I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
p
ute
r
E
ng
in
ee
ring
(
I
J
E
CE
)
Vo
l.
11
,
No
.
6
,
Dec
em
b
er
2
0
2
1
,
p
p
.
5
2
6
6
~
5
2
7
6
I
SS
N:
2
0
8
8
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v
1
1
i
6
.
pp
5
2
6
6
-
5
2
7
6
5266
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
M
ulticlas
sifica
tio
n of lice
nse pla
te
ba
sed o
n
d
eep
c
o
nv
o
lution
n
eura
l
n
etw
o
rk
s
M
a
s
a
r
Abed U
t
ha
ib
1
,
M
ua
y
a
d Sa
di
k
Cro
o
ck
2
1
Ira
q
i
Co
m
m
issio
n
f
o
r
Co
m
p
u
ters
a
n
d
I
n
f
o
rm
a
ti
c
s (ICCI),
In
f
o
rm
a
t
ics
In
stit
u
te f
o
r
P
o
stg
ra
d
u
a
te S
t
u
d
ies
,
Ba
g
h
d
a
d
,
Ira
q
2
Co
n
tr
o
l
a
n
d
S
y
ste
m
s
En
g
in
e
e
rin
g
De
p
a
rt
m
e
n
t,
Un
iv
e
rsity
o
f
Tec
h
n
o
l
o
g
y
,
Ba
g
h
d
a
d
,
Ira
q
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ju
l
11
,
2
0
20
R
ev
i
s
ed
Ma
y
1
2
,
2
0
2
1
A
cc
ep
ted
Ju
n
6
,
2
0
2
1
In
t
h
e
c
las
sif
ica
ti
o
n
o
f
li
c
e
n
se
p
l
a
te
th
e
re
a
re
so
m
e
c
h
a
ll
e
n
g
e
s
su
c
h
th
a
t
th
e
d
if
fe
re
n
t
siz
e
s
o
f
p
late
n
u
m
b
e
rs,
th
e
p
late
s'
b
a
c
k
g
ro
u
n
d
,
a
n
d
t
h
e
n
u
m
b
e
r
o
f
th
e
d
a
tas
e
t
o
f
th
e
p
late
s.
In
th
i
s
p
a
p
e
r,
a
m
u
lt
icla
ss
c
l
a
ss
i
f
ica
ti
o
n
m
o
d
e
l
e
sta
b
li
sh
e
d
u
si
n
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
(
CNN
)
to
c
las
sify
th
e
li
c
e
n
se
p
late
f
o
r
th
re
e
c
o
u
n
tri
e
s (A
r
m
e
n
ia,
Be
laru
s,
Hu
n
g
a
r
y
)
w
it
h
th
e
d
a
tas
e
t
o
f
6
0
0
im
a
g
e
s
a
s
2
0
0
im
a
g
e
s
fo
r
e
a
c
h
c
las
s
(1
6
0
f
o
r
tr
a
i
n
in
g
a
n
d
4
0
f
o
r
v
a
li
d
a
ti
o
n
se
ts
).
Be
c
a
u
se
o
f
th
e
sm
a
ll
n
u
m
b
e
rs o
f
d
a
tas
e
ts,
a
p
re
p
ro
c
e
ss
in
g
o
n
th
e
d
a
tas
e
t
is
p
e
rf
o
r
m
e
d
u
sin
g
p
ix
e
l
n
o
rm
a
li
z
a
ti
o
n
a
n
d
i
m
a
g
e
d
a
ta
a
u
g
m
e
n
tatio
n
tec
h
n
iq
u
e
s
(ro
tatio
n
,
h
o
riz
o
n
tal
f
li
p
,
z
o
o
m
ra
n
g
e
)
to
i
n
c
re
a
se
th
e
n
u
m
b
e
r
o
f
d
a
tas
e
ts.
Af
ter
th
a
t,
w
e
f
e
e
d
th
e
a
u
g
m
e
n
ted
im
a
g
e
s
in
to
th
e
c
o
n
v
o
lu
ti
o
n
lay
e
r
m
o
d
e
l
,
w
h
ich
c
o
n
sists
o
f
f
o
u
r
b
lo
c
k
s
o
f
c
o
n
v
o
lu
ti
o
n
lay
e
r.
F
o
r
c
a
lcu
latin
g
a
n
d
o
p
ti
m
izin
g
th
e
e
f
f
i
c
ien
c
y
o
f
th
e
c
las
si
f
ic
a
ti
o
n
m
o
d
e
l,
a
c
a
teg
o
rica
l
c
ro
ss
-
e
n
tro
p
y
a
n
d
A
d
a
m
o
p
ti
m
i
z
e
r
u
se
d
w
it
h
a
lea
rn
in
g
ra
te
wa
s
0.
0
0
0
1
.
T
h
e
m
o
d
e
l'
s p
e
r
f
o
r
m
a
n
c
e
sh
o
w
e
d
9
9
.
1
7
%
a
n
d
9
7
.
5
0
%
o
f
t
h
e
train
in
g
a
n
d
v
a
li
d
a
ti
o
n
se
ts
a
c
c
u
ra
c
ie
s
se
q
u
e
n
ti
a
ll
y
,
w
it
h
to
tal
a
c
c
u
ra
c
y
o
f
c
las
si
f
ica
ti
o
n
is
9
6
.
6
6
%
.
T
h
e
ti
m
e
o
f
train
in
g
is
las
ti
n
g
f
o
r
1
2
m
i
n
u
tes
.
A
n
a
n
a
c
o
n
d
a
p
y
th
o
n
3
.
7
a
n
d
Ke
ra
s
T
e
n
so
r
f
lo
w
b
a
c
k
e
n
d
a
re
u
se
d
.
K
ey
w
o
r
d
s
:
A
d
a
m
o
p
ti
m
izer
C
o
n
v
o
lu
tio
n
n
e
u
r
al
n
et
w
o
r
k
s
Data
au
g
m
e
n
tatio
n
Dr
o
p
o
u
t
L
ice
n
s
e
p
late
clas
s
i
f
icatio
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Ma
s
ar
A
b
ed
Uth
a
ib
Dep
ar
t
m
en
t o
f
C
o
m
p
u
ter
S
cie
n
ce
I
r
aq
i Co
m
m
i
s
s
io
n
f
o
r
C
o
m
p
u
t
er
s
an
d
I
n
f
o
r
m
atics (
I
C
C
I
)
,
I
n
f
o
r
m
atic
s
I
n
s
tit
u
te
f
o
r
P
o
s
tg
r
a
d
u
ate
Stu
d
ie
s
B
ag
h
d
ad
,
I
r
aq
E
m
ail: M
a
s
ar
.
u
t
h
aib
2
0
1
8
@
g
m
ail.
co
m
Mu
a
y
ad
Sad
i
k
C
r
o
o
ck
C
o
n
tr
o
l a
n
d
S
y
s
te
m
s
E
n
g
in
ee
r
in
g
Dep
ar
t
m
e
n
t
Un
i
v
er
s
it
y
o
f
T
ec
h
n
o
o
g
y
-
I
r
aq
B
ag
h
d
ad
,
ir
aq
E
m
ail:
m
u
a
y
ad
.
s
.
cr
o
o
ck
@
u
o
t
ec
h
n
o
lo
g
y
.
ed
u
.
iq
1.
I
NT
RO
D
UCT
I
O
N
I
n
all
m
a
n
u
f
ac
t
u
r
in
g
ar
ea
s
an
d
in
o
u
r
ev
er
y
d
a
y
li
f
e,
v
e
h
icl
es
ar
e
co
m
m
o
n
l
y
u
s
ed
.
An
e
f
f
ec
ti
v
e
w
a
y
o
f
d
is
ti
n
g
u
is
h
i
n
g
v
e
h
icle
s
an
d
m
ak
e
t
h
e
m
u
n
iq
u
e
b
y
t
h
e
lice
n
s
e
p
late
(
L
P
)
.
W
ith
th
e
f
a
s
t
-
g
r
o
w
i
n
g
n
u
m
b
er
o
f
ca
r
s
,
tr
af
f
ic
v
io
latio
n
s
o
cc
u
r
m
o
r
e
o
f
te
n
in
p
u
b
lic
tr
an
s
p
o
r
tatio
n
,
s
u
c
h
as
h
ig
h
w
a
y
o
r
p
ar
k
in
g
f
r
au
d
to
ll
s
,
s
p
ee
d
in
g
a
n
d
ca
r
th
e
f
t
.
T
h
er
ef
o
r
e
th
e
v
eh
icle
L
P
s
n
ee
d
t
o
b
e
id
en
tif
ied
f
o
r
p
r
o
tectio
n
.
T
h
e
in
f
o
r
m
atio
n
d
er
iv
ed
f
r
o
m
a
n
L
P
ca
n
b
e
u
s
ed
f
o
r
v
ar
io
u
s
p
u
r
p
o
s
es,
lik
e
lo
o
k
in
g
f
o
r
th
e
v
eh
ic
les
th
a
t
m
a
y
b
e
s
to
len
o
r
m
i
g
h
t
b
e
f
i
g
h
tin
g
cr
i
m
e,
m
o
n
i
to
r
in
g
cr
o
s
s
in
g
b
o
r
d
er
s
,
o
r
ac
ce
s
s
i
n
g
th
e
h
i
g
h
w
a
y
to
ll
s
tati
o
n
.
T
h
u
s
,
t
h
er
e
is
a
n
ee
d
f
o
r
a
licen
s
e
p
late
id
en
tific
atio
n
s
y
s
te
m
.
R
ec
e
n
tl
y
,
d
ee
p
lear
n
in
g
h
as
s
h
o
w
n
a
n
ex
ce
ll
en
t
p
er
f
o
r
m
an
ce
i
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
Mu
lticla
s
s
i
fica
tio
n
o
f lice
n
s
e
p
la
te
b
a
s
ed
o
n
d
ee
p
co
n
vo
lu
ti
o
n
n
eu
r
a
l
n
etw
o
r
ks (
Ma
s
a
r
A
b
ed
Uth
a
ib
)
5267
m
o
s
t
co
m
p
lex
tas
k
s
s
u
ch
as
m
ed
ical
i
m
a
g
i
n
g
an
d
c
y
b
er
s
ec
u
r
it
y
[1
]
-
[
4
]
.
C
o
n
v
o
l
u
tio
n
al
n
eu
r
al
n
et
w
o
r
k
s
(
C
NNs)
ar
e
o
n
e
o
f
t
h
e
d
ee
p
lear
n
in
g
al
g
o
r
ith
m
s
,
w
h
ic
h
i
s
wh
y
d
ee
p
lear
n
in
g
i
s
p
o
p
u
lar
[
5
]
,
[
6
]
.
T
h
e
lay
er
s
o
f
C
NNs la
y
er
s
p
r
i
m
ar
i
l
y
t
h
r
ee
l
a
y
er
s
[
7
]
:
a.
C
o
n
v
o
lu
tio
n
al
-
la
y
er
b.
P
o
o
lin
g
-
la
y
er
c.
Fu
ll
y
-
co
n
n
ec
ted
la
y
er
T
h
e
ch
allen
g
es
w
e
f
ac
ed
th
r
o
u
g
h
t
h
e
class
if
icatio
n
o
f
licen
s
e
p
lates
:
i
)
Dif
f
er
en
t
s
izes
o
f
p
lates
,
i
i
)
S
m
all
d
atase
ts
t
h
at
m
a
k
e
o
v
er
f
it
tin
g
,
i
ii
)
T
h
e
ti
m
e
f
o
r
th
e
p
r
o
ce
s
s
in
g
C
NN
's
m
o
d
el
d
u
r
i
n
g
tr
ai
n
i
n
g
.
On
t
h
e
o
t
h
er
h
a
n
d
,
s
e
v
er
al
s
t
u
d
ies
h
a
v
e
d
is
c
u
s
s
ed
lice
n
s
e
p
l
ate
class
i
f
ica
tio
n
i
n
r
ec
e
n
t
y
ea
r
s
.
J
o
s
e
et
a
l.
