I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
d
van
c
e
s
i
n
A
p
p
li
e
d
S
c
ie
n
c
e
s
(
I
JA
A
S
)
V
ol
.
14
, N
o.
3
,
S
e
pt
e
m
be
r
20
25
, pp.
838
~
848
I
S
S
N
:
2252
-
8814
,
D
O
I
:
10.11591/
ij
a
a
s
.
v14.
i
3
.
pp838
-
848
838
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
aas
.i
ae
s
c
or
e
.c
om
D
e
e
p
l
e
ar
n
i
n
g f
or
i
m
age
c
l
ass
i
f
i
c
at
i
o
n
of
su
b
m
e
r
si
b
l
e
p
u
m
p
i
m
p
e
l
l
e
r
P
h
an
N
gu
ye
n
K
y
P
h
u
c
1
, D
oan
H
u
u
C
h
an
h
1
, T
r
on
g H
ie
u
L
u
u
2
1
S
c
hool
of
I
ndus
t
r
i
a
l
E
ngi
ne
e
r
i
ng a
nd M
a
na
ge
m
e
nt
, I
nt
e
r
na
t
i
ona
l
U
ni
ve
r
s
i
t
y, V
i
e
t
N
a
m
N
a
t
i
ona
l
U
ni
ve
r
s
i
t
y,
H
o C
hi
M
i
nh C
i
t
y, V
i
e
t
na
m
2
F
a
c
ul
t
y of
A
ut
om
a
t
i
on E
ngi
ne
e
r
i
ng, C
ol
l
e
ge
of
E
ngi
ne
e
r
i
ng, C
a
n T
ho U
ni
ve
r
s
i
t
y, C
a
n T
ho C
i
t
y, V
i
e
t
na
m
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
D
e
c
16, 2024
R
e
vi
s
e
d
M
a
y 19, 2025
A
c
c
e
pt
e
d
J
un 8, 2025
This
study
presented
a
deep
learning
-
based
model
in
the
submersible
pump
impellers
quality
inspection
process.
The
proposed
method
aimed
to
relieve
worker
workload,
automate
the
system,
as
well
as
increase
the
accur
acy
in
defect
detection
and
classification.
The
proposed
approach
aims
to
be
implemented
on
systems
with
low
investment
cost
and
limited
res
ources,
i.e., small sing
le
-
board computers, enabling flexible deployment
in ind
ustrial
environm
ents.
The
model
co
nsist
ed
of
three
convolutional
neural
n
etwork
(CNN)
models,
i.e.,
visual
geometry
group
16
(
VGG16
)
,
ResNet50,
and
a
custom
model.
The
outputs
of
three
networks
were
either
synthesize
d
later
through
a
n
ensemble
stage
or
used
separately.
A
graphical
user
in
terface
(GUI)
was
also
developed
for
real
-
tim
e
inspection
and
user
-
fr
iendly
interactio
n.
The
approac
h
achieve
d
up
to
99.8%
accura
cy
in
identifying
defects,
including
surface
scratches,
corrosion,
and
geometric
irregul
arities.
The
proposed
method
improved
the
quality
assurance
process
by
re
ducing
manual
inspection
efforts.
Future
research
could
explore
advanced
techniques
like
anomaly
detection
to
further
enhance
system
perfor
mance
and versat
ility
.
K
e
y
w
o
r
d
s
:
D
e
f
e
c
t
de
te
c
ti
on
E
ns
e
m
bl
e
m
ode
l
Q
ua
li
ty
i
ns
pe
c
ti
on
R
e
s
N
e
t5
0
V
is
ua
l
ge
om
e
tr
y gr
oup
16
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
T
r
ong Hie
u L
uu
F
a
c
ul
ty
of
A
ut
om
a
ti
on E
ngi
ne
e
r
in
g, C
ol
le
ge
of
E
ngi
ne
e
r
in
g, C
a
n T
ho U
ni
ve
r
s
it
y
C
a
n T
ho C
it
y,
V
ie
tn
a
m
E
m
a
il
:
lu
ut
r
onghie
u@
c
tu
.e
du.vn
1.
I
N
T
R
O
D
U
C
T
I
O
N
I
n
m
ode
r
n
in
dus
tr
y,
m
a
nua
l
in
s
pe
c
ti
ons
a
r
e
ti
m
e
-
c
ons
um
in
g
a
nd
e
r
r
or
-
pr
one
.
I
n
or
de
r
to
pr
ovi
de
r
e
li
a
bl
e
a
nd
hi
gh
-
qua
li
ty
pr
oduc
ts
,
in
s
pe
c
ti
ons
s
houl
d
b
e
a
u
to
m
a
te
d
a
nd
a
ppl
y
in
nova
ti
ve
t
e
c
hnol
ogi
e
s
.
C
ur
r
e
nt
ly
,
c
om
put
e
r
vi
s
io
n
a
nd
de
e
p
le
a
r
ni
ng
te
c
hni
que
s
a
r
e
two
c
a
ndi
da
te
m
e
th
ods
th
a
t
a
r
e
w
id
e
ly
a
dopt
e
d
by
in
dus
tr
y
due
to
th
e
ir
pr
om
in
e
nt
a
dv
a
nt
a
ge
s
,
s
u
c
h
a
s
lo
w
c
os
t
a
nd
im
pl
e
m
e
nt
a
ti
on
s
im
pl
ic
it
y.
I
n
th
e
c
om
put
e
r
vi
s
io
n
f
ie
ld
,
th
e
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
(
C
N
N
)
ha
s
pr
ove
d
it
s
huge
va
lu
e
th
r
ough
s
e
v
e
r
a
l
a
ppl
ic
a
ti
ons
i
n both m
a
nuf
a
c
tu
r
in
g a
nd i
ns
pe
c
ti
on pr
oc
e
s
s
e
s
.
A
c
c
or
di
ng
to
s
e
v
e
r
a
l
f
or
m
e
r
s
tu
di
e
s
,
d
e
e
p
le
a
r
ni
ng
s
ys
te
m
s
ou
tp
e
r
f
or
m
e
d
r
e
gul
a
r
m
a
c
hi
ne
le
a
r
ni
ng
s
ys
te
m
s
in
pa
tt
e
r
n
r
e
c
ogni
ti
on,
c
om
put
e
r
vi
s
io
n,
a
nd
im
a
g
e
pr
oc
e
s
s
in
g.
S
im
onya
n
a
nd
Z
is
s
e
r
m
a
n
[
1]
e
xa
m
in
e
d
C
N
N
a
nd
num
e
r
ous
a
r
c
hi
te
c
tu
r
e
s
,
li
ke
a
s
L
e
N
e
t,
A
l
e
xN
e
t,
a
nd
G
oogl
e
N
e
t,
on
th
e
la
r
ge
I
m
a
ge
N
e
t
da
ta
s
e
ts
.
T
he
y de
du
c
e
d f
r
om
t
hi
s
s
tu
dy t
ha
t
th
e
a
m
ount
of
da
ta
m
a
y
di
r
e
c
tl
y i
m
pa
c
t
th
e
numbe
r
of
e
poc
hs
a
nd
a
c
c
ur
a
c
y
of
th
e
s
e
le
c
te
d
m
ode
l.
M
or
e
ov
e
r
,
He
e
t
al
.
[
2]
pr
ovi
de
d
a
m
e
th
odol
ogy
f
or
pr
oduc
t
in
s
pe
c
ti
on
a
nd
te
s
ti
ng
ba
s
e
d
on
de
e
p
le
a
r
ni
ng
a
ppr
oa
c
he
s
.
B
a
s
e
d
on
th
e
goa
ls
of
e
xi
s
ti
ng
pr
oduc
t
in
s
pe
c
ti
on
s
ys
te
m
s
,
th
e
y
pr
ovi
de
d
a
n
e
f
f
e
c
ti
ve
m
e
th
od
f
or
s
us
ta
in
in
g
a
nd
e
nha
nc
in
g
a
pr
oduc
t
in
s
pe
c
ti
on
s
ys
te
m
.
D
ue
to
th
e
a
ppr
oa
c
h
pr
ovi
de
d,
th
e
pr
opos
e
d
s
y
s
te
m
w
a
s
s
e
e
n
to
ha
ve
good
s
ys
t
e
m
m
a
in
te
na
nc
e
a
nd
s
ta
bi
li
ty
.
I
n
a
ddi
ti
on,
ut
il
iz
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
D
e
e
p l
e
ar
ni
ng f
or
i
m
ag
e
c
la
s
s
if
ic
at
io
n of
s
ubm
e
r
s
ib
le
pum
p i
m
pe
ll
e
r
(
P
han N
guy
e
n K
y
P
huc
)
839
a
c
us
to
m
iz
e
d
A
le
xN
e
t
-
ba
s
e
d
C
N
N
.
K
r
is
hna
a
nd
K
a
ll
ur
i
[
3]
s
ugge
s
te
d
a
m
e
th
od
f
or
c
la
s
s
if
yi
ng
de
f
e
c
t
s
in
ya
r
n
-
dye
d
f
a
br
ic
s
.
T
he
e
xpe
r
im
e
nt
a
l
f
in
di
ngs
de
m
ons
tr
a
te
d
s
i
gni
f
ic
a
nt
r
e
s
il
ie
nc
e
i
n
ya
r
n
-
dye
d
f
a
br
ic
f
a
ul
t
c
la
s
s
if
ic
a
ti
on
a
nd
a
pr
om
is
in
g
a
ve
r
a
ge
c
la
s
s
if
ic
a
ti
on
r
a
te
.
K
im
e
t
al
.
[
4]
a
ls
o
p
r
e
s
e
nt
e
d
a
di
s
ti
nc
ti
ve
r
e
c
ogni
ti
on
s
tr
a
te
gy
f
or
s
te
e
l
s
ur
f
a
c
e
f
a
ul
ts
ba
s
e
d
on
upgr
a
de
d
a
r
ti
f
ic
ia
l
ne
ur
a
l
ne
twor
k
a
lg
or
i
th
m
s
,
in
c
lu
di
ng
f
e
a
tu
r
e
vi
s
ua
li
z
in
g
,
a
lo
ng
w
it
h
a
c
c
ur
a
c
y
e
va
lu
a
ti
on.
T
he
s
t
e
e
l
s
ur
f
a
c
e
de
f
e
c
t
c
la
s
s
if
ic
a
ti
on
pr
obl
e
m
w
a
s
pr
e
-
tr
a
in
e
d
us
in
g
th
e
vi
s
ua
l
ge
om
e
tr
y
gr
oup
16
(
V
G
G
19
)
,
a
nd
th
e
m
a
tc
hi
ng
D
V
G
G
19
w
a
s
de
ve
lo
pe
d
to
e
xt
r
a
c
t
th
e
f
e
a
tu
r
e
pi
c
tu
r
e
s
i
n va
r
io
us
l
a
ye
r
s
f
r
om
t
he
de
f
e
c
t
w
e
ig
ht
m
o
de
l.
T
he
n, a
ne
w
vi
r
tu
a
li
z
e
d s
e
r
vi
c
e
s
di
r
e
c
to
r
y
(
V
S
D
)
ne
twor
k
w
a
s
de
ve
lo
pe
d
a
nd
ut
il
iz
e
d
to
c
la
s
s
if
y
s
te
e
l
s
u
r
f
a
c
e
f
la
w
s
.
T
he
e
xpe
r
im
e
nt
f
in
di
ngs
poi
nt
e
d
out
t
ha
t
th
e
s
ugge
s
te
d a
ppr
oa
c
h m
a
y s
ig
ni
f
ic
a
nt
ly
i
nc
r
e
a
s
e
a
ve
r
a
ge
c
la
s
s
if
ic
a
ti
on a
c
c
ur
a
c
y, a
nd t
he
m
ode
l
c
a
n
c
onve
r
ge
f
a
s
t,
w
hi
c
h
w
a
s
be
ne
f
ic
ia
l
f
or
s
te
e
l
s
ur
f
a
c
e
de
f
e
c
t
id
e
nt
if
ic
a
ti
on
ut
il
iz
in
g
a
V
S
D
ne
twor
k
m
ode
l
o
f
f
e
a
tu
r
e
vi
s
ua
li
z
a
ti
on
a
nd
qua
li
ty
e
va
lu
a
ti
on.
F
or
a
ut
om
a
ti
c
f
r
ui
t
g
r
a
de
s
c
la
s
s
if
ic
a
ti
on,
J
in
g
e
t
al
[
5]
in
ve
s
ti
ga
te
d
th
e
in
f
lu
e
nc
e
of
s
e
ve
r
a
l
c
om
pl
e
x
C
N
N
s
tr
uc
tu
r
e
s
on
th
e
r
e
li
a
bi
li
ty
of
a
s
tr
a
w
be
r
r
y
gr
a
di
ng
s
ys
te
m
(
qua
li
ty
in
s
p
e
c
ti
on)
.
T
he
n,
th
e
y
e
xa
m
in
e
d
s
e
v
e
r
a
l
ty
pe
s
of
c
ur
r
e
nt
de
e
p
C
N
N
a
r
c
hi
te
c
tu
r
e
s,
s
u
c
h
a
s
A
le
xN
e
t,
M
obi
le
N
e
t,
G
oogL
e
N
e
t,
V
G
G
N
e
t,
a
nd
X
c
e
pt
io
n
,
c
om
pa
r
e
d
to
a
two
-
la
ye
r
C
N
N
a
r
c
hi
te
c
tu
r
e
.
