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.
14
,
No
.
6
,
Dec
em
b
er
20
24
,
p
p
.
6
4
6
6
~
6
4
7
7
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
14
i
6
.
pp
6
4
6
6
-
6
4
7
7
6466
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Bo
ne f
ra
ctur
e cla
ss
ificatio
n using
c
o
nv
o
lutiona
l neural network
a
rchitec
t
ure
for h
ig
h
-
a
ccuracy
ima
g
e clas
sifica
tion
So
lik
hu
n
1
,
Ag
us
P
er
da
na
Wi
nd
a
rt
o
2
,
P
utr
a
ma
Alk
ha
iri
2
1
I
n
f
o
r
mat
i
c
s E
n
g
i
n
e
e
r
i
n
g
P
r
o
g
r
a
m,
S
TI
K
O
M
T
u
n
a
s
B
a
n
g
sa,
P
e
ma
t
a
n
g
si
a
n
t
a
r
,
I
n
d
o
n
e
s
i
a
2
I
n
f
o
r
mat
i
o
n
S
y
st
e
ms Pr
o
g
r
a
m
,
S
TI
K
O
M
T
u
n
a
s
B
a
n
g
sa,
P
e
m
a
t
a
n
g
si
a
n
t
a
r
,
I
n
d
o
n
e
si
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
y
2
1
,
2
0
2
4
R
ev
is
ed
J
u
l 1
3
,
2
0
2
4
Acc
ep
ted
J
u
l 1
7
,
2
0
2
4
Th
is
re
se
a
rc
h
i
n
tro
d
u
c
e
s
a
n
in
n
o
v
a
ti
v
e
m
e
th
o
d
f
o
r
fra
c
t
u
re
c
las
sifica
ti
o
n
u
sin
g
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
t
wo
rk
s
(CNN
)
f
o
r
h
ig
h
-
a
c
c
u
ra
c
y
ima
g
e
c
las
sifica
ti
o
n
.
T
h
e
st
u
d
y
a
d
d
re
ss
e
s
th
e
n
e
e
d
t
o
imp
ro
v
e
t
h
e
s
u
b
je
c
ti
v
it
y
a
n
d
li
m
it
e
d
a
c
c
u
ra
c
y
o
f
trad
it
io
n
a
l
m
e
th
o
d
s.
By
h
a
rn
e
ss
in
g
t
h
e
c
a
p
a
b
il
it
y
o
f
CNN
s
to
a
u
to
n
o
m
o
u
sly
e
x
trac
t
h
iera
rc
h
ica
l
fe
a
tu
re
s
fro
m
m
e
d
ica
l
ima
g
e
s,
th
is
re
se
a
rc
h
su
r
p
a
ss
e
s
th
e
li
m
it
a
ti
o
n
s
o
f
m
a
n
u
a
l
in
ter
p
re
tatio
n
a
n
d
e
x
ist
in
g
a
u
to
m
a
ted
s
y
ste
m
s.
Th
e
g
o
a
l
is
t
o
c
re
a
te
a
r
o
b
u
st
CNN
-
b
a
se
d
m
e
th
o
d
o
l
o
g
y
fo
r
p
re
c
ise
a
n
d
re
li
a
b
le
fra
c
tu
re
c
las
sifica
ti
o
n
,
p
o
te
n
ti
a
ll
y
re
v
o
l
u
ti
o
n
izin
g
c
u
rre
n
t
d
iag
n
o
st
ic
p
ra
c
ti
c
e
s.
Th
e
d
a
tas
e
t
fo
r
th
is
re
se
a
rc
h
is
so
u
r
c
e
d
fro
m
Ka
g
g
le'
s
p
u
b
li
c
m
e
d
ica
l
ima
g
e
re
p
o
sito
r
y
,
e
n
su
ri
n
g
a
d
iv
e
rse
ra
n
g
e
o
f
fra
c
tu
re
ima
g
e
s.
Th
is
stu
d
y
h
i
g
h
li
g
h
ts
CNN
s'
p
o
ten
ti
a
l
t
o
si
g
n
ifi
c
a
n
t
ly
e
n
h
a
n
c
e
d
iag
n
o
stic
p
re
c
isio
n
,
l
e
a
d
in
g
t
o
m
o
re
e
ffe
c
ti
v
e
trea
tme
n
ts
a
n
d
imp
ro
v
e
d
p
a
ti
e
n
t
c
a
re
in
o
rth
o
p
e
d
ics
.
Th
e
n
o
v
e
lt
y
l
ies
in
t
h
e
u
n
iq
u
e
a
p
p
li
c
a
ti
o
n
o
f
CNN
a
rc
h
it
e
c
tu
r
e
fo
r
fra
c
tu
re
c
las
sifica
ti
o
n
,
a
n
a
re
a
n
o
t
e
x
ten
siv
e
ly
e
x
p
lo
re
d
b
e
f
o
re
.
Tes
t
in
g
re
su
lt
s
sh
o
w
a
si
g
n
if
ica
n
t
im
p
ro
v
e
m
e
n
t
in
c
las
sifica
ti
o
n
a
c
c
u
ra
c
y
,
wit
h
t
h
e
p
ro
p
o
se
d
m
o
d
e
l
a
c
h
iev
i
n
g
a
n
a
c
c
u
ra
c
y
ra
te
o
f
0
.
9
9
2
2
c
o
m
p
a
re
d
to
Re
sN
e
t5
0
'
s
0
.
9
8
4
4
.
T
h
e
re
se
a
rc
h
su
g
g
e
sts
t
h
a
t
a
d
o
p
t
in
g
CNN
-
b
a
se
d
sy
ste
m
s
in
m
e
d
ica
l
p
ra
c
ti
c
e
c
a
n
e
n
h
a
n
c
e
d
iag
n
o
stic
a
c
c
u
ra
c
y
,
o
p
ti
m
ize
trea
tme
n
t
p
lan
s,
a
n
d
im
p
ro
v
e
p
a
ti
e
n
t
o
u
tco
m
e
s.
K
ey
w
o
r
d
s
:
B
o
n
e
f
r
ac
tu
r
es
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Dee
p
lear
n
in
g
I
m
ag
e
class
if
icatio
n
Me
d
ical
im
ag
in
g
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
:
So
lik
h
u
n
I
n
f
o
r
m
atics E
n
g
in
ee
r
in
g
Pro
g
r
am
,
STI
KOM
T
u
n
as B
an
g
s
a
Kar
tin
i
s
tr
ee
t
,
Pro
k
lam
asi
,
Sian
tar
B
ar
.
,
Pem
atan
g
s
ian
tar
,
N
o
r
th
Su
m
atr
a
2
1
1
4
3
,
I
n
d
o
n
esia
E
m
ail: so
lik
h
u
n
@
am
ik
tu
n
asb
an
g
s
a.
ac
.
id
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
tech
n
o
lo
g
ical
lan
d
s
ca
p
e
h
as
witn
ess
ed
u
n
p
r
ec
e
d
en
ted
p
r
o
g
r
ess
,
with
a
s
ig
n
if
ican
t
f
o
cu
s
o
n
ar
tific
ial
in
tellig
en
ce
(
AI
)
as
a
tr
an
s
f
o
r
m
ativ
e
f
o
r
c
e
in
v
ar
i
o
u
s
in
d
u
s
tr
ies
[
1
]
–
[
3
]
.
On
e
o
f
th
e
m
ain
d
r
iv
i
n
g
f
ac
to
r
s
b
e
h
in
d
th
is
s
u
r
g
e
is
th
e
ex
p
o
n
en
tial
g
r
o
wth
in
co
m
p
u
tin
g
p
o
wer
,
wh
ich
h
as
e
n
ab
l
ed
th
e
ex
ec
u
tio
n
o
f
co
m
p
lex
AI
alg
o
r
ith
m
s
at
s
ca
le
[
4
]
.
I
n
teg
r
atin
g
AI
in
to
ev
e
r
y
d
ay
life
is
b
ec
o
m
in
g
in
c
r
ea
s
in
g
ly
wid
esp
r
ea
d
,
with
ap
p
licatio
n
s
r
an
g
i
n
g
f
r
o
m
v
ir
tu
al
ass
is
tan
ts
an
d
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
s
to
au
to
n
o
m
o
u
s
v
eh
icles
an
d
ad
v
an
ce
d
h
ea
lth
ca
r
e
d
iag
n
o
s
tics
.
I
n
th
e
h
ea
lth
ca
r
e
s
ec
to
r
,
A
I
h
as
b
ec
o
m
e
a
b
r
ea
k
th
r
o
u
g
h
.
T
h
e
ab
ilit
y
o
f
AI
alg
o
r
ith
m
s
to
p
r
o
ce
s
s
lar
g
e
am
o
u
n
ts
o
f
m
e
d
ical
d
ata,
f
r
o
m
elec
tr
o
n
ic
h
ea
lth
r
ec
o
r
d
s
to
m
ed
ical
im
ag
in
g
,
h
as
p
av
ed
th
e
way
f
o
r
m
o
r
e
ac
c
u
r
ate
d
iag
n
o
s
tics
an
d
p
er
s
o
n
a
lized
tr
ea
tm
en
t
p
lan
s
[
5
]
,
[
6
]
.
Ma
ch
in
e
lea
r
n
in
g
m
o
d
els
ca
n
n
o
w
p
r
ed
ict
d
is
ea
s
e,
an
aly
ze
g
e
n
etic
d
ata,
an
d
ass
is
t
d
o
cto
r
s
in
m
a
k
in
g
d
ec
is
io
n
s
,
u
ltima
tely
im
p
r
o
v
in
g
p
atien
t
o
u
tco
m
es
[
7
]
,
[
8
]
.
I
n
th
e
h
ea
lth
ca
r
e
s
ec
to
r
,
AI
h
as
b
ec
o
m
e
a
b
r
ea
k
th
r
o
u
g
h
.
T
h
e
ab
ilit
y
o
f
AI
alg
o
r
ith
m
s
t
o
p
r
o
ce
s
s
lar
g
e
am
o
u
n
ts
o
f
m
ed
ical
d
ata,
f
r
o
m
elec
tr
o
n
ic
h
ea
lth
r
ec
o
r
d
s
t
o
m
ed
ical
im
a
g
in
g
,
h
as
p
av
ed
t
h
e
way
f
o
r
m
o
r
e
a
cc
u
r
ate
d
iag
n
o
s
tics
an
d
p
e
r
s
o
n
alize
d
tr
ea
tm
en
t
p
lan
s
.
Ma
ch
in
e
lear
n
in
g
m
o
d
els
ca
n
n
o
w
p
r
e
d
ict
d
is
ea
s
e,
an
aly
ze
g
en
etic
d
ata,
an
d
h
elp
d
o
c
to
r
s
m
ak
e
d
ec
is
io
n
s
,
u
ltima
tely
im
p
r
o
v
in
g
p
atien
t
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
fr
a
ctu
r
e
cla
s
s
ifica
tio
n
u
s
in
g
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
et
w
o
r
k
a
r
ch
itectu
r
e
fo
r
…
(
S
o
likh
u
n
)
6467
o
u
tco
m
es.
At
th
e
f
o
r
ef
r
o
n
t
o
f
AI
,
d
ee
p
lea
r
n
in
g
h
as
s
p
ar
k
e
d
a
r
ev
o
l
u
tio
n
in
m
ac
h
in
e
lear
n
in
g
m
eth
o
d
o
lo
g
ies
[
9
]
–
[
1
1
]
.
T
h
is
ap
p
r
o
ac
h
,
in
s
p
i
r
ed
b
y
th
e
h
u
m
a
n
b
r
ain
'
s
s
tr
u
ctu
r
e
an
d
f
u
n
ctio
n
,
in
v
o
lv
es
u
s
in
g
n
e
u
r
al
n
etwo
r
k
s
with
m
an
y
lay
er
s
(
d
ee
p
n
e
u
r
a
l
n
etwo
r
k
s
)
.
Dee
p
lear
n
in
g
e
x
ce
ls
at
f
ea
tu
r
e
ex
tr
ac
tio
n
a
n
d
p
atter
n
r
ec
o
g
n
itio
n
,
allo
win
g
s
y
s
tem
s
to
au
to
m
atic
ally
lear
n
co
m
p
lex
r
e
p
r
esen
tat
io
n
s
f
r
o
m
d
ata.
T
h
is
p
a
r
ad
ig
m
s
h
if
t
h
as
r
esu
lted
in
b
r
ea
k
th
r
o
u
g
h
s
in
im
ag
e
an
d
s
p
ee
ch
r
ec
o
g
n
itio
n
an
d
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
an
d
s
ig
n
if
ican
tly
en
h
an
ce
d
th
e
ca
p
ab
ilit
ies
o
f
AI
s
y
s
tem
s
.
W
h
en
an
al
y
zin
g
d
is
ea
s
e
u
s
in
g
d
ig
ital
im
ag
es,
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
is
cu
r
r
en
tly
h
ig
h
ly
r
ec
o
m
m
e
n
d
ed
.
Me
d
ica
l
d
is
ea
s
e
in
f
r
ac
tu
r
es
is
clo
s
el
y
r
elate
d
t
o
d
ig
ital
im
ag
es.