[
8
]
s
u
g
g
ested
an
al
g
o
r
ith
m
f
o
r
th
e
class
i
f
icatio
n
o
f
Viet
n
a
m
ese
m
u
lti
-
s
ta
n
d
ar
d
licen
s
i
n
g
p
lates
b
ased
o
n
a
co
n
v
o
lu
tio
n
n
eu
r
al
n
et
w
o
r
k
,
i
n
w
h
ich
tr
a
n
s
f
er
lear
n
in
g
(
R
e
s
id
u
al
Net)
w
a
s
u
s
ed
.
T
h
e
f
in
al
la
y
er
is
d
elete
d
an
d
ch
a
n
g
ed
w
i
th
a
n
e
w
la
y
er
o
f
class
i
f
icat
io
n
.
C
a
teg
o
r
iz
atio
n
o
f
t
h
e
m
o
d
el
in
to
t
h
r
ee
g
r
o
u
p
s
(
t
h
e
R
izal
m
o
n
u
m
en
t
s
er
ies
s
t
ick
er
f
o
r
n
e
w
v
e
h
icle
s
,
R
izal
m
o
n
u
m
en
t
s
er
ies,
2
0
1
4
s
er
ies).
C
r
o
s
s
-
e
n
tr
o
p
y
as
an
o
p
tim
izat
io
n
m
e
th
o
d
as
w
ell
a
s
0
.
0
0
1
r
ep
r
esen
ted
th
e
lear
n
i
n
g
r
ate.
T
h
e
v
a
lid
atio
n
ac
c
u
r
a
c
y
w
a
s
8
2
.
6
1
%.
I
n
[
9
]
,
a
HAAR
Feat
u
r
e
-
b
ased
C
las
s
i
f
ier
w
a
s
u
s
ed
to
d
et
ec
t
an
d
s
e
g
m
e
n
t
t
h
e
p
lates
in
to
c
h
ar
ac
ter
s
f
o
r
r
ec
o
g
n
izi
n
g
it
b
y
C
NN.
T
h
eir
tr
ain
i
n
g
a
n
d
v
alid
atio
n
ac
c
u
r
ac
ies
w
er
e
9
3
.
5
4
%
an
d
9
1
.
3
8
%,
r
esp
ec
tiv
el
y
.
T
h
e
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
w
a
s
9
0
.
9
0
%.
W
an
g
et
a
l.
[
10
]
d
o
cu
m
e
n
ted
a
w
a
y
o
f
id
e
n
ti
f
y
i
n
g
p
lates
'
ch
ar
ac
ter
s
b
y
u
s
i
n
g
a
te
m
p
lat
e
m
atc
h
in
g
tech
n
iq
u
e
an
d
ar
tif
icia
l
n
eu
r
al
n
et
w
o
r
k
s
.
I
n
th
e
b
eg
i
n
n
in
g
,
th
e
s
ec
o
n
d
ar
y
p
o
s
itio
n
i
n
g
tech
n
iq
u
e
w
as
u
s
ed
f
o
r
p
late
lo
ca
lizatio
n
.
T
h
e
p
r
ec
is
e
p
o
s
itio
n
o
f
th
e
p
late
d
ep
en
d
ed
o
n
th
e
v
er
tical
ed
g
e
a
n
d
HS
V
co
lo
r
o
f
th
e
p
late.
T
h
eir
p
r
ec
is
io
n
s
ap
p
r
o
x
i
m
atio
n
f
o
r
t
h
e
lo
ca
lizatio
n
an
d
r
ec
o
g
n
itio
n
f
o
r
th
e
n
u
m
b
er
o
f
t
h
e
p
late
w
er
e
7
5
.8
%
an
d
7
2
.
5
% seq
u
en
tiall
y
.
I
n
ar
ti
f
icial
n
eu
r
al
n
et
w
o
r
k
(
A
NN)
,
t
h
e
ac
cu
r
ac
y
o
f
r
ec
o
g
n
itio
n
w
a
s
7
5
%.
A
K
-
n
ea
r
e
s
t
n
ei
g
h
b
o
u
r
clas
s
i
f
icatio
n
w
a
s
e
m
p
lo
y
ed
f
o
r
class
i
f
y
in
g
ch
ar
ac
ter
s
i
n
th
e
lic
en
s
e
p
late
[
11
]
.
I
n
itiall
y
,
Ot
s
u
m
et
h
o
d
to
ex
tr
ac
t
p
late
i
m
a
g
e
w
as
u
s
ed
an
d
th
e
n
co
n
v
er
t
in
g
to
a
b
in
a
r
y
i
m
a
g
e.
T
h
e
test
d
ataset
w
a
s
1
0
0
im
ag
e
s
.
T
h
ey
r
ea
ch
ed
9
3
,
7
5
%,
w
h
ich
is
t
h
e
ac
cu
r
ac
y
o
f
id
e
n
ti
f
y
in
g
n
u
m
b
er
s
,
w
h
ile
9
1
.
9
2
%
ac
cu
r
ac
y
o
f
le
tter
r
ec
o
g
n
it
io
n
.
W
an
g
et
a
l.
[
1
2
]
d
esig
n
e
d
a
s
y
s
te
m
f
o
r
d
etec
tio
n
an
d
r
ec
o
g
n
izin
g
p
late
n
u
m
b
er
s
f
o
r
t
h
e
I
n
d
ian
lice
n
s
e
p
late,
w
h
er
e
t
h
e
y
u
s
ed
(
Y
o
u
O
n
l
y
L
o
o
k
O
n
ce
)
Yo
lo
v
.
3
f
o
r
tr
ain
i
n
g
t
h
e
d
ataset
th
at
co
n
s
is
t
s
o
f
3
7
cl
ass
o
f
ch
ar
ac
ter
s
i
m
ag
e
s
.
T
h
e
y
u
s
ed
au
g
m
en
tatio
n
tec
h
n
i
q
u
es
to
in
cr
ea
s
e
th
e
n
u
m
b
er
o
f
ch
ar
ac
ter
s
a
m
p
le
s
t
h
r
o
u
g
h
tr
ai
n
in
g
.
T
h
e
tr
ain
in
g
p
r
o
c
ess
w
as
ac
co
m
p
li
s
h
ed
b
y
u
s
i
n
g
Nv
id
ia
Gi
g
a
T
ex
el
Sh
ad
er
eXtr
e
m
e
(
GT
X)
1
0
8
0
.
T
h
e
ac
cu
r
ac
y
o
f
r
ec
o
g
n
i
tio
n
i
s
9
1
%.
P
atel
et
a
l.
[
13
]
in
tr
o
d
u
ce
d
th
e
n
u
m
b
er
p
late
r
ec
o
g
n
itio
n
(
NP
R
)
ap
p
r
o
ac
h
b
ased
o
n
m
o
r
p
h
o
lo
g
ical
o
p
er
atio
n
an
d
th
e
So
b
el
ed
g
e
d
etec
tio
n
m
et
h
o
d
s
.
T
h
e
b
o
u
n
d
i
n
g
b
o
x
m
et
h
o
d
w
a
s
u
s
ed
to
s
eg
m
e
n
t
th
e
n
u
m
b
er
s
an
d
letter
s
o
f
t
h
e
p
late.
T
e
m
p
late
m
atc
h
in
g
w
a
s
u
s
ed
to
r
ec
o
g
n
ize
n
u
m
b
er
s
an
d
c
h
ar
ac
ter
s
af
ter
s
e
g
m
en
tat
io
n
.
Fo
r
s
e
g
m
en
tin
g
let
ter
s
an
d
n
u
m
b
er
s
in
t
h
e
p
late
b
o
u
n
d
in
g
b
o
x
tech
n
iq
u
e
u
s
ed
.
T
h
e
m
atch
i
n
g
o
f
te
m
p
lates
w
as
u
s
e
d
to
i
d
en
tify
a
f
ter
s
eg
m
e
n
tatio
n
n
u
m
b
er
s
a
n
d
ch
ar
ac
ter
s
.
Sh
ar
m
a
in
[
14
]
h
av
e
r
ep
o
r
ted
p
late
n
u
m
b
er
id
en
ti
f
icat
i
o
n
u
s
i
n
g
p
h
ase
co
r
r
elatio
n
an
d
cr
o
s
s
-
co
r
r
elatio
n
s
tr
u
ct
u
r
ed
ap
p
r
o
a
ch
es.
T
h
e
r
eg
u
lar
cr
o
s
s
-
co
r
r
elatio
n
ap
p
r
o
ac
h
w
as
f
o
u
n
d
to
b
e
b
etter
th
an
th
e
p
h
ase
-
co
r
r
elatio
n
m
et
h
o
d
to
id
en
tify
t
h
e
v
e
h
icle
n
u
m
b
er
p
l
ate.
T
h
e
n
o
r
m
a
l
ized
cr
o
s
s
-
co
r
r
elatio
n
r
ec
o
g
n
i
tio
n
ac
cu
r
ac
y
f
o
r
th
e
p
late
w
a
s
6
7
.
9
8
%,
w
h
i
le
th
e
p
h
a
s
e
co
r
r
elatio
n
w
a
s
6
3
.
4
6
%.
Öztü
r
k
an
d
Ö
ze
n
[
15
]
p
r
o
p
o
s
ed
a
p
r
o
ce
d
u
r
e
f
o
r
id
en
ti
f
y
i
n
g
p
l
ate
ch
ar
ac
ter
s
.
Fo
r
id
en
ti
f
y
in
g
p
lates,
Ots
u
's
t
h
r
es
h
o
ld
in
g
was
u
s
ed
.
T
h
e
p
late
s
eg
m
e
n
tatio
n
w
a
s
ev
a
lu
ated
b
y
a
Ho
r
izo
n
ta
l
an
d
v
er
tica
l
h
is
to
g
r
a
m
.
E
v
en
t
u
all
y
,
f
o
r
ch
ar
ac
ter
r
ec
o
g
n
itio
n
,
p
r
o
b
a
b
ilis
tic
n
e
u
r
al
n
et
w
o
r
k
s
w
a
s
u
s
ed
.
T
h
e
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
f
o
r
th
e
c
h
ar
ac
ter
s
w
a
s
9
6
.
5
%
.
I
n
[
16
]
,
C
NN
w
a
s
u
s
ed
to
r
ec
o
g
n
ize
1
6
B
an
g
ala
lice
n
s
e
p
late
c
h
ar
ac
ter
s
,
a
n
d
th
e
m
o
d
el
te
s
ted
o
v
er
v
ar
io
u
s
i
m
a
g
e
s
a
m
p
le
s
.
T
h
ey
u
s
ed
1
7
5
0
tr
ain
in
g
s
a
m
p
les
a
n
d
3
5
0
f
o
r
test
in
g
s
a
m
p
les.
T
h
e
r
ec
o
g
n
itio
n
ac
c
u
r
ac
y
d
ep
en
d
s
o
n
t
h
e
n
u
m
b
er
o
f
ep
o
ch
s
f
r
o
m
(
1
0
0
-
1000)
an
d
th
e
s
a
m
p
le
s
’
n
u
m
b
er
f
r
o
m
(
1
3
0
0
-
1
7
5
0
)
s
o
th
at
t
h
e
r
ec
o
g
n
it
i
on
ac
cu
r
ac
y
r
an
g
e
(
7
0
-
8
8
)
%
.
B
ab
u
et
a
l.
[
1
7
]
p
r
o
p
o
s
ed
ar
tif
icial
n
eu
r
al
n
et
w
o
r
k
s
f
o
r
r
ec
o
g
n
i
zin
g
th
e
c
h
ar
ac
ter
s
o
f
t
h
e
p
late.
I
n
t
h
e
b
eg
i
n
n
i
n
g
,
t
h
e
p
late
s
eg
m
e
n
ted
b
y
u
s
in
g
t
h
e
co
n
n
ec
ted
co
m
p
o
n
en
t
s
an
a
l
y
s
is
an
d
t
h
e
v
er
tical
p
r
o
j
ec
tio
n
in
to
b
lo
ck
s
o
f
ch
ar
ac
ter
s
an
d
n
u
m
b
er
s
wh
er
e
th
e
b
lo
ck
s
r
esized
3
2
×
3
2
.
