar
ch
it
ec
tu
r
e
co
n
tai
n
s
8
la
y
er
s
.
5
la
y
er
s
o
f
th
e
Alex
Net
m
o
d
el
i
s
n
a
m
ed
as
co
n
v
o
l
u
tio
n
al
la
y
er
s
an
d
3
lay
er
s
ar
e
ca
lled
f
u
ll
y
co
n
n
ec
ted
la
y
er
.
A
d
d
to
th
at,
it
co
n
tain
s
th
r
ee
p
o
o
lin
g
la
y
er
s
an
d
f
iv
e
R
eL
U
la
y
er
s
co
m
e
af
ter
co
n
v
o
lu
tio
n
a
l la
y
er
s
.
S
of
t
m
a
x
F
C
4096
P
oo
l
I
n
p
u
t
I
m
a
ge
C
onv
512
C
on
v
512
F
C
4096
V
G
G
N
e
t
19
F
C
4096
C
on
v
512
P
oo
l
C
o
n
v
512
C
on
v
256
P
o
o
l
C
onv
256
C
on
v
128
C
on
v
128
P
oo
l
P
oo
l
C
onv
64
C
on
v
64
C
onv
256
C
on
v
512
C
on
v
512
C
on
v
512
C
on
v
512
C
onv
256
S
o
f
t
m
a
x
F
C
4096
P
oo
l
I
nput
I
m
a
ge
C
on
v
384
P
oo
l
C
o
n
v
256
C
on
v
96
F
C
4096
A
l
e
xN
e
t
P
ool
C
on
v
384
C
onv
256
F
C
1000
Fig
u
r
e
2
.
T
h
e
s
tr
u
ctu
r
es o
f
t
h
e
A
le
x
Net,
VGG1
6
an
d
VGG1
9
ar
ch
itectu
r
es
VGG1
6
is
an
o
th
er
p
r
e
-
tr
ai
n
ed
d
ee
p
m
o
d
el
w
h
ic
h
h
as
1
3
co
n
v
o
l
u
tio
n
a
l
la
y
er
s
an
d
3
f
u
ll
y
co
n
n
ec
ted
la
y
er
s
.
I
t
w
o
r
k
s
v
er
y
w
ell
an
d
h
as
g
o
o
d
p
er
f
o
r
m
a
n
ce
w
it
h
lar
g
e
d
ataset
s
,
b
ec
au
s
e
it
u
s
es
3
×3
s
m
all
co
n
v
o
lu
tio
n
f
ilter
s
in
a
ll
la
y
er
s
.
T
h
e
VGG1
9
m
o
d
el
is
co
n
s
t
r
u
cted
as
a
d
ee
p
er
v
er
s
io
n
o
f
th
e
VGG1
6
m
o
d
el
f
o
r
m
o
r
e
p
er
f
o
r
m
a
n
ce
a
n
d
b
etter
o
u
tp
u
t.
I
t
co
n
tai
n
s
1
6
co
n
v
o
l
u
tio
n
a
l
la
y
er
s
a
n
d
3
f
u
ll
y
co
n
n
ec
ted
la
y
er
s
.
An
ill
u
s
tr
ati
v
e
e
x
a
m
p
le
o
f
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
is
g
i
v
e
n
in
F
ig
u
r
e
3
.
(
a)
(
b
)
(
c)
(
d
)
Fig
u
r
e
3
.
T
h
e
ef
f
ec
t o
f
t
h
e
f
u
s
i
n
g
la
y
er
,
(
a)
T
h
e
r
esu
lt o
f
Fas
t
er
R
C
NN
m
o
d
u
le
1
,
(
b
)
T
h
e
r
esu
lt o
f
Fas
ter
R
C
N
N
m
o
d
u
le
2
,
(
c)
T
h
e
r
esu
lt o
f
Fas
ter
R
C
NN
m
o
d
u
le
3
,
(
d
)
T
h
e
r
esu
lt o
f
f
u
s
i
n
g
la
y
er
Fig
u
r
e
3
(
a)
s
h
o
w
s
th
e
o
u
tp
u
t
o
f
t
h
e
f
ir
s
t
f
a
s
ter
R
C
N
N
m
o
d
u
le.
T
h
e
l
ice
n
s
e
p
late
r
eg
io
n
i
s
i
n
d
icate
d
w
it
h
a
y
ello
w
b
o
u
n
d
i
n
g
b
o
x
.
As
s
ee
n
,
th
e
f
ir
s
t
f
a
s
ter
R
C
N
N
m
o
d
u
le
co
u
ld
n
o
t o
b
tai
n
t
h
e
e
x
ac
t lo
ca
tio
n
o
f
t
h
e
licen
s
e
p
late
r
e
g
io
n
.
O
n
e
ch
ar
ac
ter
is
n
o
t
co
v
er
ed
b
y
th
e
y
ello
w
b
o
u
n
d
i
n
g
b
o
x
.
Mo
r
eo
v
er
,
as
g
i
v
en
i
n
Fig
u
r
e
3
(
b
)
,
th
e
s
ec
o
n
d
f
aste
r
R
C
NN
m
o
d
u
le
ca
n
d
etec
t
o
n
l
y
a
s
m
all
r
eg
io
n
o
f
t
h
e
licen
s
e
p
late
r
eg
io
n
a
s
s
h
o
w
n
b
y
t
h
e
r
ed
b
o
u
n
d
in
g
b
o
x
.
T
h
e
m
o
s
t
s
u
c
c
es
s
f
u
l
d
e
t
e
ct
io
n
is
h
an
d
l
e
d
b
y
th
e
t
h
i
r
d
f
a
s
t
er
R
C
N
N
m
o
d
u
l
e
as
s
h
o
w
n
in
th
e
F
ig
u
r
e
3
(
c
)
.
A
lm
o
s
t
al
l
l
i
c
en
s
e
p
la
t
e
r
eg
i
o
n
is
c
o
v
e
r
e
d
a
s
in
d
i
c
at
e
d
b
y
th
e
g
r
e
e
n
b
o
u
n
d
in
g
b
o
x
.
F
in
a
l
ly
,
th
e
Fig
u
r
e
3
(
d
)
s
h
o
w
s
t
h
e
f
u
s
e
d
r
esu
l
ts
w
h
e
r
e
th
e
f
in
a
l
r
eg
i
o
n
is
in
d
i
c
at
e
d
b
y
a
m
ag
e
n
t
a
b
o
u
n
d
in
g
b
o
x
.
T
h
e
f
u
s
e
d
r
e
s
u
lt
i
s
a
n
n
o
u
n
c
e
d
a
s
th
e
o
u
t
p
u
t
o
f
th
e
p
r
o
p
o
s
e
d
m
e
th
o
d
.
T
h
e
f
u
s
in
g
o
p
e
r
at
i
o
n
is
g
iv
en
a
s
f
o
ll
o
w
in
g
;
(
1
)
(
2
)
(
3
)
[
[
]
[
]
]
(
4
)
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.
19
,
No
.
