I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
5
,
Octo
b
er
20
25
,
p
p
.
4
6
9
2
~
4
7
0
4
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
5
.
pp
4
6
9
2
-
4
7
0
4
4692
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Bo
ne
-
Net
:
a
para
l
lel deep
conv
o
luti
o
na
l neural net
wo
rk
-
ba
sed
bo
ne f
ra
ct
ure r
ec
o
g
nition
M
d.
H
a
s
a
n Im
a
m
B
ij
o
y
1,
2
,
Nus
ra
t
I
s
la
m
K
o
hin
o
o
r
3
,
Sy
eda
Z
a
rin Ta
s
nim
3
,
Md
.
Sa
id
ur
Ra
hm
a
n K
o
hin
o
o
r
3,
4
1
H
e
a
l
t
h
I
n
f
o
r
ma
t
i
c
s
R
e
se
a
r
c
h
La
b
(
H
I
R
L)
,
D
a
f
f
o
d
i
l
I
n
t
e
r
n
a
t
i
o
n
a
l
U
n
i
v
e
r
si
t
y
,
D
h
a
k
a
,
B
a
n
g
l
a
d
e
sh
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
D
a
f
f
o
d
i
l
I
n
t
e
r
n
a
t
i
o
n
a
l
U
n
i
v
e
r
si
t
y
,
D
h
a
k
a
,
B
a
n
g
l
a
d
e
s
h
3
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
L
e
a
d
i
n
g
U
n
i
v
e
r
si
t
y
,
S
y
l
h
e
t
,
B
a
n
g
l
a
d
e
sh
4
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
o
n
a
n
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
K
i
n
g
F
a
h
d
U
n
i
v
e
r
si
t
y
o
f
P
e
t
r
o
l
e
u
m
a
n
d
M
i
n
e
r
a
l
s,
D
h
a
h
r
a
n
,
S
a
u
d
i
A
r
a
b
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
2
3
,
2
0
2
4
R
ev
is
ed
Ma
y
2
5
,
2
0
2
5
Acc
ep
ted
J
u
n
3
0
,
2
0
2
5
M
a
n
y
p
e
o
p
le
su
ffe
r
fro
m
b
o
n
e
fra
c
tu
re
s,
wh
ich
c
a
n
re
su
lt
fr
o
m
m
in
o
r
a
c
c
id
e
n
ts,
fo
rc
e
fu
l
b
lo
ws
,
o
r
e
v
e
n
d
ise
a
se
s
li
k
e
o
ste
o
p
o
r
o
sis
o
r
b
o
n
e
c
a
n
c
e
r.
In
th
e
m
e
d
ica
l
re
a
lm,
a
c
c
u
ra
tel
y
id
e
n
ti
fy
i
n
g
b
o
n
e
fra
c
tu
re
s
fro
m
X
-
ra
y
ima
g
e
s
is
p
a
ra
m
o
u
n
t
fo
r
e
ffe
c
ti
v
e
d
iag
n
o
sis
a
n
d
trea
tme
n
t.
T
o
a
d
d
r
e
ss
th
is,
a
c
o
m
p
a
ra
ti
v
e
stu
d
y
is
c
o
n
d
u
c
ted
u
ti
li
z
i
n
g
t
h
re
e
d
isti
n
c
t
m
o
d
e
ls:
a
trad
it
io
n
a
l
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
(CNN
),
M
o
b
il
e
Ne
t
-
V2
,
a
n
d
a
n
e
wl
y
d
e
v
e
lo
p
e
d
p
a
ra
ll
e
l
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
(P
DC
NN
).
Th
e
p
rima
ry
a
im
is
to
e
v
a
lu
a
te
a
n
d
c
o
n
tras
t
t
h
e
se
m
o
d
e
ls
in
term
s
o
f
p
re
c
isio
n
,
se
n
siti
v
it
y
,
a
n
d
sp
e
c
ifi
c
it
y
fo
r
d
i
a
g
n
o
si
n
g
b
o
n
e
fra
c
tu
re
s.
X
-
ra
y
i
m
a
g
e
s
o
f
fra
c
tu
re
d
a
n
d
n
o
n
-
fra
c
tu
re
d
b
o
n
e
s
a
re
so
u
rc
e
d
fro
m
Ka
g
g
le
a
n
d
s
u
b
jec
ted
to
v
a
rio
u
s
ima
g
e
p
ro
c
e
ss
in
g
tec
h
n
i
q
u
e
s
t
o
re
c
ti
fy
a
n
o
m
a
li
e
s.
Tec
h
n
iq
u
e
s
s
u
c
h
a
s
c
ro
p
p
in
g
,
re
siz
in
g
,
c
o
n
tras
t
e
n
h
a
n
c
e
m
e
n
t,
fil
terin
g
,
a
n
d
a
u
g
m
e
n
tatio
n
a
re
a
p
p
li
e
d
,
c
u
lmi
n
a
ti
n
g
i
n
c
a
n
n
y
e
d
g
e
d
e
tec
ti
o
n
.
Th
e
se
p
ro
c
e
ss
e
d
i
m
a
g
e
s
a
re
th
e
n
u
se
d
t
o
train
a
n
d
tes
t
m
o
d
e
ls
.
Th
e
re
su
lt
s
sh
o
wc
a
se
d
th
e
su
p
e
rio
r
p
e
rfo
rm
a
n
c
e
o
f
th
e
n
e
wly
d
e
v
e
lo
p
e
d
P
DCN
N
m
o
d
e
l,
a
c
h
i
e
v
in
g
a
n
imp
re
ss
iv
e
a
c
c
u
ra
c
y
o
f
9
2
.
8
9
%
,
su
rp
a
ss
in
g
b
o
th
th
e
trad
it
io
n
a
l
CNN
a
n
d
p
re
train
e
d
M
o
b
i
leN
e
t
-
V2
m
o
d
e
ls
.
A
se
ries
o
f
a
b
latio
n
st
u
d
ies
a
re
c
o
n
d
u
c
ted
to
fi
n
e
-
tu
n
e
th
e
h
y
p
e
rp
a
ra
m
e
ters
o
f
t
h
e
P
DCN
N
m
o
d
e
l
,
f
u
rth
e
r
v
a
l
id
a
ti
n
g
it
s
e
ffica
c
y
.
Th
ro
u
g
h
o
u
t
th
e
in
v
e
stig
a
ti
o
n
,
P
DCN
N
c
o
n
siste
n
tl
y
o
u
t
p
e
rfo
rm
e
d
M
o
b
i
leN
e
t
-
V2
a
n
d
trad
it
i
o
n
a
l
CNN
,
u
n
d
e
rsc
o
rin
g
it
s
p
o
ten
t
i
a
l
a
s
a
n
a
d
v
a
n
c
e
d
t
o
o
l
fo
r
stre
a
m
li
n
i
n
g
b
o
n
e
fra
c
tu
re
id
e
n
ti
fica
ti
o
n
.
K
ey
w
o
r
d
s
:
B
o
n
e
f
r
ac
tu
r
e
C
an
n
y
ed
g
e
d
etec
tio
n
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Mo
b
ileNet
-
V2
Par
allel
d
ee
p
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
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
:
Md
.
Hasan
I
m
am
B
ijo
y
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
Daf
f
o
d
il
I
n
ter
n
atio
n
al
Un
iv
er
s
ity
Daf
f
o
d
il Sm
ar
t Ci
ty
,
Dh
ak
a
-
1
2
1
6
,
B
an
g
la
d
esh
E
m
ail: h
asan
1
5
-
1
1
7
4
3
@
d
iu
.
e
d
u
.
b
d
1.
I
NT
RO
D
UCT
I
O
N
A
m
in
er
al
m
ad
e
o
f
ca
lciu
m
is
ca
lled
b
o
n
e
[
1
]
.
Hu
m
an
b
o
n
e
p
r
o
v
i
d
es
a
m
ec
h
an
ical
s
tr
u
ct
u
r
e
f
o
r
th
e
h
u
m
an
b
o
d
y
[
2
]
.
I
t
m
ain
tai
n
s
ca
lciu
m
h
o
m
eo
s
tasi
s
an
d
s
u
b
s
tan
tial
s
to
r
ag
e
o
f
p
h
o
s
p
h
ate,
m
ag
n
esiu
m
,
p
o
tass
iu
m
,
an
d
b
icar
b
o
n
ate.
Hu
m
an
b
o
n
e
also
h
elp
s
in
m
u
s
cle
ac
tiv
ities
[
3
]
.
On
e
o
f
th
e
m
o
s
t
in
ter
esti
n
g
n
atu
r
al
co
m
p
o
s
ite
m
ater
ials
is
h
u
m
an
co
r
tical
b
o
n
e,
wh
ic
h
ac
ts
as
a
s
tr
u
ctu
r
al
f
o
u
n
d
atio
n
th
at
ca
n
with
s
tan
d
in
ju
r
y
a
n
d
allo
ws
th
e
h
u
m
a
n
b
o
d
y
to
s
elf
-
r
ep
air
.
T
h
e
m
ec
h
an
ical
ch
ar
ac
te
r
is
tics
o
f
b
o
n
e
en
ab
le
it
to
with
s
tan
d
f
r
ac
tu
r
e
u
n
d
er
d
iv
er
s
e
p
h
y
s
io
lo
g
ical
lo
ad
in
g
co
n
d
itio
n
s
[
4
]
.
A
f
ew
o
f
th
ese
co
n
d
itio
n
s
th
at
lead
to
b
o
n
e
d
is
ea
s
es
ar
e
v
itam
in
-
D
in
s
u
f
f
icien
cy
,
o
s
teo
p
o
r
o
s
is
,
Pag
et’
s
d
is
ea
s
e
o
f
th
e
b
o
n
es,
an
d
m
an
y
m
o
r
e.
T
h
e
am
o
u
n
t
an
d
q
u
ality
o
f
b
o
n
e
tis
s
u
e
in
th
e
h
u
m
a
n
s
k
eleto
n
s
tar
t
to
d
ec
lin
e
a
f
ter
r
ea
c
h
in
g
p
e
ak
b
o
n
e
m
ass
in
th
e
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:
2088
-
8
7
0
8
B
o
n
e
-
N
et:
a
p
a
r
a
llel d
ee
p
c
o
n
vo
lu
tio
n
a
l n
eu
r
a
l n
etw
o
r
k
-
b
a
s
ed
b
o
n
e
…
(
Md
.
Ha
s
a
n
I
ma
m
B
ijo
y
)
4693
th
ir
d
o
r
f
o
u
r
th
d
ec
ad
e
o
f
life
[
5
]
.
Am
o
n
g
th
e
m
o
s
t
co
m
m
o
n
in
ju
r
ies
in
to
d
ay
’
s
wo
r
ld
ar
e
b
o
n
e
f
r
ac
tu
r
es.
Ab
o
u
t
8
.
9
m
illi
o
n
p
eo
p
le
ex
p
er
ien
ce
th
is
d
is
o
r
d
e
r
ea
ch
y
ea
r
[
6
]
,
an
d
th
e
c
o
n
s
eq
u
e
n
ce
s
o
f
leav
in
g
a
f
r
ac
t
u
r
e
u
n
tr
ea
ted
m
a
y
en
d
in
d
ea
th
o
r
p
er
m
an
en
t d
am
ag
e
b
y
s
ev
er
al
ty
p
es
o
f
b
o
n
e
f
r
ac
tu
r
e
as
s
h
o
w
n
in
Fig
u
r
e
1
.
T
h
e
d
o
cto
r
s
b
ea
r
a
g
r
ea
t
d
ea
l
o
f
r
esp
o
n
s
ib
ilit
y
f
o
r
th
is
,
as
th
ey
ass
ess
lo
ts
o
f
X
-
r
ay
im
ag
e
s
d
aily
b
asis
.
T
h
e
m
ajo
r
ity
o
f
th
e
tech
n
o
lo
g
y
u
s
ed
f
o
r
t
h
e
in
itial
d
iag
n
o
s
is
is
X
-
r
ay
s
,
a
m
o
d
ality
th
at
h
as
b
e
en
ar
o
u
n
d
f
o
r
m
o
r
e
th
an
a
ce
n
tu
r
y
an
d
is
s
till
wid
e
ly
u
s
ed
.
Me
d
ical
p
r
o
f
ess
io
n
als
f
in
d
it
d
if
f
icu
lt
to
ass
ess
X
-
r
ay
im
ag
es f
o
r
m
an
y
r
ea
s
o
n
s
,
lik
e
th
e
f
ac
t
th
at
X
-
r
ay
s
m
ay
o
b
s
cu
r
e
ce
r
tain
u
n
i
q
u
e
ch
ar
ac
te
r
is
tics
o
f
th
e
b
o
n
e
o
r
th
at
it
tak
es
ex
ten
s
iv
e
tr
ain
in
g
to
ac
cu
r
ately
id
en
tify
th
e
v
a
r
io
u
s
k
in
d
s
o
f
f
r
ac
tu
r
es.