T
h
e
s
eg
m
e
n
tatio
n
an
d
r
ec
o
g
n
itio
n
ac
cu
r
ac
ies
w
e
r
e
8
5
.
4
% a
n
d
7
8
% seq
u
en
tiall
y
.
I
n
th
i
s
p
ap
er
,
an
ef
f
ic
ien
t
s
y
s
te
m
s
u
g
g
ested
f
o
r
class
i
f
y
in
g
m
u
lti
n
atio
n
al
lice
n
s
e
p
late
(
A
r
m
en
ia
,
B
elar
u
s
,
Hu
n
g
ar
y
)
co
u
n
tr
ie
s
b
as
ed
o
n
d
ee
p
C
NN
th
an
p
r
ev
io
u
s
l
y
li
ter
atu
r
e
s
t
u
d
ies
w
h
e
r
e
th
e
tr
ain
in
g
an
d
v
alid
atio
n
s
et
s
ac
cu
r
ac
ie
s
w
er
e
9
9
.
1
7
%
an
d
9
7
.
5
0
%,
r
esp
ec
tiv
el
y
.
T
h
e
o
v
er
all
p
late
cla
s
s
i
f
icatio
n
ac
c
u
r
ac
y
i
s
9
6
.
6
6
%
w
it
h
6
0
test
ed
p
late
im
ag
e
s
.
I
n
co
m
p
u
ter
v
i
s
io
n
,
t
h
e
class
if
icatio
n
o
f
i
m
a
g
es
h
a
s
an
es
s
en
t
ial
r
o
le.
C
NN
'
s
ar
e
f
r
eq
u
en
t
l
y
u
s
ed
in
d
ee
p
lear
n
in
g
m
o
d
els
wh
er
e
f
ea
t
u
r
e
ex
tr
ac
tio
n
w
it
h
o
u
t
ex
p
er
t
h
u
m
an
in
ter
v
e
n
tio
n
.
Am
o
n
g
v
ar
io
u
s
m
et
h
o
d
s
,
th
e
p
r
o
p
o
s
ed
C
NN
m
o
d
el
ac
h
ie
v
ed
h
i
g
h
p
er
f
o
r
m
a
n
ce
i
n
i
m
a
g
e
class
i
f
icatio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
6
,
Dec
em
b
er
2
0
2
1
:
5
2
6
6
-
5
2
7
6
5268
2.
RE
S
E
ARCH
M
E
T
H
O
D
Fo
r
im
p
le
m
e
n
ti
n
g
t
h
e
class
i
f
icatio
n
p
r
o
ce
s
s
,
d
ee
p
lear
n
in
g
C
NN
's
ar
e
ap
p
lied
,
as
illu
s
tr
ated
in
Fig
u
r
e
1
.
W
e
co
n
s
id
er
ed
a
p
late
i
m
a
g
e
f
o
r
th
r
ee
co
u
n
tr
ie
s
(
A
r
m
en
ia,
B
elar
u
s
,
H
u
n
g
ar
y
)
as
t
h
e
in
p
u
t
to
b
e
f
ee
d
ed
in
to
th
e
C
NN
m
o
d
el.
A
f
ter
w
ar
d
,
t
h
e
p
r
o
p
o
s
ed
m
e
th
o
d
clas
s
i
f
ies
it
i
n
to
(
A
r
m
e
n
ia)
,
(
B
elar
u
s
)
,
a
n
d
(
Hu
n
g
ar
y
)
p
late.
Fig
u
r
e
1
.
T
h
e
p
r
o
p
o
s
ed
m
u
lti
n
atio
n
al
lice
n
s
e
p
late
clas
s
i
f
icat
io
n
w
o
r
k
f
lo
w
2
.
1
.
Da
t
a
s
et
g
a
t
hering
T
h
e
ex
p
er
i
m
en
t is co
n
d
u
cted
o
n
6
0
0
i
m
ag
e
s
f
o
r
d
if
f
er
e
n
t lic
en
s
e
p
late
s
(
A
r
m
e
n
ia,
B
elar
u
s
,
Hu
n
g
ar
y
)
th
at
ar
e
co
llected
f
r
o
m
th
e
lo
ca
lizatio
n
s
tag
e
u
n
d
er
d
if
f
er
e
n
t
w
ea
t
h
er
co
n
d
itio
n
s
.
Data
s
ets
ar
e
d
is
tr
ib
u
ted
f
o
r
ea
ch
class
o
f
th
e
p
late
s
.
E
ac
h
class
h
as
2
0
0
i
m
ag
e
s
(
1
6
0
p
la
te
i
m
a
g
es
f
o
r
th
e
tr
ain
i
n
g
s
et
a
n
d
4
0
p
late
im
a
g
es
f
o
r
th
e
v
alid
atio
n
s
et,
a
s
r
ep
o
r
ted
in
F
ig
u
r
e
2
.
So
t
h
at
tr
ai
n
i
n
g
an
d
v
alid
atio
n
s
et
s
ar
e
8
0
%
an
d
2
0
%
o
f
ea
ch
class
,
r
esp
ec
tiv
e
l
y
.
Fig
u
r
e
2
.
Data
s
ets
allo
ca
tio
n
f
o
r
ea
ch
class
o
f
lice
n
s
e
p
lates
in
th
e
p
r
o
p
o
s
ed
C
NN
class
if
ic
atio
n
m
o
d
el
2
.
2
.
P
re
-
pro
ce
s
s
ing
s
t
a
g
e
A
p
r
e
-
p
r
o
ce
s
s
i
n
g
s
tep
i
s
ta
k
e
n
p
lace
b
ef
o
r
e
i
m
a
g
es
g
i
v
e
n
to
th
e
C
N
N
m
o
d
el.
T
h
is
i
s
to
m
i
n
i
m
ize
d
i
m
en
s
io
n
s
,
co
m
p
u
tat
io
n
an
d
to
s
h
o
w
b
etter
p
er
f
o
r
m
a
n
ce
.
T
h
e
o
r
ig
in
al
i
m
ag
e
i
s
a
g
r
a
y
s
ca
le
p
late
w
it
h
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
Mu
lticla
s
s
i
fica
tio
n
o
f lice
n
s
e
p
la
te
b
a
s
ed
o
n
d
ee
p
co
n
vo
lu
ti
o
n
n
eu
r
a
l
n
etw
o
r
ks (
Ma
s
a
r
A
b
ed
Uth
a
ib
)
5269
d
if
f
er
e
n
t
s
ize
s
;
w
e
u
n
i
f
o
r
m
th
e
m
to
2
0
0
×1
0
0
×1
.
T
h
e
d
a
ta
i
s
s
h
u
f
f
led
b
ef
o
r
e
s
p
litt
in
g
s
o
th
at
th
e
tr
ai
n
i
n
g
n
o
t
o
n
l
y
f
o
cu
s
es
o
n
n
ar
r
o
w
d
ata
b
u
t
al
s
o
u
n
s
o
r
ted
d
ata
in
th
e
d
ataset.
T
h
e
p
ix
el
s
o
f
th
e
i
m
a
g
e
n
o
r
m
alize
d
,
w
h
ic
h
is
th
e
ch
a
n
g
i
n
g
p
r
o
ce
s
s
f
o
r
in
t
en
s
it
y
p
ix
el
v
alu
e
s
r
an
g
e.
As
m
en
tio
n
ed
ea
r
lier
,
th
e
g
r
e
y
s
c
a
le
ch
an
n
el
h
a
s
o
n
e
ch
an
n
el
in
th
e
i
m
ag
e
(
0
-
2
5
5
)
th
at
m
a
k
es
th
e
ca
lcu
latio
n
co
m
p
lex
s
o
t
h
at
i
t
n
o
r
m
alize
d
i
m
ag
e
p
ix
el
to
b
e
i
n
th
e
r
an
g
e
(
0
-
1
)
.
Data
au
g
m
e
n
t
atio
n
tech
n
iq
u
e
s
ar
e
u
s
ed
w
it
h
ass
is
t
Ker
as
a
u
g
m
e
n
tatio
n
s
to
in
cr
ea
s
e
th
e
s
ize
an
d
also
i
m
p
r
o
v
e
s
th
e
q
u
alit
y
o
f
th
e
tr
ai
n
i
n
g
s
a
m
p
le
s
,
an
d
t
o
g
et
r
id
o
f
o
v
er
f
itt
in
g
,
w
h
er
e
o
v
er
f
itti
n
g
is
a
g
ap
b
et
w
ee
n
t
h
e
ac
c
u
r
ac
y
o
f
tr
ai
n
in
g
an
d
v
alid
atio
n
s
e
ts
.
T
h
e
y
also
i
m
p
r
o
v
e
m
o
d
el
e
f
f
icie
n
c
y
a
n
d
m
a
k
e
m
o
d
el
g
en
er
ali
t
y
[
1
8
]
,
[
1
9
]
.
Fig
u
r
e
3
s
h
o
w
s
th
e
p
r
ese
n
ted
au
g
m
en
t
e
d
i
m
ag
e
s
.
A
u
g
m
e
n
tatio
n
s
ar
e
s
et
b
ef
o
r
e
tr
ain
i
n
g
th
e
m
o
d
els.
W
e
u
s
ed
th
r
ee
au
g
m
e
n
tat
io
n
tec
h
n
iq
u
es
(
r
o
tatio
n
,
h
o
r
izo
n
tal
f
lip
,
zo
o
m
)
to
g
et
n
e
w
th
r
ee
tr
ai
n
i
n
g
s
ets
t
h
at
ex
p
an
d
t
h
e
tr
ain
i
n
g
s
a
m
p
les
d
atase
t.
T
h
e
n
e
w
l
y
g
en
er
ated
s
et
o
f
tr
ain
i
n
g
is
ess
en
tiall
y
th
e
i
n
itia
l
tr
ain
i
n
g
i
m
ag
e
s
w
it
h
th
e
au
g
m
en
ted
i
m
a
g
es
g
e
n
er
ated
b
y
th
e
ch
o
ices
o
f
au
g
m
en
tatio
n
tech
n
iq
u
e
s
.
E
ac
h
au
g
m
e
n
tatio
n
is
as
f
o
llo
w
s
[
2
0
]
-
[
2
2
]
:
a.
R
o
tatio
n
T
h
e
im
a
g
e
is
r
o
tated
2
0
° f
o
r
g
ettin
g
b
etter
class
i
f
icat
io
n
.
b.
Ho
r
izo
n
tal
Fli
p
p
in
g
I
t
is
o
n
e
o
f
t
h
e
g
eo
m
etr
ic
a
u
g
m
e
n
tatio
n
,
its
f
l
ip
s
i
m
ag
e
to
w
ar
d
s
t
h
e
h
o
r
izo
n
tal
a
x
is
.
I
ts
m
o
r
e
co
m
m
o
n
th
a
n
v
er
t
ical
f
l
ip
an
d
p
r
o
v
en
its
b
en
e
f
it i
n
C
I
F
AR
-
1
0
an
d
I
m
a
g
eNe
t d
ataset
s
.
c.
Z
o
o
m
W
e
zo
o
m
ed
i
m
a
g
es to
m
ak
e
f
ea
tu
r
es c
lear
.
(
a)
(
b
)
(
c)
(
d
)
Fig
u
r
e
3
.
Data
au
g
m
e
n
tatio
n
t
ec
h
n
iq
u
es
,
(
a
)
o
r
ig
in
al
i
m
a
g
e
,
(
b
)
r
o
tated
im
a
g
e
,
(
c)
h
o
r
izo
n
t
al
f
lip
i
m
ag
e
,
(d
)
zo
o
m
i
m
a
g
e
2
.
3
.
T
he
pro
po
s
ed
CNN
cla
s
s
if
ier
a
rc
hite
ct
ure
T
h
e
s
u
g
g
e
s
ted
p
r
o
p
o
s
ed
C
N
N
m
o
d
el
is
r
ep
o
r
ted
in
F
i
g
u
r
e
4
w
it
h
T
ab
le
1
,
w
h
ic
h
s
u
m
m
ar
izes
th
e
C
NN
la
y
er
s
.