T
h
er
e
ar
e
m
an
y
ty
p
es
o
f
m
ed
ical
d
is
ea
s
es,
s
u
ch
as
f
r
ac
tu
r
es,
an
d
C
NN
ca
n
tak
e
p
ar
t
in
th
ese
f
r
ac
tu
r
e
ca
s
es
[
1
2
]
–
[
1
4
]
.
F
r
a
c
t
u
r
es
a
r
e
a
c
o
m
m
o
n
i
n
j
u
r
y
a
n
d
h
a
v
e
s
i
g
n
i
f
i
c
a
n
t
i
m
p
l
ic
a
t
i
o
n
s
f
o
r
p
a
t
i
e
n
t
c
a
r
e
a
n
d
t
r
e
a
t
m
e
n
t
p
l
a
n
n
i
n
g
[
1
5
]
–
[
1
7
]
.
A
c
c
u
r
a
t
e
f
r
a
c
t
u
r
e
c
l
a
s
s
i
f
i
c
at
i
o
n
i
s
i
m
p
o
r
t
a
n
t
i
n
g
u
i
d
i
n
g
h
e
a
l
t
h
p
r
o
f
e
s
s
i
o
n
al
s
t
o
w
a
r
d
a
p
p
r
o
p
r
i
a
t
e
i
n
t
e
r
v
e
n
t
i
o
n
s
.
T
r
ad
i
t
i
o
n
a
l
l
y
,
t
h
i
s
c
l
as
s
i
f
i
c
a
ti
o
n
p
r
o
c
e
s
s
r
el
i
es
o
n
m
a
n
u
a
l
i
n
t
e
r
p
r
e
t
a
t
i
o
n
o
f
m
e
d
i
ca
l
i
m
a
g
i
n
g
,
w
h
i
c
h
c
a
n
b
e
t
i
m
e
-
c
o
n
s
u
m
i
n
g
a
n
d
s
u
b
je
c
t
i
v
e
.
I
n
a
d
d
i
t
i
o
n
,
e
x
is
ti
n
g
a
u
t
o
m
a
t
e
d
m
e
t
h
o
d
s
o
f
t
e
n
n
e
e
d
h
el
p
t
o
a
c
h
i
e
v
e
d
e
s
i
r
e
d
l
e
v
e
ls
o
f
a
c
c
u
r
a
c
y
,
r
e
q
u
i
r
i
n
g
m
o
r
e
s
o
p
h
i
s
t
i
c
a
t
e
d
a
p
p
r
o
a
c
h
e
s
t
o
i
m
p
r
o
v
e
d
i
a
g
n
o
s
t
i
c
y
i
e
l
d
.
I
n
r
e
s
p
o
n
s
e
t
o
t
h
e
s
e
c
h
a
l
le
n
g
e
s
,
th
e
i
n
t
e
g
r
a
t
i
o
n
o
f
A
I
a
n
d
d
e
e
p
le
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
,
e
s
p
e
ci
a
l
l
y
C
N
Ns
,
h
as
e
m
e
r
g
e
d
a
s
a
p
r
o
m
i
s
i
n
g
s
o
l
u
ti
o
n
t
o
i
m
p
r
o
v
e
t
h
e
a
c
c
u
r
a
c
y
o
f
f
r
a
c
t
u
r
e
c
l
as
s
i
f
i
c
a
ti
o
n
f
r
o
m
m
e
d
i
c
a
l
i
m
a
g
es
[
1
8
]
,
[
1
9
]
.
C
o
n
v
en
tio
n
al
m
eth
o
d
s
f
o
r
f
r
ac
tu
r
e
class
if
icatio
n
h
av
e
lim
itatio
n
s
,
in
clu
d
in
g
r
elian
ce
o
n
h
u
m
a
n
ex
p
er
tis
e
an
d
p
o
te
n
tial
in
ter
o
b
s
er
v
er
v
ar
iab
ilit
y
.
Ma
n
u
al
i
n
ter
p
r
etatio
n
is
lab
o
r
in
ten
s
iv
e
an
d
ca
n
lead
to
er
r
o
r
s
,
im
p
ac
tin
g
p
atien
t
ca
r
e.
Au
to
m
ated
s
y
s
tem
s
,
wh
il
e
o
f
f
er
in
g
a
d
e
g
r
ee
o
f
e
f
f
i
cien
cy
,
o
f
ten
f
ac
e
ch
allen
g
es
in
h
an
d
lin
g
th
e
c
o
m
p
lex
ity
an
d
v
a
r
iatio
n
i
n
h
e
r
en
t
in
m
ed
ical
im
ag
es.
As
a
r
esu
lt,
th
er
e
is
a
n
u
r
g
en
t
n
ee
d
f
o
r
ad
v
an
ce
d
tec
h
n
o
lo
g
ies
ca
p
ab
le
o
f
p
r
o
v
id
in
g
ac
cu
r
ate,
co
n
s
is
ten
t,
an
d
r
a
p
id
class
if
icatio
n
o
f
f
r
ac
tu
r
es,
th
u
s
lay
in
g
th
e
f
o
u
n
d
atio
n
f
o
r
m
o
r
e
ef
f
ec
tiv
e
t
r
ea
tm
en
t stra
teg
ies
[
2
0
]
–
[
2
4
]
.
T
h
e
em
er
g
en
ce
o
f
d
ee
p
lea
r
n
in
g
,
an
d
C
NNs
in
p
a
r
ticu
lar
,
h
as
r
ev
o
lu
tio
n
ized
m
ed
i
ca
l
im
ag
e
an
aly
s
is
.
C
NN
s
ex
ce
l
at
ca
p
tu
r
in
g
h
ier
ar
c
h
ical
f
ea
tu
r
es
f
r
o
m
im
ag
es,
m
ak
in
g
th
e
m
s
u
itab
le
f
o
r
task
s
s
u
ch
as
im
ag
e
class
if
icatio
n
[
2
5
]
,
[
2
6
]
.
I
n
th
e
co
n
tex
t
o
f
f
r
ac
tu
r
e
cla
s
s
if
icatio
n
,
u
s
in
g
C
NN
ar
ch
ite
ctu
r
e
is
p
r
o
m
is
in
g
to
o
v
er
co
m
e
th
e
s
h
o
r
tc
o
m
in
g
s
o
f
tr
ad
itio
n
al
m
eth
o
d
s
.
B
y
au
to
m
atica
lly
lear
n
in
g
co
m
p
lex
p
atter
n
s
an
d
r
ep
r
esen
tatio
n
s
f
r
o
m
m
e
d
ical
im
ag
es,
C
NNs
h
av
e
d
em
o
n
s
tr
ated
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
in
v
ar
io
u
s
m
ed
ical
f
ield
s
,
f
o
s
ter
in
g
o
p
tim
is
m
ab
o
u
t th
eir
p
o
te
n
tial to
im
p
r
o
v
e
f
r
ac
tu
r
e
class
if
icatio
n
ac
cu
r
ac
y
s
ig
n
if
ican
tly
.
Pre
v
io
u
s
r
esear
ch
co
n
d
u
cted
b
y
T
an
zi
et
a
l.
[
1
5
]
with
th
e
titl
e
"X
-
R
ay
b
o
n
e
f
r
ac
tu
r
e
class
if
icatio
n
u
s
in
g
d
ee
p
lear
n
in
g
:
a
b
aselin
e
f
o
r
d
esig
n
i
n
g
a
r
eliab
le
ap
p
r
o
ac
h
"
th
is
r
esear
ch
an
aly
ze
d
v
ar
io
u
s
s
tu
d
ies
th
at
d
is
cu
s
s
ed
b
o
n
e
f
r
ac
tu
r
e
class
if
icatio
n
u
s
in
g
a
d
ee
p
lear
n
in
g
a
p
p
r
o
ac
h
.
I
ex
am
i
n
ed
t
h
e
s
tu
d
i
es
b
ased
o
n
ce
r
tain
cr
iter
ia,
s
u
ch
as
d
ata
q
u
ality
,
im
ag
e
p
r
o
ce
s
s
in
g
m
eth
o
d
s
,
t
y
p
e
o
f
n
e
u
r
al
n
etwo
r
k
,
an
d
t
h
e
f
in
al
r
esu
lts
.
T
h
e
r
esear
ch
r
esu
lts
s
h
o
w
th
at
th
e
m
o
s
t
ef
f
ec
tiv
e
ap
p
r
o
ac
h
is
to
u
s
e
p
r
e
-
tr
ain
i
n
g
with
a
lar
g
er
d
ataset,
p
er
f
o
r
m
p
r
e
-
p
r
o
ce
s
s
in
g
co
r
r
ec
tly
,
u
s
e
d
ata
au
g
m
en
tatio
n
tech
n
iq
u
es
wis
ely
,
an
d
im
p
lem
en
t
class
ac
tiv
atio
n
m
ap
p
in
g
(
C
AM
)
to
u
n
d
er
s
tan
d
th
e
f
o
cu
s
o
f
th
e
n
e
u
r
al
n
etwo
r
k
.
Ho
we
v
er
,
th
is
s
tu
d
y
n
ee
d
s
to
b
e
s
tr
o
n
g
er
i
n
th
at
it
also
d
o
es
n
o
t
f
u
lly
u
tili
ze
v
is
u
aliza
tio
n
tech
n
iq
u
es
s
u
ch
as
Gr
ad
-
C
AM
to
u
n
d
er
s
tan
d
th
e
f
o
cu
s
o
f
n
eu
r
al
n
etwo
r
k
s
,
wh
ich
ca
n
h
elp
v
alid
ate
b
o
n
e
f
r
ac
tu
r
e
class
if
icatio
n
r
esu
lts
.
L
astl
y
,
th
is
ar
ticle
also
n
ee
d
s
to
p
r
o
v
i
d
e
m
o
r
e
in
f
o
r
m
atio
n
ab
o
u
t e
v
alu
atio
n
b
y
s
p
ec
ialis
ts
,
wh
ich
m
ay
af
f
e
ct
th
e
v
alid
ity
o
f
th
e
class
if
icatio
n
r
esu
lts
.
Me
an
wh
ile,
th
e
n
e
x
t
r
esear
ch
co
n
d
u
cted
b
y
Yo
o
n
et
a
l.
[
2
7
]
with
th
e
titl
e
"Au
to
m
atic
m
u
lti
-
class
in
ter
tr
o
ch
an
ter
ic
f
em
u
r
f
r
ac
tu
r
e
d
etec
tio
n
f
r
o
m
C
T
im
a
g
e
s
b
ased
o
n
AO/OT
A
class
if
icatio
n
u
s
in
g
f
aster
R
-
C
NN
-
B
O
m
eth
o
d
"
s
tu
d
y
ai
m
s
to
d
ev
elo
p
a
m
u
lti
-
class
f
em
u
r
f
r
ac
tu
r
e
d
etec
tio
n
m
o
d
e
l.
-
au
to
m
atic
class
u
s
es
d
ee
p
lear
n
in
g
to
class
if
y
f
r
ac
tu
r
e
ty
p
es
an
d
lo
ca
tio
n
s
in
a
s
in
g
le
f
lo
w.
B
ased
o
n
th
e
an
n
o
tated
e
x
p
er
t
d
ata,
th
e
f
em
u
r
f
r
a
ctu
r
e
im
a
g
e
is
tr
an
s
f
o
r
m
e
d
in
t
o
th
r
ee
r
e
g
io
n
s
to
tr
ain
th
e
d
ata
f
o
r
th
e
d
ee
p
,
f
aster
R
-
C
NN
m
o
d
el.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
p
er
f
o
r
m
s
b
etter
th
a
n
p
r
ev
io
u
s
s
tu
d
ies
wh
en
class
if
y
in
g
a
n
d
d
etec
tin
g
r
eg
io
n
o
f
in
ter
est
s
(
R
OI
s
)
.
T
h
is
f
em
u
r
d
etec
tio
n
m
o
d
el
ca
n
r
e
d
u
ce
d
if
f
er
en
ce
s
in
d
iag
n
o
s
is
d
u
e
to
th
e
lev
el
o
f
s
u
r
g
ical
ex
p
er
ien
ce
an
d
p
r
o
v
id
e
d
ee
p
e
r
in
s
ig
h
t
wh
e
n
s
elec
tin
g
o
r
p
la
n
n
in
g
tr
ea
tm
en
t
o
f
f
em
u
r
f
r
ac
t
u
r
es.
Ho
wev
er
,
th
e
ac
cu
r
ac
y
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
n
ee
d
s
to
b
e
im
p
r
o
v
ed
wh
en
class
if
y
in
g
f
em
o
r
al
f
r
ac
tu
r
es
in
to
ten
s
u
b
g
r
o
u
p
s
.
T
h
e
ten
d
en
cy
f
o
r
th
e
n
u
m
b
er
o
f
f
alse
p
o
s
itiv
es
(
FP
s
)
to
b
e
h
ig
h
er
th
an
tr
u
e
p
o
s
itiv
es
(
T
Ps
)
in
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
in
ad
d
itio
n
;
t
h
is
s
tu
d
y
also
s
h
o
wed
th
at
th
e
s
en
s
itiv
ity
an
d
p
o
s
itiv
e
p
r
ed
ictiv
e
v
alu
e
(
PP
V)
f
o
r
s
ev
er
al
class
if
icatio
n
s
o
f
f
e
m
u
r
f
r
ac
tu
r
e
ty
p
es
ten
d
e
d
t
o
d
ec
r
ea
s
e
as
th
e
n
u
m
b
er
o
f
class
if
ied
class
es
in
cr
ea
s
ed
.