2
,
A
u
g
u
s
t
20
2
0
:
9
7
4
-
982
978
w
h
er
e
X
a
n
d
Y
s
h
o
w
t
h
e
u
p
p
er
co
r
n
er
co
o
r
d
in
ate
o
f
t
h
e
b
o
u
n
d
in
g
b
o
x
an
d
W
a
n
d
H
s
h
o
w
t
h
e
w
id
th
an
d
h
eig
h
t
o
f
t
h
e
b
o
u
n
d
i
n
g
b
o
x
,
r
esp
ec
tiv
el
y
.
A
s
g
i
v
e
n
in
(
1
)
-
(
3
)
,
th
e
b
o
u
n
d
in
g
b
o
x
es
f
r
o
m
all
Fas
ter
R
C
NN
ar
e
s
av
ed
an
d
t
h
e
f
u
s
ed
b
o
u
n
d
i
n
g
b
o
x
is
d
ef
i
n
ed
i
n
(
4
)
.
T
h
e
av
er
ag
e
o
p
er
ato
r
is
u
s
ed
o
n
co
l
u
m
n
w
i
s
e
f
o
r
X
a
n
d
Y
co
o
r
d
in
ates
an
d
t
h
e
m
a
x
i
m
u
m
o
p
er
ato
r
is
u
s
ed
o
n
co
l
u
m
n
w
is
e
f
o
r
W
a
n
d
H
v
al
u
es.
T
h
is
f
u
s
in
g
p
r
o
ce
d
u
r
e
en
s
u
r
es t
h
e
o
p
ti
m
u
m
lo
ca
tio
n
o
f
th
e
lice
n
s
e
p
late.
I
t
is
w
o
r
t
h
o
f
m
e
n
tio
n
i
n
g
t
h
at
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
h
as
tr
ain
i
n
g
a
n
d
test
i
n
g
p
h
ases
.
I
n
itiall
y
,
a
tr
ain
i
n
g
i
m
a
g
e
s
et
i
s
u
s
ed
to
tr
ain
t
h
e
all
f
a
s
ter
R
C
NN
m
o
d
u
les.
Af
ter
tr
ain
i
n
g
o
f
th
e
all
f
a
s
ter
R
C
NN
m
o
d
u
les,
t
h
e
test
in
g
p
h
a
s
e
ca
n
b
e
ap
p
lied
o
n
th
e
test
i
n
g
i
m
a
g
es
.
3.
RE
S
E
ARCH
M
E
T
H
O
D
3
.
1
.
F
a
s
t
er
re
g
io
n
w
it
h c
o
nv
o
lutio
na
l neura
l net
w
o
rk
E
x
p
lain
i
n
g
T
h
e
f
a
s
ter
R
C
NN
ar
ch
itect
u
r
e
w
as
p
r
o
p
o
s
ed
to
r
ed
u
ce
th
e
r
u
n
n
i
n
g
ti
m
e
o
f
t
h
e
o
b
j
ec
t
d
etec
tio
n
i
n
a
g
i
v
e
n
i
m
a
g
e.
T
h
e
R
eg
io
n
al
P
r
o
p
o
s
al
Net
wo
r
k
(
R
P
N)
is
in
th
e
h
ea
r
t
o
f
th
e
Fas
ter
R
C
NN
s
tr
u
ct
u
r
e
[
2
4
,
25]
.
R
P
N
is
co
m
p
o
s
ed
o
f
f
o
u
r
la
y
er
s
s
u
c
h
as
i
n
p
u
t
la
y
er
,
r
e
g
io
n
p
r
o
p
o
s
al
la
y
er
,
f
ea
t
u
r
e
ex
tr
ac
tio
n
la
y
er
an
d
cla
s
s
i
f
ic
atio
n
la
y
er
,
r
esp
ec
ti
v
el
y
.
T
h
e
C
NN
ar
ch
i
tectu
r
e
i
s
u
s
ed
in
f
ea
t
u
r
e
ex
tr
ac
tio
n
la
y
er
.
I
n
Fas
ter
R
C
NN,
th
e
tr
ain
ed
is
ca
r
r
ied
o
u
t
w
ith
m
u
lt
ip
le
o
f
th
e
p
r
o
p
o
s
ed
r
eg
io
n
s
in
s
tea
d
o
f
a
s
in
g
le
r
eg
io
n
,
w
h
ic
h
m
ak
e
s
th
e
tr
ai
n
i
n
g
p
r
o
ce
d
u
r
e
f
aster
th
a
n
t
h
e
p
r
ev
io
u
s
ar
ch
itect
u
r
e.
I
n
f
aster
R
C
N
N,
th
e
R
P
N’
s
an
d
f
ast
R
C
NN
’
s
co
n
v
o
lu
t
io
n
al
la
y
er
s
ar
e
co
m
b
i
n
ed
i
n
o
n
e
n
et
w
o
r
k
.
Fas
t
R
C
NN
h
as
s
ev
er
al
ad
v
an
tag
e
s
,
s
u
c
h
as
h
i
g
h
d
etec
tio
n
q
u
alit
y
,
u
s
i
n
g
m
u
lt
i
tas
k
lo
s
s
i
n
tr
ai
n
in
g
s
ta
g
e,
all
la
y
er
s
i
n
n
et
w
o
r
k
ca
n
b
e
u
p
d
ate
b
y
tr
ain
i
n
g
,
f
ea
tu
r
e
s
ex
tr
ac
tio
n
n
o
n
ee
d
s
to
r
ag
e
d
is
k
an
d
f
a
s
te
r
s
p
ee
d
[2
6
,
27]
.
T
h
u
s
,
t
h
e
co
m
p
u
tatio
n
al
ti
m
e
o
f
p
r
o
p
o
s
ed
n
et
w
o
r
k
is
r
ed
u
ce
d
.
3
.
2
.
Reg
i
o
n
pro
po
s
a
l net
w
o
r
k
(
RP
N)
R
eg
io
n
p
r
o
p
o
s
al
n
et
w
o
r
k
(
R
P
N)
aim
s
to
g
e
n
er
ate
p
o
ten
t
ial
r
eg
io
n
s
a
n
d
it
e
m
p
lo
y
s
a
n
et
w
o
r
k
to
d
eter
m
i
n
e
if
t
h
e
p
o
ten
tial
r
eg
io
n
s
co
n
tai
n
an
y
o
b
j
ec
ts
[
2
4
,
2
5
]
.
T
h
e
r
eg
io
n
p
r
o
p
o
s
als
ar
e
g
en
er
ated
b
y
t
h
e
s
elec
ti
v
e
s
ea
r
c
h
al
g
o
r
ith
m
.
T
h
e
p
r
o
d
u
ce
d
r
eg
io
n
s
ar
e
r
a
n
k
ed
b
y
t
h
e
R
P
N
an
d
t
h
e
o
n
es
m
o
s
t
li
k
el
y
co
n
tain
in
g
o
b
j
ec
ts
ar
e
s
elec
ted
.
T
h
e
r
eg
io
n
p
r
o
p
o
s
al
b
o
x
es
ar
e
ca
lled
as
an
ch
o
r
s
.