T
h
e
s
y
s
te
m
co
n
t
ain
s
3
0
la
y
er
s
th
at
s
tar
ted
f
r
o
m
t
h
e
la
y
er
o
f
in
p
u
t,
w
h
ich
ca
r
r
y
i
m
a
g
es
f
r
o
m
t
h
e
au
g
m
e
n
tatio
n
m
et
h
o
d
s
in
t
h
e
p
r
ec
ed
in
g
p
r
ep
r
o
ce
s
s
in
g
p
h
as
e.
Fo
u
r
b
lo
ck
s
o
f
co
n
v
o
lu
t
io
n
lay
er
s
th
at
co
n
s
is
t
o
f
(
co
n
v
o
lu
tio
n
a
s
w
e
ll
a
s
a
R
ec
ti
f
ied
lin
ea
r
u
n
it
(
R
el
u
)
w
h
ic
h
i
s
t
h
e
f
u
n
ctio
n
o
f
ac
tiv
a
tio
n
)
,
b
atc
h
n
o
r
m
aliza
t
io
n
,
m
a
x
-
p
o
o
lin
g
,
an
d
Dr
o
p
o
u
t
r
an
g
es
(
2
0
-
25)
%
la
y
er
s
.
Af
ter
f
o
u
r
b
lo
ck
s
o
f
co
n
v
o
lu
tio
n
la
y
er
s
,
th
r
ee
f
u
ll
y
co
n
n
ec
ted
la
y
er
s
a
r
e
i
m
p
le
m
e
n
ted
,
an
d
t
h
e
n
t
h
e
last
d
r
o
p
o
u
t
w
ith
(
3
0
%)
p
r
o
b
ab
ilit
y
b
ef
o
r
e
th
e
f
i
n
al
la
y
er
(
s
o
f
t
m
a
x
-
la
y
er
)
w
it
h
t
h
r
ee
-
cla
s
s
es
o
f
p
lates
[
2
3
]
.
T
h
e
d
escr
ip
tio
n
o
f
ea
ch
la
y
er
is
ill
u
s
tr
ated
b
elo
w
[
2
4
].
2
.
3
.
1
.
Co
nv
o
lutio
n la
y
er
T
h
e
ce
n
tr
al
co
m
p
o
n
en
t
in
t
h
e
C
NN
-
m
o
d
el
i
s
th
e
co
n
v
o
l
u
tio
n
la
y
er
,
w
h
ic
h
h
a
s
lo
ca
l ties a
n
d
co
m
m
o
n
w
ei
g
h
ts
.
I
t
s
o
b
j
ec
tiv
es
ar
e
lear
n
in
g
th
e
r
ep
r
esen
ti
n
g
o
f
e
n
ter
ed
f
ea
t
u
r
es.
I
t
co
n
tai
n
s
v
ar
io
u
s
f
ea
tu
r
e
-
m
ap
s
.
T
h
e
s
i
m
ilar
it
y
o
f
t
h
e
n
e
u
r
o
n
f
ea
tu
r
es
in
d
i
v
er
s
e
lo
ca
tio
n
s
i
s
u
s
e
d
to
ex
tr
ac
t
t
h
e
lo
ca
l
p
r
o
p
r
ieties
i
n
t
h
e
p
r
ev
io
u
s
la
y
er
'
s
d
i
f
f
er
e
n
t
p
o
s
itio
n
s
.
Acc
o
r
d
in
g
to
in
d
i
v
id
u
a
l
n
e
u
r
o
n
s
,
c
h
ar
ac
ter
is
tic
s
ex
tr
ac
t
io
n
is
p
er
f
o
r
m
ed
i
n
t
h
e
s
a
m
e
f
ea
tu
r
e
m
ap
ar
ea
in
th
e
p
r
ec
ed
in
g
lay
er
.
W
e
u
s
ed
a
co
n
v
o
lu
tio
n
f
ilter
(
k
er
n
e
l)
w
it
h
d
if
f
er
en
t
s
izes
(
3
×3
,
5
×5
,
7
×7
)
o
v
er
lap
p
e
d
h
o
r
izo
n
tall
y
a
n
d
v
er
tica
ll
y
al
o
n
g
w
it
h
t
h
e
i
n
p
u
t
i
m
ag
e
to
g
et
f
ea
t
u
r
es.
T
h
e
p
ad
d
in
g
an
d
s
tr
id
e
o
n
e
p
ix
el
an
d
t
w
o
s
tep
s
,
r
esp
ec
ti
v
el
y
,
a
s
s
h
o
w
n
i
n
F
ig
u
r
e
5
,
w
h
ic
h
r
ep
o
r
ted
co
n
v
o
lu
tio
n
la
y
er
o
p
er
atio
n
s
.
Af
ter
th
at,
th
e
r
es
u
lt
lead
s
to
th
e
n
o
n
-
li
n
ea
r
ac
ti
v
atio
n
f
u
n
ctio
n
t
h
at
is
R
el
u
.
Fi
g
u
r
e
6
clar
if
ies t
h
e
w
o
r
k
o
f
R
el
u
[
2
5
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
6
,
Dec
em
b
er
2
0
2
1
:
5
2
6
6
-
5
2
7
6
5270
Fig
u
r
e
4
.
T
h
e
p
r
o
p
o
s
ed
C
NN
m
o
d
el
T
ab
le
1
.
Su
m
m
ar
y
o
f
C
NN
'
s
l
a
y
er
s
L
a
y
e
r
(
t
y
p
e
)
O
u
t
p
u
t
S
h
a
p
e
P
a
r
a
m
#
i
n
p
u
t
_
1
(
I
n
p
u
t
L
a
y
e
r
)
(
N
o
n
e
,
1
0
0
,
2
0
0
,
1
)
0
c
o
n
v
2
d
_
1
(
C
o
n
v
2
D
)
(
N
o
n
e
,
1
0
0
,
2
0
0
,
3
2
)
3
2
0
c
o
n
v
2
d
_
2
(
C
o
n
v
2
D
)
(
N
o
n
e
,
1
0
0
,
2
0
0
,
3
2
)
9
2
4
8
c
o
n
v
2
d
_
3
(
C
o
n
v
2
D
)
(
N
o
n
e
,
1
0
0
,
2
0
0
,
3
2
)
9
2
4
8
b
a
t
c
h
_
n
o
r
mal
i
z
a
t
i
o
n
_
1
(
B
a
t
c
h
(
N
o
n
e
,
1
0
0
,
2
0
0
,
3
2
)
1
2
8
max
_
p
o
o
l
i
n
g
2
d
_
1
(
M
a
x
P
o
o
l
i
n
g
2
(
N
o
n
e
,
5
0
,
1
0
0
,
3
2
)
0
d
r
o
p
o
u
t
_
1
(
D
r
o
p
o
u
t
)
(
N
o
n
e
,
5
0
,
1
0
0
,
3
2
)
0
c
o
n
v
2
d
_
4
(
C
o
n
v
2
D
)
(
N
o
n
e
,
5
0
,
1
0
0
,
6
4
)
5
1
2
6
4
c
o
n
v
2
d
_
5
(
C
o
n
v
2
D
)
(
N
o
n
e
,
5
0
,
1
0
0
,
6
4
)
1
0
2
4
6
4
c
o
n
v
2
d
_
6
(
C
o
n
v
2
D
)
(
N
o
n
e
,
5
0
,
1
0
0
,
6
4
)
1
0
2
4
6
4
b
a
t
c
h
_
n
o
r
mal
i
z
a
t
i
o
n
_
2
(
B
a
t
c
h
(
N
o
n
e
,
5
0
,
1
0
0
,
6
4
)
2
5
6
max
_
p
o
o
l
i
n
g
2
d
_
2
(
M
a
x
P
o
o
l
i
n
g
2
(
N
o
n
e
,
2
5
,
5
0
,
6
4
)
0
d
r
o
p
o
u
t
_
2
(
D
r
o
p
o
u
t
)
(
N
o
n
e
,
2
5
,
5
0
,
6
4
)
0
c
o
n
v
2
d
_
7
(
C
o
n
v
2
D
)
(
N
o
n
e
,
2
5
,
5
0
,
1
2
8
)
4
0
1
5
3
6
c
o
n
v
2
d
_
8
(
C
o
n
v
2
D
)
(
N
o
n
e
,
2
5
,
5
0
,
1
2
8
)
8
0
2
9
4
4
c
o
n
v
2
d
_
9
(
C
o
n
v
2
D
)
(
N
o
n
e
,
2
5
,
5
0
,
1
2
8
)
8
0
2
9
4
4
b
a
t
c
h
_
n
o
r
mal
i
z
a
t
i
o
n
_
3
(
B
a
t
c
h
(
N
o
n
e
,
2
5
,
5
0
,
1
2
8
)
5
1
2
max
_
p
o
o
l
i
n
g
2
d
_
3
(
M
a
x
P
o
o
l
i
n
g
2
(
N
o
n
e
,
1
2
,
2
5
,
1
2
8
)
0
d
r
o
p
o
u
t
_
3
(
D
r
o
p
o
u
t
)
(
N
o
n
e
,
1
2
,
2
5
,
1
2
8
)
0
c
o
n
v
2
d
_
1
0
(
C
o
n
v
2
D
)
(
N
o
n
e
,
1
2
,
2
5
,
3
2
)
1
0
2
4
3
2
c
o
n
v
2
d
_
1
1
(
C
o
n
v
2
D
)
(
N
o
n
e
,
1
2
,
2
5
,
3
2
)
2
5
6
3
2
b
a
t
c
h
_
n
o
r
mal
i
z
a
t
i
o
n
_
4
(
B
a
t
c
h
(
N
o
n
e
,
1
2
,
2
5
,
3
2
)
1
2
8
max
_
p
o
o
l
i
n
g
2
d
_
4
(
M
a
x
P
o
o
l
i
n
g
2
(
N
o
n
e
,
6
,
1
2
,
3
2
)
0
d
r
o
p
o
u
t
_
4
(
D
r
o
p
o
u
t
)
(
N
o
n
e
,
6
,
1
2
,
3
2
)
0
f
l
a
t
t
e
n
_
1
(
F
l
a
t
t
e
n
)
(
N
o
n
e
,
2
3
0
4
)
0
d
e
n
se
_
1
(
D
e
n
se
)
(
N
o
n
e
,
2
0
4
8
)
4
7
2
0
6
4
0
d
e
n
se
_
2
(
D
e
n
se
)
(
N
o
n
e
,
1
0
2
4
)
2
0
9
8
1
7
6
d
e
n
se
_
3
(
D
e
n
se
)
(
N
o
n
e
,
5
1
2
)
5
2
4
8
0
0
d
r
o
p
o
u
t
_
5
(
D
r
o
p
o
u
t
)
(
N
o
n
e
,
5
1
2
)
0
d
e
n
se
_
4
(
D
e
n
se
)
(
N
o
n
e
,
4
)
1
5
3
9
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
Mu
lticla
s
s
i
fica
tio
n
o
f lice
n
s
e
p
la
te
b
a
s
ed
o
n
d
ee
p
co
n
vo
lu
ti
o
n
n
eu
r
a
l
n
etw
o
r
ks (
Ma
s
a
r
A
b
ed
Uth
a
ib
)
5271
Fig
u
r
e
5
.
T
h
e
ex
a
m
p
le
f
o
r
o
p
er
atio
n
s
o
f
co
n
v
o
lu
tio
n
la
y
er
(
w
h
er
e
th
e
i
n
p
u
t_
s
h
ap
e:
3
×3
,
p
ad
d
in
g
=1
,
k
er
n
el_
s
ize=
3
×3
,
Strid
e
=
2
,
o
u
tp
u
t_
s
h
ap
e:
2
×2
)
T
h
e
eq
u
atio
n
o
f
R
el
u
is
[
2
6
]
:
(
)
=
(
0
,
)
(
1
)
w
h
er
e
(
)
=
if
:p
o
s
itiv
e
(
)
=
0
if
:n
e
g
ati
v
e
Fig
u
r
e
6
.