A
p
r
ec
is
e
class
if
ica
tio
n
o
f
th
e
f
r
ac
t
u
r
e
am
o
n
g
s
tan
d
a
r
d
ty
p
es is
cr
u
cial
f
o
r
b
o
th
th
e
f
u
tu
r
e
o
u
tlo
o
k
a
n
d
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
t
r
ea
tm
en
t
[
7
]
.
Fig
u
r
e
1
.
Sev
er
al
ty
p
es o
f
f
r
ac
tu
r
e
in
b
o
n
e
Ma
n
y
ad
v
an
ce
s
in
tech
n
o
l
o
g
y
ar
e
m
ak
in
g
it
ea
s
ier
to
d
iag
n
o
s
e
b
o
n
e
f
r
ac
tu
r
es
in
to
d
a
y
'
s
wo
r
ld
.
Au
to
m
ated
im
ag
e
p
r
o
ce
s
s
in
g
m
o
d
els,
s
u
ch
as
ar
tific
ial
in
te
llig
en
ce
(
AI
)
,
d
ee
p
lear
n
in
g
(
DL
)
,
an
d
m
ac
h
in
e
lear
n
in
g
(
ML
)
,
ar
e
q
u
ick
an
d
r
eliab
le
f
o
r
id
en
tific
atio
n
,
lo
c
aliza
tio
n
,
an
d
class
if
icatio
n
[
8
]
.
Dete
r
m
in
in
g
th
e
p
r
ec
is
e
lo
ca
tio
n
o
f
a
f
r
ac
tu
r
e
an
d
its
d
e
g
r
ee
o
f
im
p
ac
t
m
ig
h
t
b
e
b
en
ef
icial.
T
h
e
d
is
cip
lin
e
o
f
co
m
p
u
ter
-
ai
d
ed
d
iag
n
o
s
is
is
an
em
er
g
in
g
f
ield
o
f
r
esear
ch
wh
er
e
c
o
m
p
u
te
r
t
ec
h
n
o
lo
g
ies
ar
e
u
s
ed
to
o
f
f
e
r
p
r
o
m
p
t
a
n
d
p
r
ec
is
e
d
iag
n
o
s
is
.
I
t u
s
es X
-
r
ay
im
ag
e
s
an
d
p
r
ep
r
o
ce
s
s
es th
em
as n
ee
d
ed
to
d
etec
t f
r
ac
tu
r
es in
th
e
b
o
n
es
[
9
]
.
T
h
e
y
ar
e
g
ettin
g
b
etter
with
ea
c
h
p
ass
in
g
d
ay
as a
r
esu
lt o
f
lear
n
i
n
g
f
r
o
m
lab
eled
d
ata
[
1
0
]
.
Sev
er
al
wo
r
k
s
h
av
e
b
ee
n
d
is
cu
s
s
ed
,
in
clu
d
in
g
th
e
f
o
llo
wi
n
g
,
L
u
is
an
d
R
u
an
o
[
1
1
]
s
u
g
g
ested
a
co
m
p
u
ter
-
aid
ed
s
y
s
tem
f
o
r
b
o
n
e
f
r
ac
t
u
r
e
d
etec
tio
n
.
Alth
o
u
g
h
X
-
r
ay
p
ictu
r
es
ar
e
ty
p
icall
y
u
s
ed
to
d
iag
n
o
s
e
b
o
n
e
f
r
ac
tu
r
es,
a
n
d
s
u
g
g
ested
an
ap
p
r
o
ac
h
th
at
u
s
es
co
m
p
u
ter
s
to
h
elp
d
etec
t
b
o
n
e
f
r
a
ctu
r
es.
Ach
iev
ed
8
9
%
class
if
icatio
n
ac
cu
r
ac
y
b
y
u
tili
zin
g
a
v
ar
iety
o
f
tech
n
iq
u
es
f
o
r
f
r
ac
tu
r
e
r
ec
o
g
n
itio
n
,
b
o
n
e
lin
e
d
etec
tio
n
,
an
d
s
p
ec
k
le
r
ed
u
ctio
n
.
I
t
is
lim
ited
to
s
tr
ess
f
r
ac
tu
r
es,
a
n
d
th
is
co
m
p
u
te
r
-
aid
ed
ap
p
r
o
ac
h
is
u
n
ab
le
to
class
if
y
co
m
p
lex
f
r
ac
tu
r
es
ac
cu
r
ately
.
I
n
2
0
1
5
,
An
u
et
a
l.
[
1
2
]
p
r
e
s
en
ted
wo
r
k
u
s
in
g
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
es
to
r
ed
u
ce
n
o
is
e
f
r
o
m
th
e
p
h
o
to
s
an
d
tr
a
n
s
f
o
r
m
th
e
R
GB
X
-
r
ay
im
ag
es
to
g
r
a
y
s
ca
le
u
s
in
g
a
m
ed
ian
f
ilter
.
T
h
e
y
th
en
d
etec
ted
th
e
ed
g
e
o
f
th
e
i
m
ag
e
b
y
u
s
in
g
t
h
e
So
b
el
ed
g
e
d
etec
to
r
.
T
h
ey
u
s
ed
th
e
g
r
a
y
lev
el
co
-
o
cc
u
r
r
en
c
e
Ma
tr
ix
(
GL
C
M)
to
ex
tr
ac
t
t
h
e
f
ea
tu
r
e.
Fin
ally
,
th
e
d
ata
wer
e
d
iv
id
ed
in
to
f
r
ac
t
u
r
ed
an
d
n
o
n
-
f
r
ac
tu
r
ed
ca
teg
o
r
ies
u
s
in
g
a
v
ar
iety
o
f
c
lass
if
ier
ty
p
es,
in
clu
d
i
n
g
d
ec
is
io
n
tr
ee
s
(
DT
)
,
n
eu
r
al
n
etwo
r
k
s
(
NN)
,
a
n
d
m
eta
-
class
if
ier
s
.
T
h
er
e
ar
e
4
0
p
h
o
t
o
s
in
th
e
c
o
llectio
n
,
2
0
o
f
wh
ich
h
av
e
f
r
ac
tu
r
es
a
n
d
2
0
o
f
wh
ich
d
o
n
o
t.
T
h
ey
ac
h
iev
ed
th
e
b
est
ac
cu
r
ac
y
in
th
e
m
eta
-
class
if
icatio
n
alg
o
r
it
h
m
,
at
8
5
p
e
r
ce
n
t.
T
r
ip
ath
i
et
a
l.
[
1
3
]
s
tated
th
at
th
e
lo
ca
tio
n
o
f
s
m
all
o
r
h
air
li
n
e
f
em
u
r
f
r
ac
tu
r
es
is
th
e
m
ain
to
p
ic
o
f
th
is
p
ap
er
.
T
o
d
eter
m
in
e
wh
eth
er
o
r
n
o
t
th
er
e
is
a
f
r
ac
tu
r
e,
t
h
ey
em
p
l
o
y
ed
SVM.
3
0
X
-
r
a
y
im
a
g
es
co
m
p
o
s
e
t
h
e
d
ataset.
T
h
e
lo
g
ar
ith
m
ic
o
p
er
ato
r
is
em
p
lo
y
ed
to
en
h
an
ce
p
h
o
to
s
,
an
d
m
ed
ian
an
d
av
er
a
g
e
f
ilter
in
g
ar
e
u
s
ed
to
elim
in
ate
n
o
is
e
f
r
o
m
im
ag
es.
Mo
r
p
h
o
l
o
g
ical
o
p
er
atio
n
s
an
d
th
e
s
o
b
el
ed
g
e
d
etec
tio
n
ap
p
r
o
ac
h
ar
e
e
m
p
lo
y
e
d
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
h
e
y
o
b
tain
ed
8
4
.
7
p
er
ce
n
t
ac
cu
r
ac
y
in
class
if
y
in
g
t
h
e
d
ata
u
s
in
g
s
u
p
p
o
r
t
v
ec
t
o
r
m
ac
h
in
es
(
S
VM
)
in
to
f
r
ac
t
u
r
ed
an
d
n
o
n
-
f
r
ac
tu
r
ed
ca
te
g
o
r
ies.
Sin
th
u
r
a
et
a
l.
[
1
4
]
s
u
g
g
ested
a
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
to
d
etec
t
b
o
n
e
f
r
ac
tu
r
es.
T
h
ey
to
o
k
a
n
x
-
r
ay
im
a
g
e
as
in
p
u
t,
t
h
en
p
r
ep
r
o
ce
s
s
ed
it
u
s
in
g
a
m
e
d
ian
f
ilter
,
an
d
th
e
p
r
ep
r
o
ce
s
s
ed
im
ag
e
was
ap
p
lied
to
t
h
e
d
is
cr
ete
wa
v
elet
tr
a
n
s
f
o
r
m
s
(
DW
T
)
s
tag
e,
w
h
ich
is
u
s
ed
to
f
in
d
th
e
ed
g
e
in
ea
ch
c
h
an
n
el.
T
h
en
th
e
o
u
tp
u
t
is
co
m
p
a
r
ed
with
t
h
e
d
atab
ase
with
n
e
u
r
al
n
etwo
r
k
s
,
wh
ich
g
iv
es
th
e
o
u
tp
u
t.
Vasilak
ak
is
et
a
l.
[
1
5
]
s
u
g
g
ested
f
u
zz
y
p
h
r
ases
(
FP
)
f
o
r
d
etec
tio
n
.
T
h
is
wo
r
k
aim
s
to
u
s
e
t
h
e
wav
elet
f
u
zz
y
p
h
r
ases
(
W
F
P)
ap
p
r
o
ac
h
to
id
en
tify
b
o
n
e
f
r
ac
tu
r
es u
s
in
g
x
-
r
a
y
p
ictu
r
es.
T
h
e
ac
cu
r
ac
y
o
f
th
e
class
if
y
in
g
ap
p
r
o
ac
h
was
8
4
%.
Yad
a
v
an
d
San
d
ee
p
[
1
6
]
b
u
ilt
a
d
ee
p
C
NN
m
o
d
el
t
o
ac
tiv
ely
class
if
y
f
r
ac
tu
r
ed
o
r
h
ea
lth
y
b
o
n
es.
Af
ter
au
g
m
e
n
tatio
n
,
t
h
e
d
ataset
h
ad
a
s
ize
o
f
4
0
0
0
,
wh
er
e
2
0
0
0
is
a
h
ea
lth
y
b
o
n
e
an
d
2
0
0
0
is
ca
n
ce
r
o
u
s
,
th
ey
u
s
ed
f
iv
e
-
f
o
l
d
cr
o
s
s
-
v
alid
atio
n
an
d
h
av
e
t
h
e
b
est
ac
cu
r
ac
y
o
f
all,
s
co
r
in
g
9
2
.
4
4
p
er
ce
n
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
6
9
2
-
4
7
0
4
4694
R
ao
et
a
l.
[
1
7
]
p
r
o
p
o
s
ed
a
m
eth
o
d
t
o
d
etec
t
b
o
n
e
f
r
ac
tu
r
es
.
T
h
eir
d
ataset
h
a
d
3
0
0
X
-
r
ay
p
ictu
r
es,
a
n
d
th
ey
ac
h
iev
ed
9
0
%
ac
cu
r
ac
y
b
y
u
s
i
n
g
a
b
ac
k
p
r
o
p
ag
atio
n
n
eu
r
al
n
etwo
r
k
(
B
PNN)
an
d
SIFT
f
e
atu
r
e
ex
tr
ac
tio
n
.
I
n
2
0
2
0
,
Kar
im
u
n
n
is
a
et
a
l.
[
1
8
]
s
u
g
g
ested
a
m
o
d
el
u
s
in
g
9
0
0
X
-
r
ay
p
ictu
r
es,
o
f
wh
ich
4
0
0
wer
e
n
o
r
m
al
an
d
5
0
0
wer
e
f
r
ac
tu
r
e
d
.
I
n
itially
,
in
p
u
t
X
-
r
a
y
p
ictu
r
es
ar
e
tr
a
n
s
f
o
r
m
ed
i
n
to
g
r
a
y
s
ca
le
im
ag
es.
T
h
eir
B
PNN
p
r
o
v
id
es
a
n
im
p
r
o
v
e
d
class
if
icatio
n
r
ate
o
f
9
1
p
e
r
ce
n
t.
Path
ar
e
et
a
l.
[
1
9
]
em
p
lo
y
m
an
y
p
r
o
ce
s
s
in
g
ap
p
r
o
ac
h
es,
in
clu
d
i
n
g
s
eg
m
e
n
tatio
n
,
ed
g
e
d
etec
tio
n
,
an
d
f
ea
tu
r
e
ex
tr
ac
tio
n
.
B
ek
k
a
n
ti
et
a
l.
[
2
0
]
s
u
g
g
ested
th
e
Har
r
is
co
r
n
er
d
etec
tio
n
tech
n
iq
u
e,
wh
ich
is
a
tr
ad
itio
n
a
l
co
m
p
u
ter
v
is
io
n
tech
n
iq
u
e
r
elate
d
to
f
ea
tu
r
e
d
etec
tio
n
an
d
tr
ad
itio
n
al
im
a
g
e
p
r
o
ce
s
s
in
g
to
d
etec
t
b
o
n
e
f
r
ac
tu
r
es.