An
ch
o
r
s
ar
e
i
m
p
o
r
ta
n
t
i
n
Fas
ter
R
-
C
NN
ar
ch
itec
tu
r
e.
Ge
n
er
all
y
,
t
h
er
e
ar
e
9
a
n
ch
o
r
s
in
Fa
s
t
er
R
C
NN
ar
c
h
itect
u
r
e
at
a
p
o
s
itio
n
o
f
a
n
i
m
a
g
e.
T
h
e
an
ch
o
r
s
ar
e
th
en
e
x
a
m
in
ed
b
y
a
class
i
f
ier
to
ch
ec
k
th
e
p
r
o
b
ab
ilit
y
o
f
o
b
j
ec
ts
.
I
n
o
th
er
w
o
r
d
s
,
R
P
N
p
r
ed
icts
th
e
p
o
s
s
ib
ilit
y
o
f
an
a
n
ch
o
r
b
ein
g
b
ac
k
g
r
o
u
n
d
o
r
f
o
r
eg
r
o
u
n
d
,
an
d
r
ef
i
n
e
th
e
a
n
ch
o
r
.
T
h
e
tr
ain
in
g
o
f
t
h
e
Fas
ter
R
C
NN
is
ca
r
r
ied
o
u
t
w
it
h
a
tr
ain
i
n
g
d
ata
s
et
an
d
t
h
e
co
r
r
esp
o
n
d
in
g
g
r
o
u
n
d
-
tr
u
th
b
o
x
e
s
.
I
n
o
th
er
w
o
r
d
s
,
t
h
e
g
r
o
u
n
d
-
tr
u
t
h
b
o
x
es
ar
e
u
s
ed
to
lab
el
th
e
an
ch
o
r
s
.
T
h
e
b
asic
id
ea
in
Fas
ter
R
C
N
N
i
s
t
h
at
th
e
an
c
h
o
r
s
h
a
v
in
g
t
h
e
h
i
g
h
er
o
v
er
lap
s
w
it
h
g
r
o
u
n
d
-
tr
u
t
h
b
o
x
es
ar
e
lab
elled
as
f
o
r
eg
r
o
u
n
d
a
n
d
th
e
o
n
es
w
i
th
lo
w
er
o
v
er
lap
s
ar
e
lab
elled
as
b
ac
k
g
r
o
u
n
d
.
W
e
tr
ain
ed
th
r
ee
n
et
w
o
r
k
s
Al
ex
Net,
VGG1
6
an
d
VGG1
9
s
ep
ar
atel
y
,
u
s
ed
b
az
a_
s
lik
a,
w
h
ich
i
s
a
p
u
b
licl
y
a
v
a
ilab
le
d
ataset
[
2
8
,
2
9
].
4.
E
XP
E
R
I
M
E
NT
A
L
WO
RK
S
AND
RE
S
UL
T
S
T
h
e
ex
p
er
i
m
en
t
s
w
er
e
co
n
d
u
c
ted
o
n
MA
T
L
A
B
w
it
h
a
co
m
p
u
ter
h
av
in
g
a
n
I
n
tel
Xeo
n
E5
-
1
6
5
0
v
4
C
P
U
an
d
6
4
GB
m
e
m
o
r
y
.
A
p
u
b
licl
y
a
v
ailab
le
d
ata
s
et
was
u
s
ed
in
e
x
p
er
i
m
e
n
t
s
t
h
at
c
o
n
tain
s
5
0
2
i
m
a
g
es
[
2
9
]
.
T
h
e
v
eh
icle
i
m
ag
e
s
w
er
e
co
llected
u
n
d
er
v
ar
io
u
s
en
v
ir
o
n
m
e
n
tal
co
n
d
itio
n
s
s
u
ch
as
clo
u
d
y
d
a
y
,
r
ain
y
d
a
y
an
d
n
i
g
h
t
lig
h
ti
n
g
.
T
h
e
i
m
ag
e
s
co
v
e
r
d
if
f
er
e
n
t
t
y
p
es
o
f
v
e
h
icle
s
u
c
h
a
s
ca
r
s
,
tr
u
ck
s
,
b
u
s
e
s
a
n
d
m
i
n
i
b
u
s
es.
All
5
0
2
i
m
ag
e
s
w
er
e
u
s
ed
in
tr
ain
in
g
o
f
t
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
an
d
f
o
r
te
s
ti
n
g
p
r
o
ce
d
u
r
e,
r
an
d
o
m
l
y
s
elec
ted
1
0
0
im
a
g
es
w
er
e
f
li
p
p
ed
an
d
r
o
tated
in
5
an
d
1
0
d
eg
r
ee
s
.
W
h
ile
Fig
u
r
e
4
s
h
o
w
s
s
o
m
e
tr
ain
i
n
g
s
a
m
p
le
i
m
a
g
es,
Fi
g
u
r
e
5
s
h
o
ws s
o
m
e
te
s
t sa
m
p
le
i
m
a
g
e
s
.
Fig
u
r
e
4
.
So
m
e
s
a
m
p
le
i
m
a
g
e
s
th
at
w
er
e
u
s
ed
in
tr
ai
n
i
n
g
o
f
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
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
F
u
s
ed
fa
s
ter R
C
N
N
s
fo
r
ef
fici
en
t d
etec
tio
n
o
f t
h
e
licen
s
e
p
la
tes
(
N
a
a
ma
n
Oma
r
)
979
Fig
u
r
e
5
.
So
m
e
s
a
m
p
le
i
m
a
g
e
s
th
at
w
er
e
u
s
ed
in
te
s
t o
f
t
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
As
it
w
a
s
m
e
n
tio
n
ed
ea
r
lier
,
th
r
ee
p
r
e
-
tr
ain
ed
C
NN
m
o
d
el
s
w
er
e
u
s
ed
in
Fa
s
ter
R
C
NN
m
o
d
u
les.
T
h
e
tr
ain
in
g
p
ar
am
e
ter
s
o
f
t
h
e
f
aster
R
C
NN
m
o
d
u
les
s
e
t
as
f
o
llo
w
i
n
g
.
T
h
e
s
to
ch
as
tic
g
r
ad
ien
t
d
escen
t
w
i
th
m
o
m
e
n
t
u
m
o
p
ti
m
izer
w
a
s
u
t
ilized
in
t
h
e
tr
ain
in
g
p
r
o
ce
s
s
.
Ma
x
i
m
u
m
ep
o
ch
,
m
in
i
-
b
atc
h
s
ize,
a
n
d
in
it
ial
lear
n
in
g
r
ate
w
er
e
s
et
to
1
0
,
1
,
an
d
0
.
0
0
1
,
r
esp
ec
tiv
ely
.
I
n
a
d
d
itio
n
,
th
e
p
o
s
iti
v
e
a
n
d
n
e
g
a
tiv
e
o
v
er
lap
r
an
g
e
s
w
er
e
s
ca
led
to
th
e
[
0
-
0
.
3
]
a
n
d
[
0
.
6
-
1
]
r
an
g
es,
r
esp
ec
ti
v
el
y
.