R
elu
ac
ti
v
atio
n
f
u
n
ct
io
n
[
2
6
]
2
.
3
.
2
.
B
a
t
ch
no
r
m
a
liza
t
io
n
B
atch
n
o
r
m
aliza
tio
n
(
B
N)
i
s
a
s
tr
ateg
y
f
o
r
n
o
r
m
aliz
in
g
ac
ti
v
atio
n
s
in
d
ee
p
n
e
u
r
a
l
n
et
w
o
r
k
in
ter
m
ed
iate
la
y
er
s
.
B
N
is
a
f
av
o
r
ite
tech
n
iq
u
e
in
d
etail
d
u
e
to
its
ab
ilit
y
to
b
o
o
s
t
ac
cu
r
ac
y
an
d
ac
ce
ler
at
e
p
r
ep
ar
atio
n
.
I
t
en
ab
les
u
s
er
s
t
o
u
s
e
h
i
g
h
er
le
v
els
o
f
lear
n
i
n
g
r
ates,
p
r
ev
en
tin
g
m
in
o
r
p
ar
am
eter
c
h
an
g
e
s
f
r
o
m
b
o
o
s
tin
g
i
n
to
lar
g
er
a
n
d
s
u
b
o
p
ti
m
al
c
h
an
g
es
i
n
g
r
ad
ien
t
ac
t
iv
atio
n
.
Als
o
,
it
p
r
ev
en
ts
t
h
e
t
r
ain
in
g
f
r
o
m
b
ein
g
s
tu
c
k
i
n
t
h
e
s
at
u
r
ated
n
o
n
-
li
n
ea
r
it
y
r
eg
i
m
e
s
.
I
t
ac
ts
as
a
r
eg
u
lar
izer
.
W
h
e
n
B
atch
No
r
m
aliza
tio
n
is
f
o
u
n
d
d
u
r
in
g
n
et
w
o
r
k
tr
ain
in
g
,
t
h
er
e
is
co
m
m
u
n
ica
tio
n
a
m
o
n
g
a
t
r
ain
in
g
ex
a
m
p
le,
a
n
d
t
h
e
r
e
m
ain
i
n
g
ex
a
m
p
le
s
i
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
6
,
Dec
em
b
er
2
0
2
1
:
5
2
6
6
-
5
2
7
6
5272
Min
i
-
b
atch
a
n
d
th
e
g
en
er
atio
n
o
f
th
e
d
eter
m
i
n
i
s
tic
v
al
u
e
s
d
o
n
o
t
tak
e
ti
m
e
f
o
r
th
e
tr
ai
n
i
n
g
g
i
v
e
n
ex
a
m
p
le
s
o
th
at
it
s
b
en
ef
icial
f
o
r
n
e
t
w
o
r
k
g
en
er
aliza
tio
n
[
2
7
]
-
[
2
9
]
.
2
.
3
.
3
.
M
a
x
po
o
lin
g
la
y
er
T
h
e
in
f
o
r
m
atio
n
a
m
o
u
n
t
in
ev
er
y
g
at
h
er
ed
f
ea
t
u
r
ed
in
th
e
co
n
v
o
l
u
tio
n
la
y
er
i
s
d
ec
r
ea
s
ed
w
h
il
e
r
etain
i
n
g
th
e
cr
u
cia
l
d
etail
s
(
co
m
m
o
n
l
y
M
u
ltip
le
co
n
v
o
l
u
ti
o
n
s
a
n
d
p
o
o
l
la
y
er
s
r
o
u
n
d
s
)
.
Of
te
n
lar
g
e
i
m
a
g
es
ar
e
in
th
e
C
NN
m
o
d
el;
th
er
e
f
o
r
e,
w
e
n
ee
d
to
d
ec
r
ea
s
e
im
a
g
es
th
a
t
m
i
n
i
m
ize
th
e
n
u
m
b
er
o
f
th
e
p
ar
am
eter
s
.
A
ll
th
e
u
tili
ze
d
m
ax
-
p
o
o
lin
g
l
a
y
er
s
ar
e
(
2
,
2
)
,
b
esid
es
th
e
s
tr
id
e
is
(
2
,
2
)
to
m
o
v
e
h
o
r
izo
n
ta
l
l
y
an
d
v
er
tical
l
y
.
I
t
is
ac
co
m
p
lis
h
ed
b
y
d
iv
id
i
n
g
t
h
e
en
tire
i
m
a
g
e
in
to
s
m
al
ler
s
q
u
ar
es
(
2
×2
th
e
s
u
g
g
ested
m
eth
o
d
)
,
w
h
ich
p
as
s
o
v
er
th
e
i
m
ag
e
w
it
h
a
s
p
ec
i
f
ied
s
tr
in
g
(
2
×2
)
.
Af
ter
t
h
at,
t
h
e
lar
g
est
v
a
lu
e
i
n
t
h
e
m
atr
i
x
o
f
f
o
u
r
n
u
m
b
er
s
ch
o
s
en
[
3
0
]
,
[
3
1
]
.
Fig
u
r
e
7
s
h
o
w
s
t
h
e
o
p
er
atio
n
Ma
x
-
p
o
o
lin
g
la
y
er
.
I
n
(
2
)
-
(
4
)
ar
e
u
s
ed
f
o
r
Ma
x
-
p
o
o
lin
g
[
3
2
]
:
2
=
1
−
+
1
(
2
)
2
=
1
−
+
1
(
3
)
2
=
1
(
4
)
w
h
er
e:
F:
s
p
atial
e
x
te
n
d
o
f
t
h
e
f
i
lter
,
S:
s
tr
id
e,
D1
:
th
e
d
ep
th
o
f
co
n
v
o
lu
tio
n
la
y
er
,
D
2
:
d
ep
th
-
co
n
v
o
lu
tio
n
la
y
er
,
W
1
:
co
n
v
o
lu
tio
n
la
y
er
w
id
t
h
,
H1
:
co
n
v
o
l
u
tio
n
la
y
er
h
ei
g
h
t,
W
2
:
Ma
x
-
p
o
o
lin
g
la
y
er
-
w
id
t
h
,
a
n
d
H2
:
Ma
x
p
o
o
lin
g
la
y
er
-
h
ei
g
h
t.
Fig
u
r
e
7
.
T
h
e
o
p
er
atio
n
o
f
m
ax
-
p
o
o
lin
g
la
y
er
to
o
n
e
b
lo
ck
2
.
3
.
4
.
Dro
po
ut
la
y
er
Ma
n
y
ac
ti
v
atio
n
s
(
n
o
d
es)
ar
e
r
an
d
o
m
l
y
d
is
ca
r
d
ed
in
th
i
s
la
y
er
,
w
h
ic
h
o
f
te
n
d
r
a
m
atica
ll
y
a
cc
eler
ates
th
e
tr
ain
i
n
g
-
s
tag
e
a
n
d
d
ec
r
ea
s
in
g
o
v
er
f
itti
n
g
.
T
h
e
p
r
o
b
ab
il
ities
f
o
r
d
r
o
p
o
u
t
r
an
g
es
i
n
o
u
r
p
r
o
p
o
s
ed
-
m
eth
o
d
w
er
e
(
2
0
-
2
5
)
%
f
o
r
th
e
f
o
u
r
D
r
o
p
o
u
t
-
la
y
er
s
t
h
at
f
o
llo
w
m
a
x
-
p
o
o
lin
g
la
y
er
s
.
T
h
e
last
d
r
o
p
o
u
t
la
y
er
r
atio
a
f
ter
th
e
f
u
ll
y
co
n
n
ec
ted
la
y
er
is
3
0
% [
3
3
]
-
[
3
5
]
.
Fig
u
r
e
8
p
r
esen
te
d
Dr
o
p
o
u
t la
y
er
s
.
(
a)
(
b
)
Fig
u
r
e
8
.
E
x
a
m
p
le
o
f
t
h
e
d
r
o
p
o
u
t la
y
er
: (
a)
s
tan
d
ar
t
n
eu
r
al
n
et,
(
b
)
af
ter
ap
p
ly
in
g
d
r
o
p
o
u
t
[
3
6
]
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
Mu
lticla
s
s
i
fica
tio
n
o
f lice
n
s
e
p
la
te
b
a
s
ed
o
n
d
ee
p
co
n
vo
lu
ti
o
n
n
eu
r
a
l
n
etw
o
r
ks (
Ma
s
a
r
A
b
ed
Uth
a
ib
)
5273
2
.
3
.
5
.
F
ull
y
-
co
nn
ec
t
ed
-
la
y
er
s
T
h
e
ar
r
an
g
e
m
en
t
o
f
la
y
er
s
l
ik
e
n
e
u
r
o
n
s
i
n
a
t
y
p
ical
n
e
u
r
al
n
et
w
o
r
k
is
o
r
g
an
ized
.
A
n
o
d
e
in
a
f
u
l
l
y
co
n
n
ec
ted
-
la
y
er
is
attac
h
ed
d
i
r
ec
tl
y
to
ea
ch
n
o
d
e
i
n
th
e
p
r
ev
io
u
s
a
n
d
n
ex
t
la
y
er
s
,
a
s
clea
r
ed
in
F
i
g
u
r
e
9
.
T
h
e
co
n
n
ec
tio
n
b
et
w
ee
n
n
o
d
es
i
n
th
e
p
r
ev
io
u
s
f
r
a
m
e
o
f
t
h
e
p
o
o
lin
g
-
la
y
er
an
d
t
h
e
l
a
y
er
o
f
f
u
ll
y
-
co
n
n
ec
ted
is
a
v
ec
to
r
,
w
h
ic
h
r
ep
r
esen
t
s
t
h
e
f
ir
s
t
la
y
er
.
I
t
co
n
tai
n
s
p
ar
a
m
et
er
s
th
at
tak
e
lo
n
g
er
to
b
e
tr
ain
ed
.
T
h
er
ef
o
r
e,
th
e
n
u
m
b
er
o
f
n
o
d
es
an
d
li
n
k
s
a
r
e
eli
m
i
n
ated
u
s
in
g
th
e
Dr
o
p
o
u
t
tec
h
n
iq
u
e
w
it
h
p
r
o
b
ab
ilit
y
(
3
0
%)
b
ef
o
r
e
t
h
e
s
o
f
t
m
ax
la
y
e
r
.
So
f
t
m
a
x
i
s
a
m
ix
t
u
r
e
o
f
m
a
n
y
s
ig
m
o
id
ac
ti
v
a
tio
n
f
u
n
c
tio
n
s
.
Sig
m
o
id
f
u
n
cti
o
n
r
etu
r
n
s
v
al
u
es
i
n
th
e
r
a
n
g
e
(0
-
1
)
.
T
h
ese
ca
n
b
e
v
ie
w
ed
as
th
e
li
k
eli
h
o
o
d
o
f
d
ata
p
o
in
ts
o
f
a
g
i
v
en
clas
s
.
So
f
t
m
ax
ca
n
b
e
u
s
ed
f
o
r
m
u
lticla
s
s
cla
s
s
i
f
icatio
n
p
r
o
b
lem
s
,
u
n
li
k
e
s
i
g
m
o
id
f
u
n
c
tio
n
s
t
h
at
ar
e
u
s
ed
f
o
r
b
in
ar
y
class
i
f
icatio
n
.
T
h
e
f
u
n
ctio
n
r
etu
r
n
s
t
h
e
lik
e
lih
o
o
d
f
o
r
ev
er
y
d
ata
p
o
in
t o
f
all
g
r
o
u
p
s
[
3
7
]
as e
x
p
lain
ed
in
t
h
e
(
5
)
[
3
8
]
:
(
)
=
∑
=
1
f
o
r
j
=
1
,
.
.
.
,
k
.
(
5
)
K:
class
n
u
m
b
er
a
n
d
th
e
f
u
n
ctio
n
o
f
Z
o
u
tp
u
t
eq
u
al
to
o
n
e.