B
ased
o
n
p
r
ev
io
u
s
liter
atu
r
e
s
tu
d
ies
an
d
r
elate
d
r
esear
ch
o
b
jects,
th
e
f
o
cu
s
o
f
th
is
r
esear
ch
p
r
o
p
o
s
al
o
n
C
NN
ar
ch
itectu
r
e
is
d
elib
e
r
ate
an
d
s
tr
ateg
ic.
C
NNs
ar
e
d
esig
n
ed
to
m
im
ic
th
e
h
u
m
a
n
v
is
u
al
s
y
s
tem
[
2
8
]
,
[
2
9
]
,
u
tili
zin
g
co
n
v
o
lu
tio
n
al
lay
er
s
to
ex
tr
ac
t r
elev
a
n
t f
ea
tu
r
e
s
f
r
o
m
in
p
u
t im
ag
es.
T
h
is
ar
ch
itectu
r
e
allo
ws th
e
m
o
d
el
to
d
is
tin
g
u
is
h
co
m
p
le
x
p
atter
n
s
,
tex
t
u
r
es,
an
d
s
p
a
tial
r
elatio
n
s
h
ip
s
im
p
o
r
tan
t
f
o
r
ac
cu
r
ate
im
ag
e
class
if
icatio
n
.
T
h
e
in
co
r
p
o
r
ati
o
n
o
f
p
o
o
lin
g
lay
er
s
a
n
d
f
u
lly
co
n
n
ec
ted
lay
er
s
f
u
r
th
er
en
h
an
ce
s
th
is
p
r
o
ce
s
s
,
m
ak
in
g
C
NNs
h
ig
h
ly
ef
f
ec
tiv
e
in
h
an
d
lin
g
m
ed
ical
im
ag
i
n
g
d
ata.
B
y
ad
o
p
tin
g
a
C
NN
ar
ch
itectu
r
e,
th
is
s
tu
d
y
aim
s
to
h
ar
n
ess
th
e
p
o
wer
o
f
d
ee
p
lear
n
in
g
to
ac
h
ie
v
e
h
ig
h
ac
cu
r
ac
y
in
b
o
n
e
f
r
ac
tu
r
e
cla
s
s
if
icatio
n
,
th
er
eb
y
o
v
er
co
m
i
n
g
th
e
s
h
o
r
tco
m
in
g
s
o
f
ex
is
tin
g
m
eth
o
d
s
.
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.
14
,
No
.
6
,
Dec
em
b
e
r
20
24
:
6
4
6
6
-
6
4
7
7
6468
T
h
e
em
p
h
asis
o
n
C
NN
ar
c
h
itectu
r
es
u
n
d
er
s
co
r
es
th
e
c
o
m
m
itm
en
t
t
o
lev
e
r
ag
e
s
tate
-
of
-
th
e
-
ar
t
tech
n
o
lo
g
ies
to
im
p
r
o
v
e
clin
i
ca
l
o
u
tco
m
es
with
n
o
v
el
p
r
o
p
o
s
ed
C
NN
ar
ch
itectu
r
es
[
3
0
]
,
[
3
1
]
.
T
h
e
m
ain
r
ea
s
o
n
b
eh
in
d
th
is
r
esear
ch
is
to
ad
v
an
ce
t
h
e
f
ield
o
f
f
r
ac
tu
r
e
class
if
icatio
n
b
y
in
tr
o
d
u
cin
g
a
s
tate
-
of
-
th
e
-
a
r
t
C
NN
-
b
ased
ap
p
r
o
ac
h
.
T
h
e
u
lt
im
ate
g
o
al
is
to
d
ev
elo
p
a
r
o
b
u
s
t
s
y
s
tem
th
at
ac
cu
r
ately
c
ateg
o
r
izes
f
r
ac
tu
r
es
f
r
o
m
m
e
d
ical
im
ag
es,
u
ltima
t
ely
f
ac
ilit
atin
g
tim
ely
an
d
ap
p
r
o
p
r
iate
clin
ical
d
ec
is
io
n
-
m
a
k
in
g
.
T
h
r
o
u
g
h
t
h
is
r
esear
ch
,
we
aim
to
c
o
n
tr
ib
u
te
to
th
e
ev
o
l
u
tio
n
o
f
m
e
d
ical
i
m
ag
e
an
aly
s
is
,
p
a
v
in
g
th
e
wa
y
f
o
r
m
o
r
e
ef
f
icien
t
an
d
ac
cu
r
ate
d
iag
n
o
s
tic
p
r
o
ce
s
s
es in
o
r
th
o
p
ed
ic
p
r
ac
tice.
T
h
is
r
esear
ch
in
tr
o
d
u
ce
s
a
n
ew
p
ar
ad
ig
m
an
d
ar
ch
itectu
r
e
as
a
n
o
v
elty
in
th
e
f
ield
o
f
f
r
ac
t
u
r
e
class
if
icatio
n
b
y
s
tr
ateg
ically
ex
p
lo
itin
g
th
e
tr
an
s
f
o
r
m
ativ
e
p
o
ten
tial
o
f
C
NN
ar
ch
itectu
r
e
[
3
2
]
–
[
3
4
]
.
I
ts
n
o
v
elty
lies
in
th
e
d
elib
er
a
te
an
d
s
p
ec
ialized
ap
p
licatio
n
o
f
C
NNs
to
s
ig
n
if
ican
tly
im
p
r
o
v
e
im
ag
e
class
if
icatio
n
ac
cu
r
ac
y
,
s
p
ec
if
ically
o
v
er
co
m
in
g
th
e
ch
allen
g
es
in
h
er
en
t
to
tr
ad
itio
n
al
m
e
th
o
d
s
.
B
y
u
tili
zin
g
th
e
h
ier
ar
ch
ical
f
ea
tu
r
e
ex
tr
a
ctio
n
ca
p
ab
ilit
ies
o
f
C
NN,
th
is
r
esear
ch
aim
s
to
o
v
er
co
m
e
th
e
lim
itatio
n
s
o
f
m
an
u
al
in
ter
p
r
etatio
n
an
d
t
h
e
s
h
o
r
tco
m
in
g
s
o
f
ex
is
tin
g
au
t
o
m
atic
s
y
s
tem
s
.
Un
lik
e
co
n
v
e
n
tio
n
al
ap
p
r
o
ac
h
es,
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
g
o
es
b
ey
o
n
d
a
u
to
m
atio
n
,
d
elv
in
g
in
to
d
ee
p
lear
n
in
g
to
e
m
p
o
wer
s
y
s
tem
s
with
th
e
ca
p
ac
ity
to
d
is
tin
g
u
is
h
c
o
m
p
lex
p
atter
n
s
a
n
d
s
u
b
tle
n
u
an
ce
s
in
m
ed
ical
im
ag
es
a
u
to
n
o
m
o
u
s
ly
.
T
h
is
in
n
o
v
atio
n
r
e
p
r
esen
ts
a
p
ar
ad
ig
m
s
h
if
t
i
n
f
r
ac
tu
r
e
class
if
icatio
n
.
I
t
u
n
d
er
s
co
r
es
a
co
m
m
i
tm
en
t
to
ad
v
a
n
cin
g
th
e
f
ield
th
r
o
u
g
h
cu
ttin
g
-
ed
g
e
tech
n
o
l
o
g
y
,
p
av
i
n
g
th
e
way
f
o
r
m
o
r
e
p
r
ec
is
e
d
iag
n
o
s
tic
ca
p
ab
ilit
ies
in
o
r
th
o
p
e
d
ic
p
r
ac
tice.
2.
M
E
T
H
O
D
T
h
is
r
esear
ch
was
ca
r
r
ied
o
u
t
with
th
e
m
ain
aim
o
f
an
aly
z
in
g
an
d
e
v
alu
atin
g
c
o
m
p
a
r
ativ
e
m
o
d
e
l
o
p
tim
izatio
n
tech
n
iq
u
es
o
n
C
NN
in
th
e
p
r
o
ce
s
s
o
f
class
if
y
in
g
b
o
n
e
f
r
ac
tu
r
es
u
s
in
g
X
-
r
ay
C
T
s
ca
n
d
ata.
T
h
is
s
tu
d
y
is
v
er
y
im
p
o
r
tan
t
b
e
ca
u
s
e
th
e
class
if
icat
io
n
r
esu
lts
o
f
th
is
d
ata
s
er
v
e
to
o
b
tain
in
-
d
ep
th
in
f
o
r
m
atio
n
r
eg
ar
d
in
g
th
e
c
o
m
p
ar
ati
v
e
an
a
ly
s
is
o
f
th
e
v
ar
io
u
s
C
NN
m
o
d
el
o
p
tim
izatio
n
tech
n
iq
u
es
ap
p
lied
.
T
h
e
r
esear
ch
p
r
o
ce
s
s
b
eg
an
b
y
co
llectin
g
X
-
r
ay
C
T
s
ca
n
d
ata
f
r
o
m
v
ar
io
u
s
b
o
n
e
f
r
ac
tu
r
e
ca
s
es.
T
h
is
d
ata
is
th
en
p
r
o
ce
s
s
ed
an
d
u
s
ed
as
in
p
u
t
f
o
r
th
e
C
NN
m
o
d
el.
T
h
e
C
NN
m
o
d
el
u
s
ed
h
as
b
ee
n
s
p
ec
if
ically
d
esi
g
n
ed
to
d
etec
t
an
d
class
if
y
ty
p
es
o
f
b
o
n
e
f
r
ac
tu
r
es
b
ased
o
n
p
atter
n
s
an
d
f
ea
t
u
r
es
id
en
tifie
d
in
C
T
s
ca
n
im
ag
es.
T
h
e
r
esu
lts
o
f
th
is
s
tu
d
y
s
h
o
w
th
at
ap
p
ly
in
g
ap
p
r
o
p
r
iate
o
p
tim
izatio
n
tech
n
iq
u
es
ca
n
s
ig
n
if
ican
tly
im
p
r
o
v
e
th
e
p
er
f
o
r
m
an
c
e
o
f
C
NN
m
o
d
els
in
class
if
y
in
g
b
o
n
e
f
r
ac
tu
r
es.
T
h
e
co
m
p
ar
at
iv
e
an
aly
s
is
p
er
f
o
r
m
ed
p
r
o
v
i
d
es
v
alu
ab
le
in
s
ig
h
t
in
to
th
e
a
d
v
an
ta
g
es
an
d
d
is
ad
v
an
tag
es
o
f
ea
c
h
o
p
tim
izatio
n
tech
n
i
q
u
e.
T
h
u
s
,
t
h
is
r
esear
ch
n
o
t
o
n
l
y
co
n
tr
ib
u
tes
to
t
h
e
d
ev
el
o
p
m
en
t
o
f
tech
n
o
lo
g
y
in
th
e
h
e
alth
s
ec
to
r
,
esp
ec
ially
in
im
ag
e
-
b
ased
m
e
d
ical
d
iag
n
o
s
is
,
b
u
t
also
o
f
f
er
s
p
r
a
ctica
l
g
u
id
an
ce
f
o
r
r
esear
ch
e
r
s
an
d
p
r
ac
titi
o
n
er
s
wh
o
wis
h
to
o
p
tim
ize
th
ei
r
C
NN
m
o
d
els.
2
.
1
.
Da
t
a
c
o
llect
io
n m
et
ho
ds
T
h
e
d
ata
c
o
llectio
n
m
eth
o
d
u
s
ed
in
r
esear
ch
to
o
b
tain
d
ata
a
n
d
in
f
o
r
m
atio
n
is
o
p
en
-
s
o
u
r
ce
d
ata.
W
e
ca
n
ex
p
lo
r
e
th
ese
r
eso
u
r
ce
s
to
cr
ea
te
a
r
esear
ch
s
er
ies
with
r
elev
an
t
d
ata.
T
h
e
d
ata
was
r
ec
eiv
ed
f
r
o
m
th
e
Kag
g
le.
co
m
s
ite.
2
.
2
.
Arc
hite
ct
ure
CNN
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
o
r
C
NNs)
ar
e
a
s
p
ec
ial
ty
p
e
o
f
m
u
lti
-
lay
e
r
n
eu
r
al
n
etwo
r
k
o
r
d
ee
p
lear
n
in
g
a
r
ch
itectu
r
e
in
s
p
ir
ed
b
y
th
e
v
is
u
al
s
y
s
tem
s
o
f
li
v
in
g
th
i
n
g
s
.
C
NN
ar
e
d
ee
p
l
ea
r
n
in
g
alg
o
r
ith
m
s
d
esig
n
ed
to
p
r
o
ce
s
s
two
-
d
im
e
n
s
io
n
al
d
ata
[
3
5
]
–
[
3
7
]
.
C
NNs
ar
e
u
s
u
ally
u
s
ed
to
lear
n
an
d
d
etec
t
f
ea
tu
r
es
in
an
im
ag
e.
Fig
u
r
e
1
is
a
s
am
p
le
o
f
th
e
C
NN
ar
ch
itectu
r
e
with
a
d
ee
p
lear
n
i
n
g
ap
p
r
o
ac
h
.