T
h
e
r
e
wer
e
two
h
u
n
d
r
ed
n
o
n
-
f
r
ac
tu
r
ed
an
d
th
r
ee
h
u
n
d
r
ed
f
r
ac
tu
r
e
d
X
-
r
ay
im
ag
es
in
th
eir
co
llectio
n
.
T
h
e
in
p
u
t
x
-
r
a
y
im
ag
es
ar
e
p
r
ep
r
o
ce
s
s
ed
u
s
in
g
m
3
f
ilter
in
g
.
s
eg
m
e
n
ted
u
s
in
g
ca
n
n
y
e
d
g
e
d
etec
tio
n
,
f
ea
tu
r
es
ex
tr
ac
ted
u
s
in
g
Har
r
is
co
r
n
er
d
etec
tio
n
,
an
d
last
ly
ca
teg
o
r
ized
as
f
r
ac
tu
r
ed
o
r
n
o
t.
T
h
ey
c
o
n
tr
asted
th
eir
ef
f
o
r
ts
u
s
in
g
SUR
F
wi
th
B
P
N
N
an
d
ML
P
-
b
ased
B
PNN.
W
h
ile
SU
R
F
u
s
in
g
B
PNN
y
ield
s
an
ac
cu
r
ac
y
o
f
8
5
p
er
ce
n
t,
ML
P
-
b
ased
B
PNN
y
ield
s
an
ac
cu
r
ac
y
o
f
8
5
p
e
r
ce
n
t,
a
n
d
th
eir
s
u
g
g
es
ted
ap
p
r
o
ac
h
y
ield
s
a
9
4
p
e
r
ce
n
t
ac
cu
r
ac
y
.
I
n
2
0
2
2
,
Sam
o
th
ai
et
a
l.
[
2
1
]
p
r
esen
ted
a
d
v
an
ce
d
C
NN
YOL
O
m
o
d
els
in
th
is
wo
r
k
,
w
h
er
e
th
e
wid
ely
u
s
ed
YOL
O
-
X
a
n
d
YOL
O
-
R
m
o
d
els
wer
e
im
p
lem
en
te
d
.
W
ith
7
6
p
er
ce
n
t
co
n
f
id
en
tiality
,
YOL
O
-
X
ca
n
l
o
ca
te
f
r
ac
tu
r
es
with
g
r
ea
t
ac
c
u
r
ac
y
.
Ho
wev
er
,
YOL
O
-
R
ex
h
ib
its
in
ac
cu
r
ate
f
r
ac
tu
r
e
lo
ca
tio
n
d
e
tectio
n
.
Ko
s
r
at
an
d
Haw
ez
i
[
2
2
]
an
aly
ze
v
a
r
io
u
s
ML
class
if
ier
s
an
d
ass
is
t
s
u
r
g
eo
n
s
in
ac
cu
r
ately
d
iag
n
o
s
in
g
b
o
n
e
f
r
ac
tu
r
es.
T
h
ey
e
m
p
lo
y
ed
a
f
ilter
in
g
tech
n
iq
u
e
to
r
em
o
v
e
n
o
is
e
an
d
p
r
ep
r
o
ce
s
s
in
g
to
co
n
v
er
t
th
e
R
G
B
im
ag
es
to
g
r
ay
s
ca
le.
I
n
th
is
s
tu
d
y
,
SVM
i
s
f
o
u
n
d
to
b
e
th
e
b
est.
C
er
tain
is
s
u
es
p
er
s
is
t
in
th
e
c
u
r
r
en
t
s
tate
o
f
b
o
n
e
f
r
ac
tu
r
e
d
etec
tio
n
,
n
ec
ess
itatin
g
im
p
r
o
v
em
en
ts
in
ac
cu
r
ac
y
to
e
n
h
an
ce
d
etec
ti
o
n
ef
f
icac
y
,
as
o
u
tlin
ed
in
T
ab
le
1
an
d
th
e
af
o
r
em
e
n
tio
n
ed
s
tu
d
y
.
T
h
e
au
g
m
en
tatio
n
o
f
th
e
d
ataset
is
im
p
er
ativ
e,
r
eq
u
i
r
in
g
th
e
in
clu
s
io
n
o
f
h
ig
h
-
q
u
ality
p
h
o
to
s
.
T
h
is
s
tu
d
y
u
tili
ze
s
a
d
ataset
co
m
p
r
is
in
g
a
s
u
b
s
tan
tial
n
u
m
b
er
o
f
b
o
n
e
f
r
ac
tu
r
e
im
ag
es,
im
p
lem
en
tin
g
ca
n
n
y
ed
g
e
d
etec
tio
n
d
u
r
in
g
p
r
ep
r
o
ce
s
s
in
g
to
o
p
tim
ize
r
esu
lts
.
Desp
ite
th
e
g
r
o
win
g
ac
ce
s
s
ib
ilit
y
o
f
m
ac
h
in
e
-
aid
e
d
b
o
n
e
f
r
ac
tu
r
e
d
etec
tio
n
,
ac
cu
r
ac
y
r
em
ain
s
a
co
n
ce
r
n
,
o
f
ten
lead
in
g
t
o
f
alse
p
o
s
itiv
es.
T
h
e
id
en
tific
atio
n
o
f
f
r
ac
tu
r
e
s
in
th
e
ab
s
en
ce
o
f
ac
tu
al
o
cc
u
r
r
en
ce
s
is
a
p
r
ev
alen
t
p
r
o
b
lem
,
attr
ib
u
te
d
to
th
e
lim
ited
v
o
lu
m
e
o
f
d
ata
an
d
in
ac
cu
r
ac
ies
in
th
e
lab
elin
g
o
f
tr
ain
i
n
g
d
ata.
Ad
d
r
ess
in
g
th
ese
ch
allen
g
es,
th
is
r
esear
ch
f
o
cu
s
es
o
n
r
ef
in
i
n
g
t
r
ain
in
g
a
n
d
test
in
g
d
ata
b
y
in
tr
o
d
u
cin
g
a
n
ewly
c
u
s
to
m
ized
p
ar
allel
d
ee
p
lear
n
i
n
g
m
o
d
el
with
p
ar
a
m
eter
cu
s
to
m
izatio
n
.
No
tab
ly
,
p
u
b
licly
av
ailab
le
d
atasets
f
r
eq
u
en
tly
lack
f
lawless
p
r
ep
r
o
ce
s
s
in
g
an
d
ac
cu
r
ate
lab
elin
g
,
n
ec
ess
itatin
g
th
e
in
co
r
p
o
r
atio
n
o
f
ad
d
itio
n
al
la
b
eled
d
ata
to
au
g
m
en
t
in
p
u
ts
f
o
r
t
h
e
m
o
d
el.
T
o
en
s
u
r
e
an
ad
eq
u
ate
s
u
p
p
ly
o
f
h
ig
h
-
q
u
ality
d
ata,
ed
g
e
d
etec
t
io
n
alg
o
r
ith
m
s
ar
e
em
p
lo
y
ed
d
u
r
in
g
d
ata
p
r
ep
r
o
ce
s
s
in
g
.
T
h
e
co
n
t
r
ib
u
tio
n
s
o
f
th
is
s
tu
d
y
ca
n
b
e
s
u
m
m
ar
ized
as f
o
llo
ws:
a.
C
o
r
r
ec
tin
g
an
o
m
alies
in
th
e
d
ataset
th
r
o
u
g
h
s
ev
er
al
im
ag
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
e
in
v
o
l
v
es
em
p
lo
y
in
g
a
r
an
g
e
o
f
im
ag
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es,
s
u
ch
as
c
r
o
p
p
in
g
,
r
esizin
g
,
c
o
n
tr
ast
en
h
a
n
c
em
en
t,
f
ilter
in
g
,
au
g
m
en
tatio
n
,
an
d
cu
lm
in
atin
g
in
ca
n
n
y
ed
g
e
d
etec
tio
n
.
b.
T
h
e
s
tu
d
y
in
tr
o
d
u
ce
s
a
n
o
v
el
an
d
tailo
r
s
p
a
r
allel
d
ee
p
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
PDC
NN)
m
o
d
el,
d
esig
n
ed
to
s
u
r
p
ass
th
e
ex
is
tin
g
liter
atu
r
e
b
y
en
h
a
n
cin
g
ac
c
u
r
ac
y
in
b
o
n
e
f
r
ac
tu
r
e
d
etec
tio
n
.
c.
A
co
m
p
ar
ativ
e
ev
al
u
atio
n
is
c
o
n
d
u
cte
d
am
o
n
g
th
r
ee
C
NN
-
b
ased
m
o
d
els
—
C
NN,
Mo
b
ileN
et
-
V2
,
an
d
th
e
n
ewly
p
r
o
p
o
s
ed
PDC
NN.
Fo
u
r
ca
s
es
o
f
ab
latio
n
s
tu
d
ies
is
ca
r
r
ied
o
u
t
to
v
alid
ate
th
e
p
r
o
p
o
s
ed
PDC
NN
m
o
d
el.
T
h
e
o
b
jectiv
e
is
to
id
en
tify
th
e
m
o
s
t
ef
f
ec
tiv
e
m
o
d
el
f
o
r
b
o
n
e
f
r
ac
t
u
r
e
r
ec
o
g
n
itio
n
th
r
o
u
g
h
a
th
o
r
o
u
g
h
a
n
aly
s
is
o
f
th
eir
p
e
r
f
o
r
m
an
ce
.
T
ab
le
1
.
Su
m
m
a
r
ized
liter
atu
r
e
r
ev
iew
o
f
b
o
n
e
f
r
ac
tu
r
e
r
ec
o
g
n
itio
n
u
s
in
g
m
ac
h
in
e
lea
r
n
in
g
m
o
d
els
S
t
u
d
y
D
a
t
a
P
r
e
p
r
o
c
e
ss
i
n
g
Te
c
h
n
i
q
u
e
M
o
d
e
l
A
c
c
u
r
a
c
y
Li
mi
t
a
t
i
o
n
s/
R
e
m
a
r
k
s
[
1
1
]
44
C
a
n
n
y
e
d
g
e
d
e
t
e
c
t
i
o
n
S
N
A
K
E
8
9
%
O
n
l
y
d
e
t
e
c
t
s s
t
r
e
ss fr
a
c
t
u
r
e
s
a
n
d
i
s
u
n
a
b
l
e
t
o
c
l
a
ss
i
f
y
c
o
mp
l
e
x
f
r
a
c
t
u
r
e
s
a
c
c
u
r
a
t
e
l
y
.
T
h
e
r
e
a
r
e
n
o
t
ma
n
y
i
m
a
g
e
s
i
n
t
h
e
d
a
t
a
set
.
[
1
2
]
40
S
o
b
e
l
e
d
g
e
d
e
t
e
c
t
o
r
M
e
t
a
C
l
a
s
si
f
i
e
r
8
5
%
Th
e
d
a
t
a
se
t
w
a
s
n
o
t
g
o
o
d
.
[
1
3
]
30
S
o
b
e
l
e
d
g
e
d
e
t
e
c
t
o
r
S
V
M
8
7
.
5
%
Th
e
d
a
t
a
se
t
w
a
s
n
o
t
g
o
o
d
.
[
1
4
]
NM
d
i
s
c
r
e
t
e
w
a
v
e
l
e
t
t
r
a
n
sf
o
r
ms
(
D
W
T)
C
N
N
7
9
%
Th
e
d
a
t
a
se
t
w
a
s
n
o
t
g
i
v
e
n
.
[
1
5
]
3
0
0
NM
WFP
8
4
%
Th
i
s
w
a
s
t
h
e
e
a
r
l
y
s
t
a
g
e
o
f
f
u
z
z
y
p
h
r
a
ses;
a
d
d
i
t
i
o
n
a
l
d
e
v
e
l
o
p
me
n
t
i
s
r
e
q
u
i
r
e
d
b
e
f
o
r
e
t
h
e
y
c
a
n
b
e
r
e
l
i
a
b
l
y
d
e
t
e
c
t
e
d
.
[
1
7
]
3
0
0
M
3
f
i
l
t
e
r
S
I
F
T
+
B
P
N
N
9
0
%
H
TB
F
D
,
A
N
N
,
a
n
d
o
t
h
e
r
a
l
g
o
r
i
t
h
m
s
a
r
e
u
sef
u
l
t
o
o
l
s fo
r
i
mp
r
o
v
i
n
g
f
r
a
c
t
u
r
e
p
l
a
c
e
m
e
n
t
.
[
1
8
]
9
0
0
C
a
n
n
y
Ed
g
e
D
e
t
e
c
t
i
o
n
a
n
d
C
o
n
ser
v
a
t
i
v
e
F
i
l
t
e
r
i
n
g
B
P
N
N
9
1
%
Th
e
d
a
t
a
se
t
c
o
n
t
a
i
n
s m
a
n
y
t
y
p
e
s
o
f
b
o
n
e
s
h
a
p
e
s,
w
h
i
c
h
ma
y
b
e
c
o
m
e
c
o
n
f
u
s
i
n
g
.