T
h
e
o
u
tp
u
t
la
y
er
o
f
t
h
e
n
e
t
w
o
r
k
ca
n
co
n
tain
s
i
m
ilar
a
m
o
u
n
t
s
o
f
n
eu
r
o
n
s
a
s
t
h
e
n
u
m
b
er
o
f
clas
s
es
w
i
th
i
n
th
e
tar
g
et
if
w
e
cr
ea
te
a
n
et
w
o
r
k
o
r
a
m
o
d
el
f
o
r
m
u
ltip
le
cla
s
s
i
f
icat
io
n
s
.
A
cc
o
r
d
in
g
to
o
u
r
w
o
r
k
,
th
e
o
u
tp
u
t
o
f
s
o
f
t
m
a
x
is
t
h
r
ee
-
clas
s
es
(
A
r
m
en
ia
,
B
elar
u
s
,
Hu
n
g
ar
y
)
.
Fig
u
r
e
9
s
h
o
w
s
t
h
e
s
o
f
t
m
a
x
ac
ti
v
atio
n
f
u
n
ct
io
n
o
p
er
atio
n
.
I
n
th
e
e
n
d
,
w
e
h
a
v
e
u
s
ed
th
e
lo
s
s
f
u
n
ctio
n
,
w
h
ic
h
i
s
cr
o
s
s
-
e
n
tr
o
p
y
to
esti
m
ate
t
h
e
lo
s
s
o
f
class
i
f
icatio
n
an
d
to
p
r
ed
ict
th
e
lab
el
o
f
th
e
in
p
u
t
i
m
a
g
e.
T
h
e
cr
o
s
s
-
en
tr
o
p
y
eq
u
atio
n
ca
n
b
e
w
r
itte
n
as
in
(
6
)
[
3
9
]
:
L
(
,
̂
)
=
−
∑
∗
l
og
(
̂
)
=
0
(
6
)
w
h
er
e
̂
:
th
e
r
atio
o
f
th
e
p
r
ed
ic
ted
o
u
tp
u
t la
y
er
,
y
t
h
e
b
i
n
ar
y
i
n
d
icato
r
w
h
et
h
er
t
h
e
clas
s
i
f
ica
tio
n
is
co
r
r
ec
t (
1
)
o
r
n
o
t (
0
)
,
an
d
n
: th
e
clas
s
n
u
m
b
er
s
/lab
els (
t
h
e
o
u
tp
u
t
n
o
d
es
n
u
m
b
er
s
)
.
Fig
u
r
e
9
.
So
f
t
m
ax
ac
ti
v
atio
n
f
u
n
ct
io
n
ex
a
m
p
le
2
.
4
.
O
ptim
iza
t
io
n
a
lg
o
rit
h
m
T
h
e
tr
ain
in
g
o
f
C
NN
m
o
d
el
b
y
ch
a
n
g
i
n
g
o
n
th
e
iter
ativ
e
b
a
s
is
t
h
e
la
y
er
's
p
ar
a
m
eter
s
in
t
h
e
n
et
w
o
r
k
.
T
h
e
o
p
ti
m
izer
p
er
f
o
r
m
s
a
s
ig
n
i
f
ican
t
r
o
le
in
th
e
co
n
tex
t
o
f
le
ar
n
in
g
d
ee
p
C
NN
m
o
d
els.
T
h
e
lo
s
s
f
u
n
ctio
n
i
s
m
i
n
i
m
ized
b
y
t
h
e
o
p
ti
m
izer
.
W
e
u
s
ed
A
d
a
m
o
p
ti
m
izer
s
.
I
t
p
r
o
v
es
its
e
f
f
icien
c
y
i
n
t
h
e
le
ar
n
in
g
p
r
o
ce
s
s
i
n
a
s
h
o
r
t ti
m
e.
I
t ta
k
es a
s
m
al
l siz
e
o
f
m
e
m
o
r
y
.
T
h
e
co
m
p
u
tatio
n
w
it
h
th
i
s
o
p
ti
m
izer
i
s
d
o
n
e
ef
f
icien
tl
y
[
4
0
]
-
[
44]
.
3.
RE
SU
L
T
S
A
ND
D
I
SCU
SS
I
O
N
T
h
e
s
p
ec
if
icatio
n
f
o
r
h
ar
d
w
ar
e
an
d
s
o
f
t
w
ar
e
u
s
ed
in
th
i
s
s
t
u
d
y
as
f
o
llo
w
s
:
P
r
o
ce
s
s
o
r
s
:
I
n
tel
R
C
o
r
e
(
T
M)
i
7
-
9
7
5
0
H
C
P
U,
th
e
lo
g
ical
p
r
o
ce
s
s
o
r
n
u
m
b
er
s
:
1
2
,
an
d
GP
U:
NVI
DI
A
Ge
Fo
r
ce
GT
X
1
6
6
0
T
I
.
Mo
r
eo
v
er
,
w
e
e
m
p
lo
y
ees
th
e
f
o
llo
w
i
n
g
:
L
o
s
s
f
u
n
c
tio
n
:
ca
teg
o
r
ical
cr
o
s
s
-
e
n
tr
o
p
y
,
Op
ti
m
izatio
n
:
A
d
a
m
o
p
tim
izer
,
L
ea
r
n
in
g
r
ate
=0
.
0
0
0
1
,
No
.
o
f
E
p
o
ch
s
:
2
1
4
E
p
o
ch
s
,
T
i
m
e
o
f
tr
ain
i
n
g
:
1
2
m
i
n
u
tes,
P
r
o
g
r
a
m
m
i
n
g
b
y
an
ac
o
n
d
a
p
y
t
h
o
n
3
.
7
,
Ker
as
(
T
en
s
o
r
f
lo
w
b
ac
k
e
n
d
)
,
an
d
n
u
m
b
er
o
f
p
ar
a
m
eter
s
:
9
,
7
5
6
,
6
7
5
.
Fig
u
r
e
1
0
(
a
)
s
h
o
w
s
b
o
th
t
h
e
ac
c
u
r
ac
y
p
r
o
g
r
ess
d
u
r
i
n
g
t
h
e
tr
ai
n
i
n
g
a
n
d
v
alid
atio
n
p
h
ase
s
f
o
r
o
u
r
p
r
o
p
o
s
ed
m
o
d
el.
Fig
u
r
e
1
0
(
b
)
s
h
o
w
s
th
a
t
lo
s
s
h
as
b
ee
n
ac
h
ie
v
ed
r
ig
h
t
a
f
ter
2
1
4
ep
o
ch
s
.
T
h
e
lo
s
s
h
as
r
i
s
e
n
a
n
d
f
a
llen
.
Af
ter
1
9
4
ep
o
ch
s
,
it
r
ea
ch
es
s
tab
i
lit
y
,
w
h
er
e
t
h
e
c
u
r
v
e
b
eg
i
n
s
to
d
r
o
p
.
Desp
ite
a
s
m
all
n
u
m
b
er
o
f
d
ataset
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
6
,
Dec
em
b
er
2
0
2
1
:
5
2
6
6
-
5
2
7
6
5274
(
6
0
0
im
a
g
e
s
a
m
p
les),
th
e
p
r
o
p
o
s
ed
m
o
d
el
p
r
o
v
es
ef
f
ici
en
c
y
w
h
er
e
th
e
ac
cu
r
ac
y
f
o
r
th
e
tr
ain
in
g
an
d
v
alid
atio
n
s
e
ts
is
9
9
.
1
7
%
an
d
9
7
.
5
0
%,
s
eq
u
en
tiall
y
w
i
th
e
f
f
icien
t
ti
m
e
(
1
2
m
i
n
u
te
s
)
f
o
r
tr
ain
i
n
g
.
W
e
test
ed
6
0
p
late
im
a
g
es
w
h
er
e
t
h
e
o
v
er
all
clas
s
i
f
icatio
n
ac
cu
r
ac
y
is
9
6
.
6
6
%
is
co
m
p
u
ted
ac
c
o
r
d
in
g
to
(
7
)
.
T
h
e
co
m
p
ar
is
o
n
b
et
w
ee
n
th
e
p
r
o
p
o
s
ed
class
i
f
icatio
n
m
e
th
o
d
an
d
o
th
er
class
i
f
icatio
n
m
eth
o
d
s
,
as
ill
u
s
tr
ated
in
T
ab
le
2
.
T
a
b
le
3
r
ep
o
r
ted
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
C
NN
m
o
d
el
in
clas
s
i
f
y
in
g
m
u
lti
n
atio
n
a
l
licen
s
e
p
lates.
I
n
o
r
d
er
to
test
th
e
ac
cu
r
ac
y
o
f
th
e
cla
s
s
i
f
icatio
n
o
f
m
u
lti
n
atio
n
a
l lice
n
s
e
p
late
th
e
(
7
)
is
ad
o
p
te
d
[
1
1
]
:
=
×
100%
(
7
)
(
a)
(
b
)
Fig
u
r
e
1
0
.
T
r
ain
in
g
a
n
d
v
alid
a
tio
n
ac
cu
r
ac
y
a
n
d
lo
s
s
,
(
a)
tr
ain
in
g
a
n
d
v
alid
atio
n
a
cc
u
r
ac
y
i
n
v
ar
io
u
s
ep
o
ch
s
,
(
b
)
tr
ain
in
g
a
n
d
v
alid
atio
n
lo
s
s
in
v
ar
io
u
s
ep
o
ch
s
T
ab
le
2
.
C
o
m
p
ar
is
o
n
a
m
o
n
g
d
if
f
er
en
t r
ec
o
g
n
itio
n
o
f
lice
n
s
e
p
late
m
eth
o
d
s
R
e
f
e
r
e
n
c
e
A
p
p
r
o
a
c
h
R
e
c
o
g
n
i
t
i
o
n
[
6
]
C
o
n
v
o
l
u
t
i
o
n
n
e
u
r
a
l
n
e
t
w
o
r
k
s w
i
t
h
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
8
2
.
6
1
%
[
7
]
C
o
n
v
o
l
u
t
i
o
n
n
e
u
r
a
l
n
e
t
w
o
r
k
s
9
0
.
9
0
%
[
8
]
T
e
mp
l
a
t
e
mat
c
h
i
n
g
a
n
d
A
r
t
i
f
i
c
i
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
7
2
.
5
%
a
n
d
7
5
.
8
%
r
e
sp
e
c
t
i
v
e
l
y
[
9
]
K
-
n
e
a
r
e
st
n
e
i
g
h
b
o
r
s
9
3
.
7
5
[
1
0
]
Y
o
l
o
v
.
3
9
1
%
[
1
1
]
T
e
mp
l
a
t
e
mat
c
h
i
n
g
[
1
2
]
C
r
o
ss a
n
d
p
h
a
se
c
o
r
r
e
l
a
t
i
o
n
6
7
.
9
8
%
a
n
d
6
3
.
4
6
%
se
q
u
e
n
t
i
a
l
l
y
[
1
3
]
P
r
o
b
a
b
i
l
i
st
i
c
n
e
u
r
a
l
n
e
t
w
o
r
k
s
9
6
.
5
%
[
1
4
]
C
o
n
v
o
l
u
t
i
o
n
n
e
u
r
a
l
n
e
t
w
o
r
k
s
(
7
0
-
88)
%
[
1
5
]
A
r
t
i
f
i
c
i
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
7
8
%
T
h
e
p
r
o
p
o
se
d
mu
l
t
i
c
l
a
ss
i
f
i
c
a
t
i
o
n
b
y
C
N
N
9
6
.
6
6
%
T
ab
le
3
.
T
h
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
p
r
o
p
o
s
ed
C
NN
m
o
d
el
f
o
r
class
i
f
icatio
n
o
f
m
u
l
tin
a
tio
n
al
l
icen
s
e
p
late
T
h
e
t
y
p
e
o
f
l
i
c
e
n
se
p
l
a
t
e
T
o
t
a
l
t
e
st
i
mag
e
s
N
o
.
o
f
c
o
r
r
e
c
t
e
d
c
l
a
ssi
f
i
e
d
l
i
c
e
n
se
p
l
a
t
e
N
o
.
o
f
n
o
t
c
o
r
r
e
c
t
e
d
c
l
a
ssi
f
i
e
d
l
i
c
e
n
se
p
l
a
t
e
L
i
c
e
n
se
p
l
a
t
e
c
l
a
ssi
f
i
c
a
t
i
o
n
p
e
r
c
e
n
t
a
g
e
A
r
m
e
n
i
a
20
20
N
o
n
e
(
2
0
/
2
0
)
×
1
0
0
%=
1
0
0
%
B
e
l
a
r
u
s
20
19
1
(
1
9
/
2
0
)
×
1
0
0
%=
9
5
%
H
u
n
g
a
r
y
20
19
1
(
1
9
/
2
0
)
×
1
0
0
%=
9
5
%
T
o
t
a
l
80
78
2
9
6
.