I
n
Fig
u
r
e
1
m
o
d
el
o
p
tim
izatio
n
in
C
NN
r
ef
er
s
to
r
ef
in
in
g
an
d
im
p
r
o
v
in
g
a
C
NN
m
o
d
el'
s
p
er
f
o
r
m
an
ce
.
Var
io
u
s
p
ar
am
eter
s
an
d
co
n
f
ig
u
r
atio
n
s
ar
e
ad
ju
s
ted
to
im
p
r
o
v
e
th
e
n
etwo
r
k
'
s
ab
ilit
y
to
ac
cu
r
ately
u
n
d
er
s
tan
d
a
n
d
r
ep
r
esen
t
co
m
p
le
x
p
atter
n
s
in
im
ag
e
d
ata
[
3
8
]
.
R
e
f
i
n
e
m
e
n
ts
a
r
e
i
m
p
o
r
t
a
n
t
f
o
r
a
d
a
p
t
i
n
g
e
x
i
s
ti
n
g
C
N
N
m
o
d
e
ls
t
o
w
o
r
k
w
e
l
l
o
n
a
p
a
r
t
i
c
u
l
a
r
t
a
s
k
o
r
d
a
t
a
s
e
t
.
P
r
e
t
r
a
i
n
e
d
m
o
d
e
ls
p
r
o
v
i
d
e
a
f
o
u
n
d
a
t
i
o
n
f
o
r
r
ec
o
g
n
i
z
i
n
g
c
o
m
m
o
n
f
e
a
t
u
r
e
s
,
w
h
i
le
c
u
s
t
o
m
i
z
a
t
i
o
n
a
d
a
p
t
s
t
h
e
m
o
d
e
l
t
o
s
p
e
c
i
f
i
c
p
a
tt
e
r
n
s
a
n
d
i
n
f
o
r
m
a
t
i
o
n
r
e
l
e
v
a
n
t
t
o
a
p
a
r
t
i
c
u
l
a
r
a
p
p
l
i
ca
t
i
o
n
.
T
h
i
s
p
r
o
c
es
s
is
i
m
p
o
r
t
a
n
t
f
o
r
a
c
h
i
e
v
i
n
g
o
p
t
i
m
a
l
p
e
r
f
o
r
m
a
n
c
e
i
n
i
m
a
g
e
c
la
s
s
i
f
ic
a
t
i
o
n
,
o
b
j
e
c
t
d
e
t
e
ct
i
o
n
,
a
n
d
s
e
g
m
e
n
ta
t
i
o
n
t
as
k
s
.
T
h
es
e
o
p
t
i
m
i
z
a
ti
o
n
t
e
c
h
n
i
q
u
e
s
c
o
l
l
ec
t
i
v
e
l
y
c
o
n
t
r
i
b
u
t
e
t
o
C
NN
s
'
e
f
f
e
c
t
i
v
e
n
es
s
a
n
d
e
f
f
i
c
i
e
n
c
y
in
v
a
r
i
o
u
s
c
o
m
p
u
t
e
r
v
i
s
i
o
n
t
a
s
k
s
.
R
es
e
a
r
c
h
e
r
s
o
f
t
en
c
o
n
s
i
d
e
r
c
o
m
b
i
n
a
t
i
o
n
s
a
n
d
v
a
r
i
a
t
i
o
n
s
o
f
t
h
e
s
e
t
ec
h
n
i
q
u
e
s
t
o
f
i
n
d
t
h
e
o
p
t
i
m
a
l
c
o
n
f
i
g
u
r
a
t
i
o
n
f
o
r
a
p
a
r
t
i
c
u
l
a
r
a
p
p
l
i
c
a
t
i
o
n
.
T
r
a
i
n
a
C
N
N
u
s
i
n
g
d
i
f
f
e
r
e
n
t
o
p
t
i
m
i
z
a
t
i
o
n
o
p
t
i
m
i
ze
r
s
t
o
c
o
m
p
a
r
e
w
i
t
h
d
i
f
f
e
r
e
n
t
b
a
t
c
h
s
i
z
e
s
.
T
h
e
a
u
th
o
r
s
u
s
e
p
o
p
u
l
a
r
d
e
e
p
l
e
a
r
n
i
n
g
f
r
a
m
e
w
o
r
k
s
l
i
k
e
T
e
n
s
o
r
Fl
o
w
f
o
r
t
h
is
t
as
k
.
W
e
p
r
e
s
e
n
t
t
h
e
C
NN
m
o
d
e
l
o
p
t
i
m
iz
a
t
i
o
n
f
l
o
w
c
h
a
r
t
i
n
F
i
g
u
r
e
2
.
T
h
e
e
x
p
l
a
n
a
t
i
o
n
o
f
F
i
g
u
r
e
2
is
a
s
f
o
l
l
o
w
s
:
−
Star
t: Flo
wch
ar
ts
u
s
u
ally
s
tar
t w
ith
th
e
"Star
t"
s
y
m
b
o
l.
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
fr
a
ctu
r
e
cla
s
s
ifica
tio
n
u
s
in
g
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
et
w
o
r
k
a
r
ch
itectu
r
e
fo
r
…
(
S
o
likh
u
n
)
6469
−
Data
co
llectio
n
: U
s
e
r
ec
tan
g
u
l
ar
b
o
x
es to
r
ep
r
esen
t th
e
s
tep
s
wh
en
y
o
u
lo
ad
y
o
u
r
d
ata
s
et.
L
ab
el
it "L
o
ad
Data
.
"
Fig
u
r
e
s
3
(
a)
an
d
3
(
b
)
a
r
e
s
am
p
le
s
o
f
th
e
d
ata
u
s
ed
in
th
is
r
esear
ch
.
X
1
X
2
X
3
.
.
.
.
.
.
X
90
I
n
p
u
t
L
a
y
e
r
L
1
H
i
d
d
e
n
L
a
y
e
r
L
2
H
i
d
d
e
n
L
a
y
e
r
L
3
O
u
t
p
u
t
L
a
y
e
r
L
5
W
(
1
)
W
(
2
)
W
(
3
)
a
(
2
)
a
(
3
)
a(
4
)
y
0
y
1
y
2
y
3
y
4
y
5
y
6
y
7
y
8
y
9
Fig
u
r
e
1
.
C
NN
alg
o
r
ith
m
ar
ch
itectu
r
e
Fig
u
r
e
2
.
Mo
d
el
o
p
tim
izatio
n
f
lo
wch
ar
t
(
a)
(
b
)
Fig
u
r
e
3
.
Data
s
am
p
le
(
a
)
f
r
ac
t
u
r
e
an
d
(
b
)
n
o
t f
r
ac
tu
r
e
Fig
u
r
e
3
(
a)
Fra
ctu
r
e:
T
h
e
im
ag
e
s
h
o
ws
a
wr
is
t
b
o
n
e
f
r
ac
t
u
r
e
m
ar
k
e
d
b
y
ir
r
eg
u
lar
lin
e
s
o
r
clea
r
b
r
ea
k
s
,
in
d
icatin
g
d
a
m
ag
e
f
r
o
m
tr
au
m
a
o
r
p
r
ess
u
r
e.
Fig
u
r
e
3
(
b
)
No
f
r
ac
tu
r
e
:
T
h
e
im
a
g
e
d
is
p
lay
s
an
i
n
tact
wr
is
t
b
o
n
e
with
s
m
o
o
th
lin
es
an
d
n
o
s
ig
n
s
o
f
cr
ac
k
s
o
r
b
r
ea
k
s
,
in
d
icatin
g
a
h
ea
lth
y
c
o
n
d
itio
n
.
T
h
e
m
ain
d
if
f
er
en
ce
is
th
at
Fig
u
r
e
3
(
a
)
s
h
o
ws a
f
r
ac
tu
r
e,
wh
ile
Fig
u
r
e
3
(
b
)
d
ep
icts
an
u
n
d
am
ag
e
d
b
o
n
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.
14
,
No
.
6
,
Dec
em
b
e
r
20
24
:
6
4
6
6
-
6
4
7
7
6470
−
Data
p
r
ep
r
o
ce
s
s
in
g
:
ad
d
o
th
er
r
ec
tan
g
u
lar
b
o
x
es
f
o
r
d
ata
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
,
s
u
ch
as
d
at
a
n
o
r
m
aliza
tio
n
an
d
au
g
m
en
tatio
n
,
if
ap
p
lica
b
le.
L
ab
el
it "
d
ata
p
r
ep
r
o
c
ess
in
g
.
"
−
Data
a
u
g
m
en
tatio
n
:
d
ata
au
g
m
en
tatio
n
is
v
er
y
im
p
o
r
tan
t
t
o
p
r
ev
en
t
o
v
er
f
itti
n
g
an
d
im
p
r
o
v
e
th
e
m
o
d
el'
s
g
en
er
aliza
tio
n
ab
ilit
y
.
C
o
m
m
o
n
a
u
g
m
en
tatio
n
tech
n
iq
u
es
i
n
clu
d
e
r
o
tatio
n
,
f
lip
p
i
n
g
,
s
ca
lin
g
,
a
n
d
s
h
if
tin
g
.
Au
g
m
en
tatio
n
im
p
lem
en
tatio
n
s
ar
e
o
f
ten
ap
p
lied
to
tr
ai
n
in
g
d
atasets
wh
en
u
s
in
g
li
b
r
ar
ies
s
u
ch
as
T
en
s
o
r
Flo
w
o
r
Py
T
o
r
ch
.
−
Hy
p
er
p
ar
a
m
eter
t
u
n
in
g
:
lear
n
i
n
g
r
ate
(
η
)
co
n
tr
o
ls
th
e
s
tep
s
ize
d
u
r
in
g
o
p
tim
izatio
n
.
Nu
m
b
er
o
f
lay
e
r
s
an
d
co
n
v
o
l
u
tio
n
al
f
ilter
s
ar
ch
itectu
r
e
-
r
elate
d
h
y
p
er
p
ar
am
eter
s
a
f
f
ec
t
th
e
n
etwo
r
k
'
s
d
ep
th
an
d
wid
th
.
d
r
o
p
o
u
t
r
ate
is
a
r
eg
u
lar
izatio
n
tech
n
iq
u
e
to
p
r
ev
en
t
o
v
er
f
itti
n
g
.
Ker
n
el
s
ize
an
d
s
tr
id
e
:
Par
am
eter
s
th
at
d
eter
m
in
e
th
e
s
ize
o
f
th
e
co
n
v
o
l
u
tio
n
al
k
er
n
el
an
d
its
s
tr
id
e.
B
atch
n
o
r
m
aliza
tio
n
ca
n
d
eter
m
in
e
if
a
n
d
wh
er
e
a
b
atch
n
o
r
m
aliza
tio
n
lay
er
is
a
d
d
ed
.
Activ
atio
n
f
u
n
ctio
n
th
er
e
ar
e
m
an
y
o
p
tio
n
s
,
s
u
ch
as
R
eL
U,
Sig
m
o
id
,
o
r
T
an
h
,
f
o
r
th
e
ac
tiv
atio
n
f
u
n
ct
io
n
.
Nu
m
b
er
o
f
h
id
d
e
n
u
n
its
in
d
en
s
e
lay
er
s
ar
ch
itectu
r
al
o
p
tio
n
s
f
o
r
f
u
lly
co
n
n
ec
ted
la
y
er
s
.
−
Op
tim
izer
s
etu
p
:
co
m
m
o
n
o
p
ti
m
izer
s
in
clu
d
e
Ad
am
:
ad
ap
tiv
e
m
o
m
en
t e
s
tim
atio
n
,
p
o
p
u
la
r
f
o
r
its
ad
ap
tiv
e
lear
n
in
g
r
ate
an
d
m
o
m
en
tu
m
.
R
MSPr
o
p
:
r
o
o
t
m
ea
n
s
q
u
ar
e
p
r
o
p
ag
atio
n
,
s
im
ilar
to
Ad
am
b
u
t
with
d
if
f
er
en
t
u
p
d
ate
r
u
les.
L
ea
r
n
in
g
r
ate
s
ch
ed
u
le
:
Ad
ju
s
tin
g
t
h
e
lear
n
in
g
r
ate
d
u
r
in
g
tr
ain
i
n
g
(
f
o
r
ex
am
p
le,
r
ed
u
cin
g
it
o
v
er
tim
e)
ca
n
im
p
r
o
v
e
c
o
n
v
e
r
g
en
ce
.
W
eig
h
t
d
ec
ay
:
L
2
r
e
g
u
lar
izatio
n
is
ap
p
lied
to
weig
h
ts
d
u
r
in
g
o
p
tim
izatio
n
.
−
B
atch
s
ize:
im
p
ac
t
on
t
r
ain
in
g
:
lar
g
er
b
atch
s
izes
p
r
o
v
id
e
co
m
p
u
tatio
n
al
ef
f
icien
cy
b
u
t
m
a
y
r
esu
lt
in
lo
s
s
o
f
g
e
n
er
aliza
tio
n
.
Sm
aller
b
at
ch
s
izes
in
tr
o
d
u
ce
m
o
r
e
n
o
is
e
b
u
t
ca
n
im
p
r
o
v
e
g
en
er
aliza
tio
n
.