A
c
c
or
di
ng
to
th
e
r
e
s
ul
ts
,
V
G
G
N
e
t
ha
d
th
e
hi
ghe
s
t
a
c
c
ur
a
c
y,
w
he
r
e
a
s
G
oogL
e
N
e
t
ha
d
th
e
m
os
t
c
om
put
a
ti
ona
ll
y
e
f
f
ic
ie
nt
de
s
ig
n.
B
ot
h
th
e
two
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on
a
nd
th
e
f
our
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on
s
ho
w
e
d
th
e
s
a
m
e
f
in
di
ngs
.
G
ua
n
e
t
al
.
[
6]
de
ve
lo
pe
d a
n
in
je
c
ti
on
m
oul
di
ng
qua
li
ty
in
s
pe
c
ti
on
pr
oc
e
s
s
s
y
s
te
m
in
e
dg
e
in
te
ll
ig
e
nc
e
.
A
s
a
r
e
s
ul
t,
th
e
m
e
nt
io
ne
d
m
ode
l'
s
a
c
c
ur
a
c
y
w
a
s
gr
e
a
te
r
th
a
n
90%
,
de
m
ons
tr
a
ti
ng
th
a
t
th
e
s
ys
te
m
m
a
y be
u
s
e
d i
n t
he
f
ie
ld
.
I
n
r
e
c
e
nt
ye
a
r
s
,
c
om
put
e
r
pe
r
f
or
m
a
nc
e
ha
s
im
pr
ove
d
dr
a
m
a
ti
c
a
ll
y,
le
a
di
ng
to
s
ig
ni
f
ic
a
nt
a
dva
nc
e
m
e
nt
s
in
de
e
p
le
a
r
ni
ng
te
c
hnol
ogy.
D
e
e
p
le
a
r
ni
ng
is
c
a
pa
bl
e
of
a
ut
om
a
ti
c
a
ll
y
le
a
r
ni
ng
c
om
pl
e
x
f
e
a
tu
r
e
s
,
gi
vi
ng
it
s
tr
ong
ge
ne
r
a
li
z
a
ti
on
a
bi
li
ti
e
s
a
nd
m
a
ki
ng
i
t
hi
ghl
y
e
f
f
e
c
ti
ve
in
va
r
io
us
obj
e
c
t
de
te
c
ti
on
ta
s
ks
[
7]
,
[
8]
.
A
s
a
r
e
s
ul
t,
de
f
e
c
t
de
te
c
ti
on
m
e
th
ods
ba
s
e
d
on
de
e
p
le
a
r
ni
ng
a
r
e
ga
in
in
g
m
or
e
a
tt
e
nt
io
n.
T
hi
s
f
ie
ld
c
a
n
be
di
vi
de
d
in
to
obj
e
c
t
de
te
c
ti
on
a
nd
obj
e
c
t
s
e
gm
e
nt
a
t
io
n.
Du
e
t
al
.
[
9]
e
nh
a
nc
e
d
th
e
F
a
s
te
r
R
-
C
N
N
ne
twor
k
by
in
c
or
por
a
ti
ng
f
e
a
tu
r
e
pyr
a
m
id
ne
twor
ks
(
FPN
)
a
n
d
r
e
gi
on
of
in
te
r
e
s
t
(
R
oI
A
li
gn
)
,
e
na
bl
in
g
th
e
de
te
c
ti
on
of
de
f
e
c
t
s
in
X
-
r
a
y
im
a
g
e
s
of
a
ut
om
ot
iv
e
c
a
s
t
a
lu
m
in
um
pa
r
ts
.
S
im
il
a
r
ly
,
X
ue
e
t
al
.
[
10]
us
e
d
te
c
hni
que
s
li
ke
O
ve
r
la
p
a
nd
M
o
s
a
ic
to
e
xpa
nd
th
e
tr
a
in
in
g
da
ta
s
e
t,
a
c
hi
e
vi
ng
a
c
c
ur
a
te
de
t
e
c
ti
on
of
va
r
io
us
c
a
s
ti
ng
de
f
e
c
ts
w
it
h
th
e
you
onl
y
lo
ok
onc
e
,
ve
r
s
io
n
3
(
Y
O
L
O
v3
)
m
ode
l.
D
ua
n
e
t
al
.
[
11]
a
dde
d
a
n
s
pa
ti
a
l
pyr
a
m
id
pool
in
g
(
SPP
)
la
ye
r
to
Y
O
L
O
v3
be
f
o
r
e
th
e
f
in
a
l
c
onvolut
io
na
l
la
ye
r
.
E
xpe
r
im
e
nt
a
l
r
e
s
ul
ts
de
m
ons
tr
a
te
d
a
s
ig
ni
f
ic
a
nt
im
pr
ove
m
e
nt
in
th
e
m
e
a
n
a
ve
r
a
ge
p
r
e
c
is
io
n
(
m
A
P
)
f
or
r
e
c
ogni
z
in
g
c
a
s
ti
ng
di
gi
ta
l
r
a
di
ogr
a
phy
(
DR
)
im
a
ge
de
f
e
c
ts
,
r
e
a
c
hi
ng
88.02%
c
om
p
a
r
e
d
to
th
e
or
ig
in
a
l
ne
twor
k
[
12]
.
C
ha
e
t
al
.
[
13]
a
nd
C
ui
e
t
al
.
[
14]
in
tr
oduc
e
d
a
C
N
N
th
a
t
c
om
bi
ne
s
f
a
s
te
r
R
-
C
N
N
w
it
h
a
r
e
gi
on
pr
opos
a
l
ne
twor
k
(
R
P
N
)
,
e
na
bl
in
g
th
e
d
e
te
c
ti
on
of
m
ul
ti
pl
e
ty
pe
s
of
da
m
a
ge
s
im
ul
ta
ne
ou
s
ly
a
t
a
r
e
m
a
r
ka
bl
e
s
pe
e
d
of
ju
s
t
0.03
s
e
c
onds
pe
r
im
a
ge
.
O
n
th
e
ot
he
r
ha
nd,
th
e
S
D
D
N
e
t
ne
twor
k
in
c
or
por
a
te
s
a
f
e
a
tu
r
e
r
e
f
in
e
m
e
nt
m
odul
e
(
F
R
B
)
a
nd
a
s
ki
p
-
la
ye
r
c
onne
c
ti
on
m
odul
e
to
ha
ndl
e
va
r
io
us
te
xt
ur
e
de
f
e
c
ts
[
15]
,
[
16]
.
H
ow
e
ve
r
,
th
is
m
ode
l
ha
s
li
m
it
a
ti
ons
, s
uc
h a
s
m
is
s
e
d de
te
c
ti
ons
w
he
n de
a
li
ng w
it
h t
a
r
ge
ts
t
ha
t
ha
ve
s
tr
ong ba
c
kgr
ound nois
e
or
unc
le
a
r
te
xt
ur
e
de
ta
il
s
.
I
t
a
ls
o
s
uf
f
e
r
s
f
r
om
lo
w
s
e
gm
e
nt
a
ti
on
a
c
c
ur
a
c
y
a
nd
li
m
it
e
d
ge
ne
r
a
li
z
a
ti
on
a
bi
li
ty
w
he
n
id
e
nt
if
yi
ng both wor
kpi
e
c
e
s
a
nd de
f
e
c
ti
ve
c
ondi
ti
ons
.
T
hi
s
s
tu
dy
lo
oke
d
in
to
th
e
e
f
f
e
c
ts
of
C
N
N
a
r
c
hi
te
c
tu
r
e
on
in
dus
tr
ia
l
pr
oduc
t
in
s
pe
c
ti
on
a
nd
c
la
s
s
if
ic
a
ti
on. W
hi
le
pr
e
vi
ous
s
tu
di
e
s
i
nve
s
ti
ga
te
d t
he
i
m
pa
c
t
of
de
e
p l
e
a
r
ni
ng t
e
c
hni
que
s
on
th
e
s
e
t
a
s
ks
, t
he
y
di
d
not
e
xpl
ic
it
ly
a
ddr
e
s
s
th
e
ir
in
f
lu
e
nc
e
on
s
y
s
te
m
de
pl
oym
e
nt
us
in
g
s
in
gl
e
-
boa
r
d c
om
put
e
r
s
. T
hr
ough
th
e
s
e
s
tu
di
e
s
,
C
N
N
a
nd
de
e
p
le
a
r
ni
ng
te
c
hni
que
s
ha
ve
pr
ove
n
e
f
f
e
c
ti
ve
f
or
pr
oduc
t
in
s
pe
c
ti
on
a
nd
c
la
s
s
if
ic
a
ti
on.
H
ow
e
ve
r
,
th
e
s
e
a
ppr
oa
c
h
e
s
r
e
qui
r
e
s
y
s
te
m
s
w
it
h
f
a
s
t
c
om
p
ut
a
ti
ona
l
c
a
pa
bi
li
ti
e
s
a
nd
hi
gh
R
A
M
,
of
te
n
le
a
di
ng
to
hi
gh
in
ve
s
tm
e
nt
c
os
ts
a
nd
pow
e
r
c
on
s
um
pt
io
n,
w
hi
c
h
hi
nde
r
th
e
ir
in
dus
tr
ia
l
a
ppl
ic
a
ti
ons
.
T
o
a
ddr
e
s
s
th
is
li
m
it
a
ti
on,
th
is
s
tu
dy
f
oc
us
e
d
on
C
N
N
a
r
c
hi
te
c
tu
r
e
s
w
it
h
a
ppr
opr
ia
te
s
iz
e
s
,
e
na
bl
in
g
e
a
s
y
de
pl
oym
e
nt
on s
in
gl
e
-
boa
r
d c
om
put
e
r
s
s
uc
h a
s
B
e
a
gl
e
B
one
, R
a
s
pbe
r
r
y P
i,
or
O
r
a
nge
P
i.
T
he
s
e
opt
im
iz
a
ti
ons
e
nha
nc
e
th
e
pr
a
c
ti
c
a
li
ty
of
C
N
N
a
nd
de
e
p
le
a
r
ni
ng
f
or
in
dus
tr
ia
l
in
s
pe
c
ti
on.
T
he
pr
opos
e
d
m
e
c
ha
ni
s
m
a
ls
o
e
ns
ur
e
s
s
y
s
te
m
s
u
s
ta
in
a
bi
li
ty
,
w
hi
c
h
is
a
c
r
it
ic
a
l
f
a
c
to
r
in
de
te
r
m
in
in
g
th
e
a
dopt
io
n
of
th
e
m
e
th
od.
S
in
c
e
th
e
pr
opos
e
d
s
ys
te
m
op
e
r
a
te
s
in
de
p
e
nde
nt
ly
of
th
e
m
a
in
s
ys
t
e
m
a
nd
e
a
c
h
m
odul
e
f
unc
ti
ons
a
ut
onomou
s
ly
,
th
e
in
s
pe
c
ti
on
s
ys
te
m
r
e
m
a
in
s
ope
r
a
ti
ona
l
e
v
e
n
if
a
m
odul
e
f
a
il
s
or
ne
e
ds
te
s
ti
ng
be
f
or
e
in
te
gr
a
ti
on.
F
ig
ur
e
1
il
lu
s
tr
a
te
s
t
he
s
ys
t
e
m
de
s
ig
n a
ppl
ie
d i
n t
hi
s
s
tu
dy.
We
f
oc
us
e
d
on
a
ut
om
a
ti
ng
in
s
p
e
c
ti
ons
of
s
ubm
e
r
s
ib
le
pum
p
im
pe
ll
e
r
s
,
w
hi
c
h
a
r
e
e
r
r
or
-
pr
one
due
to
va
r
io
us
c
a
s
ti
ng de
f
e
c
ts
. B
y i
m
pl
e
m
e
nt
in
g di
f
f
e
r
e
nt
C
N
N
a
r
c
hi
t
e
c
tu
r
e
s
s
uc
h a
s
V
G
G
16
[
7]
, R
e
s
N
e
t5
0 [
8]
,
a
nd
one
c
us
to
m
m
ode
l,
th
e
s
tu
dy
in
ve
s
ti
ga
te
d
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
us
in
g
a
no
n
-
pr
e
tr
a
in
e
d
C
N
N
unde
r
th
e
c
ir
c
um
s
ta
nc
e
of
li
m
it
e
d
da
ta
,
w
hi
c
h
is
ve
r
y
c
om
m
on
in
in
du
s
tr
y.
M
or
e
ove
r
,
dur
in
g
r
e
s
e
a
r
c
h,
th
e
tr
a
ns
f
e
r
le
a
r
ni
ng a
nd e
ns
e
m
bl
e
m
e
th
od
s
w
e
r
e
a
ls
o c
on
s
id
e
r
e
d. U
nde
r
t
r
a
ns
f
e
r
l
e
a
r
ni
ng, only t
he
out
put
l
a
ye
r
is
t
r
a
in
e
d
w
hi
le
ot
he
r
la
ye
r
s
a
r
e
f
r
oz
e
n.
T
hi
s
s
tr
a
te
gy
r
e
duc
e
s
ove
r
f
it
ti
ng,
th
e
tr
a
in
in
g
ti
m
e
,
a
nd
ut
il
iz
e
s
th
e
f
or
m
e
r
w
e
ll
pr
e
-
tr
a
in
e
d
m
ode
l.