T
h
e
n
u
m
b
er
o
f
r
eg
io
n
p
r
o
p
o
s
als
to
r
an
d
o
m
l
y
s
a
m
p
le
f
r
o
m
ea
c
h
tr
ai
n
i
n
g
i
m
ag
e
w
a
s
s
elec
ted
as
[
2
5
6
1
2
8
]
.
B
o
x
p
y
r
a
m
id
s
ca
le,
w
h
ich
was
n
a
m
ed
as
a
n
c
h
o
r
b
o
x
p
y
r
a
m
id
s
ca
le
f
ac
to
r
,
is
also
1
.
2
.
T
ab
le
1
.
T
h
e
tr
ain
in
g
p
r
o
ce
d
u
r
e
o
f
th
e
f
a
s
ter
R
C
NN
w
i
th
VG
G1
6
m
o
d
el
Ep
o
c
h
I
t
e
r
a
t
i
o
n
T
i
me
El
a
p
se
d
M
i
n
i
-
b
a
t
c
h
L
o
ss
M
i
n
i
-
b
a
t
c
h
A
c
c
u
r
a
c
y
M
i
n
i
-
b
a
t
c
h
R
M
S
E
B
a
se
L
e
a
r
n
i
n
g
R
a
t
e
10
4
5
0
0
0
0
:
2
2
:
5
3
0
.
0
1
3
1
1
0
0
.
0
0
%
0
.
3
7
0
.
0
0
1
0
10
4
5
5
0
0
0
:
2
3
:
0
8
0
.
0
2
1
9
1
0
0
.
0
0
%
0
.
5
4
0
.
0
0
1
0
10
4
6
0
0
0
0
:
2
3
:
2
2
0
.
0
4
6
8
9
9
.
2
2
%
0
.
7
6
0
.
0
0
1
0
10
4
6
5
0
0
0
:
2
3
:
3
6
0
.
0
1
3
8
1
0
0
.
0
0
%
0
.
3
6
0
.
0
0
1
0
10
4
7
0
0
0
0
:
2
3
:
5
0
0
.
0
2
3
3
1
0
0
.
0
0
%
0
.
3
9
0
.
0
0
1
0
10
4
7
5
0
0
0
:
2
4
:
0
4
0
.
0
2
0
0
1
0
0
.
0
0
%
0
.
3
6
0
.
0
0
1
0
10
4
8
0
0
0
0
:
2
4
:
1
8
0
.
0
1
7
2
1
0
0
.
0
0
%
0
.
3
9
0
.
0
0
1
0
10
4
8
5
0
0
0
:
2
4
:
3
2
0
.
0
3
0
5
9
9
.
6
1
%
0
.
4
7
0
.
0
0
1
0
10
4
9
0
0
0
0
:
2
4
:
4
6
0
.
0
2
7
0
1
0
0
.
0
0
%
0
.
4
4
0
.
0
0
1
0
10
4
9
5
0
0
0
:
2
5
:
0
0
0
.
0
1
5
3
1
0
0
.
0
0
%
0
.
4
4
0
.
0
0
1
0
T
ab
le
1
s
h
o
w
s
t
h
e
tr
ain
i
n
g
iter
atio
n
s
o
f
th
e
Fas
ter
R
C
NN.
T
h
e
VGG1
6
m
o
d
el
w
a
s
u
s
ed
in
Fas
ter
R
C
NN
s
tr
u
ct
u
r
e.
T
h
e
co
lu
m
n
s
o
f
t
h
e
T
ab
le
1
s
h
o
w
ep
o
ch
,
iter
atio
n
,
ti
m
e
elap
s
e
d
,
Min
i
-
b
atc
h
lo
s
s
,
Min
i
-
b
atch
ac
cu
r
ac
y
,
Mi
n
i
-
b
a
tch
R
MSE
a
n
d
b
ase
lear
n
in
g
r
ate,
r
esp
ec
tiv
el
y
.
A
s
s
ee
n
i
n
T
ab
le
1
,
th
e
tr
ain
in
g
p
r
o
ce
d
u
r
e
r
ea
ch
ed
th
e
m
a
x
i
m
u
m
ac
c
u
r
ac
y
(
1
0
0
%)
at
1
0
ep
o
ch
an
d
4
9
5
0
iter
atio
n
s
.
T
h
e
m
i
n
i
-
b
atch
R
MSE
v
alu
e
w
a
s
0
.
4
4
.
T
h
e
lear
n
in
g
r
ate
w
a
s
n
o
t
ch
a
n
g
ed
d
u
r
in
g
th
e
iter
atio
n
s
.
T
h
e
o
b
tain
ed
test
r
esu
lt
s
ar
e
g
iv
e
n
i
n
Fig
u
r
es.
6
,
7
,
8
an
d
9
,
r
esp
ec
tiv
el
y
.
W
h
ile
Fig
u
r
e
6
s
h
o
w
s
t
h
e
A
le
x
Net
b
a
s
ed
f
as
t
er
-
R
C
NN
’
s
r
es
u
lt
s
,
Fig
u
r
es
7
,
8
an
d
9
s
h
o
w
t
h
e
VGG1
6
b
ased
f
aster
R
C
NN
’
s
r
es
u
lt
s
,
VGG1
9
b
ased
f
ast
er
R
C
NN
’
s
r
es
u
lts
,
an
d
th
e
f
u
s
ed
r
esu
lt
s
.
T
h
e
d
e
tecte
d
licen
s
e
p
late
r
eg
io
n
s
w
er
e
i
n
d
icate
d
w
it
h
co
lo
r
ed
r
ec
tan
g
le
s
f
o
r
ea
ch
m
et
h
o
d
.
Fig
u
r
e
6
s
h
o
w
s
t
h
e
te
s
t
r
es
u
lt
s
f
o
r
p
r
e
-
tr
ain
ed
Alex
Net
m
o
d
el
b
ased
f
a
s
ter
-
R
C
N
N.
A
s
s
ee
n
i
n
g
iv
e
n
s
a
m
p
le
i
m
ag
e
s
,
th
e
“y
ello
w
”
r
ec
tan
g
le
s
w
er
e
u
s
ed
to
lo
ca
te
th
e
licen
s
e
p
late
r
e
g
io
n
s
.
W
h
ile
all
ca
r
’
s
licen
s
e
p
lates
w
er
e
d
etec
ted
co
r
r
ec
tl
y
,
th
e
licen
s
e
p
lates
f
o
r
tr
ac
k
s
wer
e
n
o
t d
etec
ted
.
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.
19
,
No
.
2
,
A
u
g
u
s
t
20
2
0
:
9
7
4
-
982
980
Fig
u
r
e
6
.
Sh
o
w
s
t
h
e
test
i
n
g
r
esu
lt
s
o
f
t
h
e
A
le
x
Net
m
o
d
el
Fig
u
r
e
7
s
h
o
w
s
th
e
d
etec
tio
n
r
esu
lts
w
it
h
“
r
ed
”
r
ec
tan
g
les
th
at
w
er
e
o
b
tain
ed
w
it
h
VG
G1
6
b
ased
f
aster
R
C
NN.