Selectio
n
:
o
f
ten
d
e
p
en
d
s
o
n
av
ailab
le
c
o
m
p
u
tin
g
r
eso
u
r
ce
s
an
d
s
p
ec
if
i
c
d
ata
s
et
ch
ar
ac
ter
is
tics
.
−
Per
f
o
r
m
an
ce
a
n
aly
s
is
:
tr
ain
i
n
g
an
d
v
alid
atio
n
l
o
s
s
cu
r
v
es
:
p
lo
ttin
g
l
o
s
s
o
v
er
tim
e
h
elp
s
ass
ess
co
n
v
er
g
en
ce
a
n
d
id
e
n
tify
o
v
e
r
f
itti
n
g
.
Acc
u
r
ac
y
m
etr
ics:
ch
ec
k
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
al
l,
an
d
F1
-
s
co
r
e
o
n
tr
ain
i
n
g
a
n
d
v
alid
atio
n
s
ets.
C
o
n
f
u
s
io
n
m
atr
ix
:
p
r
o
v
id
es
i
n
s
ig
h
t
in
to
m
o
d
el
p
er
f
o
r
m
a
n
c
e
ac
r
o
s
s
class
es
.
Vis
u
al
i
n
s
p
ec
tio
n
:
an
aly
ze
m
is
class
if
ied
s
am
p
les an
d
u
n
d
er
s
t
an
d
m
o
d
el
is
s
u
es.
−
R
ef
in
em
en
t:
I
ter
ativ
e
p
r
o
ce
s
s
:
h
y
p
e
r
p
ar
a
m
eter
r
ef
in
em
en
ts
,
ar
ch
itectu
r
al
ad
j
u
s
tm
en
ts
,
a
n
d
o
p
tim
izatio
n
ch
o
ices
m
ay
n
ee
d
to
b
e
it
er
ativ
ely
r
ef
i
n
ed
.
R
eg
u
lar
iza
tio
n
t
ec
h
n
iq
u
es:
Ap
p
ly
d
r
o
p
o
u
t
o
r
b
atch
n
o
r
m
aliza
tio
n
tech
n
iq
u
es
to
im
p
r
o
v
e
g
en
er
aliza
tio
n
.
E
a
r
ly
s
to
p
:
tr
ai
n
in
g
s
to
p
s
wh
en
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
o
n
th
e
v
alid
atio
n
s
et
n
o
lo
n
g
e
r
im
p
r
o
v
es.
−
Def
in
e
C
NN
a
r
ch
itectu
r
e:
cr
ea
te
a
r
ec
tan
g
u
lar
b
o
x
to
d
ef
in
e
th
e
C
NN
ar
ch
itectu
r
e,
in
clu
d
in
g
th
e
n
u
m
b
er
o
f
lay
er
s
,
f
ilter
s
,
an
d
ac
tiv
atio
n
f
u
n
ctio
n
s
.
L
ab
el
it
"De
f
in
e
C
NN
ar
ch
itectu
r
e
".
Dete
r
m
in
e
b
atch
s
izes
:
u
s
e
a
d
iam
o
n
d
s
h
ap
e
(
d
ec
is
io
n
s
y
m
b
o
l)
t
o
in
d
icate
th
at
y
o
u
will
m
ak
e
d
ec
is
io
n
s
b
ased
o
n
d
if
f
er
e
n
t
b
atch
s
izes.
L
ab
el
it
"Selec
t
b
atch
s
ize
".
B
atch
s
ize
=
1
2
8
:
Ad
d
a
n
ar
r
o
w
p
o
in
tin
g
to
th
e
r
ec
tan
g
u
lar
b
o
x
with
"Batch
s
ize
=
1
2
8
.
"
I
n
th
is
b
o
x
,
y
o
u
ca
n
in
d
icate
th
at
m
o
d
e
l
tr
ain
in
g
was
d
o
n
e
with
a
b
atch
s
ize
o
f
1
2
8
.
B
atch
s
ize
=
2
5
6
:
Ad
d
an
ar
r
o
w
p
o
in
tin
g
to
an
o
th
e
r
r
ec
ta
n
g
u
lar
b
o
x
with
"Batch
s
ize
=
2
5
6
.
"
I
n
t
h
is
b
o
x
,
y
o
u
ca
n
in
d
icate
t
h
at
m
o
d
el
tr
ain
in
g
h
as
b
ee
n
d
o
n
e
wit
h
a
b
atch
s
ize
2
5
6
.
I
n
Fig
u
r
es
4
an
d
5
,
th
e
f
o
llo
win
g
is
th
e
s
tan
d
ar
d
R
es
N
et5
0
m
o
d
el
with
t
h
e
p
r
o
p
o
s
e
d
m
o
d
el
.
Z
e
ro p
a
ddi
ng
Conv
(
64
,
(
3
,
3
))
256
Ba
t
c
h S
i
z
e
,
64
Re
l
u
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
2
,
2
))
125
Co
nv
Bl
oc
k
127
Id B
l
oc
k x
3
Con
v Bl
oc
k
254
Id Bl
oc
k x
2
A
ve
ra
ge
P
ool
F
l
a
t
t
e
n
62
D
e
ns
e
,
32
F
ul
l
y C
on
ne
c
t
e
d
I
n
p
u
t
I
m
a
g
e
b
o
n
e
5
1
2
x
5
1
2
2
v
a
r
i
a
b
l
e
s
O
u
t
p
u
t
C
l
a
s
s
s
i
f
i
c
a
t
i
o
n
Z
e
ro
pa
ddi
ng
Conv
(
64
,
(
3
,
3
))
Ba
t
c
h S
i
z
e
Re
l
u
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
2
,
2
))
Co
nv
Bl
oc
k
Id B
l
oc
k
A
ve
ra
ge
P
ool
F
l
a
t
t
e
n
D
e
ns
e
,
32
F
ul
l
y
Con
ne
c
t
e
d
Z
e
r
o
P
a
d
d
i
n
g
C
o
n
v
o
l
u
t
i
o
n
a
l
La
y
e
r
B
a
t
c
h
N
o
r
m
a
l
i
z
a
t
i
o
n
A
c
t
i
v
a
t
i
o
n
M
a
x
P
o
o
l
i
n
g
C
o
n
v
o
l
u
t
i
o
n
a
l
B
l
o
c
k
I
d
e
n
t
i
t
y
B
l
o
c
k
A
v
e
r
a
g
e
P
o
o
l
F
l
a
t
t
e
n
D
e
n
s
e
F
u
l
l
y
C
o
n
n
e
c
t
e
d
F
e
a
t
ur
e
s
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
3
,
3
))
62
RM
S
P
rop
RM
S
P
rop
O
p
t
i
m
a
z
a
t
i
o
n
Fig
u
r
e
4
.
Stan
d
a
r
d
R
es
N
et5
0
m
o
d
el
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
fr
a
ctu
r
e
cla
s
s
ifica
tio
n
u
s
in
g
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
et
w
o
r
k
a
r
ch
itectu
r
e
fo
r
…
(
S
o
likh
u
n
)
6471
Z
e
ro p
a
ddi
ng
Conv
(
64
,
(
3
,
3
))
256
Ba
t
c
h
S
i
z
e
,
128
Re
l
u
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
2
,
2
))
125
Conv B
l
oc
k
128
Id B
l
oc
k x
3
Conv
Bl
oc
k
256
Id Bl
oc
k x
2
A
ve
ra
ge
P
ool
F
l
a
t
t
e
n
62
D
e
ns
e
,
32
F
ul
l
y C
on
ne
c
t
e
d
I
n
p
u
t
I
m
a
g
e
B
o
n
e
5
1
2
x
5
1
2
3
v
a
r
i
a
b
l
e
s
O
u
t
p
u
t
C
l
a
s
s
s
i
f
i
c
a
t
i
o
n
Z
e
ro p
a
ddi
ng
Conv
(
64
,
(
3
,
3
))
Ba
t
c
h S
i
z
e
Re
l
u
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
2
,
2
))
Co
nv
Bl
oc
k
Id B
l
oc
k
A
ve
ra
ge
P
ool
F
l
a
t
t
e
n
D
e
ns
e
,
32
F
ul
l
y Con
ne
c
t
e
d
Ze
r
o
P
a
d
d
i
n
g
C
o
n
v
o
l
u
t
i
o
n
a
l
La
y
e
r
B
a
t
c
h
N
o
r
m
a
l
i
z
a
t
i
o
n
A
c
t
i
v
a
t
i
o
n
M
a
x
P
o
o
l
i
n
g
C
o
n
v
o
l
u
t
i
o
n
a
l
B
l
o
c
k
I
d
e
n
t
i
t
y
B
l
o
c
k
A
v
e
r
a
g
e
P
o
o
l
F
l
a
t
t
e
n
D
e
n
s
e
F
u
l
l
y
C
o
n
n
e
c
t
e
d
P
r
opo
s
e
d M
e
t
hod
M
a
x
P
ool
(
pool
_
s
i
z
e
=(
3
,
3
))
62
S
of
t
m
a
x
S
oft
m
a
x
S
o
f
t
m
a
x
F
e
a
t
ur
e
s
O
pt
i
m
i
z
a
t
i
on
A
D
A
M
O
pt
i
m
i
z
a
t
i
on
A
D
A
M
O
p
t
i
m
a
z
a
t
i
o
n
Fig
u
r
e
5
.
Pro
p
o
s
ed
R
es
N
et
m
o
d
el
T
r
ain
th
e
m
o
d
el:
i)
C
r
ea
te
a
r
ec
tan
g
u
lar
b
o
x
to
r
ep
r
esen
t
th
e
m
o
d
el
tr
ain
in
g
s
tep
s
;
ii)
L
ab
el
it
"T
r
ain
m
o
d
el
"
;
iii)
T
h
is
b
o
x
s
h
o
ws
th
e
tr
ain
in
g
p
r
o
ce
s
s
with
ep
o
ch
s
,
lo
s
s
,
an
d
ac
cu
r
ac
y
;
a
n
d
iv
)
Use
ar
r
o
ws
to
co
n
n
ec
t th
ese
b
o
x
es to
th
e
t
h
r
ee
b
atch
s
ize
o
p
tio
n
s
,
i
n
d
icatin
g
th
at
th
e
m
o
d
el
was tr
ain
ed
with
d
if
f
er
en
t sizes.
−
Mo
d
el
e
v
alu
atio
n
: A
d
d
a
r
ec
ta
n
g
u
lar
b
o
x
lab
eled
"
m
o
d
el
ev
a
lu
atio
n
"
to
s
h
o
w
h
o
w
th
e
m
o
d
el
p
er
f
o
r
m
a
n
ce
was
ev
alu
ated
u
s
in
g
v
alid
atio
n
o
r
test
d
ata.
I
n
th
e
ev
alu
atio
n
s
tag
e,
th
e
p
er
f
o
r
m
an
ce
o
f
al
l
2
class
if
ier
s
in
class
if
y
in
g
th
e
b
o
n
e
d
ata
s
et
i
n
to
"b
o
n
e
f
r
ac
tu
r
e"
a
n
d
"n
o
n
-
f
r
ac
tu
r
e"
was
ass
ess
ed
.
T
h
is
p
r
o
ce
s
s
in
v
o
lv
es
th
e
u
tili
za
tio
n
o
f
co
n
f
u
s
io
n
m
atr
ices,
R
O
C
cu
r
v
es,
an
d
E
q
.
T
h
e
R
OC
cu
r
v
e
an
d
co
n
f
u
s
io
n
m
atr
ix
ar
e
b
u
ilt
u
s
in
g
th
e
f
alse
p
o
s
itiv
e
r
ate
a
n
d
tr
u
e
p
o
s
itiv
e
r
ate
o
b
tain
ed
f
r
o
m
ca
lcu
latio
n
s
with
th
e
c
o
n
f
u
s
io
n
m
atr
ix
.
Per
f
o
r
m
an
ce
is
d
eter
m
in
ed
b
y
th
e
AUC o
f
th
e
R
OC
cu
r
v
e,
as sh
o
wn
in
Alg
o
r
ith
m
s
1
an
d
2
.
T
h
e
lar
g
er
th
e
ar
ea
u
n
d
e
r
th
e
c
u
r
v
e
in
d
icate
s
b
etter
p
er
f
o
r
m
an
ce
.
Acc
u
r
ac
y
=
+
+
+
+
(
1
)
Pre
cisi
o
n
=
+
(
2
)
R
ec
all=
+
(
3
)
−
Per
f
o
r
m
an
ce
a
n
aly
s
is
:
C
r
ea
te
a
r
ec
tan
g
u
lar
b
o
x
lab
eled
"
p
er
f
o
r
m
an
ce
a
n
aly
s
is
.
"
I
n
th
i
s
b
o
x
,
y
o
u
ca
n
in
d
icate
th
at
y
o
u
an
aly
ze
d
a
n
d
co
m
p
ar
ed
p
e
r
f
o
r
m
an
ce
m
etr
i
cs,
s
u
ch
as
ac
cu
r
ac
y
,
f
o
r
d
if
f
er
en
t
b
atch
s
izes.