[
1
9
]
20
NM
H
TM
7
5
%
A
n
X
-
r
a
y
o
f
t
h
e
s
k
u
l
l
,
p
e
l
v
i
s,
a
n
d
s
p
i
n
a
l
c
o
l
u
mn
w
a
s
n
o
t
a
b
l
e
t
o
b
e
o
b
t
a
i
n
e
d
.
[
2
2
]
2
7
0
C
a
n
n
y
e
d
g
e
d
e
t
e
c
t
i
o
n
S
V
M
NM
I
t
o
n
l
y
c
o
mp
a
r
e
d
t
h
e
m
o
d
e
l
s
.
a.
N
M
:
N
o
t
me
n
t
i
o
n
e
d
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:
2088
-
8
7
0
8
B
o
n
e
-
N
et:
a
p
a
r
a
llel d
ee
p
c
o
n
vo
lu
tio
n
a
l n
eu
r
a
l n
etw
o
r
k
-
b
a
s
ed
b
o
n
e
…
(
Md
.
Ha
s
a
n
I
ma
m
B
ijo
y
)
4695
2.
RE
S
E
ARCH
M
E
T
H
O
D
T
h
is
s
tu
d
y
ce
n
ter
s
o
n
th
e
d
etec
tio
n
o
f
b
o
n
e
f
r
ac
tu
r
es
in
X
-
r
ay
im
ag
es
th
r
o
u
g
h
th
e
u
tili
z
atio
n
o
f
a
C
NN
-
b
ased
m
o
d
el.
T
h
e
d
ev
elo
p
m
en
t
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
tem
in
v
o
lv
es
s
eq
u
en
tial
s
t
ep
s
,
in
clu
d
in
g
d
ata
co
llectio
n
,
im
ag
e
p
r
e
-
p
r
o
ce
s
s
in
g
,
an
d
f
r
ac
tu
r
e
r
ec
o
g
n
itio
n
u
s
in
g
C
NN
m
o
d
els.
T
h
e
s
ch
em
atic
r
ep
r
esen
tatio
n
o
f
th
e
wo
r
k
in
g
p
r
o
ce
s
s
is
o
u
tlin
ed
in
Fig
u
r
e
2
.
Fig
u
r
e
2
.
T
h
e
wo
r
k
f
lo
w
d
iag
r
am
o
f
p
r
o
p
o
s
ed
s
tu
d
y
2
.
1
.
Da
t
a
s
et
des
cr
iptio
n
T
h
is
s
tu
d
y
m
ak
es
u
s
e
o
f
a
p
u
b
lic
d
ataset
o
f
b
o
n
e
f
r
ac
tu
r
e
a
n
d
n
o
n
-
f
r
ac
tu
r
ed
x
-
r
a
y
im
ag
e
s
.
T
o
k
ee
p
th
in
g
s
s
im
p
le,
t
h
e
p
u
b
licly
av
ailab
le
d
ataset
o
f
b
in
a
r
y
-
class
b
o
n
e
f
r
ac
tu
r
e
a
n
d
n
o
n
-
f
r
ac
tu
r
e
x
-
r
a
y
im
a
g
es
was
o
b
tain
ed
u
s
in
g
th
e
Kag
g
le
p
la
tf
o
r
m
;
f
o
r
th
e
s
ak
e
o
f
th
is
p
a
p
er
,
th
is
d
ata
is
r
e
f
er
r
ed
to
as
th
e
“
b
o
n
e
f
r
ac
tu
r
e
d
ataset
”
[
2
3
]
.
I
t
is
em
p
lo
y
e
d
in
th
e
cr
ea
tio
n
o
f
an
im
ag
e
cla
s
s
if
ier
th
at
id
en
tifie
s
b
o
n
e
f
r
ac
tu
r
es
in
g
iv
en
x
-
r
ay
p
ictu
r
es.
T
h
er
e
ar
e
1
8
9
9
p
h
o
t
o
s
id
en
tifie
d
in
th
e
test
in
g
cla
s
s
(
f
r
ac
tu
r
e
an
d
n
o
n
-
f
r
ac
tu
r
e
)
an
d
8
,
8
8
4
im
a
g
es
in
th
e
tr
ain
in
g
class
(
f
r
ac
tu
r
e
an
d
n
o
n
-
f
r
ac
tu
r
e)
.
T
h
e
im
ag
es
wer
e
n
o
t
th
e
s
am
e
s
ize,
an
d
s
o
m
e
o
f
th
em
in
clu
d
ed
an
o
m
alo
u
s
d
ata
in
ad
d
itio
n
to
b
ein
g
u
n
clea
r
.
T
h
e
s
am
p
le
im
ag
e
o
f
f
r
ac
tu
r
ed
b
o
n
e
an
d
f
r
esh
b
o
n
e
o
f
x
-
r
ay
im
ag
e
is
g
iv
en
in
Fig
u
r
e
3
.
Fig
u
r
e
3
(
a)
r
ep
r
esen
ts
th
e
f
r
ac
tu
r
ed
b
o
n
e,
wh
ile
Fig
u
r
e
3
(
b
)
d
ep
icts
th
e
n
o
n
-
f
r
ac
tu
r
ed
b
o
n
e
.
(
a)
(
b
)
Fig
u
r
e
3
.
T
h
e
s
am
p
le
im
ag
e
o
f
(
a)
f
r
ac
t
u
r
ed
b
o
n
e,
an
d
(
b
)
n
o
n
-
f
r
ac
tu
r
ed
b
o
n
e
2
.
2
.
Da
t
a
prepro
ce
s
s
ing
I
n
th
e
r
ea
lm
o
f
b
o
n
e
f
r
ac
tu
r
e
id
en
tific
atio
n
,
a
cr
u
cial
s
tep
in
th
e
r
esear
ch
ap
p
licatio
n
s
o
f
co
m
p
u
ter
v
is
io
n
an
d
im
ag
e
p
r
o
ce
s
s
in
g
in
v
o
lv
es
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
.
T
h
is
p
r
o
ce
s
s
s
er
v
es
to
en
h
a
n
ce
th
e
q
u
ality
o
f
im
ag
es,
r
ed
u
ce
n
o
is
e
lev
els,
r
ec
tify
d
is
to
r
tio
n
,
an
d
p
r
ep
ar
e
th
e
im
ag
es
f
o
r
s
u
b
s
eq
u
e
n
t
an
aly
s
is
[
1
6
]
.
T
h
e
f
o
llo
win
g
ar
e
k
ey
s
tr
ateg
ies
(
s
ee
o
u
tco
m
e
o
f
Fig
u
r
e
4
(
a)
to
4
(
h
)
)
em
p
lo
y
ed
in
th
is
s
tu
d
y
,
alo
n
g
with
ex
p
lan
atio
n
s
.
2
.
1
.
1
.
Cro
pp
ing
I
n
b
o
n
e
f
r
ac
tu
r
e
i
d
en
tific
atio
n
,
cr
o
p
p
in
g
is
a
p
r
ev
ale
n
t
im
ag
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
e
u
tili
ze
d
to
ex
tr
ac
t
th
e
r
eg
io
n
o
f
in
ter
est
(
R
OI
)
with
in
an
im
ag
e
[
2
4
]
.
T
h
is
f
ac
ilit
ates
th
e
is
o
latio
n
o
f
r
elev
an
t
an
ato
m
ical
s
tr
u
ctu
r
es
f
o
r
m
o
r
e
ac
cu
r
ate
an
aly
s
is
,
d
is
r
eg
ar
d
in
g
u
n
n
ec
e
s
s
ar
y
elem
en
ts
.
T
h
e
o
u
tp
u
t
o
f
cr
o
p
p
in
g
f
r
o
m
th
e
r
aw
p
ictu
r
e
o
f
Fig
u
r
e
4
(
a)
,
ca
n
b
e
s
ee
n
in
Fig
u
r
e
4
(
b
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
6
9
2
-
4
7
0
4
4696
2
.
1
.
2
.
Resizing
R
esizin
g
p
lay
s
a
v
ital
r
o
le
in
s
tan
d
ar
d
izin
g
th
e
d
im
en
s
io
n
s
o
f
im
ag
es
with
in
a
d
ataset.
T
h
is
en
s
u
r
es
u
n
if
o
r
m
ity
,
f
ac
ilit
atin
g
ef
f
icie
n
t
p
r
o
ce
s
s
in
g
an
d
an
aly
s
is
[
2
5
]
.
I
t
also
aid
s
in
ac
co
m
m
o
d
atin
g
v
ar
iatio
n
s
in
im
ag
e
r
eso
lu
tio
n
s
co
m
m
o
n
ly
en
co
u
n
ter
ed
in
m
ed
ical
im
a
g
in
g
.
R
esizin
g
is
em
p
lo
y
ed
to
s
tan
d
ar
d
ize
th
e
d
im
en
s
io
n
s
o
f
im
ag
es,
en
s
u
r
i
n
g
u
n
if
o
r
m
ity
ac
r
o
s
s
th
e
d
ataset
with
2
2
7
×2
2
7
p
ix
els
an
d
th
e
o
u
tp
u
t
s
h
o
wn
in
Fig
u
r
e
4
(
c)
.
2
.
1
.
3
.
E
nh
a
ncing
co
ntr
a
s
t
E
n
h
an
cin
g
c
o
n
tr
ast
is
cr
u
cial
to
ac
c
en
tu
ate
s
u
b
tle
d
etails
with
in
th
e
i
m
ag
es
[
2
6
]
.
T
h
is
allo
ws
f
o
r
b
etter
d
if
f
er
en
tiatio
n
b
etwe
en
h
ea
lth
y
a
n
d
f
r
ac
tu
r
ed
b
o
n
e
s
tr
u
ctu
r
es,
en
h
an
cin
g
th
e
o
v
er
all
in
ter
p
r
etab
ilit
y
o
f
th
e
im
ag
es.
Af
ter
en
h
an
cin
g
th
e
im
ag
e
q
u
ality
,
Fig
u
r
e
4
(
d
)
d
is
p
lay
s
th
e
im
ag
e
with
im
p
r
o
v
ed
v
is
u
al
clar
ity
.
2
.
1
.
4
.
F
ilte
ring
Fil
ter
in
g
is
em
p
lo
y
ed
to
e
m
p
h
asize
r
elev
an
t
s
tr
u
ctu
r
al
d
etails
wh
ile
s
u
p
p
r
ess
in
g
n
o
is
e.
B
y
s
elec
tiv
ely
en
h
an
cin
g
ce
r
tai
n
im
ag
e
ch
ar
ac
ter
is
tics
,
f
ilter
in
g
co
n
tr
ib
u
tes
to
th
e
im
p
r
o
v
e
m
e
n
t
o
f
im
ag
e
clar
ity
an
d
th
e
e
x
tr
ac
tio
n
o
f
ess
en
tial
in
f
o
r
m
atio
n
[
1
8
]
.
Af
te
r
f
il
ter
in
g
an
d
en
h
a
n
cin
g
t
h
e
ess
en
tial
f
ea
tu
r
es,
th
e
im
ag
e
is
d
is
p
lay
ed
in
Fig
u
r
e
4
(
e)
.
2
.
1
.
5
.
Ca
nn
y
edg
e
det
ec
t
io
n
C
an
n
y
ed
g
e
d
etec
tio
n
(
C
E
D)
is
an
ad
v
an
ce
d
im
ag
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
e
u
s
ed
in
co
m
p
u
te
r
v
is
io
n
to
id
en
tify
ed
g
es
an
d
b
o
u
n
d
a
r
ie
s
with
in
an
im
ag
e
[
2
7
]
.
I
t
w
as
d
ev
elo
p
ed
to
a
d
d
r
ess
th
e
ch
allen
g
es
o
f
ed
g
e
d
etec
tio
n
b
y
p
r
o
v
id
i
n
g
ac
cu
r
a
te
an
d
r
eliab
le
r
esu
lts
wh
ile
m
in
im
izin
g
f
alse
p
o
s
itiv
es.
I
n
th
e
co
n
tex
t
o
f
b
o
n
e
f
r
ac
tu
r
e
r
ec
o
g
n
itio
n
u
s
in
g
c
o
m
p
u
ter
v
is
io
n
an
d
im
a
g
e
p
r
o
c
ess
in
g
,
ca
n
n
y
e
d
g
e
d
etec
tio
n
p
lay
s
a
cr
u
cial
r
o
le
in
h
ig
h
lig
h
tin
g
p
r
o
m
in
en
t
co
n
to
u
r
s
an
d
ed
g
es
with
in
m
ed
ical
i
m
ag
es.
T
h
is
tech
n
iq
u
e
en
h
an
c
es
th
e
v
is
u
aliza
tio
n
o
f
s
tr
u
ctu
r
al
d
etails in
b
o
n
e
im
ag
es,
m
ak
in
g
it p
ar
ticu
lar
l
y
v
a
lu
ab
le
f
o
r
p
i
n
p
o
in
tin
g
f
r
ac
tu
r
e
s
an
d
ir
r
eg
u
lar
ities
in
th
e
b
o
n
e
s
tr
u
ctu
r
e
w
h
ich
is
d
is
p
lay
ed
in
Fig
u
r
e
4
(
f
).