6
6
%
4.
CO
NCLU
SI
O
N
An
e
f
f
ec
ti
v
e
ap
p
r
o
ac
h
f
o
r
m
u
lti
-
n
a
tio
n
alit
y
v
e
h
icle
p
la
te
class
if
ica
tio
n
b
ased
o
n
C
NNs
w
as
p
r
o
p
o
s
ed
.
I
t
f
o
cu
s
ed
o
n
th
e
p
r
ep
r
o
ce
s
s
in
g
f
o
r
th
e
p
late
(
d
a
ta
au
g
m
e
n
tatio
n
t
h
at
i
n
cr
ea
s
e
d
th
e
d
ataset)
an
d
r
eg
u
lar
izatio
n
u
s
i
n
g
(
b
atch
n
o
r
m
aliza
tio
n
an
d
d
r
o
p
o
u
t
la
y
er
s
)
to
eli
m
i
n
ate
o
v
er
f
itti
n
g
a
n
d
m
a
k
e
g
en
er
ali
t
y
to
th
e
m
o
d
el.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
co
n
s
i
s
ted
o
f
th
r
e
e
s
ta
g
es:
p
r
ep
r
o
ce
s
s
in
g
,
th
e
C
NN
m
o
d
el
ar
ch
itect
u
r
e,
a
n
d
p
r
ed
ictio
n
o
f
th
e
clas
s
o
f
l
icen
s
e.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
h
ad
b
ee
n
p
r
o
v
ed
as a
n
e
f
f
icie
n
t
m
e
th
o
d
in
p
er
f
o
r
m
i
n
g
th
e
p
late
clas
s
i
f
icatio
n
o
f
t
h
e
m
u
lti
n
atio
n
al
l
y
-
lice
n
s
ed
p
l
ate
f
o
r
th
e
t
h
r
ee
co
u
n
tr
ies
(
A
r
m
en
ia,
B
elar
u
s
,
Hu
n
g
ar
y
)
w
h
er
e
th
e
ac
cu
r
ac
ie
s
o
f
tr
ain
in
g
an
d
v
alid
atio
n
s
e
ts
ar
e
9
9
.
1
7
%
an
d
9
7
.
5
0
%,
r
e
s
p
ec
tiv
el
y
,
w
ith
t
h
e
o
v
er
all
class
i
f
icat
io
n
ac
cu
r
ac
y
is
9
6
.
6
6
%
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
Mu
lticla
s
s
i
fica
tio
n
o
f lice
n
s
e
p
la
te
b
a
s
ed
o
n
d
ee
p
co
n
vo
lu
ti
o
n
n
eu
r
a
l
n
etw
o
r
ks (
Ma
s
a
r
A
b
ed
Uth
a
ib
)
5275
RE
F
E
R
E
NC
E
S
[1
]
L
.
A
lzu
b
a
id
i,
M
.
A
.
F
a
d
h
e
l,
O.
A
l
-
S
h
a
m
m
a
,
J.
Zh
a
n
g
,
a
n
d
Y.
D
u
a
n
,
“
De
e
p
L
e
a
rn
i
n
g
M
o
d
e
ls
f
o
r
Clas
sif
ic
a
ti
o
n
o
f
Re
d
Blo
o
d
Ce
ll
s
i
n
M
icro
sc
o
p
y
Im
a
g
e
s
to
A
id
in
S
ick
le
Ce
ll
A
n
e
m
ia
Dia
g
n
o
sis,”
El
e
c
tro
n
ics
,
v
o
l.
9
,
n
o
.
3
,
2
0
2
0
,
A
rt.
n
o
.
4
27
,
d
o
i.
1
0
.
3
3
9
0
/ele
c
tro
n
ics
9
0
3
0
4
2
7
.
[2
]
L
.
A
lzu
b
a
id
i
e
t
a
l.
,
“
No
v
e
l
T
ra
n
sf
e
r
L
e
a
rn
in
g
A
p
p
ro
a
c
h
f
o
r
M
e
d
ica
l
Im
a
g
in
g
w
it
h
L
i
m
it
e
d
L
a
b
e
led
Da
ta
,”
Ca
n
c
e
rs
,
v
o
l.
13
,
n
o
.
7
,
2
0
2
1
,
A
rt.
n
o
.
1
5
9
0
d
o
i
:
1
0
.
3
3
9
0
/ca
n
c
e
rs1
3
0
7
1
5
9
0
.
[3
]
M
.
N.
Ya
sir
a
n
d
M
.
S
.
Cro
o
c
k
,
“
S
o
f
twa
re
e
n
g
in
e
e
rin
g
b
a
se
d
se
l
f
-
c
h
e
c
k
in
g
p
ro
c
e
ss
f
o
r
c
y
b
e
r
s
e
c
u
rit
y
s
y
st
e
m
in
V
A
NET
,
”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
El
e
c
trica
l
a
n
d
Co
mp
u
ter
En
g
i
n
e
e
rin
g
(
IJ
ECE
)
,
v
o
l.
1
0
,
n
o
.
6
,
p
p
.
5
8
4
4
–
5
8
5
2
,
2
0
2
0
,
d
o
i:
1
0
.
1
1
5
9
1
/i
jec
e
.
v
1
0
i
6
.
p
p
5
8
4
4
-
5
8
5
2
.
[4
]
M
.
N.
Ya
sir
a
n
d
M
.
S
.
Cro
o
c
k
,
“
C
y
b
e
r
Do
S
a
tt
a
c
k
-
b
a
s
e
d
se
c
u
rit
y
sim
u
lato
r
f
o
r
V
A
NE
T
,
”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
El
e
c
trica
l
a
n
d
Co
m
p
u
ter
E
n
g
i
n
e
e
rin
g
(
IJ
ECE
),
v
o
l.
1
0
,
n
o
.
6
,
p
p
.
5
8
3
2
–
5
8
4
3
,
2
0
2
0
,
d
o
i:
1
0
.
1
1
5
9
1
/i
jec
e
.
v
1
0
i
6
.
p
p
5
8
3
2
-
5
8
4
3
.
[5
]
L
.
A
lzu
b
a
id
i,
O.
A
l
-
S
h
a
m
m
a
,
M
.
A
.
F
a
d
h
e
l,
L
.
F
a
rh
a
n
,
J.
Zh
a
n
g
,
a
n
d
Y.
Du
a
n
,
“
Op
ti
m
izin
g
th
e
P
e
rf
o
rm
a
n
c
e
o
f
Bre
a
st
Ca
n
c
e
r
Clas
si
f
ica
ti
o
n
b
y
Em
p
lo
y
in
g
th
e
S
a
m
e
Do
m
a
in
T
r
a
n
sf
e
r
Lea
rn
in
g
f
ro
m
H
y
b
rid
De
e
p
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
tw
o
rk
M
o
d
e
l,
”
El
e
c
tro
n
ics
,
v
o
l.
9
,
n
o
.
3
,
2
0
2
0
,
A
rt.
n
o
.
4
4
5
,
d
o
i:
1
0
.
3
3
9
0
/ele
c
tro
n
ics
9
0
3
0
4
4
5
.
[6
]
L
.
A
lzu
b
a
id
i
e
t
a
l
.
,
“
T
o
w
a
rd
s
a
Be
tt
e
r
Un
d
e
rsta
n
d
i
n
g
o
f
T
ra
n
sfe
r
L
e
a
rn
in
g
f
o
r
M
e
d
ica
l
I
m
a
g
in
g
:
A
C
a
se
S
tu
d
y
,
”
Ap
p
l
.
S
c
i,
v
o
l.
1
0
,
n
o
.
1
3
,
2
0
2
0
,
A
rt.
n
o
.
4
5
2
3
,
d
o
i:
1
0
.
3
3
9
0
/ap
p
1
0
1
3
4
5
2
3
.
[7
]
L
.
A
lzu
b
a
id
i
e
t
a
l.
,
“
Re
v
ie
w
o
f
d
e
e
p
lea
rn
in
g
:
c
o
n
c
e
p
ts,
CNN
a
rc
h
it
e
c
tu
re
s,
c
h
a
ll
e
n
g
e
s,
a
p
p
li
c
a
ti
o
n
s,
f
u
t
u
re
d
irec
ti
o
n
s
,”
J
o
u
rn
a
l
o
f
B
ig
D
a
ta
,
v
o
l.
8
,
n
o
.
1,
pp.
1
–
74
,
2
0
2
1
,
d
o
i:
1
0
.
1
1
8
6
/s
4
0
5
3
7
-
0
2
1
-
0
0
4
4
4
-
8
.
[8
]
J.
A
.
C.
Jo
se
e
t
a
l
.
,
“
Ca
teg
o
rizin
g
L
ice
n
se
P
late
s
Us
in
g
Co
n
v
o
lu
ti
o
n
a
l
Ne
u
ra
l
Ne
tw
o
rk
w
it
h
Re
sid
u
a
l
L
e
a
rn
in
g
,
”
i
n
2
0
1
9
4
t
h
Asia
-
P
a
c
if
ic
Co
n
f
e
re
n
c
e
o
n
In
telli
g
e
n
t
R
o
b
o
t
S
y
ste
ms
(
ACIR
S
)
,
2
0
1
9
,
p
p
.
2
3
1
–
2
3
4
,
1
0
.
1
1
0
9
/A
CIRS
.
2
0
1
9
.
8
9
3
5
9
9
7
.
[9
]
M
.
A
ti
k
u
z
z
a
m
a
n
,
M
.
A
s
a
d
u
z
z
a
m
a
n
,
a
n
d
M
.
Z.
Isla
m
,
“
V
e
h
icle
N
u
m
b
e
r
P
late
De
tec
ti
o
n
a
n
d
Ca
teg
o
riza
ti
o
n
Us
in
g
CNN
s,”
in
2
0
1
9
I
n
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
S
u
sta
in
a
b
le
T
e
c
h
n
o
lo
g
ies
f
o
r
In
d
u
stry
4
.
0
(
S
T
I)
,
2
0
1
9
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
1
1
0
9
/
S
T
I4
7
6
7
3
.
2
0
1
9
.
9
0
6
8
0
4
9
.
[1
0
]
J.
W
a
n
g
,
B.
B
a
c
ic,
a
n
d
W
.
Q.
Y
a
n
,
“
A
n
e
ff
e
c
ti
v
e
m
e
th
o
d
f
o
r
p
late
n
u
m
b
e
r
re
c
o
g
n
it
io
n
,
”
M
u
lt
ime
d
.
T
o
o
ls
Ap
p
l.
,
v
o
l.
7
7
,
n
o
.
2
,
p
p
.
1
6
7
9
–
1
6
9
2
,
2
0
1
8
,
d
o
i:
1
0
.
1
0
0
7
/s
1
1
0
4
2
-
0
17
-
4
3
5
6
-
z
.
[1
1
]
M
.
R.
Hid
a
y
a
h
,
I.
Ak
h
li
s,
a
n
d
E.
S
u
g
ih
a
rti
,
“
Re
c
o
g
n
it
i
o
n
N
u
m
b
e
r
o
f
T
h
e
V
e
h
icle
P
late
Us
in
g
Otsu
M
e
th
o
d
a
n
d
K
-
Ne
a
re
st Ne
i
g
h
b
o
u
r
Clas
sif
ica
ti
o
n
,
”
S
c
i.
J
.
I
n
fo
rm
a
ti
c
s
,
v
o
l.
4
,
n
o
.
1
,
p
p
.