T
he
r
e
tu
r
ne
d
r
e
s
ul
ts
of
a
ll
m
ode
ls
w
e
r
e
f
us
e
d
la
te
r
in
th
e
e
ns
e
m
bl
e
s
ta
ge
to
e
nha
nc
e
th
e
a
c
c
ur
a
c
y
a
nd
e
f
f
ic
ie
nc
y
of
de
f
e
c
t
de
te
c
ti
on
a
nd
c
la
s
s
if
ic
a
ti
on.
F
r
om
th
e
in
dus
tr
ia
l
vi
e
w
,
e
a
c
h
m
ode
l
pl
a
ye
d
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
.
14
, N
o.
3
,
S
e
pt
e
m
be
r
20
25
:
838
-
848
840
r
ol
e
a
s
a
m
odul
e
t
ha
t
c
ont
r
ib
ut
e
d di
r
e
c
tl
y t
o t
he
s
uc
c
e
s
s
of
t
he
c
la
s
s
if
ie
r
. D
ur
in
g t
he
ope
r
a
ti
on, i
ne
f
f
e
c
ti
ve
a
nd
in
a
c
c
ur
a
te
m
odul
e
s
c
a
n be
c
he
c
ke
d a
nd r
e
pl
a
c
e
d w
it
hout
c
r
e
a
t
i
ng
a
ny s
e
r
io
us
e
f
f
e
c
t
on
th
e
w
hol
e
s
ys
te
m
.
F
ig
ur
e
1. T
he
s
ys
te
m
m
ode
l
s
A
ddi
ti
ona
ll
y
,
a
gr
a
phi
c
a
l
u
s
e
r
in
te
r
f
a
c
e
(
G
U
I
)
w
a
s
d
e
ve
lo
pe
d
to
im
pr
ove
us
a
bi
li
ty
a
nd
r
e
a
l
-
ti
m
e
de
c
is
io
n
-
m
a
ki
ng
c
a
pa
bi
li
ti
e
s
.
T
hi
s
r
e
s
e
a
r
c
h
a
ddr
e
s
s
e
d
th
e
li
m
it
a
ti
ons
of
m
a
nua
l
in
s
pe
c
ti
on,
of
f
e
r
e
d
a
de
e
p
le
a
r
ni
ng
-
ba
s
e
d
a
ppr
oa
c
h
f
or
qua
li
ty
in
s
pe
c
ti
on.
T
he
in
te
gr
a
ti
on
of
C
N
N
m
ode
ls
a
nd
G
U
I
te
c
hnol
ogy
c
ont
r
ib
ut
e
d
to
m
or
e
e
f
f
ic
ie
nt
a
nd
a
c
c
ur
a
te
de
f
e
c
t
id
e
nt
if
ic
a
ti
o
n,
pa
v
in
g
th
e
w
a
y
f
or
s
m
a
r
te
r
pr
oduc
ti
on
a
nd
in
s
pe
c
ti
on
s
y
s
te
m
s
in
th
e
e
r
a
of
I
ndus
tr
y
4.0
a
nd
m
a
nuf
a
c
tu
r
in
g
di
gi
ti
z
a
ti
on.
I
n
th
e
r
e
s
t
of
th
e
pa
p
e
r
,
th
e
s
e
c
ond
s
e
c
ti
on
m
e
nt
io
ne
d
pr
ope
r
li
te
r
a
tu
r
e
r
e
vi
e
w
s
,
w
hi
le
th
e
ne
xt
s
e
c
ti
on
de
s
c
r
ib
e
d
th
e
us
e
d
da
ta
s
e
t
a
nd
pr
opos
e
d m
e
th
odol
ogy. T
he
c
on
c
lu
s
io
n w
a
s
pr
ovi
de
d i
n t
he
f
in
a
l
s
e
c
ti
on.
2.
R
E
S
E
A
R
C
H
M
E
T
H
O
D
2.1
.
D
at
as
e
t
i
n
f
or
m
at
io
n
T
he
da
ta
s
e
t
u
s
e
d
in
th
is
s
tu
dy
c
om
pr
is
e
s
7
,
348
im
a
ge
s
w
it
h
di
m
e
ns
io
ns
of
300
×
300
a
nd
w
as
obt
a
in
e
d
f
r
om
th
e
"
R
eal
-
li
f
e
in
dus
tr
ia
l
da
ta
s
e
t
of
c
a
s
ti
ng
pr
od
uc
t"
on
K
a
ggl
e
[
17]
.
A
ll
of
th
e
6
,
633
im
a
ge
s
w
e
r
e
us
e
d
f
or
tr
a
in
in
g
a
nd
v
a
li
da
ti
on,
w
it
h
3
,
758
r
e
pr
e
s
e
nt
in
g de
f
e
c
ti
ve
a
nd
2
,
875
r
e
pr
e
s
e
nt
in
g
non
-
de
f
e
c
ti
ve
pum
p
im
pe
ll
e
r
s
.
A
n
80/
20
s
pl
it
w
a
s
us
e
d
to
di
vi
de
th
e
tr
a
in
in
g
a
nd
va
li
da
ti
on
da
ta
.
T
he
te
s
t
f
ol
de
r
c
ont
a
in
e
d
453 im
a
ge
s
of
f
a
ul
ty
pump i
m
pe
ll
e
r
s
a
nd 262
i
m
a
ge
s
of
non
-
f
a
ul
ty
pump i
m
pe
ll
e
r
s
.
F
ig
ur
e
2
pr
ovi
de
s
a
s
hor
t
da
ta
s
e
t
s
um
m
a
r
y.
T
hi
s
da
t
a
s
e
t
w
as
a
lr
e
a
dy
us
e
d
in
s
om
e
w
or
ks
,
s
uc
h
a
s
E
ka
m
ba
r
a
n
a
nd
P
onnus
a
m
y
[
18]
,
A
lf
a
r
iz
i
e
t
al
.
[
19]
,
W
a
ng
a
nd
J
in
g
[
20
]
,
S
unda
r
a
m
a
nd
Z
e
id
[
21]
,
o
r
H
u
e
t
al
.
[
22]
,
to
na
m
e
a
f
e
w
.
E
ka
m
ba
r
a
n
a
nd
P
onnus
a
m
y
[
18]
us
e
d
m
ul
ti
pa
th
D
e
ns
e
N
e
t
a
nd
R
e
s
N
e
t3
4
f
or
pr
oduc
t
c
la
s
s
if
ic
a
ti
on.
A
lf
a
r
iz
i
e
t
al
.
[
19]
c
om
pa
r
e
d
th
e
a
c
c
ur
a
c
y
of
k
-
ne
a
r
e
s
t
ne
ig
hbor
s
(
K
N
N
)
a
nd
n
a
iv
e
B
a
ye
s
a
lg
or
it
hm
s
in
de
te
c
ti
ng
de
f
e
c
ts
in
im
pe
ll
e
r
pr
oduc
ts
.
U
s
in
g
t
-
di
s
tr
ib
ut
e
d
s
to
c
ha
s
ti
c
ne
ig
hbor
e
m
be
ddi
ng
(
t
-
S
N
E
)
vi
s
ua
li
z
a
ti
on,
th
e
s
tu
dy
c
onc
lu
de
d
th
a
t
K
N
N
w
a
s
m
or
e
r
e
li
a
bl
e
f
or
de
f
e
c
t
de
te
c
ti
on
in
in
dus
tr
ia
l
a
ppl
ic
a
ti
ons
.
W
a
ng
a
nd
J
in
g
[
20]
in
tr
oduc
e
d
th
e
c
oor
di
na
te
a
tt
e
nt
io
n
m
e
c
h
a
ni
s
m
in
to
th
e
ba
c
kbone
n
e
twor
k
to
a
ll
oc
a
te
m
or
e
a
tt
e
nt
io
n t
o t
he
de
f
e
c
t
ta
r
ge
t.
T
he
r
e
s
e
a
r
c
h a
ls
o us
e
d t
he
b
id
ir
e
c
ti
ona
l
w
e
ig
ht
e
d f
e
a
tu
r
e
pyr
a
m
id
ne
twor
k
in
th
e
f
e
a
tu
r
e
f
us
io
n
ne
twor
k
to
r
e
pl
a
c
e
th
e
or
ig
in
a
l
pa
th
a
ggr
e
ga
ti
on
ne
twor
k,
im
pr
ovi
ng
th
e
m
ode
l’
s
a
bi
li
t
y
to
f
us
e
f
e
a
tu
r
e
s
of
di
f
f
e
r
e
nt
s
iz
e
s
.
S
unda
r
a
m
a
nd
Z
e
id
[
21
]
in
t
r
oduc
e
d
th
e
qua
li
ty
c
ont
r
ol
s
ys
te
m
us
in
g
one
c
us
to
m
C
N
N
m
ode
l
f
or
in
s
pe
c
ti
on
a
nd
a
c
om
put
e
r
a
ppl
ic
a
ti
on
th
a
t
c
a
n
be
de
pl
oye
d
on
th
e
s
hop
f
lo
or
.
Hu
e
t
al
.
[
22]
a
dopt
e
d
th
e
X
c
e
pt
io
n
m
ode
l
to
c
r
e
a
te
a
r
obus
t
c
la
s
s
if
ic
a
ti
on
s
y
s
te
m
.
T
he
s
tu
dy
a
ls
o
a
ppl
ie
d
da
ta
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
to
e
nha
nc
e
th
e
d
a
ta
s
e
t
in
F
ig
u
r
e
3,
a
ll
ow
in
g
th
e
m
ode
l
to
ge
ne
r
a
li
z
e
m
or
e
e
f
f
e
c
ti
ve
ly
a
nd
im
pr
ove
it
s
de
f
e
c
t
r
e
c
ogni
ti
on
c
a
pa
bi
li
ti
e
s
.
A
ugm
e
nt
a
ti
on
te
c
hni
que
s
ha
ve
be
e
n
a
ppl
ie
d
to
a
ll
th
e
im
a
ge
s
in
th
e
da
ta
s
e
t
to
e
nha
nc
e
th
e
di
ve
r
s
it
y
a
nd
va
r
ia
bi
li
ty
of
th
e
da
ta
.
T
he
im
a
ge
s
w
e
r
e
la
be
ll
e
d
w
it
h
ta
gs
in
di
c
a
ti
ng
w
he
th
e
r
th
e
y
a
r
e
c
la
s
s
if
ie
d
a
s
"
ok"
(
nor
m
a
l
,
a
s
s
how
n
in
F
ig
ur
e
3
(
a
)
)
or
"
de
f
"
(
de
f
e
c
t/
a
nom
a
ly
,
a
s
s
how
n i
n
F
ig
ur
e
3
(
b
)
).
2.2
.
D
at
as
e
t
t
r
ai
n
in
g m
od
e
l
I
n
th
is
s
tu
dy,
V
G
G
16,
R
e
s
N
e
t5
0,
a
nd
a
c
u
s
to
m
m
ode
l
ba
s
e
d
o
n
a
C
N
N
s
tr
uc
tu
r
e
w
e
r
e
a
dopt
e
d
f
or
tr
a
in
in
g.
T
he
r
e
tu
r
ne
d
out
put
s
f
r
om
e
a
c
h
ne
twor
k
w
e
r
e
s
e
nt
la
te
r
to
th
e
e
ns
e
m
bl
e
s
ta
ge
f
or
m
a
ki
ng
f
in
a
l
de
c
is
io
ns
or
us
e
d
s
e
pa
r
a
te
ly
.
I
n
th
e
e
ns
e
m
bl
e
s
ta
ge
,
th
e
f
in
a
l
la
be
l
w
a
s
de
c
id
e
d
by
m
a
jo
r
it
y
r
ul
e
[
23]
.
T
he
m
a
in
us
e
of
th
e
e
ns
e
m
bl
e
te
c
hni
que
w
a
s
to
im
pr
ove
th
e
ove
r
a
ll
pe
r
f
or
m
a
nc
e
of
th
e
e
nt
i
r
e
s
ys
te
m
if
it
w
e
r
e
r
e
qui
r
e
d
.
B
y
c
om
bi
ni
ng
s
e
ve
r
a
l
in
de
p
e
nde
nt
w
e
a
k
c
la
s
s
if
ie
r
s
,
th
e
e
n
s
e
m
bl
e
m
e
th
od
c
a
n
c
r
e
a
te
a
s
tr
ong
c
la
s
s
if
ie
r
w
it
h hi
ghe
r
s
e
ns
it
iv
it
y. T
h
e
pr
opos
e
d s
tr
uc
tu
r
e
of
t
he
c
la
s
s
if
ie
r
i
s
s
how
n
i
n F
ig
ur
e
4.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
D
e
e
p l
e
ar
ni
ng f
or
i
m
ag
e
c
la
s
s
if
ic
at
io
n of
s
ubm
e
r
s
ib
le
pum
p i
m
pe
ll
e
r
(
P
han N
guy
e
n K
y
P
huc
)
841
I
n
a
d
di
ti
on,
P
yQ
t5
w
a
s
e
m
p
lo
ye
d
to
c
r
e
a
te
a
G
U
I
.
T
he
G
U
I
f
unc
ti
one
d
a
s
a
n
in
te
r
a
c
t
iv
e
im
a
ge
c
la
s
s
i
f
ic
a
t
io
n
t
oo
l,
a
ll
ow
in
g
us
e
r
s
to
in
p
ut
i
m
a
ge
s
o
f
p
um
p
i
m
pe
ll
e
r
s
a
nd
r
e
c
e
iv
e
r
e
a
l
-
t
i
m
e
o
ut
c
om
e
s
f
or
c
la
s
s
i
f
ic
a
t
io
n.