As
s
ee
n
in
Fi
g
u
r
e
7
,
all
v
e
h
icle
’
s
lice
n
s
e
p
lat
e
r
eg
io
n
s
w
er
e
d
etec
ted
co
r
r
e
ctl
y
b
y
t
h
e
V
GG1
6
b
ased
f
aster
R
C
N
N.
Fi
g
u
r
e
8
also
s
h
o
w
s
t
h
e
d
etec
tio
n
r
es
u
lts
w
it
h
“g
r
ee
n
”
r
ec
ta
n
g
le
s
f
o
r
VGG1
9
b
ased
f
aster
R
C
NN.
As
s
ee
n
i
n
Fig
u
r
e
8
,
ex
ce
p
t
o
n
e
tr
ac
k
,
al
l
o
t
h
er
v
e
h
icle
’
s
lice
n
s
e
p
late
r
eg
io
n
s
w
er
e
d
etec
ted
co
r
r
ec
tly
b
y
t
h
e
VG
G1
9
b
ased
f
aster
R
C
NN.
Fi
n
all
y
,
Fi
g
u
r
e
9
s
h
o
w
s
th
e
f
u
s
ed
r
es
u
lts
f
r
o
m
p
r
ev
io
u
s
f
aster
R
C
N
N
m
o
d
els.
T
h
e
“m
a
g
en
ta
”
co
lo
r
w
a
s
u
s
ed
f
o
r
d
eter
m
in
in
g
th
e
e
x
a
ct
lo
ca
tio
n
o
f
th
e
li
ce
n
s
e
p
lates
f
o
r
all
v
eh
ic
les.
A
s
s
ee
n
i
n
Fi
g
u
r
e
9
,
all
v
eh
icles lice
n
s
e
p
late
lo
ca
ti
o
n
s
w
er
e
d
etec
ted
co
r
r
ec
tly
.
Fig
u
r
e
7
.
Sh
o
w
s
t
h
e
test
i
n
g
r
esu
lt
s
o
f
t
h
e
VGG1
6
m
o
d
el
Fig
u
r
e
8
.
Sh
o
w
s
t
h
e
test
i
n
g
r
esu
lt
s
o
f
t
h
e
VGG1
9
m
o
d
el
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
F
u
s
ed
fa
s
ter R
C
N
N
s
fo
r
ef
fici
en
t d
etec
tio
n
o
f t
h
e
licen
s
e
p
la
tes
(
N
a
a
ma
n
Oma
r
)
981
Fig
u
r
e
9
.
Sh
o
w
s
t
h
e
test
i
n
g
r
esu
lt
s
o
f
t
h
e
f
u
s
ed
m
o
d
el
T
ab
le
2
s
h
o
w
s
t
h
e
o
b
tain
ed
r
esu
lt
w
it
h
i
n
d
i
v
id
u
al
Fas
ter
-
R
C
NN
m
o
d
u
le
s
a
n
d
f
u
s
ed
r
es
u
lt.
A
s
s
ee
n
in
T
ab
le
2
,
th
e
A
le
x
Net
m
o
d
el
p
r
o
d
u
ce
s
7
4
%
ac
cu
r
ac
y
s
co
r
e
w
h
ic
h
is
t
h
e
w
o
r
s
t
a
m
o
n
g
all
r
esu
l
ts
.
T
h
e
s
ec
o
n
d
w
o
r
s
t
r
esu
lt
8
7
%
i
s
o
b
tain
ed
b
y
t
h
e
VGG1
9
m
o
d
el.
VGG1
6
m
o
d
el
p
r
o
d
u
ce
s
9
3
%
ac
cu
r
ac
y
s
co
r
e
w
h
ic
h
is
b
etter
r
esu
lt
th
a
n
A
le
x
Net
a
n
d
VGG1
9
m
o
d
el
s
.
T
h
e
Fu
s
in
g
L
a
y
er
p
r
o
d
u
ce
s
th
e
b
est
r
esu
lt
w
h
ic
h
is
9
7
%.
T
ab
le
2
.
P
er
f
o
r
m
a
n
ce
co
m
p
ar
i
s
o
n
o
f
i
n
d
iv
id
u
al
Fas
ter
-
R
C
N
N
m
o
d
u
les
a
n
d
f
u
s
i
n
g
la
y
er
Pre
-
t
r
a
i
n
e
d
C
N
N
mo
d
e
l
s
A
c
c
u
r
a
c
y
(
%)
A
l
e
x
N
e
t
74
V
G
G
1
6
93
V
G
G
1
9
87
F
u
si
n
g
L
a
y
e
r
97
5.
CO
NCLU
SI
O
N
T
h
is
p
ap
er
p
r
o
p
o
s
es
a
n
o
v
el
m
et
h
o
d
f
o
r
d
etec
tio
n
o
f
th
e
li
ce
n
s
e
p
lates
i
n
a
g
i
v
en
v
e
h
icl
e
i
m
ag
e
s
.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
is
b
ased
o
n
d
ee
p
f
aster
R
C
NN
m
o
d
els
.
R
ec
en
tl
y
,
d
ee
p
lear
n
in
g
h
a
s
b
ec
o
m
e
a
h
o
t
to
p
ic
in
co
m
p
u
ter
v
is
io
n
an
d
i
m
a
g
e
p
r
o
ce
s
s
in
g
co
m
m
u
n
it
ies.
W
e
o
p
ted
to
u
s
e
m
u
ltip
le
f
as
ter
R
C
NN
m
o
d
els
w
h
er
e
p
r
e
-
tr
ain
ed
d
ee
p
C
N
N
m
o
d
e
ls
ar
e
u
s
ed
.
T
h
r
ee
p
r
e
-
tr
ai
n
e
d
C
NN
m
o
d
els
n
a
m
el
y
A
le
x
Net,
V
GG1
6
an
d
VGG1
9
w
er
e
u
s
ed
i
n
f
aster
R
C
NN
m
o
d
u
le
s
.
T
h
e
r
esu
lt
s
f
r
o
m
all
d
ee
p
m
o
d
el
s
w
er
e
t
h
en
f
u
s
ed
f
o
r
o
b
tain
i
n
g
a
f
in
al
lic
e
n
s
e
p
late
r
eg
io
n
.
T
h
e
ex
p
er
i
m
e
n
t
s
o
n
th
e
p
u
b
licl
y
a
v
ailab
le
v
e
h
icle
lice
n
s
e
p
l
ate
d
ataset
s
h
o
w
ed
th
at
p
r
o
p
o
s
ed
f
u
s
ed
r
esu
lts
o
u
tp
er
f
o
r
m
ed
in
d
iv
id
u
al
f
as
ter
R
C
N
N
m
o
d
u
les.
T
h
e
o
b
tain
ed
to
tal
ac
cu
r
ac
y
w
as
9
7
%.
I
n
o
u
r
f
u
tu
r
e
w
o
r
k
s
,
w
e
ar
e
p
lan
n
in
g
to
ap
p
ly
o
t
h
er
d
ee
p
m
o
d
els
i
n
licen
s
e
p
late
d
etec
tio
n
.