2
.
1
.
6
.
Aug
m
ent
a
t
io
n
Au
g
m
en
tatio
n
is
cr
u
cial
f
o
r
en
h
an
cin
g
th
e
r
o
b
u
s
tn
ess
o
f
b
o
n
e
f
r
ac
tu
r
e
id
en
tific
atio
n
m
o
d
els.
B
y
ex
p
o
s
in
g
th
e
m
o
d
el
to
d
iv
er
s
e
o
r
ien
tatio
n
s
,
s
ca
les,
an
d
p
er
s
p
ec
tiv
es,
it
b
ec
o
m
es
m
o
r
e
ad
ep
t
at
ac
cu
r
ately
id
en
tify
in
g
f
r
ac
tu
r
es
u
n
d
e
r
a
r
an
g
e
o
f
co
n
d
itio
n
s
,
co
n
t
r
ib
u
t
in
g
to
im
p
r
o
v
ed
g
en
e
r
aliza
tio
n
an
d
p
er
f
o
r
m
a
n
ce
[
2
8
]
.
I
n
th
is
s
tu
d
y
,
s
o
m
e
p
r
im
ar
y
im
ag
e
au
g
m
en
tatio
n
m
eth
o
d
is
u
s
ed
s
u
ch
as r
o
tatio
n
,
f
li
p
p
in
g
,
z
o
o
m
in
g
,
an
d
s
h
ea
r
in
g
an
d
th
e
o
u
tco
m
es
o
f
th
e
f
ilter
ed
im
ag
es
an
d
C
E
D
ar
e
d
is
p
lay
ed
in
Fig
u
r
e
4
(
g
)
an
d
Fig
u
r
e
4
(
h
)
,
r
esp
ec
tiv
ely
.
Af
ter
u
n
d
e
r
g
o
in
g
ex
te
n
s
iv
e
im
ag
e
p
r
o
ce
s
s
in
g
,
th
e
r
aw
im
a
g
e
is
s
u
b
jecte
d
to
c
o
n
tr
ast
en
h
an
ce
m
en
t,
as
illu
s
tr
ated
in
Fig
u
r
e
4
(
a)
to
4
(
h
)
.
T
h
e
d
ataset
co
m
p
r
is
es
m
o
r
e
th
an
1
0
,
7
8
3
im
a
g
e
s
ca
teg
o
r
ized
in
to
f
r
ac
tu
r
ed
an
d
n
o
n
-
f
r
ac
t
u
r
ed
cl
ass
es,
ex
h
ib
itin
g
a
n
ea
r
b
alan
ce
b
etwe
en
th
e
two
.
Fo
llo
win
g
an
8
5
:1
5
s
p
lit
f
o
r
tr
ain
in
g
an
d
test
in
g
,
th
e
d
ataset
co
n
s
is
ts
o
f
8
8
8
4
im
a
g
es in
th
e
tr
ain
in
g
s
et
an
d
1
8
9
9
im
ag
es
in
th
e
test
s
et.
(
a)
(
b
)
(
c)
(
d
)
(
e)
(f)
(
g
)
(
h
)
Fig
u
r
e
4
.
Sev
er
al
im
ag
e
p
r
o
ce
s
s
in
g
ap
p
lied
in
to
(
a
)
r
aw
im
a
g
e
an
d
g
ettin
g
(
b
)
cr
o
p
p
in
g
im
ag
e,
(
c)
r
esize
im
ag
e,
(
d
)
en
h
a
n
ce
co
n
tr
ast,
(
e)
f
ilter
in
g
,
(
f
)
a
u
g
m
en
tatio
n
o
n
f
ilter
ed
im
ag
e
,
an
d
(
g
)
ca
n
n
y
ed
g
e
d
etec
tio
n
an
d
(
h
)
au
g
m
en
ted
C
E
D
im
ag
e
to
in
cr
ea
s
e
th
e
s
ize
o
f
t
h
e
d
a
taset
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:
2088
-
8
7
0
8
B
o
n
e
-
N
et:
a
p
a
r
a
llel d
ee
p
c
o
n
vo
lu
tio
n
a
l n
eu
r
a
l n
etw
o
r
k
-
b
a
s
ed
b
o
n
e
…
(
Md
.
Ha
s
a
n
I
ma
m
B
ijo
y
)
4697
2
.
3
.
M
o
del
im
plem
ent
a
t
io
n
Fo
llo
win
g
th
e
p
r
ep
r
o
ce
s
s
in
g
p
h
ase,
th
is
s
tu
d
y
en
d
ea
v
o
r
s
to
co
n
s
tr
u
ct
a
b
o
n
e
f
r
ac
t
u
r
e
r
ec
o
g
n
itio
n
s
y
s
tem
b
y
em
p
l
o
y
in
g
C
NN
b
ased
m
o
d
els.
I
n
p
u
r
s
u
it
o
f
h
ig
h
ac
c
u
r
ac
y
,
we
ex
p
lo
r
e
a
v
ar
iety
o
f
m
o
d
els
in
clu
d
in
g
th
e
tr
ad
itio
n
al
C
NN
m
o
d
el,
a
C
NN
-
b
ased
tr
an
s
f
e
r
lear
n
in
g
m
o
d
el,
Mo
b
ileNet
-
V2
,
an
d
o
u
r
n
o
v
el
PDC
NN
m
o
d
el.
C
NN,
Mo
b
ileNet,
an
d
th
e
n
ewly
p
r
o
p
o
s
ed
PDC
NN
ex
h
ib
it
r
em
ar
k
ab
le
ef
f
icac
y
in
id
en
tify
in
g
b
o
n
e
f
r
ac
tu
r
es f
r
o
m
X
-
r
ay
im
a
g
es.
T
h
e
s
u
cc
in
ct
d
escr
ip
tio
n
o
f
ea
ch
m
o
d
el
is
g
iv
en
b
elo
w:
2
.
3
.
1
.
C
o
nv
o
lutio
na
l
n
eura
l
n
et
wo
rk
C
o
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
[
2
9
]
,
o
r
C
NNs,
ar
e
wid
ely
u
tili
ze
d
in
d
ee
p
lear
n
in
g
n
etwo
r
k
m
o
d
els
an
d
co
m
p
u
ter
v
is
io
n
alg
o
r
ith
m
s
.
B
ec
au
s
e
it
ca
n
id
e
n
tify
p
atter
n
s
in
im
a
g
es,
th
is
k
in
d
o
f
ar
tific
ial
n
eu
r
al
n
etwo
r
k
is
u
tili
ze
d
f
o
r
im
a
g
e
r
ec
o
g
n
itio
n
an
d
p
r
o
ce
s
s
in
g
.
C
o
n
v
o
lu
ti
o
n
al,
p
o
o
lin
g
,
an
d
f
u
l
ly
co
n
n
ec
ted
lay
er
s
ar
e
s
o
m
e
o
f
th
e
lay
er
s
th
at
m
ak
e
it
u
p
.
T
h
e
co
n
v
o
lu
tio
n
al
lay
er
,
wh
ic
h
m
a
k
es
u
p
th
e
m
ajo
r
ity
o
f
C
NN,
is
wh
er
e
ch
a
r
ac
ter
is
tics
lik
e
f
o
r
m
s
,
ed
g
es,
an
d
p
atter
n
s
ar
e
e
x
tr
ac
ted
f
r
o
m
th
e
in
p
u
t
im
ag
e
b
y
ap
p
ly
in
g
f
ilter
s
.
On
e
o
r
m
o
r
e
f
u
lly
co
n
n
ec
ted
l
ay
er
s
ar
e
th
en
a
p
p
lied
to
t
h
e
o
u
tp
u
t
o
f
th
e
p
o
o
lin
g
lay
er
s
to
class
if
y
o
r
p
r
e
d
ict
th
e
im
ag
e.
2
.
3
.
2
.
M
o
bil
eNe
t
-
V2
Mo
b
ileNet
-
V2
is
a
lig
h
tweig
h
t
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
d
esig
n
ed
f
o
r
m
o
b
ile
an
d
em
b
ed
d
e
d
d
ev
ices.
I
t
im
p
r
o
v
es
ef
f
icien
cy
with
d
ep
t
h
wis
e
s
ep
ar
ab
le
co
n
v
o
lu
tio
n
s
an
d
in
v
e
r
ted
r
esid
u
als
with
lin
ea
r
b
o
ttlen
ec
k
s
[
3
0
]
.
I
t
also
in
c
o
r
p
o
r
ates
ex
p
an
s
io
n
an
d
s
q
u
ee
z
e
-
ex
citatio
n
m
o
d
u
les
f
o
r
b
ette
r
f
ea
tu
r
e
lear
n
in
g
.
W
ith
its
s
tr
ea
m
lin
ed
ar
ch
itectu
r
e,
Mo
b
ileNet
-
V2
ac
h
iev
es
h
ig
h
ac
cu
r
ac
y
wh
ile
m
in
im
i
zin
g
co
m
p
u
tatio
n
al
r
eso
u
r
ce
s
,
m
ak
in
g
it id
ea
l f
o
r
m
o
b
ile
ap
p
licatio
n
s
r
e
q
u
ir
in
g
f
ast an
d
ef
f
icien
t im
a
g
e
p
r
o
ce
s
s
in
g
.
2
.
3
.
3
.
P
a
ra
llel dee
p c
o
nv
o
luti
o
na
l neura
l net
wo
rk
(
P
DCN
N)
A
p
ar
allel
ar
c
h
itectu
r
e
with
two
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
PD
C
NNs)
is
p
r
esen
ted
to
h
an
d
le
t
h
e
r
ec
o
g
n
itio
n
a
n
d
class
if
icatio
n
o
f
b
o
n
e
f
r
ac
tu
r
es
in
X
-
r
ay
p
ictu
r
es.
T
h
e
s
u
g
g
ested
ar
ch
i
tectu
r
e
en
tails
th
e
f
o
llo
win
g
s
eq
u
e
n
ce
o
f
ev
e
n
ts
:
X
-
r
ay
im
a
g
es
with
b
o
n
e
f
r
ac
tu
r
es
ar
e
f
ed
in
to
t
h
e
PDC
NNs
[
3
1
]
i
n
p
u
t
la
y
er
.
T
o
r
ed
u
ce
th
e
co
m
p
u
tatio
n
al
co
m
p
lex
ity
,
th
ese
im
ag
es
a
r
e
p
r
ep
r
o
ce
s
s
ed
.
T
h
e
i
n
p
u
t
i
m
ag
es
h
av
e
b
ee
n
n
o
r
m
alize
d
t
o
a
2
7
7
×
2
2
7
-
p
ix
el
r
eso
lu
tio
n
f
o
r
tr
ain
i
n
g
,
allo
win
g
f
o
r
d
if
f
e
r
en
ce
s
in
p
i
x
el
wid
th
s
an
d
h
eig
h
ts
.
T
h
e
s
u
b
s
eq
u
en
t
p
r
o
ce
d
u
r
e
t
h
at
h
elp
s
to
s
im
p
lify
co
m
p
lex
i
ty
is
to
co
n
v
er
t
t
h
e
in
p
u
t
im
ag
es
to
ca
n
n
y
ed
g
e
d
etec
tio
n
.
T
h
e
ar
ch
itectu
r
e
o
f
th
e
PDC
NN
is
th
en
u
s
ed
to
class
if
y
in
p
u
t
X
-
r
ay
p
ictu
r
es
b
y
co
m
b
in
in
g
,
o
u
tp
u
t
,
lo
ca
l,
an
d
g
lo
b
al
p
ath
.
T
h
e
So
f
tMa
x
f
u
n
cti
o
n
is
u
s
ed
in
th
e
o
u
tp
u
t
p
ath
way
to
ca
r
r
y
o
u
t
t
h
e
class
if
icatio
n
o
f
b
o
n
e
f
r
ac
tu
r
es.
Fig
u
r
e
5
s
h
o
w
s
th
e
PDC
NN’
s
s
tr
u
ctu
r
e.
Fig
u
r
e
5
.
T
h
e
d
iag
r
am
o
f
p
r
o
p
o
s
ed
PDC
NN
m
o
d
el
th
at
co
n
t
ain
s
f
o
u
r
s
tag
es wh
ich
ar
e
lo
c
al
p
ath
,
g
lo
b
al
p
ath
,
m
er
g
in
g
t
h
em
,
an
d
o
u
tp
u
t stag
es in
o
r
d
er
t
o
id
en
tif
y
th
e
b
o
n
e
f
r
ac
tu
r
e
f
r
o
m
t
h
e
in
p
u
ted
x
-
r
a
y
im
ag
e
T
h
e
p
r
o
p
o
s
ed
PDC
NN
m
o
d
el
f
o
r
b
o
n
e
f
r
ac
tu
r
e
id
en
tific
atio
n
in
teg
r
ates
v
ar
io
u
s
lay
er
s
to
f
ac
ilit
ate
ac
cu
r
ate
r
ec
o
g
n
itio
n
o
f
f
r
ac
t
u
r
es
in
X
-
r
ay
im
ag
es.