I
n
t
hi
s
r
e
s
e
a
r
c
h
,
t
he
da
ta
s
e
t
w
a
s
t
r
a
i
ne
d
us
in
g
v
a
r
i
ous
C
N
N
a
r
c
hi
te
c
tu
r
e
s
,
in
c
lu
di
n
g
V
G
G
16
,
R
e
s
N
e
t
50,
a
nd
a
c
us
to
m
m
o
de
l
.
T
he
c
us
to
m
m
o
de
l
c
ons
is
te
d
of
m
a
n
y
c
on
vo
lu
t
io
na
l
la
ye
r
s
a
c
ti
va
te
d
by
R
e
L
U
,
f
ol
lo
w
e
d
b
y
m
a
x
po
ol
in
g
f
o
r
d
im
e
ns
i
on
r
e
d
uc
t
io
n.
T
he
f
i
r
s
t
la
ye
r
e
m
p
lo
ye
d
3
2
3×
3
f
i
lt
e
r
s
w
it
h
R
e
L
U
a
c
t
iv
a
ti
on.
I
n
2×
2
w
i
ndo
w
s
,
o
ut
pu
ts
w
e
r
e
r
o
ut
e
d
t
hr
o
ugh
m
a
x
poo
l
in
g
.
T
hi
s
m
e
th
o
d
w
a
s
r
e
pe
a
t
e
d
two
m
o
r
e
t
i
m
e
s
.
T
he
f
in
a
l
m
a
xi
m
u
m
p
ool
in
g
o
ut
p
ut
s
w
e
r
e
f
la
tt
e
ne
d
a
nd
s
e
nt
to
a
de
ns
e
la
ye
r
w
it
h
s
i
gm
oi
d
a
c
ti
va
t
io
n
to
r
e
d
uc
e
d
im
e
ns
i
ona
li
ty
.
F
o
r
p
r
ob
a
b
il
it
y
di
s
tr
ib
u
ti
on
o
ve
r
c
la
s
s
e
s
,
th
e
l
a
s
t
de
ns
e
la
y
e
r
e
m
p
lo
y
e
d
S
o
f
tM
a
x
a
c
t
iv
a
ti
on
.
F
ig
ur
e
2. D
a
ta
di
s
pe
r
s
io
n
(
a
)
(
b)
F
ig
ur
e
3. R
a
ndom i
m
a
ge
s
of
pump i
m
pe
ll
e
r
s
f
r
om
th
e
da
ta
s
e
t
of
(
a
)
de
f
e
c
ti
ve
a
nd
(
b)
nor
m
a
l
F
ig
ur
e
4. T
he
pr
opos
e
d s
tr
uc
tu
r
e
of
t
he
c
la
s
s
if
ie
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
.
14
, N
o.
3
,
S
e
pt
e
m
be
r
20
25
:
838
-
848
842
I
n
ge
ne
r
a
l,
a
ll
th
r
e
e
c
ons
id
e
r
e
d
m
ode
ls
c
ont
a
in
e
d
s
e
v
e
r
a
l
two
-
di
m
e
ns
io
na
l
c
onvolut
io
n
la
y
e
r
s
.
T
he
y
a
r
e
th
e
ke
y
op
e
r
a
ti
ons
in
C
N
N
th
a
t
a
r
e
r
e
s
pon
s
ib
le
f
or
ge
ne
r
a
ti
ng
a
f
e
a
tu
r
e
m
a
p
th
a
t
c
a
pt
ur
e
s
im
por
ta
nt
pa
tt
e
r
ns
or
f
e
a
tu
r
e
s
i
n t
he
i
nput
i
m
a
ge
. T
he
op
e
r
a
ti
on c
a
n be
r
e
pr
e
s
e
nt
e
d m
a
th
e
m
a
ti
c
a
ll
y a
s
(
1)
.
(
∗
)
[
,
]
=
∑
[
,
]
.
[
−
,
−
]
∞
,
=
−
∞
(
1)
W
he
r
e
f
is
th
e
in
put
im
a
ge
a
nd
g
is
th
e
ke
r
ne
l,
x
a
nd
y
a
r
e
th
e
in
di
c
e
s
or
c
oor
di
na
te
s
of
th
e
out
put
f
unc
ti
o
n
r
e
s
ul
ti
ng
f
r
om
th
e
c
onvolut
io
n.
T
he
y
de
te
r
m
in
e
th
e
s
pe
c
if
ic
lo
c
a
ti
on
in
th
e
out
put
w
he
r
e
th
e
c
onvolut
io
n
ope
r
a
ti
on
is
a
ppl
ie
d.
A
c
ti
va
ti
on
f
unc
ti
ons
f
or
di
f
f
e
r
e
nt
ne
twor
ks
in
th
is
s
tu
dy
in
c
lu
de
r
e
c
ti
f
ie
d
li
ne
a
r
uni
t
(
R
e
L
U
)
[
24]
a
nd
s
ig
m
oi
d
[
25]
f
unc
ti
ons
,
w
hi
c
h
a
r
e
s
how
n
in
(
2)
-
(
3)
,
r
e
s
pe
c
ti
ve
ly
.
F
ur
th
e
r
m
or
e
,
a
t
th
e
out
put
la
ye
r
,
th
e
S
of
tM
a
x
[
26]
a
c
ti
va
ti
on f
unc
ti
on w
a
s
a
dopt
e
d a
s
i
n (
4
)
.
(
)
=
(
0
,
)
(
2)
(
)
=
1
1
+
−
=
+
1
(
3)
(
)
=
∑
=
1
(
4)
W
he
r
e
(
)
r
e
pr
e
s
e
nt
s
th
e
out
put
pr
oba
bi
li
ty
of
th
e
ℎ
c
la
s
s
,
is
th
e
e
xpone
nt
ia
l
f
unc
ti
on
a
ppl
ie
d
to
th
e
ℎ
c
la
s
s
out
put
,
a
nd
∑
=
1
r
e
pr
e
s
e
nt
s
th
e
s
um
of
e
xpon
e
nt
ia
l
f
unc
ti
ons
a
ppl
ie
d
to
a
ll
c
l
a
s
s
out
put
s
.
T
w
o
di
s
ti
nc
t
f
unc
ti
ons
w
e
r
e
u
s
e
d
to
c
om
put
e
th
e
lo
s
s
f
unc
ti
on.
T
he
f
ir
s
t
is
s
pa
r
s
e
c
a
te
gor
ic
a
l
c
r
o
s
s
-
e
nt
r
opy,
w
hi
c
h
is
a
ppr
opr
ia
te
f
or
m
ul
t
i
-
c
la
s
s
c
la
s
s
if
ic
a
ti
on j
obs
w
he
r
e
t
he
t
a
r
ge
t
va
r
ia
bl
e
i
s
i
nt
e
ge
r
e
nc
ode
d. T
hi
s
w
a
s
u
s
e
d i
n
th
e
c
us
to
m
m
ode
l
s
ta
te
d i
n (
5)
.
=
−
∑
∗
l
o
g
(
)
=
1
(
5)
W
he
r
e
L
r
e
pr
e
s
e
nt
s
th
e
lo
s
s
va
lu
e
,
N
is
th
e
num
be
r
of
c
la
s
s
e
s
,
is
th
e
tr
ue
la
be
l
a
nd
is
th
e
pr
e
di
c
te
d
pr
oba
bi
li
ty
di
s
tr
ib
ut
io
n
ove
r
ℎ
th
e
c
la
s
s
e
s
.
T
he
s
e
c
ond
lo
s
s
f
unc
ti
on
is
bi
na
r
y
c
r
os
s
-
e
nt
r
opy,
w
hi
c
h
is
of
te
n
us
e
d
f
or
bi
na
r
y
c
la
s
s
if
ic
a
ti
on
ta
s
ks
a
nd
i
s
e
m
pl
oye
d
in
V
G
G
16
a
nd
R
e
s
N
e
t5
0.
T
he
bi
na
r
y
c
r
os
s
-
e
nt
r
opy
lo
s
s
f
unc
ti
on i
s
m
a
th
e
m
a
ti
c
a
ll
y r
e
pr
e
s
e
nt
e
d by (
6)
.
=
−
1
∑
∗
=
1
(
̂
)
+
(
1
−
)
∗
(
1
−
̂
)
(
6
)
W
he
r
e
is
th
e
ta
r
ge
t
la
be
l
f
or
th
e
ℎ
s
a
m
pl
e
a
nd
̂
is
th
e
pr
e
di
c
te
d
pr
oba
bi
li
ty
by
th
e
m
ode
l
f
or
th
e
ℎ
s
a
m
pl
e
be
lo
ngi
ng t
o t
he
pos
it
iv
e
c
la
s
s
.
T
o
ut
il
iz
e
th
e
pow
e
r
of
th
e
s
e
s
uc
c
e
s
s
f
ul
ne
twor
ks
a
s
w
e
ll
a
s
to
r
e
duc
e
th
e
tr
a
in
in
g
ti
m
e
of
th
e
s
e
m
ode
ls
,
tr
a
ns
f
e
r
le
a
r
ni
ng
a
nd
e
ns
e
m
bl
e
te
c
hni
que
s
w
e
r
e
a
ppl
i
e
d
in
la
te
r
s
te
ps
.
F
or
V
G
G
16
a
nd
R
e
s
N
e
t5
0
,
onl
y
th
e
la
s
t
out
put
l
a
ye
r
w
a
s
r
e
tr
a
in
e
d
w
hi
le
th
e
ir
ot
he
r
la
y
e
r
s
w
e
r
e
f
r
oz
e
n.
S
to
c
h
a
s
ti
c
gr
a
di
e
nt
de
s
c
e
nt
(
S
G
D
)
a
nd
A
da
m
opt
im
iz
a
ti
on
m
e
th
ods
w
e
r
e
us
e
d
to
upda
t
e
ne
twor
k
pa
r
a
m
e
te
r
s
dur
in
g
tr
a
in
in
g.
S
G
D
c
ha
nge
s
p
a
r
a
m
e
te
r
s
ba
s
e
d
on
gr
a
di
e
nt
s
de
te
r
m
in
e
d
on
a
s
ub
s
e
t
of
tr
a
in
in
g
da
ta
,
w
he
r
e
a
s
A
da
m
c
om
bi
ne
s
a
da
pt
iv
e
le
a
r
ni
ng
r
a
te
s
a
nd
m
om
e
nt
um
.
T
r
a
in
in
g
us
e
d
a
ba
tc
h
s
iz
e
of
32
to
m
a
xi
m
iz
e
c
om
put
a
ti
ona
l
e
f
f
ic
ie
nc
y.
E
a
c
h
a
r
c
hi
te
c
tu
r
e
w
a
s
tr
a
in
e
d
f
or
ni
ne
e
poc
hs
,
a
ll
ow
in
g
th
e
ne
twor
k
to
le
a
r
n
a
nd
m
odi
f
y
w
e
ig
ht
s
a
nd
bi
a
s
e
s
ove
r
ti
m
e
.
A
n
e
poc
h
w
a
s
one
tr
ip
th
r
ough
th
e
c
om
pl
e
te
tr
a
in
in
g
da
ta
s
e
t.
T
hi
s
m
e
th
od
opt
im
iz
e
d
th
e
l
os
s
f
unc
ti
on, c
ha
nge
d
p
a
r
a
m
e
te
r
s
, a
nd i
nc
r
e
a
s
e
d
c
onv
e
r
ge
n
c
e
s
pe
e
d.
M
or
e
ove
r
,
t
he
G
U
I
c
r
e
a
te
d
w
it
h
P
y
Q
t
5
w
a
s
a
ls
o
de
v
e
l
ope
d
f
or
c
o
ns
t
r
uc
ti
ng
i
nt
e
r
a
c
ti
ve
a
nd
us
e
r
-
f
r
ie
ndl
y
p
r
og
r
a
m
s
.
T
he
G
U
I
a
ll
ow
s
t
he
us
e
r
t
o
u
pl
o
a
d
ph
ot
os
a
nd
s
e
le
c
t
th
e
de
s
ir
e
d
a
r
c
h
it
e
c
t
ur
e
f
o
r
in
s
pe
c
t
io
n
. W
he
n
t
he
a
pp
li
c
a
t
io
n
is
l
a
un
c
he
d,
th
e
us
e
r
i
s
p
r
e
s
e
nt
e
d
w
it
h a
n
in
tu
i
t
iv
e
i
n
te
r
f
a
c
e
i
n
w
hi
c
h
th
e
y
c
a
n
b
r
ow
s
e
a
nd
up
lo
a
d
i
m
a
ge
s
f
r
o
m
th
e
i
r
l
oc
a
l
s
ys
te
m
.
T
he
s
e
le
c
t
e
d
p
ho
to
s
w
e
r
e
th
e
n
i
ns
pe
c
te
d
us
i
ng
t
he
f
a
v
or
e
d
a
r
c
h
it
e
c
t
ur
e
of
c
ho
ic
e
(
e
.
g.
,
V
G
G
16
,
R
e
s
N
e
t5
0
,
t
h
e
c
us
t
om
m
ode
l
,
or
e
ns
e
m
bl
e
m
o
de
ls
)
.