Acc
u
r
ac
y
is
a
m
etr
ic
th
at
m
e
asu
r
es
h
o
w
well
a
m
o
d
el
p
r
e
d
icts
o
v
er
all,
e
x
p
r
ess
ed
as
th
e
p
er
ce
n
ta
g
e
o
f
co
r
r
ec
t
p
r
e
d
ictio
n
s
(
tr
u
e
p
o
s
itiv
es
an
d
tr
u
e
n
e
g
ativ
es
)
co
m
p
ar
ed
to
th
e
t
o
tal
n
u
m
b
er
o
f
f
o
r
ec
asts
(
tr
u
e
p
o
s
itiv
es,
f
alse
p
o
s
itiv
es,
f
alse
n
eg
ativ
es,
an
d
tr
u
e
n
e
g
a
tiv
es).
I
n
co
n
tr
ast,
p
r
ec
is
io
n
m
ea
s
u
r
es
th
e
p
r
o
p
o
r
tio
n
o
f
tr
u
e
p
o
s
itiv
e
p
r
ed
ictio
n
s
o
u
t
o
f
all
p
o
s
itiv
e
p
r
ed
ictio
n
s
m
ad
e
b
y
th
e
m
o
d
e
l,
in
d
icatin
g
its
"p
r
ec
is
io
n
"
o
r
ac
c
u
r
ac
y
in
cl
ass
if
y
in
g
p
o
s
itiv
e
la
b
els.
R
ec
all,
also
k
n
o
wn
as
s
en
s
itiv
ity
,
m
ea
s
u
r
es
th
e
p
r
o
p
o
r
tio
n
o
f
all
p
o
s
itiv
e
ca
s
es
a
m
o
d
el
s
u
cc
ess
f
u
lly
d
etec
ts
,
r
ef
lectin
g
its
ab
ilit
y
to
"r
e
ca
ll"
o
r
"d
etec
t"
p
o
s
itiv
e
ca
s
es.
Dec
is
io
n
:
a.
Use th
e
d
ec
is
io
n
s
y
m
b
o
l to
d
e
ter
m
in
e
wh
eth
er
m
o
r
e
b
atch
s
izes m
u
s
t b
e
ev
alu
ated
.
b.
I
f
y
es,
lo
o
p
b
ac
k
to
th
e
"Selec
t
B
atch
Size"
d
ec
is
io
n
s
y
m
b
o
l.
c.
I
f
n
o
t,
co
n
tin
u
e
to
th
e
n
ex
t step
.
E
n
d
: A
d
d
a
n
"E
n
d
"
s
y
m
b
o
l to
in
d
icate
th
e
en
d
o
f
th
e
f
lo
wch
ar
t.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
B
ased
o
n
th
e
r
esu
lts
o
f
tr
ain
i
n
g
,
v
alid
atio
n
,
an
d
test
in
g
,
v
a
r
io
u
s
h
y
p
er
p
a
r
am
eter
s
an
d
ar
ch
itectu
r
es
ar
e
u
s
ed
to
s
ee
th
e
r
esu
lts
o
f
b
o
n
e
im
ag
e
class
if
icatio
n
.
T
h
e
tr
ain
in
g
p
r
o
ce
s
s
in
v
o
lv
es
th
e
u
s
e
o
f
a
b
o
n
e
im
ag
e
d
ataset
th
at
is
d
iv
id
ed
in
to
th
r
ee
s
u
b
s
ets:
tr
ain
in
g
d
ata,
v
alid
atio
n
d
ata,
an
d
test
in
g
d
ata.
E
ac
h
s
u
b
s
et
s
er
v
es
a
d
if
f
er
en
t
p
u
r
p
o
s
e,
n
am
ely
t
r
ain
in
g
th
e
m
o
d
el,
tu
n
in
g
h
y
p
e
r
p
ar
am
eter
s
,
an
d
ev
alu
atin
g
t
h
e
f
in
al
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
el.
Du
r
in
g
tr
ain
in
g
,
th
e
C
NN
m
o
d
el
is
test
ed
wit
h
v
ar
io
u
s
co
m
b
in
atio
n
s
o
f
h
y
p
er
p
ar
am
eter
s
s
u
ch
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.
14
,
No
.
6
,
Dec
em
b
e
r
20
24
:
6
4
6
6
-
6
4
7
7
6472
as
b
atch
s
ize,
n
u
m
b
er
o
f
ep
o
ch
s
,
lear
n
i
n
g
r
ate,
a
n
d
n
et
wo
r
k
lay
e
r
s
tr
u
ctu
r
e.
T
h
e
u
s
e
o
f
v
ar
io
u
s
C
NN
ar
ch
itectu
r
es,
r
an
g
in
g
f
r
o
m
s
im
p
le
to
co
m
p
lex
,
was
also
ex
p
lo
r
ed
to
d
eter
m
in
e
th
eir
im
p
ac
t
o
n
b
o
n
e
im
ag
e
class
if
icatio
n
ac
cu
r
ac
y
.
3
.
1
.
Resul
t
T
h
e
m
o
d
el
tr
ai
n
in
g
r
esu
lts
p
r
e
s
en
ted
in
th
is
ar
ticle
in
clu
d
e
s
o
m
e
im
p
o
r
tan
t
in
f
o
r
m
atio
n
th
at
ex
p
lain
s
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
o
p
tim
ized
C
NN
m
o
d
el
f
o
r
f
r
ac
tu
r
e
class
if
icatio
n
.
I
t
us
es
th
e
C
NN
a
r
ch
itectu
r
e
f
o
r
Hig
h
ac
cu
r
ac
y
im
ag
e
class
if
icatio
n
with
R
esNet5
0
Plu
s
m
o
d
el
as
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
an
d
R
esNet5
0
Stan
d
ar
d
,
as sh
o
wn
i
n
Fig
u
r
e
5
.
T
h
e
f
o
llo
win
g
is
an
e
x
p
lan
a
tio
n
o
f
th
e
r
esu
lts
f
o
r
ea
c
h
ep
o
ch
:
T
h
e
R
esNet5
0
p
lu
s
m
o
d
el
is
o
p
tim
ized
to
im
p
r
o
v
e
th
e
ac
c
u
r
ac
y
o
f
b
o
n
e
im
ag
e
class
if
icati
o
n
,
ta
k
in
g
in
to
ac
co
u
n
t
f
ac
to
r
s
s
u
c
h
as
n
etwo
r
k
d
e
p
th
,
n
u
m
b
e
r
o
f
f
ilter
s
,
an
d
r
e
g
u
latio
n
tec
h
n
iq
u
es
u
s
ed
d
u
r
in
g
tr
ain
in
g
.
I
n
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
th
e
m
o
d
el
is
test
ed
o
v
e
r
m
u
ltip
le
ep
o
ch
s
,
w
h
er
e
ea
c
h
ep
o
ch
in
v
o
lv
es
c
o
m
p
lete
iter
atio
n
th
r
o
u
g
h
th
e
en
tire
tr
ain
in
g
d
ataset.
Fig
u
r
e
5
d
is
p
l
ay
s
a
g
r
ap
h
o
f
th
e
m
o
d
el'
s
p
er
f
o
r
m
a
n
ce
d
u
r
i
n
g
tr
ain
in
g
,
s
h
o
win
g
im
p
o
r
tan
t
m
etr
ics
s
u
ch
as
ac
cu
r
ac
y
an
d
lo
s
s
o
n
tr
ain
in
g
an
d
v
alid
atio
n
d
ata.
At
ea
ch
ep
o
ch
,
th
e
m
o
d
el'
s
ac
cu
r
ac
y
is
m
ea
s
u
r
ed
to
ass
ess
h
o
w
well
th
e
m
o
d
el
ca
n
r
ec
o
g
n
ize
p
atter
n
s
in
b
o
n
e
im
a
g
es.
L
o
s
s
,
wh
ich
in
d
icate
s
h
o
w
well
th
e
m
o
d
el
f
its
th
e
tr
ain
in
g
d
at
a,
is
also
m
o
n
ito
r
ed
to
en
s
u
r
e
th
e
m
o
d
el
is
n
o
t
o
v
er
f
itti
n
g
o
r
u
n
d
er
f
itti
n
g
.
T
h
e
tr
ain
in
g
p
r
o
ce
s
s
ab
o
v
e,
Fig
u
r
e
6
s
h
o
ws
th
at
af
ter
th
e
3
0
th
e
p
o
ch
,
th
e
tr
ain
in
g
p
r
o
ce
s
s
was
s
to
p
p
ed
b
ec
au
s
e
th
e
lo
s
s
v
alu
e
d
id
n
o
t
d
ec
r
ea
s
e
f
o
r
5
co
n
s
e
cu
tiv
e
ep
o
ch
s
,
wh
ich
in
d
icate
s
th
at
th
e
m
o
d
el
h
a
s
r
ea
ch
ed
co
n
v
er
g
en
ce
o
r
is
p
o
s
s
ib
ly
to
o
f
it.
T
o
tal
tr
ain
in
g
t
im
e
is
3
0
m
in
u
tes
f
o
r
R
esNet5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
an
d
2
5
m
i
n
u
tes
f
o
r
R
esNet5
0
.
T
h
is
s
h
o
ws
th
at
th
e
d
if
f
er
en
ce
in
tr
ain
in
g
tim
e
is
q
u
ite
d
if
f
er
en
t.
T
h
is
will r
esu
lt in
tim
e
in
ef
f
icien
cies in
tr
ain
in
g
,
wh
ic
h
is
to
o
tim
e
-
co
n
s
u
m
in
g
.
Epoch 25
Train_cost=0.0348
|
Test_cost=0.0326
|
Train_score=0.9899
| Test_score=0.9950 |
Train:
100%
33/33
[00:30<00:00,
2.44it/s]
Test:
100%
4/4
[00:04<00:00,
1.13it/s]
Epoch 26
Train_cost=0.0326
|
Test_cost=0.0465
|
Train_score=0.9879
| Test_score=0.9799 |
==> EarlyStop patience=1 | Best test_score: 0.9950
Train:
100%
33/33
[00:32<00:00,
1.80it/s]
Test:
100%
4/4
[00:03<00:00,
1.63it/s]
Epoch 27
Train_cost=0.0417
|
Test_cost=0.0196
|
Train_score=0.9865
| Test_score=0.9925 |
==> EarlyStop patience=2 | Best test_score: 0.9950
Train:
100%
33/33
[00:35<00:00,
2.03it/s]
Test:
100%
4/4
[00:03<00:00,
1.39it/s]
Epoch 28
Train_cost=0.0340
|
Test_cost=0.0199
|
Train_score=0.9891
| Test_score=0.9900 |
==> EarlyStop patience=3 | Best test_score: 0.9950
Train:
100%
33/33
[00:31<00:00,
2.35it/s]
Test:
100%
4/4
[00:03<00:00,
1.09s/it]
Epoch 29
Train_cost=0.0325
|
Test_cost=0.0197
|
Train_score=0.9908
| Test_score=0.9950 |
==> EarlyStop patience=4 | Best test_score: 0.9950
Train:
100%
33/33
[00:50<00:00,
1.54it/s]
Test:
100%
4/4
[00:03<00:00,
1.04it/s]
Epoch 30
Train_cost=0.0183
|
Test_cost=0.0165
|
Train_score=0.9947
| Test_score=0.9925 |
==> EarlyStop patience=5 | Best test_score: 0.9950
==> Execute Early Stopping at epoc
h: 30 | Best test_score:
0.9950
==> Best model is saved at model
Epoch 25
Train_cost=0.1060
|
Test_cost=0.0541
|
Train_score=0.9645
| Test_score=0.9724 |
==> EarlyStop patience=1 | Best test_score: 0.9875
Train:
100%
33/33
[00:36<00:00,
2.63it/s]
Test:
100%
4/4
[00:03<00:00,
1.26it/s]
Epoch 26
Train_cost
=
0.0286
|
Test_cost
=
0.0313
|
Train_score
=
0.9918 | Test_score = 0.9875 |
==> EarlyStop patience = 2 | Best test_score: 0.9875
Train:
100%
33/33
[00:31<00:00,
1.85it/s]
Test:
100%
4/4
[00:04<00:00,
1.25it/s]
Epoch 27
Train_cost
=
0.0246
|
Test_cost
=
0.0587
|
Train_score
=
0.9918 | Test_score = 0.9875 |
==> EarlyStop patience = 3 | Best test_score: 0.9875
Train:
100%
33/33
[00:30<00:00,
1.72it/s]
Test:
100%
4/4
[00:03<00:00,
1.24it/s]
Epoch 28
Train_cost
=
0.1039
|
Test_cost
=
0.0333
|
Train_score
=
0.9642 | Test_score = 0.9850 |
==> EarlyStop patience = 4 | Best test_score: 0.9875
Train:
100%
33/33
[00:30<00:00,
2.61it/s]
Test:
100%
4/4
[00:03<00:00,
1.08s/it]
Epoch 29
Train_cost
=
0.0230
|
Test_cost
=
0.0368
|
Train_score
=
0.9942 | Test_score = 0.9875 |
==> EarlyStop patience = 5 | Best test_score: 0.9875
==> Execute Early Stopping at epoc
h: 29 | Best test_score:
0.9875
==> Best model is saved at model
(
a)
(
b
)
Fig
u
r
e
6
.