I
n
itially
,
2
D
co
n
v
o
lu
tio
n
al
lay
e
r
s
ar
e
em
p
lo
y
ed
t
o
ex
tr
ac
t
p
er
tin
en
t
f
ea
tu
r
es
f
r
o
m
th
e
in
p
u
t
im
ag
es.
Su
b
s
eq
u
en
tly
,
r
e
ctif
ied
lin
ea
r
u
n
it
(
R
eL
U)
lay
er
s
in
tr
o
d
u
ce
n
o
n
-
lin
ea
r
ity
to
th
e
n
etwo
r
k
,
en
h
a
n
cin
g
its
ab
ilit
y
to
ca
p
tu
r
e
co
m
p
lex
p
atter
n
s
.
A
2
D
cr
o
s
s
-
ch
an
n
el
n
o
r
m
aliza
tio
n
lay
er
n
o
r
m
alize
s
ac
tiv
atio
n
s
ac
r
o
s
s
ch
an
n
els,
co
n
tr
ib
u
tin
g
t
o
im
p
r
o
v
e
d
m
o
d
el
p
er
f
o
r
m
a
n
ce
.
Fo
llo
win
g
th
is
,
2
D
m
ax
p
o
o
lin
g
lay
er
s
r
ed
u
ce
th
e
d
im
e
n
s
io
n
ality
o
f
f
ea
tu
r
e
m
ap
s
wh
ile
r
etain
in
g
ess
en
tial
in
f
o
r
m
ati
o
n
[
3
2
]
.
C
o
n
ca
ten
atio
n
m
e
r
g
es
f
e
atu
r
es
lear
n
ed
f
r
o
m
d
if
f
er
e
n
t
p
ath
way
s
with
in
th
e
n
et
wo
r
k
,
en
r
ich
in
g
t
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
6
9
2
-
4
7
0
4
4698
r
ep
r
esen
tatio
n
o
f
f
r
ac
t
u
r
e
-
r
el
ated
f
ea
tu
r
es.
B
atch
No
r
m
aliza
tio
n
is
ap
p
lied
t
o
ac
ce
ler
at
e
an
d
s
tab
ilize
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
lead
in
g
to
f
a
s
ter
co
n
v
er
g
en
ce
a
n
d
im
p
r
o
v
e
d
g
en
e
r
aliza
tio
n
.
Fu
lly
c
o
n
n
e
cted
lay
er
s
p
r
o
ce
s
s
h
ig
h
-
lev
el
f
ea
tu
r
es
ex
tr
ac
ted
b
y
co
n
v
o
lu
tio
n
al
lay
e
r
s
,
en
ab
lin
g
th
e
n
etwo
r
k
to
g
en
e
r
ate
p
r
ed
ictio
n
s
f
o
r
f
r
ac
tu
r
e
id
en
tific
atio
n
.
Dr
o
p
o
u
t
r
eg
u
lar
izatio
n
is
u
tili
ze
d
to
m
itig
ate
o
v
er
f
itti
n
g
b
y
r
an
d
o
m
ly
d
r
o
p
p
in
g
u
n
its
d
u
r
in
g
tr
ai
n
in
g
.
Fin
ally
,
So
f
tm
ax
ac
tiv
atio
n
f
ac
ilit
ates
r
o
b
u
s
t
clas
s
if
icatio
n
d
ec
is
io
n
s
b
y
allo
win
g
th
e
n
etwo
r
k
to
p
r
e
d
ict
class
es
b
ased
o
n
f
ea
tu
r
es
e
x
tr
ac
ted
t
h
r
o
u
g
h
v
ar
io
u
s
p
ar
allel
p
ath
s
.
T
a
b
le
2
p
r
esen
ts
th
e
s
p
ec
if
icatio
n
s
o
f
ea
ch
lay
er
in
th
e
p
r
o
p
o
s
ed
PDC
NN
m
o
d
el,
in
clu
d
in
g
th
e
la
y
er
ty
p
e,
p
r
o
p
er
ties
,
ac
tiv
atio
n
f
u
n
ctio
n
,
lear
n
ab
le
p
r
o
p
er
ty
,
a
n
d
n
u
m
b
er
o
f
lear
n
ab
le
p
a
r
am
eter
s
.
T
ab
le
2
.
T
h
e
s
p
ec
if
icatio
n
s
o
f
p
r
o
p
o
s
ed
PDC
NN
m
o
d
el
La
y
e
r
n
o
.
La
y
e
r
t
y
p
e
P
r
o
p
e
r
t
i
e
s
A
c
t
i
v
a
t
i
o
n
Le
a
r
n
a
b
l
e
p
r
o
p
e
r
t
y
N
u
mb
e
r
o
f
l
e
a
r
n
i
n
g
s
1
I
mag
e
I
n
p
u
t
2
2
7
×
2
2
7
×
3
i
m
a
g
e
s w
i
t
h
‘
z
e
r
o
c
e
n
t
e
r
’
n
o
r
m
a
l
i
z
a
t
i
o
n
2
2
7
(
S
)
×
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
(
C
)
×
1
(
B
)
-
0
2
2
-
D
c
o
n
v
o
l
u
t
i
o
n
3
2
3
×
3
c
o
n
v
o
l
u
t
i
o
n
s
w
i
t
h
s
t
r
i
d
e
[
1
×
1
]
a
n
d
p
a
d
d
i
n
g
‘
same
’
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
x
×
1
(
B
)
W
e
i
g
h
t
:
3
×
3
×
3
B
i
a
s
:
1
×
1
×
32
8
9
6
3
R
e
LU
R
e
LU
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
4
C
r
o
ss
-
c
h
a
n
n
e
l
n
o
r
m
a
l
i
z
a
t
i
o
n
C
r
o
ss
-
c
h
a
n
n
e
l
n
o
r
m
a
l
i
z
a
t
i
o
n
w
i
t
h
5
c
h
a
n
n
e
l
s
p
e
r
e
l
e
me
n
t
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
5
2
-
D
M
a
x
P
o
o
l
i
n
g
5
×
5
m
a
x
p
o
o
l
i
n
g
w
i
t
h
s
t
r
i
d
e
[
1
×
1
]
a
n
d
p
a
d
d
i
n
g
‘
same
’
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
6
2
-
D
c
o
n
v
o
l
u
t
i
o
n
3
2
3
×
3
c
o
n
v
o
l
u
t
i
o
n
s
w
i
t
h
s
t
r
i
d
e
[
1
×
1
]
a
n
d
p
a
d
d
i
n
g
‘
same
’
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
W
e
i
g
h
t
:
3
×
3
×
32
B
i
a
s
:
1
×
1
×
32
9
2
4
8
7
R
e
LU
R
e
LU
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
8
C
r
o
ss
-
c
h
a
n
n
e
l
N
o
r
mal
i
z
a
t
i
o
n
C
r
o
ss
-
c
h
a
n
n
e
l
n
o
r
m
a
l
i
z
a
t
i
o
n
w
i
t
h
5
c
h
a
n
n
e
l
s
p
e
r
e
l
e
me
n
t
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
9
2
-
D
M
a
x
P
o
o
l
i
n
g
5
×
5
m
a
x
p
o
o
l
i
n
g
w
i
t
h
s
t
r
i
d
e
[
1
×
1
]
a
n
d
p
a
d
d
i
n
g
‘
same
’
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
10
2
-
D
c
o
n
v
o
l
u
t
i
o
n
3
2
3
×
3
c
o
n
v
o
l
u
t
i
o
n
s
w
i
t
h
s
t
r
i
d
e
[
1
×
1
]
a
n
d
p
a
d
d
i
n
g
‘
same
’
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
W
e
i
g
h
t
:
3
×
3
×
32
B
i
a
s
:
1
×
1
×
32
9
2
4
8
11
R
e
LU
R
e
LU
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
12
C
r
o
ss
-
c
h
a
n
n
e
l
n
o
r
m
a
l
i
z
a
t
i
o
n
C
r
o
ss
-
c
h
a
n
n
e
l
n
o
r
m
a
l
i
z
a
t
i
o
n
w
i
t
h
5
c
h
a
n
n
e
l
s
p
e
r
e
l
e
me
n
t
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
13
2
-
D
M
a
x
P
o
o
l
i
n
g
5
×
5
m
a
x
p
o
o
l
i
n
g
w
i
t
h
s
t
r
i
d
e
[
1
×
1
]
a
n
d
p
a
d
d
i
n
g
‘
same
’
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
14
2
-
D
c
o
n
v
o
l
u
t
i
o
n
3
2
3
×
3
c
o
n
v
o
l
u
t
i
o
n
s
w
i
t
h
s
t
r
i
d
e
[
1
×
1
]
a
n
d
p
a
d
d
i
n
g
‘
same
’
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
W
e
i
g
h
t
:
3
×
3
×
32
B
i
a
s
:
1
×
1
×
32
8
9
6
15
R
e
LU
R
e
LU
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
16
C
r
o
ss
-
c
h
a
n
n
e
l
n
o
r
m
a
l
i
z
a
t
i
o
n
C
r
o
ss
-
c
h
a
n
n
e
l
n
o
r
m
a
l
i
z
a
t
i
o
n
w
i
t
h
5
c
h
a
n
n
e
l
s
p
e
r
e
l
e
me
n
t
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
17
2
-
D
M
a
x
P
o
o
l
i
n
g
5
×
5
m
a
x
p
o
o
l
i
n
g
w
i
t
h
s
t
r
i
d
e
[
1
×
1
]
a
n
d
p
a
d
d
i
n
g
‘
same
’
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
18
2
-
D
c
o
n
v
o
l
u
t
i
o
n
3
2
3
×
3
c
o
n
v
o
l
u
t
i
o
n
s
w
i
t
h
s
t
r
i
d
e
[
1
×
1
]
a
n
d
p
a
d
d
i
n
g
‘
same
’
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
W
e
i
g
h
t
:
3
×
3
×
32
B
i
a
s
:
1
×
1
×
32
9
2
4
8
19
R
e
LU
R
e
LU
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
20
C
r
o
ss
-
c
h
a
n
n
e
l
n
o
r
m
a
l
i
z
a
t
i
o
n
C
r
o
ss
-
c
h
a
n
n
e
l
n
o
r
m
a
l
i
z
a
t
i
o
n
w
i
t
h
5
c
h
a
n
n
e
l
s
p
e
r
e
l
e
me
n
t
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
21
2
-
D
M
a
x
P
o
o
l
i
n
g
5
×
5
m
a
x
p
o
o
l
i
n
g
w
i
t
h
s
t
r
i
d
e
[
1
×
1
]
a
n
d
p
a
d
d
i
n
g
‘
same
’
2
2
7
(
S
)
×
22
×
(S)
×
3
2
(
C
)
×
1
(
B
)
-
0
22
2
-
D
c
o
n
v
o
l
u
t
i
o
n
3
2
3
×
3
c
o
n
v
o
l
u
t
i
o
n
s
w
i
t
h
s
t
r
i
d
e
[
1
×
1
]
a
n
d
p
a
d
d
i
n
g
‘
same
’
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
W
e
i
g
h
t
:
3
×
3
×
3
2
…
B
i
a
s
:
1
×
1
×
32
9
2
4
8
23
R
e
LU
R
e
LU
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
24
C
r
o
ss
-
c
h
a
n
n
e
l
n
o
r
m
a
l
i
z
a
t
i
o
n
C
r
o
ss
-
c
h
a
n
n
e
l
n
o
r
m
a
l
i
z
a
t
i
o
n
w
i
t
h
5
c
h
a
n
n
e
l
s
p
e
r
e
l
e
me
n
t
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
25
2
-
D
M
a
x
P
o
o
l
i
n
g
5
×
5
m
a
x
p
o
o
l
i
n
g
w
i
t
h
s
t
r
i
d
e
[
1
×
1
]
a
n
d
p
a
d
d
i
n
g
‘
same
’
2
2
7
(
S
)
×
2
2
7
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
26
C
o
n
c
a
t
e
n
a
t
i
o
n
C
o
n
c
a
t
e
n
a
t
i
o
n
o
f
2
i
n
p
u
t
s
a
l
o
n
g
t
h
e
d
i
m
e
n
s
i
o
n
a
l
v
a
l
u
e
2
2
7
(
S
)
×
4
5
4
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
27
B
a
t
c
h
N
o
r
mal
i
z
a
t
i
o
n
B
a
t
c
h
n
o
r
ma
l
i
z
a
t
i
o
n
2
2
7
(
S
)
×
4
5
4
(
S
)
×
3
2
(
C
)
×
1
(
B
)
O
f
f
set
:
1
×
1
×
32
S
c
a
l
e
:
1
×
1
×
32
64
28
R
e
LU
R
e
LU
2
2
7
(
S
)
×
4
5
4
(
S
)
×
3
2
(
C
)
×
1
(
B
)
-
0
29
F
u
l
l
y
c
o
n
n
e
c
t
e
d
1
0
f
u
l
l
y
c
o
n
n
e
c
t
e
d
l
a
y
e
r
s
1
(
S
)
×
1
(
S
)
×
1
0
(
C
)
×
1
(
B
)
W
e
i
g
h
t
s
10
×
3
2
9
7
…
B
i
a
s
1
0
×
1
3
2
9
7
8
5
7
0
30
R
e
LU
R
e
LU
1
(
S
)
×
1
(
S
)
×
1
0
(
C
)
×
1
(
B
)
-
0
31
D
r
o
p
o
u
t
5
0
%
d
r
o
p
o
u
t
1
(
S
)
×
1
(
S
)
×
1
0
(
C
)
×
1
(
B
)
-
0
32
F
u
l
l
y
C
o
n
n
e
c
t
e
d
1
0
f
u
l
l
y
c
o
n
n
e
c
t
e
d
l
a
y
e
r
s
1
(
S
)
×
1
(
S
)
×
1
0
(
C
)
×
1
(
B
)
W
e
i
g
h
t
s
1
0
×
10
B
i
a
s
1
0
×
1
1
1
0
33
S
o
f
t
ma
x
S
o
f
t
ma
x
1
(
S
)
×
1
(
S
)
×
1
0
(
C
)
×
1
(
B
)
-
0
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:
2088
-
8
7
0
8
B
o
n
e
-
N
et:
a
p
a
r
a
llel d
ee
p
c
o
n
vo
lu
tio
n
a
l n
eu
r
a
l n
etw
o
r
k
-
b
a
s
ed
b
o
n
e
…
(
Md
.