B
y
ut
i
li
z
i
ng
t
he
t
r
a
i
ne
d
m
ode
ls
w
i
th
in
th
e
G
U
I
,
us
e
r
s
c
a
n
e
a
s
i
ly
e
va
l
ua
te
th
e
c
ond
it
io
n
of
pu
m
p
i
m
pe
ll
e
r
s
b
y
s
im
p
ly
u
pl
oa
d
in
g
a
n
i
m
a
ge
.
T
he
c
h
os
e
n
m
ode
l
w
i
ll
pr
o
c
e
s
s
t
he
i
m
a
g
e
a
nd
p
r
o
vi
de
t
he
p
r
e
d
ic
t
e
d
c
la
s
s
i
f
ic
a
t
io
n,
i
nd
ic
a
ti
n
g
w
h
e
th
e
r
th
e
im
p
e
l
le
r
is
de
f
e
c
ti
ve
o
r
non
-
de
f
e
c
t
iv
e
.
F
i
gu
r
e
5
s
how
s
th
e
d
e
ve
lo
p
e
d
GUI
,
a
n
d
t
he
c
la
s
s
i
f
ic
a
ti
o
n
lo
g
i
s
s
how
n
i
n
T
a
b
le
1
.
A
f
te
r
f
in
is
hi
ng
th
e
i
ns
pe
c
ti
on,
th
e
G
U
I
gi
ve
s
t
he
us
e
r
th
e
opt
io
n
t
o
e
x
po
r
t
t
he
r
e
s
u
lt
s
to
a
n
E
xc
e
l
f
i
le
.
T
he
p
r
o
duc
e
d
E
xc
e
l
f
i
le
c
o
m
p
r
is
e
s
t
he
r
e
c
o
gn
iz
e
d
c
ond
it
io
ns
o
f
t
he
t
e
s
te
d
p
r
o
duc
ts
,
m
a
k
in
g
i
t
e
a
s
y
f
o
r
f
u
r
th
e
r
a
na
ly
s
is
a
nd
r
e
c
o
r
d
-
ke
e
p
in
g.
T
he
P
yQ
t
5
-
ba
s
e
d
G
U
I
p
r
o
vi
de
d
a
s
m
oo
th
a
n
d
in
te
r
a
c
t
iv
e
e
xpe
r
ie
nc
e
,
a
l
lo
w
in
g
us
e
r
s
to
qui
c
k
ly
s
ub
m
i
t
ph
ot
os
,
c
h
oos
e
t
he
a
r
c
h
it
e
c
t
ur
e
,
a
nd
e
xp
or
t
in
s
pe
c
t
io
n
r
e
s
ul
ts
.
I
t
i
m
p
r
o
ve
d
us
a
bi
li
ty
a
nd
a
ll
ow
e
d
f
o
r
m
o
r
e
e
f
f
e
c
t
iv
e
da
ta
ha
nd
li
ng
a
n
d a
n
a
l
ys
is
f
or
qu
a
l
it
y c
ont
r
ol
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
D
e
e
p l
e
ar
ni
ng f
or
i
m
ag
e
c
la
s
s
if
ic
at
io
n of
s
ubm
e
r
s
ib
le
pum
p i
m
pe
ll
e
r
(
P
han N
guy
e
n K
y
P
huc
)
843
T
a
bl
e
1
.
T
he
c
la
s
s
if
ic
a
ti
on l
og
P
r
od.
ID
T
i
m
e
D
a
t
e
VGG
R
e
s
N
e
t
C
us
t
om
E
ns
e
m
bl
e
I
m
pe
l
l
e
r
973
16:
45
31
-
05
-
2023
Ok
Ok
Ok
Ok
I
m
pe
l
l
e
r
27
16:
45
31
-
05
-
2023
D
e
f
D
e
f
D
e
f
D
e
f
I
m
pe
l
l
e
r
50
16:
45
31
-
05
-
2023
D
e
f
D
e
f
D
e
f
D
e
f
I
m
pe
l
l
e
r
95
16:
45
31
-
05
-
2023
D
e
f
D
e
f
D
e
f
D
e
f
I
m
pe
l
l
e
r
71
16:
46
31
-
05
-
2023
Ok
Ok
Ok
Ok
I
m
pe
l
l
e
r
118
16:
46
31
-
05
-
2023
Ok
Ok
Ok
Ok
I
m
pe
l
l
e
r
26
16:
46
31
-
05
-
2023
D
e
f
D
e
f
D
e
f
D
e
f
...
...
...
...
...
...
...
...
F
ig
ur
e
5.
T
he
de
ve
lo
pe
d G
U
I
3.
R
E
S
U
L
T
S
3.1.
M
od
e
l
p
e
r
f
or
m
an
c
e
s
T
he
e
va
lu
a
ti
on
of
th
e
pr
opos
e
d
a
r
c
hi
te
c
tu
r
e
s
w
a
s
c
onduc
te
d
us
in
g
a
c
onf
us
io
n
m
a
tr
ix
,
a
c
c
ur
a
c
y,
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
F
1
-
s
c
or
e
.
T
h
e
s
e
e
v
a
lu
a
ti
on
m
e
tr
ic
s
pr
ovi
de
d
a
c
om
pr
e
he
ns
iv
e
pe
r
f
or
m
a
nc
e
of
th
e
m
ode
ls
,
va
lu
a
bl
e
in
s
ig
ht
s
a
bout
th
e
ir
c
a
pa
bi
li
ti
e
s
,
a
s
w
e
ll
a
s
t
he
ir
s
tr
e
ngt
hs
a
nd
w
e
a
kne
s
s
e
s
.
T
he
a
c
c
ur
a
c
y,
pr
e
c
is
io
n, r
e
c
a
ll
,
a
nd F
1
-
s
c
or
e
a
r
e
c
om
put
e
d by (
7)
-
(
10)
, r
e
s
pe
c
ti
ve
ly
.
=
+
+
+
+
(
7)
=
+
(
8)
=
+
(
9)
1
−
=
2
∗
∗
+
(
10)
F
ig
ur
e
6
pr
ov
id
e
d
de
ta
il
s
a
bout
th
e
tr
a
in
in
g
pr
oc
e
s
s
of
th
r
e
e
m
ode
ls
.
T
a
b
le
2
s
um
m
a
r
iz
e
s
th
e
pe
r
f
or
m
a
nc
e
of
th
r
e
e
m
ode
ls
a
f
te
r
9
e
poc
hs
.
W
e
f
ound
th
a
t
tr
a
in
in
g
a
c
c
ur
a
c
y
c
or
r
e
la
te
s
w
it
h
th
e
us
e
of
pr
e
-
tr
a
in
e
d
m
ode
ls
,
a
s
tr
a
ns
f
e
r
le
a
r
ni
ng
e
na
bl
e
s
f
a
s
te
r
c
onve
r
ge
nc
e
.
T
he
pr
opos
e
d
m
e
th
od
in
th
is
s
tu
dy
te
nde
d
to
ha
ve
a
n
in
or
di
na
te
ly
hi
ghe
r
pr
opor
ti
on
of
tr
a
in
in
g
e
f
f
ic
ie
nc
y
,
a
s
V
G
G
16
a
c
hi
e
ve
d
100%
tr
a
in
in
g
a
c
c
ur
a
c
y
a
nd
99.24%
va
li
da
ti
on
a
c
c
ur
a
c
y
a
t
th
e
9t
h
e
po
c
h,
w
hi
le
R
e
s
N
e
t5
0
r
e
a
c
he
d
99.92%
tr
a
in
in
g
a
c
c
ur
a
c
y
a
nd
99.47%
va
li
da
ti
on
a
c
c
ur
a
c
y
a
t
th
e
s
a
m
e
e
poc
h.
A
t
th
is
poi
nt
,
th
e
c
us
to
m
m
ode
l
obt
a
in
e
d
98.79%
tr
a
in
in
g
a
c
c
ur
a
c
y
a
nd
99.10%
va
li
da
ti
on
a
c
c
ur
a
c
y.
V
G
G
16
a
nd
R
e
s
N
e
t5
0
c
onve
r
ge
d
m
or
e
qui
c
kl
y
due
to
tr
a
ns
f
e
r
le
a
r
ni
ng,
w
he
r
e
a
s
th
e
c
us
to
m
m
ode
l
r
e
qui
r
e
d
a
lo
nge
r
tr
a
in
in
g
pe
r
io
d
s
in
c
e
a
ll
pa
r
a
m
e
t
e
r
s
ha
d
to
be
tr
a
in
e
d
f
r
om
s
c
r
a
tc
h, a
nd t
he
da
ta
s
e
t
u
s
e
d w
a
s
not
a
s
c
om
pr
e
he
ns
iv
e
a
s
t
he
one
a
ppl
ie
d f
or
V
G
G
16 a
nd R
e
s
N
e
t5
0.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
.
14
, N
o.
3
,
S
e
pt
e
m
be
r
20
25
:
838
-
848
844
T
a
bl
e
2
.
S
um
m
a
r
iz
e
s
th
e
p
e
r
f
or
m
a
nc
e
of
t
hr
e
e
m
ode
ls
a
f
te
r
9 e
poc
hs
M
ode
l
T
r
a
i
ni
ng
l
os
s
V
a
l
i
da
t
i
on
l
os
s
T
r
a
i
ni
ng
acc
ur
a
c
y
V
a
l
i
da
t
i
on
a
cc
ur
a
c
y
V
G
G
16
0.002
0.030
1.000
0.992
R
e
s
N
e
t
50
0.003
0.008
0.999
0.995
C
us
t
om
m
ode
l
0.007
0.010
0.987
0.971
F
r
om
th
e
F
ig
u
r
e
6,
a
ll
V
G
G
16
(
F
ig
ur
e
6
(
a
)
)
,
R
e
s
N
e
t5
0
(
F
ig
ur
e
6
(
b
)
)
,
a
nd
th
e
c
us
to
m
m
ode
l
(
F
ig
ur
e
6
(
c
)
)
a
ll
ha
d
hi
gh
pe
r
f
or
m
a
nc
e
in
pi
c
tu
r
e
c
la
s
s
if
ic
a
ti
on.
C
ons
e
que
nt
ly
,
a
ll
m
ode
ls
c
a
n
s
e
r
ve
a
s
a
ba
c
kbone
w
hi
c
h
a
r
e
r
e
s
pons
ib
le
f
or
f
e
a
tu
r
e
e
xt
r
a
c
ti
on.
T
h
e
ir
out
put
la
ye
r
c
a
n
be
s
ub
s
ti
tu
te
d
w
it
h
a
ny
pr
ope
r
c
la
s
s
if
ic
a
ti
on
m
e
th
od.
I
n
a
ddi
ti
on,
th
e
f
e
a
tu
r
e
ve
c
to
r
s
f
r
om
ba
c
kbone
s
c
a
n
a
ls
o
be
s
ta
c
k
e
d
to
f
or
m
a
uni
que
f
e
a
tu
r
e
ve
c
to
r
,
w
hi
c
h
c
a
n
be
us
e
d
a
s
in
put
f
or
a
ny
c
la
s
s
if
ie
r
.
D
e
pe
ndi
ng
on
th
e
a
ppl
ie
d
m
e
th
od
of
th
e
c
la
s
s
if
ie
r
,
in
de
pe
nde
nt
m
ode
ls
c
a
n
in
te
r
a
c
t
w
it
h
ot
he
r
s
in
s
e
ve
r
a
l
w
a
ys
a
nd
e
nha
nc
e
th
e
a
c
c
ur
a
c
y
of
th
e
s
ynt
he
s
is
m
ode
l.
T
hi
s
s
tu
dy
onl
y
e
xa
m
in
e
d
th
e
m
o
s
t
c
om
m
on
w
a
y
to
s
ynt
he
s
i
ze
th
e
s
e
r
e
s
ul
ts
by
u
s
in
g
th
e
ba
ggi
ng
m
e
th
od,
i.
e
.
,
th
e
m
a
jo
r
it
y
vot
in
g
r
ul
e
.
T
hi
s
ha
s
th
e
a
dva
nt
a
ge
of
e
a
s
e
w
he
n
th
e
m
odi
f
ic
a
ti
ons
a
r
e
r
e
qui
r
e
d.
T
he
in
c
lu
s
io
n
a
nd
e
xc
lu
s
io
n
of
a
ne
twor
k
f
r
om
t
he
e
ns
e
m
bl
e
c
a
n
be
done
ve
r
y
qui
c
kl
y
a
nd
c
onve
ni
e
nt
ly
w
it
hout
a
ny r
e
tr
a
in
e
d a
c
ti
vi
ty
. T
hi
s
c
a
nnot
be
a
c
hi
e
ve
d e
f
f
e
c
ti
ve
ly
i
f
ot
he
r
m
e
th
ods
a
r
e
a
dopt
e
d.
(
a
)
(
b)
(
c
)
F
ig
ur
e
6. T
r
a
in
in
g l
os
s
a
nd a
c
c
ur
a
c
y of
t
hr
e
e
m
ode
ls
of
(
a
)
t
r
a
in
in
g l
os
s
a
nd a
c
c
ur
a
c
y of
t
he
R
e
s
N
e
t
50
m
ode
l
, (
b)
t
r
a
in
in
g l
os
s
a
nd a
c
c
ur
a
c
y of
t
he
V
G
G
16 mode
l
, a
nd
(
c
)
t
r
a
in
in
g l
os
s
a
nd a
c
c
ur
a
c
y of
th
e
c
us
to
m
m
ode
l
3.2.