T
r
ain
in
g
p
r
o
ce
s
s
co
m
p
ar
is
o
n
b
etwe
en
(
a)
R
esNet5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
an
d
(
b
)
R
esNet5
0
s
tan
d
ar
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
fr
a
ctu
r
e
cla
s
s
ifica
tio
n
u
s
in
g
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
et
w
o
r
k
a
r
ch
itectu
r
e
fo
r
…
(
S
o
likh
u
n
)
6473
Fig
u
r
e
6
is
a
g
r
ap
h
th
at
co
m
p
ar
es
th
e
p
er
f
o
r
m
a
n
ce
o
f
two
a
lg
o
r
ith
m
s
,
in
Fig
u
r
e
6
(
a)
R
esNet5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
an
d
Fig
u
r
e
6
(
b
)
R
esNet5
0
,
with
3
0
an
d
2
9
ep
o
c
h
s
,
r
esp
ec
tiv
ely
.
On
t
h
e
co
s
t
(
lo
s
s
)
g
r
ap
h
o
n
a
lo
g
ar
ith
m
ic
s
ca
le,
R
esNet5
0
p
l
u
s
s
h
o
ws
a
d
ec
r
ea
s
e
f
r
o
m
ar
o
u
n
d
1
(
ep
o
ch
0
)
t
o
ar
o
u
n
d
0
.
0
2
(
ep
o
ch
3
0
)
f
o
r
tr
ain
in
g
a
n
d
test
in
g
d
ata
with
ea
r
ly
s
to
p
p
atien
ce
=
5
,
B
est
te
s
t
s
co
r
e:
0
.
9
9
5
0
,
E
x
ec
u
te
ea
r
ly
s
to
p
p
in
g
at
ep
o
ch
:
3
0
,
B
est
te
s
t
s
co
r
e:
0
.
9
9
5
0
,
with
r
elativ
ely
s
m
all
f
lu
ctu
atio
n
s
.
R
esNet5
0
also
s
h
o
ws
a
d
ec
r
ea
s
e
in
co
s
t
f
r
o
m
a
r
o
u
n
d
1
to
ar
o
u
n
d
0
.
0
2
with
ea
r
ly
s
to
p
p
atien
ce
=
5
,
B
est
tes
t
s
co
r
e:
0
.
9
8
7
5
,
E
x
ec
u
te
ea
r
ly
s
to
p
p
i
n
g
a
t
ep
o
ch
:
2
9
,
B
est
test
s
co
r
e:
0
.
9
8
7
5
,
b
u
t
with
lar
g
e
r
f
lu
ctu
atio
n
s
.
On
th
e
s
co
r
e
(
ac
c
u
r
ac
y
)
g
r
ap
h
,
R
esNet5
0
p
lu
s
im
p
r
o
v
es f
r
o
m
ar
o
u
n
d
0
.
5
(
ep
o
ch
0
)
to
ar
o
u
n
d
1
(
e
p
o
ch
3
0
)
f
o
r
b
o
th
tr
ain
i
n
g
an
d
test
d
ata,
with
s
m
o
o
th
er
an
d
m
o
r
e
co
n
s
is
ten
t
lin
es.
Me
an
wh
ile,
R
esNe
t5
0
in
cr
ea
s
es
f
r
o
m
ar
o
u
n
d
0
.
5
to
1
with
lar
g
er
f
lu
ctu
atio
n
s
,
esp
ec
ially
in
th
e
ea
r
ly
ep
o
c
h
.
R
esNet5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
s
h
o
ws
a
m
o
r
e
s
tab
le
an
d
co
n
s
is
ten
t
p
er
f
o
r
m
an
ce
t
h
an
R
esNet5
0
.
Fig
u
r
e
s
7
(
a
)
an
d
7
(
b
)
s
h
o
w
s
th
e
co
n
f
u
s
io
n
m
atr
ix
o
f
th
e
two
ar
ch
itectu
r
es
test
ed
:
R
es
Net5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
an
d
R
esNet5
0
s
tan
d
ar
d
.
(
a)
(
b
)
Fig
u
r
e
7
.
L
o
s
s
lear
n
in
g
g
r
ap
h
an
d
s
co
r
e
lear
n
in
g
c
u
r
v
e
g
r
ap
h
f
r
ac
tu
r
e
im
ag
e
class
if
icatio
n
tr
ain
in
g
p
r
o
ce
s
s
u
s
in
g
ar
ch
itectu
r
e:
(
a)
R
esNet5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
a
n
d
(
b
)
R
esNet5
0
s
tan
d
ar
d
Fig
u
r
e
8
s
h
o
ws
th
e
co
n
f
u
s
io
n
m
atr
ix
f
o
r
two
alg
o
r
ith
m
s
,
Fig
u
r
e
8
(
a
)
R
esNet5
0
p
l
u
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
an
d
Fig
u
r
e
8
(
b
)
s
tan
d
ar
d
R
esNet5
0
,
wh
ich
class
if
y
d
ata
in
to
two
ca
teg
o
r
ies:
f
r
ac
tu
r
ed
an
d
n
o
t
f
r
ac
tu
r
ed
.
I
n
th
e
p
r
o
p
o
s
ed
m
eth
o
d
,
th
e
al
g
o
r
ith
m
co
r
r
ec
tl
y
class
if
ies
5
7
o
u
t
o
f
5
7
f
r
ac
tu
r
ed
im
a
g
es
(
T
r
u
e
Po
s
itiv
e)
an
d
in
co
r
r
ec
tly
class
if
ies
1
n
o
t
f
r
ac
tu
r
ed
im
ag
e
as
f
r
ac
tu
r
ed
(
Fals
e
Po
s
itiv
e)
,
with
7
0
o
u
t
o
f
7
1
n
o
t
f
r
ac
tu
r
ed
im
ag
es
b
ei
n
g
c
o
r
r
e
ctly
class
if
ied
(
T
r
u
e
Neg
ativ
e)
.
I
n
co
n
tr
ast,
th
e
s
tan
d
ar
d
m
eth
o
d
co
r
r
ec
tly
class
if
ies
6
4
o
u
t
o
f
6
4
f
r
ac
t
u
r
ed
im
ag
es
(
T
r
u
e
Po
s
itiv
e)
a
n
d
in
co
r
r
ec
tly
class
if
ies
2
n
o
t
f
r
ac
tu
r
ed
im
ag
es
as
f
r
ac
tu
r
ed
(
Fals
e
Po
s
itiv
e)
,
with
6
2
o
u
t
o
f
6
4
n
o
t
f
r
ac
t
u
r
ed
i
m
ag
es
b
ein
g
co
r
r
ec
tly
class
if
ied
(
T
r
u
e
Neg
ativ
e
)
.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
s
h
o
ws h
ig
h
er
ac
cu
r
ac
y
with
f
ewe
r
m
is
class
if
icatio
n
s
.
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.
14
,
No
.
6
,
Dec
em
b
e
r
20
24
:
6
4
6
6
-
6
4
7
7
6474
R
esNet5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
o
d
el)
:
R
esNet5
0
Stan
d
ar
d
:
=
57
+
70
57
+
70
+
1
+
0
=
127
128
≈
0
.
9922
=
57
57
+
1
=
57
58
≈
0
.
9828
=
57
57
+
0
=
57
57
≈
1
.
0
1
=
2
.
(
0
.
9828
.
1
.
0
)
0
.
9828
.
1
.
0
=
1
.
9656
1
.
9828
≈
0
.
9922
=
64
+
62
64
+
62
+
2
+
0
=
126
128
≈
0
.
9844
=
64
64
+
2
=
64
68
≈
0
.
9697
=
64
64
+
0
=
64
64
≈
1
.
0
1
=
2
.
(
0
.
9697
.
1
.
0
)
0
.
9697
.
1
.
0
=
1
.
9394
1
.
9697
≈
0
.
9847
I
n
R
esNet5
0
p
lu
s
,
o
f
th
e
5
8
f
r
ac
tu
r
ed
ca
s
es,
5
7
wer
e
co
r
r
e
ctly
class
if
ied
(
tr
u
e
p
o
s
itiv
e
)
,
an
d
1
was
in
co
r
r
ec
tly
class
if
ied
as
f
r
ac
t
u
r
ed
(
f
alse
p
o
s
itiv
e
)
,
wh
ile
o
f
th
e
7
0
n
o
t
f
r
ac
tu
r
e
d
ca
s
es,
all
wer
e
co
r
r
ec
tly
class
if
ied
(
tr
u
e
n
eg
ativ
e
)
.
T
h
i
s
alg
o
r
ith
m
p
r
o
d
u
ce
s
a
n
ac
cu
r
ac
y
o
f
0
.
9
9
2
2
,
p
r
ec
is
io
n
o
f
0
.
9
8
2
8
,
r
ec
all
o
f
1
.
0
,
an
d
F1
-
s
co
r
e
o
f
0
.
9
9
2
2
.
I
n
R
esNet5
0
,
o
f
th
e
6
6
f
r
ac
tu
r
ed
c
ases
,
6
4
wer
e
co
r
r
ec
tly
class
if
ied
(
tr
u
e
p
o
s
itiv
e
)
,
an
d
2
we
r
e
in
co
r
r
ec
tly
class
if
ied
as
f
r
ac
tu
r
e
d
(
f
alse
p
o
s
itiv
e
)
,
wh
ile
o
f
th
e
6
2
n
o
t
f
r
ac
tu
r
ed
ca
s
es,
all
wer
e
co
r
r
ec
tly
class
if
ied
(
tr
u
e
n
eg
ativ
e
)
.
T
h
is
alg
o
r
ith
m
p
r
o
d
u
c
es
an
ac
cu
r
ac
y
o
f
0
.
9
8
4
4
,
p
r
ec
is
io
n
o
f
0
.
9
6
9
7
,
r
ec
all
o
f
1
.
0
,
an
d
F1
-
s
co
r
e
o
f
0
.
9
8
4
7
.
T
h
ese
r
esu
lts
s
h
o
w
t
h
at
R
esNet5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
p
er
f
o
r
m
s
s
lig
h
tly
b
etter
ac
cu
r
ac
y
an
d
F1
-
s
co
r
e
th
an
R
esNet5
0
.
T
h
e
c
lass
if
icatio
n
r
esu
lts
ca
n
b
e
s
ee
n
in
th
e
Fig
u
r
e
9.
Fig
u
r
e
9
(
a)
R
esNet5
0
p
lu
s
an
d
Fig
u
r
e
9
(
b
)
R
esNet5
0
Stan
d
ar
d
illu
s
tr
ate
th
e
class
if
icatio
n
r
esu
lts
o
f
im
a
g
e
p
ar
am
eter
p
r
ed
ictio
n
s
o
f
f
r
ac
t
u
r
ed
a
n
d
n
o
n
-
f
r
ac
tu
r
e
d
b
o
n
es.
T
h
e
g
r
ee
n
m
ar
k
s
at
th
e
b
o
tt
o
m
o
f
th
e
f
o
n
t
ar
e
co
r
r
ec
t p
r
e
d
ictio
n
s
; if
th
ey
ar
e
r
ed
,
th
e
y
ar
e
in
c
o
r
r
ec
t
o
r
in
ac
cu
r
ate.
(
a)
(
b
)
Fig
u
r
e
8
.
C
o
n
f
u
s
io
n
m
atr
i
x
(
a)
R
esNet5
0
p
lu
s
d
an
(
b
)
R
esNet5
0
s
tan
d
ar
d
(
a)
(
b
)
Fig
u
r
e
9
.
C
lass
if
icatio
n
r
esu
lts
(
a)
R
esNet5
0
p
lu
s
an
d
(
b
)
R
e
s
Net5
0
s
tan
d
ar
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
fr
a
ctu
r
e
cla
s
s
ifica
tio
n
u
s
in
g
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
et
w
o
r
k
a
r
ch
itectu
r
e
fo
r
…
(
S
o
likh
u
n
)
6475
3
.
2
.
Dis
cus
s
io
n
T
h
is
s
tu
d
y
aim
s
to
co
m
p
ar
e
two
C
NN
ar
ch
itectu
r
al
m
o
d
e
ls
f
o
r
f
r
ac
tu
r
e
class
if
icatio
n
with
h
ig
h
ac
cu
r
ac
y
.
T
h
e
two
m
o
d
els
t
ested
ar
e
R
esNet5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
an
d
R
esNet5
0
.
I
n
T
ab
le
1
,
th
e
class
if
icatio
n
co
m
p
ar
is
o
n
r
esu
lts
o
f
th
e
two
C
NN
ar
ch
itectu
r
es
ar
e
p
r
esen
ted
.
T
h
e
R
esNet5
0
p
lu
s
m
o
d
el
was
d
ev
elo
p
e
d
with
s
ev
er
al
im
p
r
o
v
em
en
ts
an
d
o
p
tim
izatio
n
s
to
im
p
r
o
v
e
p
er
f
o
r
m
an
ce
in
class
if
y
in
g
b
o
n
e
f
r
ac
tu
r
e
im
ag
es.
Me
an
wh
ile,
th
e
R
esN
et5
0
m
o
d
el
is
u
s
ed
as
a
b
aseli
n
e
o
r
s
tan
d
ar
d
m
o
d
el
f
o
r
e
v
alu
atio
n
.