Ha
s
a
n
I
ma
m
B
ijo
y
)
4699
2
.
4
.
P
er
f
o
r
m
a
nce
ev
a
lua
t
io
n
Af
ter
im
p
lem
en
tin
g
t
h
e
m
o
d
el
,
v
ar
io
u
s
p
e
r
f
o
r
m
an
ce
m
etr
ics ar
e
ca
lcu
lated
to
ass
ess
it
s
ef
f
ec
tiv
en
ess
in
a
b
in
ar
y
class
if
icatio
n
task
o
f
id
en
tify
i
n
g
b
o
n
e
f
r
ac
tu
r
es in
X
-
r
ay
im
ag
es.
T
h
e
f
ir
s
t step
in
v
o
lv
es g
en
er
atin
g
a
co
n
f
u
s
io
n
m
atr
ix
,
t
h
en
k
e
y
m
etr
ics
s
u
ch
as
tr
u
e
-
p
o
s
itiv
e
r
ate,
f
alse
-
n
eg
ativ
e
r
ate,
f
alse
-
p
o
s
itiv
e
r
ate,
an
d
tr
u
e
-
n
eg
ativ
e
r
ate
ar
e
d
e
r
iv
e
d
.
Fo
llo
win
g
th
is
,
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
F1
s
co
r
e,
an
d
er
r
o
r
ar
e
co
m
p
u
ted
to
d
eter
m
in
e
th
e
o
p
tim
al
m
o
d
el
f
o
r
b
o
n
e
f
r
ac
t
u
r
e
id
e
n
tific
atio
n
.
T
h
eir
eq
u
atio
n
s
“
(1
)
-
(
8)
”
ar
e
g
iv
en
b
el
o
w:
=
+
.
×
100%
(
1
)
=
+
×
100%
(
2
)
=
+
×
100%
(
3
)
=
+
×
100%
(
4
)
=
+
×
100%
(
5
)
=
+
×
100%
(
6
)
1
=
2
×
×
+
×
100%
(
7
)
=
+
.
×
100%
(
8
)
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
th
is
s
tu
d
y
,
th
r
ee
d
ee
p
lear
n
i
n
g
m
o
d
els
ex
p
lo
r
e
th
e
d
etec
tio
n
o
f
b
o
n
e
f
r
ac
tu
r
es
f
r
o
m
d
ig
i
tal
im
ag
es.
R
ig
o
r
o
u
s
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
is
ap
p
lied
to
tr
ain
an
d
v
alid
at
e
th
e
m
o
d
els
u
s
in
g
a
test
in
g
d
ataset.
T
o
m
ea
s
u
r
e
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
d
e
v
el
o
p
ed
m
o
d
els,
s
ev
er
al
p
er
f
o
r
m
an
ce
m
etr
ics
ar
e
co
m
p
u
ted
to
f
in
d
th
e
o
p
tim
al
s
o
lu
tio
n
.
T
h
e
th
r
ee
m
o
d
els
a
r
e
tr
ain
ed
with
8
,
8
8
4
im
ag
es
an
d
v
alid
ated
with
1
,
8
9
9
i
m
ag
es.
Du
r
in
g
th
e
im
p
lem
en
tatio
n
p
h
ase,
s
o
m
e
ca
s
es
o
f
ab
latio
n
s
tu
d
y
ar
e
ca
r
r
ied
o
u
t
to
d
eter
m
in
e
th
e
b
est
-
s
u
ited
p
ar
am
eter
s
f
o
r
th
e
p
r
o
p
o
s
ed
m
o
d
el.
T
h
e
m
o
d
els
ar
e
im
p
lem
en
ted
u
s
in
g
1
0
0
ep
o
ch
s
an
d
a
b
atc
h
s
ize
o
f
6
4
.
Ho
wev
e
r
,
in
th
e
ca
s
e
o
f
th
e
tr
a
d
itio
n
al
C
NN
an
d
th
e
p
r
et
r
ain
ed
Mo
b
ileN
et
-
V2
m
o
d
el,
p
e
r
f
o
r
m
an
ce
is
n
o
t
s
atis
f
ac
to
r
y
,
a
n
d
is
s
u
es
ar
is
e
d
u
r
in
g
th
e
im
p
le
m
en
tatio
n
p
h
ase.
C
o
n
s
eq
u
en
t
ly
,
a
n
ewly
d
ev
elo
p
e
d
p
r
o
p
o
s
ed
PDC
NN
m
o
d
el
p
er
f
o
r
m
s
v
er
y
well.
Fo
u
r
a
b
latio
n
ca
s
e
s
tu
d
ies
ar
e
co
n
s
i
d
er
ed
to
v
alid
ate
th
e
p
r
o
p
o
s
ed
m
o
d
el
f
o
r
b
o
n
e
f
r
ac
tu
r
e
id
e
n
tific
atio
n
.
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
ese
s
tu
d
ies is
d
etailed
b
elo
w:
3
.
1
.
P
er
f
o
r
m
a
nce
o
f
CNN
a
nd
M
o
bi
leNe
t
-
V2
I
n
ter
m
s
o
f
p
er
f
o
r
m
an
ce
,
b
o
t
h
th
e
tr
ad
itio
n
al
C
NN
an
d
M
o
b
ileNet
-
V2
m
o
d
els
ex
h
ib
it
f
lu
ctu
atin
g
r
esu
lts
,
in
d
icatin
g
u
n
d
e
r
f
itti
n
g
is
s
u
es.
T
h
e
c
o
n
f
u
s
io
n
m
atr
ices
o
f
th
e
C
NN
an
d
M
o
b
ile
Net
-
V2
m
o
d
els
a
r
e
p
r
esen
ted
in
T
ab
le
3
an
d
p
er
f
o
r
m
an
ce
m
etr
ices
s
h
wo
n
in
T
ab
le
4
.
Fro
m
T
ab
le
4
,
C
NN
p
r
o
v
id
e
7
4
.
5
1
%
ac
cu
r
ac
y
w
h
ile
Mo
b
ileNet
-
V2
g
ain
ed
8
1
.
2
0
%
ac
c
u
r
ac
y
.
E
x
a
m
in
atio
n
o
f
Fig
u
r
e
6
(
a)
a
n
d
6
(
b
)
an
d
Fig
u
r
e
7
(
a)
an
d
7
(
b
)
r
ev
ea
ls
ac
cu
r
ac
y
(
Fig
u
r
e
6
(
a)
a
n
d
Fig
u
r
e
7
(
a
)
an
d
lo
s
s
cu
r
v
es
(
Fig
u
r
e
6
(
b
)
an
d
Fig
u
r
e
7
(
b
)
th
at
f
u
r
th
er
em
p
h
asize
th
e
u
n
d
e
r
f
itti
n
g
p
r
o
b
lem
s
ex
p
er
ien
ce
d
b
y
b
o
t
h
m
o
d
els
C
NN
a
n
d
Mo
v
ileNet
-
V2
,
r
esp
ec
tv
iely
.
Un
d
er
f
itti
n
g
o
cc
u
r
s
wh
en
a
m
o
d
el
is
u
n
ab
le
to
ca
p
tu
r
e
th
e
u
n
d
er
ly
in
g
p
at
ter
n
s
in
th
e
d
ata,
r
esu
ltin
g
in
p
o
o
r
p
e
r
f
o
r
m
an
c
e
an
d
lo
w
ac
cu
r
ac
y
.
I
n
th
e
ca
s
e
o
f
th
e
tr
ad
itio
n
al
C
NN
an
d
Mo
b
ileNet
-
V2
m
o
d
els,
th
is
u
n
d
er
f
itti
n
g
p
h
e
n
o
m
en
o
n
is
ev
id
en
t
in
th
eir
in
a
b
ilit
y
to
ad
e
q
u
ately
lear
n
f
r
o
m
th
e
tr
ain
i
n
g
d
ata,
lead
in
g
to
in
c
o
n
s
is
ten
t a
n
d
s
u
b
o
p
tim
al
r
esu
lts
.
T
ab
le
3
.
C
o
n
f
u
s
io
n
m
atr
ix
f
o
r
ap
p
lied
th
r
ee
m
o
d
els
M
o
d
e
l
TP
FN
FP
TN
C
N
N
7
6
9
3
7
0
1
1
4
6
4
6
M
o
b
i
l
e
N
e
t
-
V2
8
5
3
2
6
6
1
0
5
6
8
9
P
D
C
N
N
9
4
1
84
51
8
2
3
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
6
9
2
-
4
7
0
4
4700
Tab
le 4
.
P
e
rfo
rm
a
n
c
e
m
e
tri
c
e
s fo
r
CNN
a
n
d
M
o
b
il
e
Ne
t
-
V2
m
o
d
e
l
M
o
d
e
l
A
c
c
u
r
a
c
y
TPR
F
N
R
FPR
TN
R
P
r
e
c
i
s
i
o
n
F
1
S
c
o
r
e
Er
r
o
r
r
a
t
e
C
N
N
7
4
.
5
1
6
7
.
5
2
3
2
.
4
8
1
5
.
0
0
8
5
.
0
0
8
7
.
0
9
7
6
.
0
6
2
5
.
4
9
M
o
b
i
l
e
N
e
t
-
V2
8
1
.
2
0
7
6
.
2
3
2
3
.
7
7
1
3
.
2
2
8
6
.
7
8
8
9
.
0
4
8
2
.
1
4
1
9
.
5
4
(
a)
(
b
)
Fig
u
r
e
6
.
T
h
e
(
a)
ac
cu
r
ac
y
g
r
a
p
h
an
d
(
b
)
lo
s
s
g
r
ap
h
f
o
r
tr
ad
it
io
n
al
C
NN
m
o
d
el
(
a)
(
b
)
Fig
u
r
e
7
.
T
h
e
(
a)
ac
cu
r
ac
y
g
r
a
p
h
an
d
(
b
)
lo
s
s
g
r
ap
h
f
o
r
Mo
b
i
leNe
t
-
V2
m
o
d
el
3
.
2
.
P
er
f
o
r
m
a
nce
o
f
pro
po
s
ed
P
DCCN
m
o
del w
it
h a
bla
t
io
n study
I
n
th
e
p
r
ec
ed
i
n
g
s
ec
tio
n
,
b
o
th
m
o
d
els
f
ailed
to
d
eliv
er
s
atis
f
ac
to
r
y
r
esu
lts
.
No
w,
it'
s
im
p
er
ativ
e
to
v
alid
ate
th
e
p
r
o
p
o
s
ed
m
o
d
el
to
d
em
o
n
s
tr
ate
th
at
th
e
n
ewly
d
ev
elo
p
e
d
PDC
NN
m
o
d
el
s
er
v
es
as
th
e
o
p
tim
al
s
o
lu
tio
n
f
o
r
b
o
n
e
f
r
ac
tu
r
e
d
etec
tio
n
.
I
n
th
e
v
alid
atio
n
p
r
o
c
ess
,
f
o
u
r
d
etailed
ca
s
e
s
tu
d
ie
s
ar
e
co
n
d
u
cted
b
y
alter
in
g
h
y
p
e
r
p
ar
am
eter
s
s
u
ch
as
k
er
n
el
s
ize,
lo
s
s
f
u
n
ctio
n
,
p
o
o
lin
g
lay
e
r
,
an
d
o
p
tim
izer
.
T
h
e
r
esu
lts
o
f
th
e
ab
latio
n
s
tu
d
y
ar
e
p
r
esen
ted
i
n
T
ab
le
5
.