T
e
s
t
in
g ap
p
li
c
at
io
n
I
n
th
e
te
s
ti
ng
ph
a
s
e
,
th
e
p
e
r
f
or
m
a
nc
e
of
th
e
tr
a
in
e
d
m
ode
ls
w
a
s
e
v
a
lu
a
te
d
on
a
s
e
pa
r
a
te
da
ta
s
e
t
c
ont
a
in
in
g
453
im
a
ge
s
of
f
a
ul
ty
pum
p
im
pe
ll
e
r
s
a
nd
262
im
a
g
e
s
of
non
-
f
a
ul
ty
pum
p
im
pe
ll
e
r
s
.
O
ur
f
in
di
ngs
in
di
c
a
te
th
a
t
hi
ghe
r
f
e
a
tu
r
e
s
e
ns
it
iv
it
y
is
not
a
s
s
oc
ia
te
d
w
it
h
po
or
pe
r
f
or
m
a
nc
e
in
c
la
s
s
if
yi
ng
pum
p
im
pe
ll
e
r
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
D
e
e
p l
e
ar
ni
ng f
or
i
m
ag
e
c
la
s
s
if
ic
at
io
n of
s
ubm
e
r
s
ib
le
pum
p i
m
pe
ll
e
r
(
P
han N
guy
e
n K
y
P
huc
)
845
a
c
c
ur
a
te
ly
.
T
he
pr
opos
e
d
m
e
th
od
m
a
y
be
ne
f
it
f
r
om
in
c
or
por
a
ti
ng
e
ns
e
m
bl
e
s
tr
a
te
gi
e
s
w
it
hout
ne
ga
ti
ve
ly
a
f
f
e
c
ti
ng
it
s
ge
ne
r
a
li
z
a
ti
on
c
a
p
a
bi
li
ty
.
T
he
s
e
te
s
t
im
a
ge
s
,
r
e
pr
e
s
e
nt
in
g
r
e
a
l
-
li
f
e
s
a
m
pl
e
s
not
u
s
e
d
dur
in
g
tr
a
in
in
g
or
va
li
da
ti
on,
pr
ovi
de
d
a
n
unbi
a
s
e
d
be
nc
hm
a
r
k
f
or
a
s
s
e
s
s
in
g
th
e
m
ode
ls
'
a
bi
li
ty
to
a
c
c
ur
a
te
ly
di
s
ti
ngui
s
h
be
twe
e
n
f
a
ul
ty
a
nd
non
-
f
a
ul
ty
im
pe
ll
e
r
s
ba
s
e
d
o
n
th
e
le
a
r
ne
d
f
e
a
tu
r
e
s
a
nd
pa
tt
e
r
ns
.
F
ig
ur
e
7
s
how
s
th
e
c
onf
us
io
n m
a
tr
ix
of
a
ll
m
ode
ls
.
D
ur
in
g
th
e
te
s
ti
ng
pha
s
e
,
th
e
V
G
G
16
m
ode
l
e
xhi
bi
te
d
e
xc
e
pt
io
na
l
pe
r
f
or
m
a
nc
e
in
a
na
ly
z
in
g
th
e
te
s
ti
ng
phot
os
.
I
t
a
c
hi
e
ve
d
im
pr
e
s
s
iv
e
a
c
c
ur
a
c
y,
c
or
r
e
c
tl
y
p
r
e
di
c
ti
ng
450
out
of
453
f
a
ul
ty
phot
os
a
nd
a
c
c
ur
a
te
ly
id
e
nt
if
yi
ng
a
ll
non
-
de
f
e
c
ti
ve
im
a
ge
s
(
F
ig
ur
e
7(
a
)
)
.
T
hi
s
hi
ghl
ig
ht
e
d
th
e
m
ode
l'
s
e
f
f
e
c
ti
ve
ne
s
s
in
di
s
ti
ngui
s
hi
ng
be
twe
e
n
th
e
two
c
la
s
s
e
s
.
I
n
F
ig
ur
e
7(
b)
,
th
e
R
e
s
N
e
t5
0
m
ode
l
de
m
ons
tr
a
te
d
e
xc
e
pt
io
na
l
a
c
c
ur
a
c
y
in
pr
e
di
c
ti
ng
th
e
te
s
t
phot
os
.
I
t
a
c
c
ur
a
te
ly
id
e
nt
if
ie
d
451
out
of
453
f
a
ul
ty
phot
os
a
nd
c
or
r
e
c
tl
y
c
la
s
s
if
ie
d
a
ll
non
-
de
f
e
c
ti
ve
im
a
ge
s
,
s
how
c
a
s
in
g
it
s
e
f
f
e
c
ti
ve
ne
s
s
in
di
s
ti
ngui
s
hi
ng
be
twe
e
n
th
e
two
c
la
s
s
e
s
.
T
he
s
e
num
be
r
s
f
or
th
e
c
u
s
to
m
m
ode
l
w
e
r
e
452
out
of
453
in
th
e
de
f
e
c
ti
ve
c
la
s
s
a
nd
247
out
of
263
in
th
e
non
-
de
f
e
c
ti
ve
c
la
s
s
(
F
ig
ur
e
7(
c
)
)
,
w
hi
le
th
e
e
ns
e
m
bl
e
m
ode
l
ha
d
451
out
of
453
in
th
e
de
f
e
c
ti
ve
c
la
s
s
a
nd
263 out of
263 in t
he
non
-
de
f
e
c
ti
ve
c
la
s
s
(
F
ig
ur
e
7(
d)
)
.
(
a
)
(
b)
(
c
)
(
d)
F
ig
ur
e
7
.
T
r
a
in
in
g l
os
s
a
nd a
c
c
ur
a
c
y of
a
ll
m
ode
ls
of
(
a
)
c
onf
us
io
n m
a
tr
ix
of
t
he
V
G
G
m
ode
l
,
(
b)
c
onf
us
io
n
m
a
tr
ix
of
t
he
R
e
s
N
e
t5
0 m
ode
l
, (
c
)
c
onf
us
io
n m
a
tr
ix
of
t
he
c
us
to
m
m
ode
l
, a
nd (
d)
c
onf
us
io
n m
a
tr
ix
of
t
he
e
ns
e
m
bl
e
m
ode
l
T
hi
s
s
tu
dy
in
v
e
s
ti
ga
te
d
a
c
om
pr
e
he
ns
iv
e
e
va
lu
a
ti
on
a
nd
a
na
ly
s
is
of
th
e
pr
opos
e
d
s
y
s
te
m
.
H
ow
e
v
e
r
,
a
ddi
ti
ona
l
a
nd
in
-
de
pt
h
r
e
s
e
a
r
c
h
m
a
y
be
r
e
qui
r
e
d
to
c
onf
ir
m
it
s
lo
ng
-
te
r
m
e
f
f
e
c
ti
ve
ne
s
s
,
pa
r
ti
c
ul
a
r
ly
r
e
ga
r
di
ng
it
s
s
c
a
la
bi
li
ty
in
di
ve
r
s
e
in
dus
tr
ia
l
e
nvi
r
onm
e
nt
s
.
A
lt
hough
th
e
c
us
to
m
m
ode
l
s
di
d
not
pe
r
f
or
m
s
o
w
e
ll
c
om
pa
r
e
d
to
V
G
G
16
a
nd
R
e
s
N
e
t5
0
due
to
th
e
la
c
k
of
pr
e
-
tr
a
in
e
d
da
ta
,
th
e
pr
opos
e
d
s
ys
te
m
ha
s
a
ve
r
y
hi
gh
s
e
ns
it
iv
it
y
f
or
th
e
r
e
m
a
in
in
g
c
a
s
e
s
.
T
hi
s
c
a
n
be
a
tt
r
ib
ut
e
d
to
th
e
r
obus
t
s
e
ns
it
iv
it
y
of
th
e
two
pr
e
-
t
r
a
in
e
d
ne
twor
ks
.
G
iv
e
n
a
da
ta
poi
nt
be
lo
ngi
ng
to
c
la
s
s
C
,
if
a
ll
ne
two
r
ks
a
r
e
in
de
pe
nde
nt
of
ot
he
r
ne
twor
ks
a
nd
th
e
s
e
ns
it
iv
it
y
of
ne
twor
k
i
r
e
ga
r
di
ng
to
c
la
s
s
C
is
de
not
e
d
a
s
,
i.
e
.,
=
(
|
∈
)
,
th
e
s
e
ns
it
iv
it
y of
t
he
s
ys
te
m
w
h
e
n us
in
g t
he
e
n
s
e
m
bl
e
m
ode
l
r
e
ga
r
d
in
g t
o c
la
s
s
C
i
s
c
om
put
e
d a
s
in
(
11)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
.
14
, N
o.
3
,
S
e
pt
e
m
be
r
20
25
:
838
-
848
846
=
1
2
3
+
(
1
−
1
)
2
3
+
1
(
1
−
2
)
3
+
1
2
(
1
−
3
)
(
11)
I
n
di
s
c
us
s
io
n,
E
ka
m
ba
r
a
n
a
nd
P
onnus
a
m
y
[
18]
ut
il
iz
e
d
V
G
G
19
a
nd
R
e
s
N
e
t3
4,
a
c
hi
e
vi
ng
a
n
F1
-
s
c
or
e
of
99.54%
w
it
h
a
c
la
s
s
if
ic
a
ti
on
ti
m
e
of
454
m
s
.
S
im
il
a
r
ly
,
A
lf
a
r
iz
i
e
t
al
.
[
19
]
c
onduc
te
d
a
c
om
pa
r
a
ti
ve
a
na
ly
s
is
of
K
N
N
a
nd
na
ïv
e
B
a
ye
s
ia
n, de
m
ons
tr
a
ti
ng t
ha
t
K
N
N
a
c
hi
e
ve
d a
n a
c
c
ur
a
c
y of
98.11%
,
w
he
r
e
a
s
na
ïv
e
B
a
ye
s
ia
n only r
e
a
c
he
d 85.38%
. T
hi
s
f
in
di
ng
s
ugge
s
ts
t
ha
t
K
N
N
i
s
s
ig
ni
f
ic
a
nt
ly
m
or
e
e
f
f
e
c
ti
ve
f
or
f
a
ul
t
de
te
c
ti
on s
ys
te
m
s
. F
ur
th
e
r
m
or
e
, H
u
e
t
a
l
.
[
22]
e
m
pl
oy
e
d t
he
X
c
e
pt
io
n A
ug mode
l
to
e
nha
nc
e
s
ys
te
m
s
ta
bi
li
ty
.
T
he
s
tu
dy
c
om
pa
r
e
d
C
N
N
,
I
nc
e
pt
io
n
V
3,
a
nd
X
c
e
pt
io
n
A
ug,
r
e
por
ti
ng
a
c
c
ur
a
c
y
r
a
te
s
of
98.02%
,
98.48%
,
a
nd
99.16%
,
r
e
s
pe
c
ti
ve
ly
.
B
a
s
e
d
on
th
e
s
e
r
e
s
ul
t
s
,
th
e
a
ut
hor
c
onc
lu
de
d
th
a
t
th
e
X
c
e
pt
io
n
m
ode
l
is
th
e
m
os
t
s
ui
ta
bl
e
f
or
e
ns
ur
in
g
s
ys
te
m
s
ta
bi
li
ty
.
I
n
our
c
ont
r
ib
ut
io
n,
w
e
c
om
pa
r
e
V
G
G
16,
R
e
s
N
e
t5
0,
a
nd
a
c
us
to
m
m
ode
l.
T
he
f
in
di
ngs
in
di
c
a
te
th
a
t
bot
h
V
G
G
16
a
nd
R
e
s
N
e
t5
0
a
c
hi
e
ve
d
a
tr
a
in
in
g
a
c
c
ur
a
c
y
of
99%
.
A
ddi
ti
ona
ll
y,
w
e
de
ve
lo
pe
d
a
n
a
ppl
ic
a
ti
on
c
a
pa
bl
e
of
r
unni
ng
on
e
m
be
dde
d
c
om
put
in
g
de
vi
c
e
s
,
pr
ovi
di
ng
a
pr
a
c
ti
c
a
l
a
nd c
os
t
-
e
f
f
e
c
ti
ve
c
la
s
s
if
ic
a
ti
on s
ol
ut
io
n f
or
i
ndus
tr
ia
l
m
a
nuf
a
c
tu
r
in
g e
nvi
r
onm
e
nt
s
.
4.
C
O
N
C
L
U
S
I
O
N
T
he
s
tu
dy
u
s
e
d
V
G
G
16,
R
e
s
N
e
t5
0,
a
nd
th
e
c
u
s
to
m
m
ode
l
to
a
c
c
ur
a
te
ly
c
a
te
gor
iz
e
phot
o
s
of
c
a
s
ti
ng
pr
oduc
ts
,
s
pe
c
if
ic
a
ll
y
pum
p
im
pe
ll
e
r
s
.
O
ur
f
in
di
ngs
of
f
e
r
de
f
in
it
iv
e
pr
oof
th
a
t
th
is
phe
nom
e
non
is
li
nke
d
to
s
ubt
le
a
lt
e
r
a
ti
ons
in
pr
oduc
t
qua
li
ty
,
r
a
th
e
r
th
a
n
be
in
g
c
a
us
e
d
by
in
c
r
e
a
s
e
d
qua
nt
it
ie
s
of
im
a
ge
da
ta
.