I
n
ad
d
itio
n
,
licen
s
e
p
late
r
ec
o
g
n
itio
n
w
ill
b
e
in
o
u
r
f
u
tu
r
e
w
o
r
k
s
.
RE
F
E
R
E
NC
E
S
[1
]
Zee
b
a
re
e
,
D.
Q.,
H
a
ro
n
,
H.,
a
n
d
A
b
d
u
laz
e
e
z
,
A
.
M
.
,
“
Ge
n
e
S
e
lec
t
io
n
a
n
d
Clas
sif
ica
ti
o
n
o
f
M
icro
a
rra
y
D
a
ta
U
sin
g
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
tw
o
rk
,
”
p
re
se
n
ted
a
t
th
e
In
tern
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
A
d
v
a
n
c
e
d
S
c
ien
c
e
a
n
d
E
n
g
in
e
e
rin
g
(ICOA
S
E)
(p
p
.
1
4
5
-
1
5
0
).
I
EE
E.
,
2
0
1
8
.
[2
]
O.
A
h
m
e
d
a
n
d
A
.
Bri
f
c
a
n
i,
“
G
e
n
e
Ex
p
re
ss
io
n
C
las
si
f
ica
ti
o
n
Ba
se
d
o
n
De
e
p
L
e
a
rn
in
g
,
”
p
re
se
n
ted
a
t
th
e
2
0
1
9
4
t
h
S
c
ien
ti
f
ic In
tern
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
Na
ja
f
(S
ICN),
2
0
1
9
,
p
p
.
1
4
5
–
1
4
9
.
[3
]
M
.
S
.
A
l
-
S
h
e
m
a
rr
y
,
Y.
L
i,
a
n
d
S
.
A
b
d
u
ll
a
,
“
A
n
e
ff
ici
e
n
t
tex
tu
re
d
e
sc
rip
to
r
f
o
r
th
e
d
e
tec
ti
o
n
o
f
li
c
e
n
se
p
late
s
f
ro
m
ve
h
icle
ima
g
e
s in
d
if
f
icu
lt
c
o
n
d
i
ti
o
n
s,”
IEE
E
T
ra
n
s.
I
n
tell.
T
ra
n
sp
.
S
y
st.
,
2
0
1
9
.
[4
]
Y.
Yu
a
n
,
W
.
Z
o
u
,
Y.
Z
h
a
o
,
X
.
W
a
n
g
,
X
.
Hu
,
a
n
d
N
.
Ko
m
o
d
a
k
is,
“
A
ro
b
u
st
a
n
d
e
f
f
ici
e
n
t
a
p
p
r
o
a
c
h
to
li
c
e
n
se
p
late
d
e
tec
ti
o
n
,
”
I
EE
E
T
r
a
n
s.
Ima
g
e
Pro
c
e
ss
.
,
v
o
l.
2
6
,
n
o
.
3
,
p
p
.
1
1
02
–
1
1
1
4
,
2
0
1
6
.
[5
]
B.
V
.
Ka
k
a
n
i,
D.
G
a
n
d
h
i,
a
n
d
S
.
Ja
n
i,
“
Im
p
ro
v
e
d
OCR
b
a
se
d
a
u
to
m
a
ti
c
v
e
h
icle
n
u
m
b
e
r
p
late
re
c
o
g
n
it
io
n
u
sin
g
f
e
a
tu
re
s
train
e
d
n
e
u
ra
l
n
e
tw
o
rk
,
”
p
re
se
n
ted
a
t
t
h
e
2
0
1
7
8
th
I
n
tern
a
ti
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Co
m
p
u
ti
n
g
,
Co
m
m
u
n
ica
ti
o
n
a
n
d
Ne
tw
o
rk
in
g
T
e
c
h
n
o
lo
g
ies
(ICCCNT
),
2
0
1
7
,
p
p
.
1
–
6.
[6
]
B.
S
in
g
h
,
M
.
Ka
u
r,
D.
S
i
n
g
h
,
a
n
d
G
.
S
i
n
g
h
,
“
A
u
to
m
a
ti
c
n
u
m
b
e
r
p
late
re
c
o
g
n
it
i
o
n
sy
ste
m
b
y
c
h
a
ra
c
ter
p
o
siti
o
n
m
e
th
o
d
,
”
I
n
t
J
C
o
mp
u
t
Vi
s.
Ro
b
o
t
,
v
o
l.
6
,
n
o
.
1
/
2
,
p
p
.
9
4
–
1
1
2
,
2
0
1
6
.
[7
]
H.
L
i
a
n
d
C.
S
h
e
n
,
“
Re
a
d
i
n
g
c
a
r
li
c
e
n
se
p
late
s
u
sin
g
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s
a
n
d
lstm
s,
”
ArXi
v
Pre
p
r.
ArXi
v
1
6
0
1
0
5
6
1
0
,
2
0
1
6
.
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.
19
,
No
.
2
,
A
u
g
u
s
t
20
2
0
:
9
7
4
-
982
982
[8
]
N.
S
u
laim
a
n
,
S
.
N.
H.
M
.
Ja
lan
i,
M
.
M
u
sta
f
a
,
a
n
d
K.
Ha
w
a
ri,
“
D
e
v
e
lo
p
m
e
n
t
o
f
a
u
to
m
a
ti
c
v
e
h
icle
p
late
d
e
tec
ti
o
n
s
y
ste
m
,
”
p
re
se
n
ted
a
t
th
e
2
0
1
3
I
EE
E
3
r
d
In
tern
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
S
y
ste
m
En
g
in
e
e
rin
g
a
n
d
T
e
c
h
n
o
l
o
g
y
,
2
0
1
3
,
p
p
.
1
3
0
–
1
3
5
.
[9
]
W
.
W
ih
a
rto
,
H.
Ku
sn
a
n
to
,
a
n
d
H.
He
rian
to
,
“
S
y
ste
m
Dia
g
n
o
sis
o
f
Co
ro
n
a
ry
He
a
rt
Dis
e
a
se
Us
in
g
a
Co
m
b
in
a
ti
o
n
o
f
Dim
e
n
sio
n
a
l
Re
d
u
c
ti
o
n
a
n
d
Da
ta
M
in
i
n
g
T
e
c
h
n
iq
u
e
s:
A
Re
v
i
e
w
,
”
In
d
o
n
e
s.
J
.
El
e
c
tr.
En
g
.
C
o
mp
u
t.
S
c
i.
,
v
o
l.
7
,
n
o
.
2
,
p
p
.
5
1
4
–
5
2
3
,
2
0
1
7
.
[1
0
]
S
.
A
z
a
m
a
n
d
M
.
M
.
Isla
m
,
“
A
u
to
m
a
ti
c
li
c
e
n
se
p
late
d
e
tec
ti
o
n
i
n
h
a
z
a
rd
o
u
s
c
o
n
d
it
i
o
n
,
”
J
.
Vi
s.
C
o
mm
u
n
.
Ima
g
e
Rep
re
se
n
t.
,
v
o
l.
3
6
,
p
p
.