T
h
is
co
m
p
r
eh
en
s
i
v
e
an
aly
s
is
aim
s
to
ascer
tain
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
in
d
etec
tin
g
b
o
n
e
f
r
ac
tu
r
es
an
d
to
id
e
n
tify
th
e
o
p
tim
al
co
n
f
ig
u
r
atio
n
f
o
r
ac
h
iev
i
n
g
th
e
h
ig
h
est p
er
f
o
r
m
an
ce
.
Fro
m
th
e
ab
latio
n
s
tu
d
ies
i
n
T
ab
le
5
,
v
ar
i
o
u
s
co
n
v
o
l
u
tio
n
al
lay
e
r
k
e
r
n
el
s
izes
h
av
e
b
ee
n
in
v
esti
g
ated
.
Fo
u
r
k
er
n
el
s
iz
es
(
2
,
3
,
4
,
an
d
5
)
ar
e
co
m
p
iled
an
d
ev
alu
ated
.
No
tab
ly
,
a
k
er
n
el
s
ize
o
f
3
ac
h
iev
ed
th
e
m
a
x
im
u
m
ac
c
u
r
ac
y
,
r
ea
c
h
in
g
8
8
.
6
1
%,
with
a
r
elativ
ely
lo
w
p
e
r
-
ep
o
c
h
tr
ain
in
g
tim
e
o
f
1
2
7
s
ec
o
n
d
s
.
As
a
r
esu
lt,
a
k
er
n
el
s
ize
o
f
3
is
s
elec
ted
f
o
r
im
p
lem
en
tatio
n
in
th
e
PDC
NN
m
o
d
el.
T
o
o
p
tim
iz
e
p
er
f
o
r
m
an
ce
,
d
if
f
er
en
t l
o
s
s
f
u
n
ctio
n
s
ar
e
ev
alu
ated
,
in
clu
d
in
g
b
in
ar
y
cr
o
s
s
-
en
tr
o
p
y
,
ca
teg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
,
an
d
m
ea
n
s
q
u
ar
ed
e
r
r
o
r
.
T
h
e
m
o
d
el
attain
ed
its
h
ig
h
est test
ac
cu
r
ac
y
o
f
8
9
.
6
5
% wh
en
u
tili
zin
g
th
e
ca
teg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
f
u
n
ctio
n
.
C
o
n
s
eq
u
en
tly
,
th
is
lo
s
s
f
u
n
ctio
n
is
ch
o
s
en
f
o
r
i
n
teg
r
a
tio
n
in
to
th
e
f
in
al
m
o
d
el.
Fu
r
th
e
r
ex
p
er
im
en
tatio
n
in
v
o
lv
ed
co
m
p
ar
in
g
m
a
x
-
p
o
o
lin
g
a
n
d
av
er
a
g
e
p
o
o
lin
g
lay
er
s
.
I
t
is
f
o
u
n
d
th
at
th
e
m
o
d
el
ac
h
iev
e
d
its
p
e
ak
p
er
f
o
r
m
an
ce
with
th
e
m
ax
-
p
o
o
lin
g
lay
e
r
,
r
esu
ltin
g
in
an
ac
cu
r
ac
y
o
f
9
0
.
4
7
%.
Mo
r
eo
v
er
,
th
e
o
p
tim
izatio
n
p
r
o
ce
s
s
in
clu
d
ed
test
in
g
f
o
u
r
d
is
tin
ct
o
p
tim
izer
s
—
SGD,
Ad
am
,
R
MSp
r
o
p
,
an
d
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:
2088
-
8
7
0
8
B
o
n
e
-
N
et:
a
p
a
r
a
llel d
ee
p
c
o
n
vo
lu
tio
n
a
l n
eu
r
a
l n
etw
o
r
k
-
b
a
s
ed
b
o
n
e
…
(
Md
.
Ha
s
a
n
I
ma
m
B
ijo
y
)
4701
Ad
am
ax
—
ea
ch
with
a
lear
n
in
g
r
ate
o
f
0
.
0
0
1
a
n
d
a
b
atch
s
ize
o
f
6
4
.
R
em
ar
k
ab
l
y
,
th
e
h
ig
h
est
test
ac
cu
r
ac
y
o
f
9
2
.
8
9
% is
ac
h
iev
e
d
b
y
ad
am
o
p
tim
izer
,
s
u
r
p
ass
in
g
all
p
r
e
v
io
u
s
tu
n
in
g
ef
f
o
r
ts
.
Af
ter
m
eticu
lo
u
s
p
ar
am
ete
r
tu
n
in
g
a
n
d
s
elec
tio
n
,
th
e
f
in
al
m
o
d
el
s
h
o
wca
s
ed
s
atis
f
ac
to
r
y
p
er
f
o
r
m
an
ce
.
Su
b
s
eq
u
en
tly
,
f
in
e
-
tu
n
in
g
p
ar
am
eter
s
ar
e
co
n
s
id
er
ed
to
r
ef
in
e
t
h
e
PDC
N
N
m
o
d
el,
an
d
th
e
p
er
f
o
r
m
an
ce
m
etr
ics
ar
e
p
r
es
en
ted
in
T
ab
le
6
.
Fig
u
r
e
8
illu
s
tr
ates
th
e
ac
cu
r
ac
y
an
d
lo
s
s
cu
r
v
es
f
o
r
th
e
PDC
NN
m
o
d
el,
d
em
o
n
s
tr
atin
g
a
lack
o
f
o
v
er
f
itti
n
g
o
r
u
n
d
e
r
f
itti
n
g
d
u
r
in
g
tr
ain
in
g
,
s
ee
F
ig
u
r
e
8
(
a)
an
d
Fig
u
r
e
8
(
b
)
.
B
o
th
th
e
tr
ain
in
g
an
d
v
alid
atio
n
cu
r
v
es
s
m
o
o
t
h
ly
c
o
n
v
er
g
e,
with
m
in
im
al
d
is
p
a
r
ity
b
etwe
en
t
h
em
.
Fu
r
th
er
m
o
r
e
,
th
e
lo
s
s
cu
r
v
es
s
tead
ily
d
ec
r
ea
s
e
f
r
o
m
th
e
in
i
tial
to
th
e
f
in
al
ep
o
ch
,
m
ain
t
ain
in
g
a
s
m
all
g
ap
th
r
o
u
g
h
o
u
t.
T
h
is
s
tab
ilit
y
in
th
e
co
n
v
er
g
en
ce
o
f
ac
cu
r
ac
y
an
d
lo
s
s
cu
r
v
es
u
n
d
er
s
co
r
es
th
e
r
o
b
u
s
tn
ess
an
d
ef
f
ec
tiv
en
ess
o
f
th
e
d
ev
elo
p
e
d
PDC
NN
m
o
d
el
in
b
o
n
e
f
r
ac
tu
r
e
d
etec
tio
n
.
T
ab
le
5
.
Per
f
o
r
m
an
ce
r
esu
lts
o
f
f
o
u
r
ca
s
e
o
f
a
b
alatio
n
s
tu
d
ies
A
b
l
a
t
i
o
n
S
t
u
d
y
C
o
n
f
i
g
u
r
a
t
i
o
n
P
a
r
a
me
t
e
r
Ep
o
c
h
×
T
i
me
Te
st
A
c
c
u
r
a
c
y
F
i
n
d
i
n
g
C
h
a
n
g
i
n
g
k
e
r
n
e
l
si
z
e
1
2
1
0
0
×
1
2
7
s
8
7
.
3
4
%
I
mp
r
o
v
e
d
2
3
1
0
0
×
1
2
7
s
8
8
.
6
1
%
I
mp
r
o
v
e
d
3
4
1
0
0
×
1
5
2
s
8
6
.
5
1
%
D
r
o
p
e
d
4
5
1
0
0
×
1
6
3
s
8
2
.
4
7
%
D
r
o
p
e
d
C
h
a
n
g
i
n
g
t
h
e
l
o
ss
f
u
n
c
t
i
o
n
1
B
i
n
a
r
y
c
r
o
ss
-
e
n
t
r
o
p
y
l
o
ss
1
0
0
×
1
2
4
s
8
7
.
1
9
%
D
r
o
p
e
d
2
C
a
t
e
g
o
r
i
c
a
l
c
r
o
ss
-
e
n
t
r
o
p
y
1
0
0
×
1
2
4
s
8
9
.
6
5
%
I
mp
r
o
v
e
d
3
M
e
a
n
sq
u
a
r
e
d
e
r
r
o
r
1
0
0
×
1
2
4
s
8
4
.
2
7
%
D
r
o
p
e
d
C
h
a
n
g
i
n
g
p
o
o
l
i
n
g
l
a
y
e
r
1
M
a
x
1
0
0
×
1
2
4
s
9
0
.
4
7
%
I
mp
r
o
v
e
d
2
A
v
e
r
a
g
e
1
0
0
×
1
2
4
s
8
8
.
1
4
%
D
r
o
p
e
d
C
h
a
n
g
i
n
g
o
p
t
i
m
i
z
e
r
1
S
G
D
1
0
0
×
1
2
4
s
8
8
.
3
1
%
D
r
o
p
2
A
d
a
m
1
0
0
×
1
2
4
s
9
2
.
8
9
%
H
i
g
h
e
s
t
3
R
M
S
p
r
o
p
1
0
0
×
1
2
4
s
9
1
.
5
6
%
D
r
o
p
e
d
4
A
d
a
ma
x
1
0
0
×
1
2
4
s
9
0
.
4
7
%
D
r
o
p
e
d
Tab
le 6
.
P
e
rfo
rm
a
n
c
e
m
e
tri
c
e
s fo
r
p
ro
p
o
se
d
P
DCN
N m
o
d
e
l
M
o
d
e
l
A
c
c
u
r
a
c
y
TPR
F
N
R
FPR
TN
R
P
r
e
c
i
s
i
o
n
F
1
S
c
o
r
e
Er
r
o
r
r
a
t
e
P
D
C
N
N
9
2
.
8
9
9
1
.
8
0
8
.
2
0
5
.
8
4
9
4
.
1
6
9
4
.
8
6
9
3
.
3
1
7
.
1
1
(
a)
(
b
)
Fig
u
r
e
8
.
T
h
e
(
a)
ac
cu
r
ac
y
g
r
a
p
h
an
d
(
b
)
lo
s
s
g
r
ap
h
f
o
r
p
r
o
p
o
s
ed
PDC
NN
m
o
d
el
3.
3
.
Co
m
pera
t
iv
e
a
na
ly
s
is
a
nd
dis
c
us
s
io
ns
I
n
co
m
p
ar
is
o
n
with
p
r
e
v
io
u
s
s
tu
d
ies
p
r
esen
ted
i
n
T
ab
le
7
,
o
u
r
s
tu
d
y
s
u
r
p
ass
es
th
e
p
er
f
o
r
m
a
n
ce
ac
h
iev
ed
b
y
ea
r
lier
wo
r
k
s
in
b
o
n
e
f
r
ac
tu
r
e
r
ec
o
g
n
itio
n
.
No
ta
b
ly
,
o
u
r
p
r
o
p
o
s
ed
PDC
NN
m
o
d
el
o
u
tp
er
f
o
r
m
s
all
p
r
ev
io
u
s
ly
r
e
p
o
r
ted
m
eth
o
d
s
.
T
h
is
r
em
ar
k
ab
le
ac
h
ie
v
em
en
t
u
n
d
er
s
co
r
es
th
e
ef
f
ec
tiv
en
ess
an
d
ad
v
a
n
ce
m
en
t
o
f
o
u
r
ap
p
r
o
ac
h
in
ac
cu
r
ately
d
etec
tin
g
b
o
n
e
f
r
ac
tu
r
e
s
.
T
o
m
ain
tain
m
eth
o
d
o
lo
g
ic
al
co
h
er
e
n
ce
,
we
in
co
r
p
o
r
ated
t
h
e
ca
n
n
y
ed
g
e
d
etec
tio
n
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
e
alo
n
g
s
id
e
a
u
g
m
en
ted
im
ag
es.
Ad
d
itio
n
ally
,
we
co
n
d
u
cted
r
i
g
o
r
o
u
s
v
alid
atio
n
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
th
r
o
u
g
h
ab
latio
n
s
tu
d
ies.
T
h
ese
ef
f
o
r
ts
co
llectiv
ely
co
n
tr
i
b
u
ted
to
s
i
g
n
if
ican
t
im
p
r
o
v
em
en
ts
in
a
cc
u
r
ac
y
r
an
g
i
n
g
f
r
o
m
1
.
8
9
%
to
1
7
.
8
9
%.
Su
ch
s
u
b
s
tan
tial
en
h
an
ce
m
en
ts
h
ig
h
lig
h
t
th
e
s
u
cc
ess
o
f
o
u
r
s
tu
d
y
in
p
u
s
h
in
g
th
e
b
o
u
n
d
a
r
ie
s
o
f
b
o
n
e
f
r
ac
t
u
r
e
d
etec
tio
n
ca
p
ab
ilit
ies.
Evaluation Warning : The document was created with Spire.PDF for Python.