6
6
–
7
5
,
2
0
1
7
,
d
o
i
:
1
0
.
1
5
2
9
4
/sji
.
v
4
i1
.
9
5
0
3
.
[1
2
]
J.
W
a
n
g
,
X
.
L
iu
,
A
.
L
iu
,
a
n
d
J.
Xia
o
,
“
A
d
e
e
p
le
a
rn
in
g
-
b
a
se
d
m
e
th
o
d
f
o
r
v
e
h
icle
li
c
e
n
se
p
late
re
c
o
g
n
it
io
n
i
n
n
a
t
u
ra
l
sc
e
n
e
,
”
AP
S
IPA
T
r
a
n
s.
S
i
g
n
a
l
I
n
f
.
Pro
c
e
ss
.
,
v
o
l.
8
,
2
0
1
9
,
d
o
i:
1
0
.
1
0
1
7
/A
T
S
IP
.
2
0
1
9
.
8
.
[1
3
]
F
.
P
a
tel,
J.
S
o
lan
k
i,
V.
Ra
jg
u
r
u
,
a
n
d
A
.
S
a
x
e
n
a
,
“
Re
c
o
g
n
it
io
n
o
f
v
e
h
icle
n
u
m
b
e
r
p
late
u
sin
g
im
a
g
e
p
ro
c
e
ss
in
g
tec
h
n
iq
u
e
,
”
Co
n
tro
l
S
y
st.
E
n
g
,
v
o
l.
2
,
n
o
.
1
,
p
p
.
1
–
7
,
2
0
1
8
.
[1
4
]
G
.
S
h
a
r
m
a
,
“
P
e
rf
o
r
m
a
n
c
e
a
n
a
l
y
sis
o
f
v
e
h
icle
n
u
m
b
e
r
p
late
re
c
o
g
n
it
io
n
sy
ste
m
u
sin
g
te
m
p
late
m
a
tch
in
g
tec
h
n
iq
u
e
s,”
J
.
In
f
.
T
e
c
h
n
o
l.
S
o
ft
w.
En
g
.
,
v
o
l.
8
,
n
o
.
2
,
p
p
.
1
0
–
4
1
7
2
,
2
0
1
8
,
d
o
i:
1
0
.
4
1
7
2
/
2
1
6
5
-
7
8
6
6
.
1
0
0
0
2
3
2
.
[1
5
]
F
.
Öz
tü
rk
a
n
d
F
.
Öz
e
n
,
“
A
n
e
w
li
c
e
n
se
p
late
r
e
c
o
g
n
it
io
n
sy
ste
m
b
a
se
d
o
n
p
ro
b
a
b
il
isti
c
n
e
u
ra
l
n
e
tw
o
rk
s,”
Pro
c
e
d
ia
T
e
c
h
n
o
l
.
,
v
o
l
.
1
,
p
p
.
1
2
4
–
1
2
8
,
2
0
1
2
,
d
o
i:
1
0
.
1
0
1
6
/
j.
p
r
o
tcy
.
2
0
1
2
.
0
2
.
0
2
4
.
[1
6
]
M
.
M
.
S
.
Ra
h
m
a
n
,
M
.
M
o
sta
k
im
,
M
.
S
.
Na
srin
,
a
n
d
M
.
Z.
A
lo
m
,
“
Ba
n
g
la
L
i
c
e
n
se
P
late
Re
c
o
g
n
it
io
n
Us
i
n
g
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
t
w
o
rk
s
(CNN
),
”
in
2
0
1
9
2
2
n
d
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Co
mp
u
ter
a
n
d
In
f
o
rm
a
ti
o
n
T
e
c
h
n
o
l
o
g
y
(
ICCIT
)
,
2
0
1
9
,
p
p
.
1
–
6
,
d
o
i:
1
0
.
1
1
0
9
/ICCIT
4
8
8
8
5
.
2
0
1
9
.
9
0
3
8
5
9
7
.
[1
7
]
R.
N.
Ba
b
u
,
V
.
S
o
w
m
y
a
,
a
n
d
K.
P
.
S
o
m
a
n
,
“
In
d
ia
n
Ca
r
Nu
m
b
e
r
P
late
Re
c
o
g
n
it
io
n
u
sin
g
De
e
p
L
e
a
r
n
in
g
,
”
in
2
0
1
9
2
n
d
I
n
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
In
telli
g
e
n
t
Co
m
p
u
ti
n
g
,
In
stru
me
n
ta
ti
o
n
a
n
d
Co
n
tro
l
T
e
c
h
n
o
l
o
g
ies
,
2
0
1
9
,
d
o
i:
1
0
.
1
1
0
9
/ICICICT
4
6
0
0
8
.
2
0
1
9
.
8
9
9
3
2
3
8
.
[1
8
]
A
.
M
i
k
o
łajc
z
y
k
a
n
d
M
.
G
ro
c
h
o
w
s
k
i,
“
Da
t
a
a
u
g
m
e
n
tatio
n
f
o
r
im
p
ro
v
in
g
d
e
e
p
lea
rn
in
g
in
i
m
a
g
e
c
la
ss
i
f
ica
ti
o
n
p
ro
b
lem
,
”
in
2
0
1
8
in
ter
n
a
ti
o
n
a
l
in
ter
d
isc
ip
li
n
a
ry
Ph
D
wo
rk
sh
o
p
(
IIP
h
DW
)
,
2
0
1
8
,
p
p
.
1
1
7
–
1
2
2
,
1
0
.
1
1
0
9
/II
P
HD
W
.
2
0
1
8
.
8
3
8
8
3
3
8
.
[1
9
]
A
.
A
sp
e
rti
a
n
d
C.
M
a
stro
n
a
r
d
o
,
“
T
h
e
e
ff
e
c
ti
v
e
n
e
ss
o
f
d
a
ta
a
u
g
m
e
n
tatio
n
f
o
r
d
e
tec
ti
o
n
o
f
g
a
stro
in
tes
ti
n
a
l
d
ise
a
se
s
f
ro
m
e
n
d
o
sc
o
p
ica
l
im
a
g
e
s,”
a
rXiv P
re
p
r.
a
rXiv1
7
1
2
.
0
3
6
8
9
,
2
0
1
7
,
d
o
i:
1
0
.
5
2
2
0
/0
0
0
6
7
3
0
9
0
1
9
9
0
2
0
5
.
[2
0
]
W
.
L
i,
C.
Ch
e
n
,
M
.
Zh
a
n
g
,
H.
L
i,
a
n
d
Q.
Du
,
“
Da
ta
a
u
g
m
e
n
tatio
n
f
o
r
h
y
p
e
rsp
e
c
tral
i
m
a
g
e
c
las
si
f
ic
a
ti
o
n
w
it
h
d
e
e
p
CNN
,
”
IEE
E
Ge
o
sc
i.
Rem
o
te
S
e
n
s.
L
e
tt
.
,
v
o
l.
1
6
,
n
o
.
4
,
p
p
.
5
9
3
–
5
9
7
,
2
0
1
8
,
d
o
i:
1
0
.
1
1
0
9
/L
G
RS
.
2
0
1
8
.
2
8
7
8
7
7
3
.
[2
1
]
C.
S
h
o
rten
a
n
d
T
.
M
.
K
h
o
s
h
g
o
f
taa
r,
“
A
su
rv
e
y
o
n
im
a
g
e
d
a
ta
a
u
g
m
e
n
tatio
n
f
o
r
d
e
e
p
lea
rn
in
g
,
”
J
.
Bi
g
Da
ta
,
v
o
l.
6
,
n
o
.
1
,
p
.
6
0
,
2
0
1
9
,
d
o
i:
1
0
.
1
1
8
6
/s
4
0
5
3
7
-
0
1
9
-
0
1
9
7
-
0
.
[2
2
]
T.
C.
P
h
a
m
,
C.
M
.
L
u
o
n
g
,
M
.
V
isa
n
i,
a
n
d
V
.
D.
H
o
a
n
g
,
“
De
e
p
CNN
a
n
d
d
a
ta
a
u
g
m
e
n
tatio
n
f
o
r
sk
in
les
io
n
c
las
si
f
ica
ti
o
n
,
”
in
Asia
n
Co
n
fer
e
n
c
e
o
n
In
telli
g
e
n
t
In
f
o
rm
a
ti
o
n
a
n
d
Da
t
a
b
a
se
S
y
ste
ms
,
2
0
1
8
,
p
p
.
5
7
3
–
5
8
2
,
d
o
i:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
3
1
9
-
7
5
4
2
0
-
8
_
5
4
.
[2
3
]
S
.
A
lb
a
w
i,
T
.
A
.
M
o
h
a
m
m
e
d
,
a
n
d
S
.
A
l
-
Za
w
i,
“
Un
d
e
rsta
n
d
i
n
g
o
f
a
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
t
w
o
rk
,
”
in
2
0
1
7
In
ter
n
a
t
io
n
a
l
Co
n
fer
e
n
c
e
o
n
En
g
in
e
e
rin
g
a
n
d
T
e
c
h
n
o
l
o
g
y
(
ICET
)
,
2
0
1
7
,
p
p
.
1
–
6
,
d
o
i:
1
0
.
1
1
0
9
/IC
En
g
T
e
c
h
n
o
l.
2
0
1
7
.
8
3
0
8
1
8
6
.
[2
4
]
F
.
G
a
o
,
T
.
Hu
a
n
g
,
J.
W
a
n
g
,
J.
S
u
n
,
A
.
Hu
ss
a
in
,
a
n
d
E
.
Ya
n
g
,
“
Du
a
l
-
b
ra
n
c
h
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
n
e
u
ra
l
n
e
tw
o
rk
f
o
r
p
o
larim
e
tri
c
S
A
R
i
m
a
g
e
c
la
ss
i
f
ic
a
ti
o
n
,
”
A
p
p
l.
S
c
i.
,
v
o
l
.
7
,
n
o
.
5
,
p
.
4
4
7
,
2
0
1
7
,
d
o
i:
1
0
.
3
3
9
0
/a
p
p
7
0
5
0
4
4
7
.
[2
5
]
A
.
A
.
M
.
A
l
-
S
a
ff
a
r,
H.
Tao
,
a
n
d
M
.
A
.
T
a
lab
,
“
Re
v
ie
w
o
f
d
e
e
p
c
o
n
v
o
lu
ti
o
n
n
e
u
ra
l
n
e
tw
o
rk
in
im
a
g
e
c
las
si
f
ica
ti
o
n
,
”
in
2
0
1
7
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Ra
d
a
r,
An
ten
n
a
,
M
icr
o
wa
v
e
,
El
e
c
tro
n
ics
,
a
n
d
T
e
lec
o
mm
u
n
ic
a
ti
o
n
s
(
ICRA
M
ET
)
,
2
0
1
7
,
p
p
.
2
6
–
31
,
d
o
i:
1
0
.
1
1
0
9
/ICRA
M
ET
.
2
0
1
7
.
8
2
5
3
1
3
9
.
[2
6
]
H.
H.
S
u
lt
a
n
,
N.
M
.
S
a
lem
,
a
n
d
W
.
A
l
-
A
tab
a
n
y
,
“
M
u
lt
i
-
c
las
sif
ic
a
ti
o
n
o
f
b
ra
in
t
u
m
o
r
im
a
g
e
s
u
sin
g
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
,
”
IEE
E
Acc
e
ss
,
v
o
l.
7
,
p
p
.
6
9
2
1
5
–
6
9
2
2
5
,
2
0
1
9
,
d
o
i:
1
0
.
1
1
0
9
/A
CCES
S
.
2
0
1
9
.
2
9
1
9
1
2
2
.
[2
7
]
N.
Bjo
rc
k
,
C.
P
.
G
o
m
e
s,
B.
S
e
lma
n
,
a
n
d
K.
Q.
W
e
in
b
e
rg
e
r,
“
Un
d
e
rsta
n
d
in
g
b
a
tch
n
o
rm
a
li
z
a
ti
o
n
,
”
in
A
d
v
a
n
c
e
s
i
n
Evaluation Warning : The document was created with Spire.PDF for Python.