T
he
s
tu
dy
e
m
pl
oye
d
V
G
G
16,
R
e
s
N
e
t5
0,
a
nd
a
c
u
s
to
m
m
ode
l
to
a
c
c
ur
a
te
ly
c
a
te
gor
iz
e
phot
os
of
c
a
s
ti
ng
pr
oduc
ts
-
s
pe
c
if
ic
a
ll
y,
pum
p
im
pe
ll
e
r
s
.
E
a
c
h
m
od
e
l
pe
r
f
or
m
e
d
a
dm
ir
a
bl
y,
a
c
hi
e
vi
ng
hi
gh
a
c
c
ur
a
c
ie
s
on
bot
h
tr
a
in
in
g
a
nd
te
s
ti
ng
da
ta
s
e
ts
.
A
ddi
ti
ona
ll
y,
a
us
e
r
-
f
r
ie
ndl
y
gr
a
phi
c
a
l
in
te
r
f
a
c
e
w
a
s
de
v
e
lo
pe
d
us
in
g
P
yQ
t5
,
e
na
bl
in
g
us
e
r
s
to
in
put
phot
os
,
s
e
le
c
t
di
f
f
e
r
e
nt
a
r
c
hi
te
c
tu
r
e
s
,
a
nd
e
xpor
t
th
e
c
a
te
gor
iz
a
ti
on
r
e
s
ul
ts
to
a
n
E
xc
e
l
f
il
e
.
T
he
pr
opos
e
d
de
s
ig
n
is
w
e
ll
-
s
ui
te
d
f
or
in
dus
tr
ia
l
in
s
pe
c
ti
on
a
nd
e
ns
ur
e
s
s
y
s
te
m
s
us
ta
in
a
bi
li
ty
,
a
s
it
ope
r
a
te
s
in
de
pe
nde
nt
ly
of
th
e
m
a
in
s
ys
te
m
w
it
h
e
a
c
h
m
odul
e
f
unc
ti
on
in
g
a
ut
onomous
ly
.
M
or
e
ove
r
,
a
s
id
e
f
r
om
th
e
lo
w
in
ve
s
tm
e
nt
c
os
t,
th
e
s
y
s
te
m
'
s
ove
r
a
ll
a
c
c
ur
a
c
y
in
it
s
in
it
ia
l
s
ta
ge
-
w
he
n
da
ta
a
r
e
li
m
it
e
d
-
c
a
n
be
e
nh
a
nc
e
d
by
le
ve
r
a
gi
ng
w
e
ll
-
known
pr
e
-
tr
a
in
e
d
m
ode
ls
in
pa
r
a
ll
e
l
w
it
h
c
us
to
m
m
ode
ls
.
T
he
e
ns
e
m
bl
e
m
e
th
od
a
ls
o
pl
a
ye
d
a
n
im
por
ta
nt
r
ol
e
in
in
c
r
e
a
s
in
g
th
e
a
c
c
ur
a
c
y
of
th
e
m
od
e
l
in
th
e
in
it
ia
l
s
ta
g
e
.
S
o,
f
ut
ur
e
r
e
s
e
a
r
c
h
m
a
y
lo
ok
in
to
va
r
io
us
da
ta
s
e
t
s
be
us
e
d
to
tr
a
in
th
e
m
ode
ls
. T
hi
s
w
il
l
a
id
in
e
va
lu
a
ti
ng
th
e
m
ode
ls
'
p
e
r
f
or
m
a
nc
e
on
a
va
r
ie
ty
of
m
a
nuf
a
c
tu
r
in
g
pr
oduc
ts
a
nd
d
e
te
r
m
in
e
th
e
ir
ge
ne
r
a
li
z
a
ti
on
c
a
pa
bi
li
ti
e
s
.
F
ur
th
e
r
m
or
e
,
in
c
r
e
a
s
in
g
th
e
G
U
I
f
unc
ti
ons
,
s
uc
h
a
s
a
ddi
ng
r
e
a
l
-
ti
m
e
im
a
ge
c
a
pt
ur
in
g
a
nd
of
f
e
r
in
g
vi
s
ua
li
z
a
ti
on
to
ol
s
f
or
i
m
pr
ove
d
m
ode
l
in
te
r
pr
e
ta
ti
on,
w
oul
d
im
p
r
ove
th
e
us
e
r
e
xpe
r
ie
nc
e
e
ve
n
f
ur
th
e
r
.
T
hi
s
s
tu
dy
hi
ghl
ig
ht
s
th
e
s
uc
c
e
s
s
f
ul
im
pl
e
m
e
nt
a
ti
on
of
C
N
N
a
r
c
hi
te
c
tu
r
e
s
a
nd
e
ns
e
m
bl
e
m
e
th
od
s
in
in
dus
tr
ia
l
qua
li
ty
c
ont
r
ol
a
nd
e
xhi
bi
ts
th
e
pos
s
ib
il
it
ie
s
of
us
in
g G
U
I
t
e
c
hnol
ogy f
or
e
f
f
ic
ie
nt
a
nd us
e
r
-
f
r
ie
ndl
y pi
c
tu
r
e
c
a
te
gor
iz
a
ti
on s
ys
te
m
s
.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
A
ut
hor
s
s
ta
te
no f
undi
ng i
nvol
ve
d.
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
di
T
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on.
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
P
ha
n N
guye
n K
y P
huc
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
D
oa
n H
uu C
ha
nh
✓
✓
✓
✓
✓
✓
✓
T
r
ong Hie
u L
uu
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
l
i
z
a
t
i
on
M
:
M
e
t
hodol
ogy
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
da
t
i
on
Fo
:
Fo
r
m
a
l
a
na
l
ys
i
s
I
:
I
nve
s
t
i
ga
t
i
on
R
:
R
e
s
our
c
e
s
D
:
D
a
t
a
C
ur
a
t
i
on
O
:
W
r
i
t
i
ng
-
O
r
i
gi
na
l
D
r
a
f
t
E
:
W
r
i
t
i
ng
-
R
e
vi
e
w
&
E
di
t
i
ng
Vi
:
Vi
s
ua
l
i
z
a
t
i
on
Su
:
Su
pe
r
vi
s
i
on
P
:
P
r
oj
e
c
t
a
dm
i
ni
s
t
r
a
t
i
on
Fu
:
Fu
ndi
ng a
c
qui
s
i
t
i
on
C
O
N
F
L
I
C
T
O
F
I
N
T
E
R
E
S
T
S
T
A
T
E
M
E
N
T
A
ut
hor
s
s
ta
te
no c
onf
li
c
t
of
i
nt
e
r
e
s
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
D
e
e
p l
e
ar
ni
ng f
or
i
m
ag
e
c
la
s
s
if
ic
at
io
n of
s
ubm
e
r
s
ib
le
pum
p i
m
pe
ll
e
r
(
P
han N
guy
e
n K
y
P
huc
)
847
D
A
T
A
A
V
A
I
L
A
B
I
L
I
T
Y
T
he
da
ta
th
a
t
s
uppor
t
th
e
f
in
di
ngs
of
th
i
s
s
tu
dy
a
r
e
ope
nl
y
a
va
il
a
bl
e
in
[
K
a
ggl
e
]
a
t
ht
tp
s
:/
/ww
w
.ka
ggl
e
.c
om
/d
a
ta
s
e
ts
/r
a
vi
r
a
js
in
h45/
r
e
a
l
-
li
f
e
-
in
dus
tr
ia
l
-
da
ta
s
e
t
-
of
-
c
a
s
ti
ng
-
pr
oduc
t/
da
ta
,
r
e
f
e
r
e
nc
e
num
be
r
[
17
].
R
E
F
E
R
E
N
C
E
S
[
1]
K
.
S
i
m
onya
n
a
nd
A
.
Z
i
s
s
e
r
m
a
n,
“
V
e
r
y
de
e
p
c
onvol
ut
i
ona
l
ne
t
w
or
ks
f
or
l
a
r
ge
-
s
c
a
l
e
i
m
a
ge
r
e
c
ogni
t
i
on,”
i
n
3
r
d
I
nt
e
r
nat
i
ona
l
C
onf
e
r
e
nc
e
on
L
e
ar
ni
ng
R
e
pr
e
s
e
nt
at
i
ons
,
I
C
L
R
2015
-
C
onf
e
r
e
nc
e
T
r
ac
k
P
r
oc
e
e
di
ngs
,
2015,
pp.
1
–
14.
[
O
nl
i
ne
]
.
A
va
i
l
a
bl
e
:
ht
t
ps
:
/
/
s
hor
t
l
i
nkur
l
.i
d/
8z
2o8
[
2]
K
.
H
e
,
X
.
Z
ha
ng,
S
.
R
e
n,
a
nd
J
.
S
un,
“
D
e
e
p
r
e
s
i
dua
l
l
e
a
r
ni
ng
f
or
i
m
a
ge
r
e
c
ogni
t
i
on,”
i
n
2016
I
E
E
E
C
onf
e
r
e
nc
e
on
C
o
m
put
e
r
V
i
s
i
on and P
at
t
e
r
n R
e
c
ogni
t
i
on
, I
E
E
E
,
J
un. 2016, pp. 770
–
778. doi
:
10.1109/
C
V
P
R
.2016.90.
[
3]
S
.
T
.
K
r
i
s
hna
a
nd
H
.
K
.
K
a
l
l
ur
i
,
“
D
e
e
p
l
e
a
r
ni
ng
a
nd
t
r
a
ns
f
e
r
l
e
a
r
ni
ng
a
ppr
oa
c
he
s
f
or
i
m
a
ge
c
l
a
s
s
i
f
i
c
a
t
i
on,”
I
nt
e
r
nat
i
onal
J
ou
r
nal
of
R
e
c
e
nt
T
e
c
hnol
ogy
and
E
ngi
ne
e
r
i
ng
,
vol
.
7,
no.
5,
pp.
427
–
432,
2019,
[
O
nl
i
ne
]
.
A
va
i
l
a
bl
e
:
ht
t
ps
:
/
/
w
w
w
.i
j
r
t
e
.or
g/
w
p
-
c
ont
e
nt
/
upl
oa
ds
/
pa
pe
r
s
/
v7i
5s
4/
E
10900275S
419.pdf
[
4]
T
.
H
.
K
i
m
,
H
.
R
.
K
i
m
,
a
nd
Y
. J
.
C
ho,
“
P
r
oduc
t
i
ns
pe
c
t
i
on
m
e
t
hodol
ogy
vi
a
de
e
p
l
e
a
r
ni
ng:
a
n
ove
r
vi
e
w
,”
Se
ns
or
s
,
vol
.
21,
no.
15,
p. 5039, J
ul
. 2021, doi
:
10.3390/
s
21155039.
[
5]
J
.
J
i
ng,
A
.
D
ong,
P
.
L
i
,
a
nd
K
.
Z
ha
ng,
“
Y
a
r
n
-
dye
d
f
a
br
i
c
de
f
e
c
t
c
l
a
s
s
i
f
i
c
a
t
i
on
ba
s
e
d
on
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k,”
O
pt
i
c
al
E
ngi
ne
e
r
i
ng
, vol
. 56, no. 09, p. 1, S
e
p. 2017, doi
:
10.1117/
1.oe
.56.9.093104.
[
6]
S
.
G
ua
n,
M
.
L
e
i
,
a
nd
H
.
L
u,
“
A
s
t
e
e
l
s
ur
f
a
c
e
de
f
e
c
t
r
e
c
ogni
t
i
on
a
l
gor
i
t
hm
ba
s
e
d
on
i
m
pr
ove
d
de
e
p
l
e
a
r
ni
ng
ne
t
w
or
k
m
ode
l
us
i
ng
f
e
a
t
ur
e
vi
s
ua
l
i
z
a
t
i
on a
nd qua
l
i
t
y e
va
l
ua
t
i
on,”
I
E
E
E
A
c
c
e
s
s
, vol
. 8, pp. 49885
–
49895, 2020, doi
:
10.1109/
A
C
C
E
S
S
.2020.2979755.
[
7]
T
.
W
a
ng,
Y
.
C
he
n,
M
.
Q
i
a
o,
a
nd
H
.
S
nous
s
i
,
“
A
f
a
s
t
a
nd
r
obus
t
c
onvol
ut
i
on
a
l
ne
ur
a
l
ne
t
w
or
k
-
ba
s
e
d
de
f
e
c
t
de
t
e
c
t
i
on
m
ode
l
i
n
pr
oduc
t
qua
l
i
t
y
c
ont
r
ol
,”
T
he
I
nt
e
r
nat
i
onal
J
our
nal
of
A
dv
anc
e
d
M
anuf
ac
t
ur
i
ng
T
e
c
hnol
ogy
,
vol
.
94,
no.
9
–
12,
pp.
3465
–
3471,
F
e
b. 2018, doi
:
10.1007/
s
00170
-
017
-
0882
-
0.
[
8]
C
.
D
e
w
i
,
R
.
C
.
C
he
n,
a
nd
H
.
Y
u,
“
W
e
i
ght
a
na
l
ys
i
s
f
or
va
r
i
ous
pr
ohi
bi
t
or
y
s
i
gn
de
t
e
c
t
i
on
a
nd
r
e
c
ogni
t
i
on
us
i
ng
de
e
p
l
e
a
r
ni
ng,”
M
ul
t
i
m
e
di
a T
ool
s
and A
ppl
i
c
at
i
ons
, vol
. 79, no. 43
–
44, pp. 32897
–
32915, N
ov.