1
7
2
–
1
8
6
,
2
0
1
6
.
[1
1
]
A
.
N
a
i
m
i,
Y.
Ke
s
se
n
ti
n
i,
a
n
d
M
.
Ha
m
m
a
m
i,
“
M
u
lt
i
-
n
a
ti
o
n
a
n
d
m
u
lt
i
-
n
o
rm
li
c
e
n
se
p
late
s
d
e
tec
ti
o
n
in
re
a
l
tra
f
f
i
c
su
rv
e
il
lan
c
e
e
n
v
iro
n
m
e
n
t
u
sin
g
d
e
e
p
lea
rn
in
g
,
”
p
re
se
n
ted
a
t
th
e
In
tern
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
Ne
u
ra
l
In
f
o
r
m
a
ti
o
n
P
r
o
c
e
ss
in
g
,
2
0
1
6
,
p
p
.
4
6
2
–
4
6
9
.
[1
2
]
T
.
A
ja
n
t
h
a
n
,
P
.
K
a
m
a
l
a
r
u
b
a
n
,
a
n
d
R
.
R
o
d
r
i
g
o
,
“
A
u
t
o
m
a
t
i
c
n
u
m
b
e
r
p
l
a
t
e
r
e
c
o
g
n
i
t
i
o
n
i
n
l
o
w
q
u
a
l
i
ty
v
id
e
o
s
,
”
p
r
e
se
n
t
e
d
a
t
t
h
e
2
0
1
3
I
E
E
E
8
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
d
u
s
t
r
i
a
l
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
y
s
tem
s
,
2
0
1
3
,
p
p
.
5
6
6
–
5
7
1
.
[1
3
]
Y.
Zh
a
o
,
J.
G
u
,
C.
L
iu
,
S
.
Ha
n
,
Y.
G
a
o
,
a
n
d
Q.
Hu
,
“
L
ice
n
se
p
late
lo
c
a
ti
o
n
b
a
se
d
o
n
h
a
a
r
-
li
k
e
c
a
s
c
a
d
e
c
las
si
f
iers
a
n
d
e
d
g
e
s,” p
re
se
n
ted
a
t
th
e
2
0
1
0
S
e
c
o
n
d
W
RI
G
lo
b
a
l
Co
n
g
re
ss
o
n
In
telli
g
e
n
t
S
y
ste
m
s,
2
0
1
0
,
v
o
l
.
3
,
p
p
.
1
0
2
–
1
0
5
.
[1
4
]
D.
Zan
g
,
Z.
Ch
a
i,
J.
Zh
a
n
g
,
D.
Zh
a
n
g
,
a
n
d
J.
Ch
e
n
g
,
“
V
e
h
icle
li
c
e
n
se
p
late
re
c
o
g
n
it
io
n
u
sin
g
v
isu
a
l
a
tt
e
n
ti
o
n
m
o
d
e
l
a
n
d
d
e
e
p
lea
rn
in
g
,
”
J
.
E
lec
tro
n
.
I
ma
g
in
g
,
v
o
l.
2
4
,
n
o
.
3
,
p
.
0
3
3
0
0
1
,
2
0
1
5
.
[1
5
]
P
.
P
ra
b
h
a
k
a
r,
P
.
A
n
u
p
a
m
a
,
a
n
d
S
.
Re
s
m
i,
“
A
u
to
m
a
ti
c
v
e
h
icle
n
u
m
b
e
r
p
late
d
e
tec
ti
o
n
a
n
d
re
c
o
g
n
it
io
n
,
”
p
re
se
n
te
d
a
t
th
e
2
0
1
4
I
n
tern
a
ti
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Co
n
tro
l
,
I
n
stru
m
e
n
tatio
n
,
C
o
m
m
u
n
ica
ti
o
n
a
n
d
C
o
m
p
u
tatio
n
a
l
T
e
c
h
n
o
lo
g
ies
(ICCICCT
),
2
0
1
4
,
p
p
.
1
8
5
–
1
9
0
.
[1
6
]
S
.
Kim
,
H.
Je
o
n
,
a
n
d
H.
Ko
o
,
“
De
e
p
-
lea
rn
in
g
-
b
a
se
d
li
c
e
n
se
p
late
d
e
tec
ti
o
n
m
e
th
o
d
u
sin
g
v
e
h
icle
re
g
io
n
e
x
trac
ti
o
n
,
”
El
e
c
tro
n
.
L
e
tt
.
,
v
o
l.
5
3
,
n
o
.
1
5
,
p
p
.
1
0
3
4
–
1
0
3
6
,
2
0
1
7
.
[1
7
]
N.
O.
Ya
se
e
n
,
S
.
G
.
S
.
A
l
-
A
li
,
a
n
d
A
.
S
e
n
g
u
r,
“
A
n
Ef
f
icie
n
t
M
o
d
e
l
f
o
r
A
u
to
m
a
ti
c
Nu
m
b
e
r
P
late
De
tec
ti
o
n
u
si
n
g
HO
G
F
e
a
tu
re
f
ro
m
Ne
w
No
rth
Ira
q
V
e
h
icle
Im
a
g
e
s
D
a
tas
e
t,
”
p
re
se
n
ted
a
t
th
e
2
0
1
9
1
st
In
tern
a
ti
o
n
a
l
In
f
o
rm
a
ti
c
s
a
n
d
S
o
f
twa
re
En
g
in
e
e
rin
g
Co
n
f
e
re
n
c
e
(UBMYK),
2
0
1
9
,
p
p
.
1
–
6.
[1
8
]
N.
O.
Ya
s
e
e
n
,
S
.
G
.
S
.
A
l
-
A
li
,
a
n
d
A
.
S
e
n
g
u
r,
“
De
v
e
lo
p
m
e
n
t
o
f
Ne
w
A
n
p
r
Da
tas
e
t
f
o
r
A
u
to
m
a
ti
c
Nu
m
b
e
r
P
late
De
tec
ti
o
n
a
n
d
Re
c
o
g
n
it
io
n
in
No
rth
o
f
Ira
q
,
”
p
re
se
n
ted
a
t
th
e
2
0
1
9
1
st
In
tern
a
ti
o
n
a
l
In
f
o
rm
a
ti
c
s
a
n
d
S
o
f
twa
re
En
g
in
e
e
rin
g
Co
n
f
e
re
n
c
e
(UBMY
K),
2
0
1
9
,
p
p
.
1
–
6.
[1
9
]
N.
Om
a
r,
A
.
S
e
n
g
u
r,
a
n
d
S
.
G
.
S
.
A
l
-
A
li
,
“
Ca
sc
a
d
e
d
De
e
p
Lea
r
n
in
g
-
Ba
se
d
Ef
f
ici
e
n
t
A
p
p
ro
a
c
h
f
o
r
L
ice
n
se
P
late
De
tec
ti
o
n
a
n
d
Re
c
o
g
n
it
io
n
,
”
Ex
p
e
rt S
y
st.
A
p
p
l.
,
p
.
1
1
3
2
8
0
,
2
0
2
0
.
[2
0
]
W
.
T
.