B
o
th
m
o
d
els
wer
e
test
ed
u
s
in
g
th
e
s
am
e
d
ataset,
an
d
th
eir
p
er
f
o
r
m
a
n
ce
was
co
m
p
ar
ed
b
ased
o
n
v
ar
io
u
s
m
etr
ics
s
u
ch
as
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
-
s
co
r
e.
T
h
e
co
m
p
ar
is
o
n
r
esu
lts
s
h
o
wn
in
T
ab
le
1
s
h
o
w
th
e
d
if
f
er
en
ce
s
in
p
er
f
o
r
m
a
n
ce
b
etwe
e
n
th
e
two
m
o
d
els.
T
h
e
R
esNet5
0
p
lu
s
m
o
d
el,
wh
ich
is
th
e
p
r
o
p
o
s
ed
m
eth
o
d
,
s
h
o
ws s
ig
n
if
ican
t im
p
r
o
v
em
en
t in
ter
m
s
o
f
class
if
icatio
n
ac
cu
r
ac
y
co
m
p
ar
ed
to
th
e
s
tan
d
ar
d
R
esNet5
0
m
o
d
el.
Ad
d
itio
n
ally
,
im
p
r
o
v
em
en
ts
in
o
th
e
r
m
etr
ics s
u
ch
as p
r
ec
is
io
n
,
r
ec
a
ll,
an
d
F1
-
s
co
r
e
a
r
e
also
v
is
ib
l
e
in
th
e
R
esNet5
0
p
lu
s
m
o
d
el.
T
ab
le
1
.
C
o
m
p
a
r
is
o
n
r
esu
lts
o
f
th
e
class
if
icatio
n
o
f
two
C
NN
ar
ch
itectu
r
es
O
p
t
i
mi
z
e
B
a
t
c
h
si
z
e
O
p
t
i
mi
z
a
t
i
o
n
Ep
o
c
h
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
A
c
c
u
r
a
c
y
R
e
sN
e
t
5
0
p
l
u
s
(
p
r
o
p
o
se
d
m
o
d
e
l
)
1
2
8
A
d
a
m
30
0
.
9
8
2
8
1
.
0
0
0
0
0
.
9
9
2
2
0
.
9
9
2
2
R
e
sN
e
t
5
0
64
R
M
S
P
r
o
p
29
0
.
9
6
9
7
1
.
0
0
0
0
0
.
9
8
4
7
0
.
9
8
4
4
T
h
e
r
esear
ch
r
esu
lts
s
h
o
w
th
at
R
esNet5
0
Plu
s
u
s
e
s
a
b
atch
s
ize
o
f
1
2
8
,
Ad
am
o
p
tim
iz
er
,
an
d
is
tr
ain
ed
f
o
r
3
0
ep
o
c
h
s
.
T
h
is
m
o
d
el
ac
h
iev
ed
a
p
r
ec
is
io
n
o
f
0
.
9
8
2
8
,
a
r
ec
all
o
f
1
.
0
0
0
0
,
an
F1
-
s
co
r
e
o
f
0
.
9
9
2
2
,
an
d
an
ac
cu
r
ac
y
o
f
0
.
9
9
2
2
.
O
n
th
e
o
th
er
h
an
d
,
R
esNet5
0
u
s
es
a
b
atch
s
ize
o
f
6
4
,
R
MSPr
o
p
o
p
tim
izer
,
an
d
is
tr
ain
ed
f
o
r
2
9
ep
o
c
h
s
.
T
h
is
m
o
d
el
ac
h
iev
ed
a
p
r
ec
is
io
n
o
f
0
.
9
6
9
7
,
a
r
ec
all
o
f
1
.
0
0
0
0
,
an
F1
-
s
co
r
e
o
f
0
.
9
8
4
7
,
an
d
a
n
ac
c
u
r
ac
y
o
f
0
.
9
8
4
4
.
T
h
e
r
esu
lts
s
h
o
w
th
at
R
esNet5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
p
e
r
f
o
r
m
s
s
lig
h
tly
b
etter
th
an
R
esNet5
0
,
with
a
h
ig
h
er
p
r
ec
is
io
n
v
alu
e
o
f
0
.
9
8
2
8
co
m
p
ar
ed
to
0
.
9
6
9
7
i
n
R
esNet5
0
.
E
v
en
t
h
o
u
g
h
b
o
th
m
o
d
els
ac
h
iev
ed
a
p
er
f
ec
t
r
ec
all
v
alu
e
(
1
.
0
0
0
0
)
,
th
e
F1
-
s
co
r
e
an
d
ac
c
u
r
ac
y
o
f
R
esNet5
0
Plu
s
(
0
.
9
9
2
2
)
wer
e
also
s
u
p
er
io
r
to
R
esNet5
0
(
F1
-
s
co
r
e
0
.
9
8
4
7
an
d
ac
c
u
r
ac
y
0
.
9
8
4
4
)
.
T
h
e
s
u
cc
ess
o
f
R
es
Net5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
in
ac
h
ie
v
in
g
b
etter
r
esu
lts
ca
n
b
e
attr
ib
u
ted
to
u
s
in
g
lar
g
er
b
atch
s
izes
an
d
th
e
Ad
am
o
p
tim
izer
,
w
h
ich
i
s
k
n
o
wn
t
o
h
a
v
e
b
etter
ad
ap
tiv
e
ca
p
ab
ilit
ies
in
u
p
d
atin
g
p
a
r
am
eter
s
.
T
h
is
s
h
o
ws
th
at
s
elec
tin
g
ap
p
r
o
p
r
i
ate
p
ar
am
eter
s
an
d
o
p
tim
iza
tio
n
m
eth
o
d
s
ca
n
im
p
r
o
v
e
m
o
d
el
p
er
f
o
r
m
an
ce
i
n
co
m
p
lex
im
ag
e
class
if
icatio
n
task
s
s
u
ch
as
b
o
n
e
f
r
ac
tu
r
e
id
en
tific
atio
n
.
T
h
is
s
tu
d
y
'
s
r
esu
lts
co
n
f
ir
m
th
at
R
esNet5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
is
m
o
r
e
ef
f
ec
tiv
e
in
p
er
f
o
r
m
in
g
b
o
n
e
f
r
ac
tu
r
e
class
if
icatio
n
with
h
ig
h
ac
c
u
r
ac
y
c
o
m
p
ar
e
d
with
R
esNet5
0
,
s
o
it
ca
n
b
e
a
b
etter
ch
o
ice
f
o
r
m
e
d
ical
ap
p
licatio
n
s
th
at
r
eq
u
ir
e
a
h
i
g
h
lev
el
o
f
p
r
ec
is
io
n
.
4.
CO
NCLU
SI
O
N
B
ased
o
n
th
e
r
esear
ch
r
esu
lts
co
m
p
ar
in
g
two
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
ar
c
h
itectu
r
e
m
o
d
els
in
b
o
n
e
f
r
ac
tu
r
e
class
if
icatio
n
,
i
t
ca
n
b
e
c
o
n
clu
d
ed
t
h
at
R
esNet5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
s
h
o
ws
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
co
m
p
ar
ed
to
R
esNet5
0
.
R
esNet5
0
Plu
s
,
wh
ich
u
s
es
a
b
atch
s
ize
o
f
1
2
8
an
d
th
e
Ad
am
o
p
tim
izer
f
o
r
3
0
ep
o
ch
s
,
ac
h
iev
es
a
p
r
e
cisi
o
n
o
f
0
.
9
8
2
8
,
r
ec
all
o
f
1
.
0
0
0
0
,
F1
-
s
co
r
e
o
f
0
.
9
9
2
2
,
an
d
ac
cu
r
ac
y
o
f
0
.
9
9
2
2
.
On
th
e
o
th
er
h
an
d
,
R
esNet5
0
,
wh
ich
u
s
es
a
b
atch
s
ize
o
f
6
4
an
d
th
e
R
MSPr
o
p
o
p
tim
izer
f
o
r
2
9
ep
o
c
h
s
,
ac
h
iev
es
a
p
r
ec
is
io
n
o
f
0
.
9
6
9
7
,
r
ec
all
o
f
1
.
0
0
0
0
,
F1
-
s
co
r
e
o
f
0
.
9
8
4
7
,
an
d
ac
cu
r
ac
y
o
f
0
.
9
8
4
4
.
T
h
e
s
u
cc
ess
o
f
R
esNet5
0
Plu
s
in
attain
in
g
b
etter
r
esu
lts
ca
n
b
e
attr
ib
u
ted
to
u
s
in
g
a
lar
g
er
b
atch
s
ize
an
d
th
e
Ad
am
o
p
tim
izer
,
wh
ich
h
as
b
etter
ad
ap
tiv
e
ca
p
a
b
ilit
ies
in
u
p
d
a
tin
g
p
ar
am
ete
r
s
.
T
h
ese
r
esu
lt
s
d
em
o
n
s
tr
ate
th
at
ap
p
r
o
p
r
iate
p
ar
am
ete
r
s
elec
tio
n
an
d
o
p
tim
izatio
n
m
eth
o
d
s
ca
n
im
p
r
o
v
e
m
o
d
el
p
er
f
o
r
m
an
ce
in
co
m
p
lex
im
a
g
e
class
if
icatio
n
task
s
,
s
u
ch
as
b
o
n
e
f
r
ac
tu
r
e
id
e
n
tific
atio
n
.
Ov
er
all,
th
is
s
tu
d
y
co
n
f
ir
m
s
th
at
R
esNet5
0
p
lu
s
(
p
r
o
p
o
s
ed
m
eth
o
d
)
is
m
o
r
e
ef
f
ec
tiv
e
in
p
er
f
o
r
m
in
g
b
o
n
e
f
r
ac
tu
r
e
class
if
icatio
n
with
h
ig
h
ac
cu
r
ac
y
th
a
n
R
esNet5
0
.
T
h
u
s
,
R
esNet5
0
Plu
s
ca
n
b
e
a
b
etter
c
h
o
ice
f
o
r
m
ed
ical
ap
p
licatio
n
s
r
e
q
u
ir
in
g
h
ig
h
p
r
ec
is
io
n
an
d
ac
cu
r
ac
y
.
RE
F
E
R
E
NC
E
S
[
1
]
P
.
A
l
k
h
a
i
r
i
a
n
d
A
.
P
.
W
i
n
d
a
r
t
o
,
“
C
l
a
ssi
f
i
c
a
t
i
o
n
a
n
a
l
y
si
s
o
f
b
a
c
k
p
r
o
p
a
g
a
t
i
o
n
-
o
p
t
i
mi
z
e
d
C
N
N
p
e
r
f
o
r
ma
n
c
e
i
n
i
mag
e
p
r
o
c
e
s
si
n
g
,
”
J
o
u
rn
a
l
o
f
S
y
st
e
m
s
E
n
g
i
n
e
e
r
i
n
g
a
n
d
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
(
J
O
S
EI
T
)
,
v
o
l
.
2
,
n
o
.
1
,
M
a
r
.
2
0
2
3
,
d
o
i
:
1
0
.
2
9
2
0
7
/
j
o
sei
t
.
v
2
i
1
.
5
0
1
5
.
[
2
]
A
.
P
.
W
i
n
d
a
r
t
o
,
T.
H
e
r
a
w
a
n
,
a
n
d
P
.
A
l
k
h
a
i
r
i
,
“
Ea
r
l
y
d
e
t
e
c
t
i
o
n
o
f
b
r
e
a
s
t
c
a
n
c
e
r
b
a
se
d
o
n
p
a
t
i
e
n
t
sy
m
p
t
o
m
d
a
t
a
u
si
n
g
n
a
i
v
e
B
a
y
e
s
a
l
g
o
r
i
t
h
m
o
n
g
e
n
o
m
i
c
d
a
t
a
,
”
i
n
T
h
e
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
S
m
a
rt
E
n
v
i
ro
n
m
e
n
t
,
2
0
2
4
,
p
p
.
4
7
8
–
4
8
4
.
[
3
]
A
.
P
.
W
i
n
d
a
r
t
o
,
T.
H
e
r
a
w
a
n
,
a
n
d
P
.
A
l
k
h
a
i
r
i
,
“
P
r
e
d
i
c
t
i
o
n
o
f
k
i
d
n
e
y
d
i
s
e
a
se
p
r
o
g
r
e
s
si
o
n
u
si
n
g
k
-
m
e
a
n
s
a
l
g
o
r
i
t
h
m
a
p
p
r
o
a
c
h
o
n
h
i
s
t
o
p
a
t
h
o
l
o
g
y
d
a
t
a
,
”
i
n
T
h
e
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
S
m
a
r
t
E
n
v
i
r
o
n
m
e
n
t
,
2
0
2
4
,
p
p
.
4
9
2
–
4
9
7
.
[
4
]
S
.
D
e
f
i
t
,
A
.
P
.
W
i
n
d
a
r
t
o
,
a
n
d
P
.
A
l
k
h
a
i
r
i
,
“
C
o
mp
a
r
a
t
i
v
e
a
n
a
l
y
si
s
o
f
c
l
a
ss
i
f
i
c
a
t
i
o
n
me
t
h
o
d
s
i
n
se
n
t
i
me
n
t
a
n
a
l
y
s
i
s:
t
h
e
i
m
p
a
c
t
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.