2020, doi
:
10.1007/
s
11042
-
020
-
09509
-
x.
[
9]
W
.
D
u,
H
.
S
he
n,
J
.
F
u,
G
.
Z
ha
ng,
a
nd
Q
.
H
e
,
“
A
ppr
oa
c
he
s
f
or
i
m
pr
ove
m
e
nt
of
t
he
X
-
r
a
y
i
m
a
ge
de
f
e
c
t
de
t
e
c
t
i
on
of
a
ut
om
obi
l
e
c
a
s
t
i
ng
a
l
um
i
num
pa
r
t
s
ba
s
e
d
on
d
e
e
p
l
e
a
r
ni
ng,”
N
D
T
and
E
I
nt
e
r
nat
i
onal
,
vol
.
107,
p.
102144,
O
c
t
.
2019,
doi
:
10.1016/
j
.ndt
e
i
nt
.2019.102144.
[
10]
L
.
X
ue
,
J
.
H
e
i
,
X
.
C
he
n,
Y
.
X
u,
a
nd
L
.
Z
he
ng,
“
A
n
e
f
f
i
c
i
e
nt
m
e
t
hod
of
c
a
s
t
i
ng
de
f
e
c
t
s
de
t
e
c
t
i
on
ba
s
e
d
on
de
e
p
l
e
a
r
ni
ng,”
i
n
P
r
oc
e
e
di
ngs
of
2020
I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
A
dv
anc
e
s
i
n
E
l
e
c
t
r
i
c
al
E
ngi
ne
e
r
i
ng
and
C
om
put
e
r
A
ppl
i
c
at
i
ons
,
A
E
E
C
A
2020
, I
E
E
E
, A
ug. 2020, pp. 480
–
483. doi
:
10.1109/
A
E
E
C
A
49918.2020.9213492.
[
11]
L
.
D
ua
n,
K
.
Y
a
ng,
a
nd
L
.
R
ua
n,
“
R
e
s
e
a
r
c
h
on
a
ut
om
a
t
i
c
r
e
c
ogni
t
i
on
of
c
a
s
t
i
ng
de
f
e
c
t
s
ba
s
e
d
on
de
e
p
l
e
a
r
ni
ng,”
I
E
E
E
A
c
c
e
s
s
,
vol
. 9, pp. 12209
–
12216, 2021, doi
:
10.1109/
A
C
C
E
S
S
.2020.3048432.
[
12]
Z
.
G
uo,
C
.
W
a
ng,
G
.
Y
a
ng,
H
ua
ng,
a
nd
G
.
L
i
,
“
M
S
F
T
-
Y
O
L
O
:
i
m
pr
ove
d
Y
O
L
O
v5
ba
s
e
d
on
t
r
a
n
s
f
or
m
e
r
f
or
de
t
e
c
t
i
ng
de
f
e
c
t
s
of
s
t
e
e
l
s
ur
f
a
c
e
,”
Se
n
s
or
s
, vol
. 22, no. 9, p. 3467, M
a
y 2022, doi
:
10.3390/
s
22093
467.
[
13]
Y
.
J
.
C
ha
,
W
.
C
hoi
,
G
.
S
uh,
S
.
M
a
hm
oudkha
ni
,
a
nd
O
.
B
üyüköz
t
ür
k,
“
A
ut
o
nom
ous
s
t
r
uc
t
ur
a
l
vi
s
ua
l
i
ns
pe
c
t
i
on
us
i
ng
r
e
gi
on
-
ba
s
e
d
de
e
p
l
e
a
r
ni
ng
f
or
de
t
e
c
t
i
ng
m
ul
t
i
pl
e
da
m
a
ge
t
ype
s
,”
C
om
put
e
r
-
A
i
de
d
C
i
v
i
l
and
I
nf
r
as
t
r
uc
t
u
r
e
E
ngi
ne
e
r
i
ng
,
vol
.
33,
no.
9
,
pp. 731
–
747, S
e
p. 2018, doi
:
10.1111/
m
i
c
e
.12334.
[
14]
L
.
C
ui
,
X
.
J
i
a
ng,
M
.
X
u,
W
.
L
i
,
P
.
L
v,
a
nd
B
.
Z
hou,
“
S
D
D
N
e
t
:
a
f
a
s
t
a
nd
a
c
c
ur
a
t
e
ne
t
w
or
k
f
or
s
ur
f
a
c
e
de
f
e
c
t
d
e
t
e
c
t
i
on,”
I
E
E
E
T
r
ans
ac
t
i
ons
on I
ns
t
r
um
e
nt
at
i
on and M
e
as
ur
e
m
e
nt
, vol
. 70, pp. 1
–
13, 2021, do
i
:
10.1109/
T
I
M
.2021.3056744.
[
15]
R
.
S
us
t
i
ka
,
A
.
S
ube
kt
i
,
H
.
F
.
P
a
r
de
de
,
E
.
S
ur
ya
w
a
t
i
,
O
.
M
a
h
e
ndr
a
,
a
nd
S
.
Y
uw
a
na
,
“
E
va
l
ua
t
i
on
of
de
e
p
c
onvol
ut
i
ona
l
n
e
ur
a
l
ne
t
w
or
k
a
r
c
hi
t
e
c
t
ur
e
s
f
or
s
t
r
a
w
be
r
r
y
qua
l
i
t
y
i
n
s
pe
c
t
i
on,”
I
nt
e
r
nat
i
onal
J
our
n
al
of
E
ngi
ne
e
r
i
ng
and
T
e
c
hnol
ogy
(
U
A
E
)
,
vol
.
7,
no. 4, pp. 75
–
80, D
e
c
. 2018, doi
:
10.14419/
i
j
e
t
.v7i
4.40.24080.
[
16]
H
.
H
a
a
nd
J
.
J
e
ong,
“
C
N
N
-
ba
s
e
d
de
f
e
c
t
i
ns
pe
c
t
i
on
f
or
i
nj
e
c
t
i
on
m
ol
di
ng
us
i
ng
e
dge
c
om
put
i
ng
a
nd
i
ndus
t
r
i
a
l
I
oT
s
ys
t
e
m
s
,”
A
ppl
i
e
d Sc
i
e
nc
e
s
, vol
. 11, no. 14, p. 6378, J
ul
. 2021, doi
:
10.3390/
a
pp11146378.
[
17]
D
.
R
a
vi
r
a
j
s
i
nh,
“
C
a
s
t
i
ng
pr
oduc
t
i
m
a
ge
d
a
t
a
f
or
qua
l
i
t
y
i
ns
pe
c
t
i
on,”
K
a
ggl
e
.
A
c
c
e
s
s
e
d:
J
ul
.
01,
2025.
[
O
nl
i
ne
]
.
A
va
i
l
a
bl
e
:
ht
t
ps
:
/
/
w
w
w
.ka
ggl
e
.c
om
/
da
t
a
s
e
t
s
/
r
a
vi
r
a
j
s
i
nh45/
r
e
a
l
-
l
i
f
e
-
i
ndus
t
r
i
a
l
-
da
t
a
s
e
t
-
of
-
c
a
s
t
i
ng
-
pr
oduc
t
?
r
e
s
our
c
e
=dow
nl
oa
d
-
di
r
e
c
t
or
y
[
18]
D
.
E
ka
m
ba
r
a
m
a
nd
V
.
P
onnus
a
m
y,
“
I
de
nt
i
f
i
c
a
t
i
on
o
f
de
f
e
c
t
s
i
n
c
a
s
t
i
ng
pr
odu
c
t
s
by
us
i
ng
a
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k,”
I
E
I
E
T
r
ans
ac
t
i
ons
on Sm
a
r
t
P
r
oc
e
s
s
i
ng and C
om
put
i
ng
, vol
. 11, no. 3, pp. 149
–
155,
J
un. 2022, doi
:
10.5573/
I
E
I
E
S
P
C
.2022.11.3.149.
[
19]
M
.
F
.
A
l
f
a
r
i
z
i
,
M
.
F
a
t
c
ha
n,
a
nd
W
.
H
a
di
kr
i
s
t
a
nt
o,
“
C
om
pa
r
i
s
on
of
de
f
e
c
t
i
ve
c
a
s
t
i
ng
pr
oduc
t
c
l
a
s
s
i
f
i
c
a
t
i
on
r
e
s
ul
t
s
us
i
ng
t
he
k
-
ne
a
r
e
s
t
ne
i
ghbor
s
a
l
gor
i
t
hm
,”
I
nt
e
r
nat
i
onal
J
our
nal
of
A
pp
l
i
e
d
R
e
s
e
ar
c
h
and
Sus
t
ai
nabl
e
Sc
i
e
nc
e
s
,
vol
.
2,
no.
6,
pp.
417
–
426,
2024, doi
:
10.59890/
i
j
a
r
s
s
.v2i
6.1968.
[
20]
P
.
W
a
ng
a
nd
P
.
J
i
ng,
“
D
e
e
p
l
e
a
r
ni
ng
-
ba
s
e
d
m
e
t
hods
f
or
de
t
e
c
t
i
ng
de
f
e
c
t
s
i
n
c
a
s
t
i
r
on
pa
r
t
s
a
nd
s
ur
f
a
c
e
s
,”
I
E
T
I
m
age
P
r
oc
e
s
s
i
ng
,
vol
. 18, no. 1, pp. 47
–
58, J
a
n. 2024, doi
:
10.1049/
i
pr
2.12932.
[
21]
S
.
S
unda
r
a
m
a
nd
A
.
Z
e
i
d,
“
A
r
t
i
f
i
c
i
a
l
i
nt
e
l
l
i
ge
nc
e
-
ba
s
e
d
s
m
a
r
t
qua
l
i
t
y
i
ns
pe
c
t
i
on
f
or
m
a
nuf
a
c
t
ur
i
ng,”
M
i
c
r
om
ac
hi
ne
s
,
vol
.
14,
no. 3, p. 570, F
e
b. 2023, doi
:
10.3390/
m
i
14030570.
[
22]
H
.
H
u,
S
.
L
i
,
J
.
H
ua
ng,
B
.
L
i
u,
a
nd
C
.
C
he
,
“
C
a
s
t
i
ng
pr
oduc
t
i
m
a
ge
da
t
a
f
or
qua
l
i
t
y
i
ns
pe
c
t
i
on
w
i
t
h
X
c
e
pt
i
on
a
nd
da
t
a
a
ugm
e
nt
a
t
i
on,”
J
our
nal
of
T
he
or
y
and
P
r
ac
t
i
c
e
of
E
ngi
ne
e
r
i
ng
Sc
i
e
nc
e
,
vol
.
3,
no.
10,
pp.
42
–
46,
O
c
t
.
2023,
doi
:
10.53469/
j
t
pe
s
.2023.03(
10)
.06.
[
23]
L
.
R
oka
c
h,
“
E
ns
e
m
bl
e
-
ba
s
e
d
c
l
a
s
s
i
f
i
e
r
s
,”
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
R
e
v
i
e
w
,
vol
.
33,
no.
1
–
2,
pp.
1
–
39,
F
e
b.
2010,
doi
:
10.1007/
s
10462
-
009
-
9124
-
7.
[
24]
V
.
N
a
i
r
a
nd
G
.
E
.
H
i
nt
on,
“
R
e
c
t
i
f
i
e
d
l
i
ne
a
r
uni
t
s
i
m
pr
ove
r
e
s
t
r
i
c
t
e
d
bol
t
z
m
a
nn
m
a
c
hi
ne
s
,”
I
C
M
L
2010
-
P
r
oc
e
e
di
ngs
,
27t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
M
ac
hi
ne
L
e
a
r
ni
ng
,
pp.
807
–
814,
2010,
[
O
nl
i
ne
]
.
A
v
a
i
l
a
bl
e
:
ht
t
ps
:
/
/
w
w
w
.c
s
.t
or
ont
o.e
du/
~
f
r
i
t
z
/
a
bs
ps
/
r
e
l
uI
C
M
L
.pdf
[
25]
D
.
E
.
R
um
e
l
ha
r
t
,
G
.
E
.
H
i
nt
on,
a
nd
R
.
J
.
W
i
l
l
i
a
m
s
,
“
L
e
a
r
ni
ng
r
e
pr
e
s
e
nt
a
t
i
ons
by
ba
c
k
-
pr
opa
ga
t
i
ng
e
r
r
or
s
,”
N
at
ur
e
,
vol
.
323,
no. 6088, pp. 533
–
536, O
c
t
. 1986, doi
:
10.1038/
323533a
0.
[
26]
J
.
S
.
B
r
i
dl
e
,
“
P
r
oba
bi
l
i
s
t
i
c
i
nt
e
r
pr
e
t
a
t
i
on
of
f
e
e
df
or
w
a
r
d
c
l
a
s
s
i
f
i
c
a
t
i
on
ne
t
w
o
r
k
out
put
s
,
w
i
t
h
r
e
l
a
t
i
ons
hi
ps
t
o
s
t
a
t
i
s
t
i
c
a
l
pa
t
t
e
r
n
r
e
c
ogni
t
i
on,”
i
n
N
e
ur
oc
om
put
i
ng
,
B
e
r
l
i
n,
H
e
i
de
l
be
r
g:
S
pr
i
nge
r
B
e
r
l
i
n
H
e
i
de
l
be
r
g,
1990,
pp.
227
–
236.
doi
:
10.1007/
978
-
3
-
642
-
76153
-
9_28.
Evaluation Warning : The document was created with Spire.PDF for Python.