H
o
,
H
.
W
.
L
im
,
a
n
d
Y
.
H
.
T
a
y
,
“
T
w
o
-
s
t
a
g
e
l
ic
e
n
se
p
l
a
t
e
d
e
t
e
c
t
i
o
n
u
s
i
n
g
g
e
n
t
l
e
A
d
a
b
o
o
s
t
a
n
d
S
I
F
T
-
S
VM
,
”
p
r
e
s
e
n
t
e
d
a
t
t
h
e
2
0
0
9
F
i
r
s
t
A
s
i
a
n
C
o
n
f
e
re
n
c
e
o
n
I
n
t
e
l
l
i
g
e
n
t
I
n
f
o
rm
a
ti
o
n
a
n
d
D
a
t
a
b
a
s
e
S
y
s
t
e
m
s
,
2
0
0
9
,
p
p
.
1
0
9
–
1
1
4
.
[2
1
]
S.
Kta
ta,
F
.
Be
n
z
a
rti
,
a
n
d
H.
Am
iri
,
“
L
ice
n
se
p
late
lo
c
a
li
z
a
ti
o
n
u
sin
g
G
a
b
o
r
f
il
ters
a
n
d
n
e
u
r
a
l
n
e
tw
o
rk
s,”
J
.
Co
mp
u
t
.
S
c
i
.
,
v
o
l.
9
,
n
o
.
1
0
,
p
.
1
3
4
1
,
2
0
1
3
.
[2
2
]
S
.
Z.
M
a
so
o
d
,
G
.
S
h
u
,
A
.
De
h
g
h
a
n
,
a
n
d
E.
G
.
Ortiz,
“
L
ice
n
se
p
late
d
e
tec
ti
o
n
a
n
d
re
c
o
g
n
it
i
o
n
u
sin
g
d
e
e
p
ly
lea
rn
e
d
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s,”
ArXi
v
Pre
p
r.
ArXi
v
1
7
0
3
0
7
3
3
0
,
2
0
1
7
.
[2
3
]
M
.
A
.
L
a
li
m
i,
S
.
G
h
o
f
r
a
n
i,
a
n
d
D.
M
c
L
e
rn
o
n
,
“
A
v
e
h
icle
li
c
e
n
s
e
p
late
d
e
tec
ti
o
n
m
e
th
o
d
u
sin
g
r
e
g
io
n
a
n
d
e
d
g
e
b
a
se
d
m
e
th
o
d
s,”
Co
mp
u
t.
El
e
c
tr.
En
g
.
,
v
o
l.
3
9
,
n
o
.
3
,
p
p
.
8
3
4
–
8
4
5
,
2
0
1
3
.
[2
4
]
S
.
Re
n
,
K.
He
,
R.
G
irsh
ic
k
,
a
n
d
J.
S
u
n
,
“
F
a
ste
r
r
-
c
n
n
:
T
o
w
a
rd
s
re
a
l
-
ti
m
e
o
b
jec
t
d
e
tec
ti
o
n
w
it
h
re
g
io
n
p
ro
p
o
sa
l
n
e
tw
o
rk
s,” p
re
se
n
ted
a
t
th
e
A
d
v
a
n
c
e
s in
n
e
u
ra
l
in
f
o
rm
a
ti
o
n
p
r
o
c
e
ss
in
g
sy
ste
m
s,
2
0
1
5
,
p
p
.
9
1
–
9
9
.
[2
5
]
J
.
E
.
E
s
p
i
n
o
s
a
,
S
.
A
.
Ve
l
a
s
t
i
n
,
a
n
d
J
.
W
.
B
ra
n
c
h
,
“
Ve
h
i
c
l
e
d
e
t
e
c
t
i
o
n
u
s
i
n
g
a
le
x
n
e
t
a
n
d
f
a
s
te
r
R
-
C
NN
d
e
e
p
le
a
r
n
i
n
g
m
o
d
e
ls
:
a
c
o
m
p
a
r
a
t
iv
e
s
t
u
d
y
,
”
p
r
e
s
e
n
t
e
d
a
t
t
h
e
I
n
t
e
r
n
a
t
i
o
n
a
l
Vi
s
u
a
l
I
n
f
o
rm
a
t
i
c
s
C
o
n
f
e
r
e
n
c
e
,
2
0
1
7
,
p
p
.
3
–
15.
[2
6
]
R.
G
irsh
ick
,
“
F
a
st
r
-
c
n
n
,
”
p
re
se
n
ted
a
t
t
h
e
P
ro
c
e
e
d
i
n
g
s
o
f
th
e
IE
EE
in
ter
n
a
ti
o
n
a
l
c
o
n
f
e
re
n
c
e
o
n
c
o
m
p
u
ter
v
isio
n
,
2
0
1
5
,
p
p
.
1
4
4
0
–
1
4
4
8
.
[2
7
]
H.
Qa
ss
i
m
,
A
.
V
e
r
m
a
,
a
n
d
D.
F
e
in
z
im
e
r,
“
Co
m
p
re
ss
e
d
re
sid
u
a
l
-
V
GG
1
6
CNN
m
o
d
e
l
f
o
r
b
ig
d
a
t
a
p
lac
e
s
i
m
a
g
e
re
c
o
g
n
it
io
n
,
”
p
re
se
n
ted
a
t
t
h
e
2
0
1
8
I
EE
E
8
th
A
n
n
u
a
l
C
o
m
p
u
ti
n
g
a
n
d
C
o
m
m
u
n
ica
ti
o
n
W
o
rk
sh
o
p
a
n
d
C
o
n
f
e
re
n
c
e
(CCW
C),
2
0
1
8
,
p
p
.
1
6
9
–
1
7
5
.
[2
8
]
D.
Q.
Zee
b
a
re
e
,
H.
Ha
ro
n
,
A
.
M
.
A
b
d
u
laz
e
e
z
,
a
n
d
D.
A
.
Zeb
a
ri,
“
M
a
c
h
in
e
lea
rn
i
n
g
a
n
d
Re
g
io
n
G
ro
w
in
g
f
o
r
Bre
a
st
Ca
n
c
e
r
S
e
g
m
e
n
tatio
n
,
”
p
re
se
n
ted
a
t
th
e
2
0
1
9
In
tern
a
t
io
n
a
l
Co
n
f
e
r
e
n
c
e
o
n
A
d
v
a
n
c
e
d
S
c
ien
c
e
a
n
d
E
n
g
in
e
e
rin
g
(ICOA
S
E),
2
0
1
9
,
p
p
.
8
8
–
93.
[
29
]
L
P
D
a
t
a
b
a
s
e
.
A
c
c
e
s
s
e
d
:
J
u
l
.
2
0
1
6
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
:
/
/
w
w
w
.
z
e
m
r
i
s
.
f
e
r
.
h
r
/
p
r
o
j
e
c
t
s
/
L
i
c
e
n
s
e
P
l
a
t
e
s
/
e
n
g
l
i
s
h
/
b
a
z
a
_
s
l
i
k
a
.
z
i
p
.
Evaluation Warning : The document was created with Spire.PDF for Python.