I
AE
S
I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
Ar
t
if
icial
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
Vol.
14,
No.
5:
Oc
tober
2025
,
pp.
4382
~
4389
I
S
S
N:
2252
-
8938
,
DO
I
:
10
.
11591/i
jai
.
v
14
.i
5
.
pp
43
82
-
4389
4382
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
ai
.
iaes
c
or
e
.
c
om
A
c
c
e
le
r
a
t
in
g sol
d
e
r
j
oi
n
t
c
la
ssi
f
ic
at
io
n
u
si
n
g ge
n
e
r
at
iv
e
a
r
t
i
f
ic
ia
l
in
t
e
ll
ig
e
n
c
e
f
or
d
at
a a
u
gm
e
n
t
at
io
n
T
e
n
g
Ye
ow
On
g
,
P
in
g
Chow
T
e
oh
,
Koon
T
at
t
T
an
S
c
hool
of
T
e
c
hnol
ogy a
nd E
ngi
ne
e
r
in
g S
c
ie
nc
e
, W
a
w
a
s
a
n
O
pe
n U
ni
ve
r
s
it
y,
P
e
na
ng
,
M
a
la
y
s
ia
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
De
c
31,
2024
R
e
vis
e
d
Aug
4,
2025
Ac
c
e
pted
S
e
p
7,
2025
D
es
p
i
t
e
ad
v
an
ceme
n
t
s
i
n
co
m
p
u
t
er
v
i
s
i
o
n
,
d
e
p
l
o
y
i
n
g
d
eep
l
earn
i
n
g
al
g
o
ri
t
h
m
s
fo
r
au
t
o
ma
t
ed
o
p
t
i
ca
l
i
n
s
p
ec
t
i
o
n
(
A
O
I)
i
n
p
ri
n
t
e
d
ci
rc
u
i
t
b
o
ar
d
(PCB)
man
u
fact
u
ri
n
g
remai
n
s
ch
a
l
l
e
n
g
i
n
g
d
u
e
t
o
t
h
e
n
e
ed
fo
r
l
arg
e,
d
i
v
ers
e
,
an
d
h
i
g
h
-
q
u
a
l
i
t
y
t
ra
i
n
i
n
g
d
a
t
as
e
t
s
.
A
O
I
p
r
o
g
ram
s
mu
s
t
b
e
d
e
v
el
o
p
e
d
q
u
i
ck
l
y
,
o
ft
e
n
as
s
o
o
n
as
t
h
e
fi
rs
t
PCB
i
s
as
s
em
b
l
e
d
,
t
o
meet
t
i
g
h
t
p
ro
d
u
c
t
i
o
n
t
i
me
l
i
n
es
.
H
o
w
ev
er,
d
eep
l
ear
n
i
n
g
mo
d
el
s
req
u
i
re
e
x
t
e
n
s
i
v
e
d
at
a
s
et
s
o
f
d
efect
i
ma
g
es
,
w
h
i
c
h
are
b
o
t
h
s
carce
an
d
t
i
me
-
c
o
n
s
u
mi
n
g
t
o
co
l
l
e
ct
.
A
s
a
res
u
l
t
,
A
O
I
s
o
f
t
w
are
d
ev
e
l
o
p
ers
fre
q
u
e
n
t
l
y
res
o
rt
t
o
t
rad
i
t
i
o
n
a
l
ru
l
e
-
b
as
e
d
met
h
o
d
s
.
T
h
i
s
s
t
u
d
y
i
n
t
ro
d
u
ce
s
a
n
o
v
el
frame
w
o
r
k
t
h
at
l
ev
era
g
es
g
e
n
era
t
i
v
e
A
I
an
d
d
i
s
cr
i
mi
n
at
i
v
e
A
I
t
o
ad
d
res
s
d
a
t
as
e
t
l
i
m
i
t
a
t
i
o
n
s
.
By
a
p
p
l
y
i
n
g
a
d
i
ff
u
s
i
o
n
mo
d
el
t
o
s
y
s
t
ema
t
i
ca
l
l
y
ad
d
an
d
remo
v
e
G
au
s
s
i
an
n
o
i
s
e,
t
h
e
framew
o
r
k
g
en
era
t
es
rea
l
i
s
t
i
c
d
efect
i
mag
e
s
,
ex
p
a
n
d
i
n
g
t
h
e
av
a
i
l
a
b
l
e
t
rai
n
i
n
g
d
a
t
a.
T
h
i
s
d
at
a
au
g
men
t
at
i
o
n
accel
era
t
es
t
h
e
l
earn
i
n
g
p
ro
ce
s
s
o
f
d
eep
l
earn
i
n
g
mo
d
el
s
,
en
h
an
c
i
n
g
t
h
ei
r
r
o
b
u
s
t
n
e
s
s
an
d
g
en
era
l
i
za
b
i
l
i
t
y
.
E
x
p
eri
me
n
t
a
l
res
u
l
t
s
d
emo
n
s
t
rat
e
t
h
at
t
h
i
s
ap
p
r
o
ach
i
mp
ro
v
es
A
O
I
s
y
s
t
e
m
p
erfo
rma
n
ce
b
y
p
ro
d
u
c
i
n
g
b
al
a
n
ced
d
a
t
as
e
t
s
acro
s
s
v
a
ri
o
u
s
d
efect
cl
a
s
s
e
s
.
T
h
e
framew
o
rk
s
h
o
r
t
en
s
t
ra
i
n
i
n
g
t
i
me
s
w
h
i
l
e
ma
i
n
t
ai
n
i
n
g
h
i
g
h
i
n
s
p
ec
t
i
o
n
accu
racy
,
fac
i
l
i
t
a
t
i
n
g
fa
s
t
er
d
e
p
l
o
y
me
n
t
o
f
A
O
I
s
y
s
t
e
ms
i
n
man
u
fact
u
ri
n
g
.
T
h
i
s
ad
v
an
ceme
n
t
en
h
an
ce
s
q
u
a
l
i
t
y
co
n
t
r
o
l
p
ro
ce
s
s
e
s
,
co
n
t
ri
b
u
t
i
n
g
t
o
mo
re
effi
ci
e
n
t
,
an
d
rel
i
a
b
l
e
ma
s
s
p
r
o
d
u
c
t
i
o
n
o
f
PCBs
.
K
e
y
w
o
r
d
s
:
Automate
d
opti
c
a
l
ins
pe
c
ti
on
C
onvolut
ional
ne
ur
a
l
ne
twor
ks
Da
tas
e
t
a
ugmenta
ti
on
Ge
ne
r
a
ti
ve
dif
f
us
ion
model
S
older
joi
nt
c
las
s
if
ica
ti
on
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
T
e
ng
Ye
ow
Ong
S
c
hool
of
T
e
c
hnology
a
nd
E
nginee
r
ing
S
c
ienc
e
,
W
a
wa
s
a
n
Ope
n
Unive
r
s
it
y
54,
J
a
lan
S
ult
a
n
Ahma
d
S
ha
h,
P
e
na
ng
-
10
0
50
,
M
a
lays
ia
E
mail:
tyong@wou
.
e
du.
my
1.
I
NT
RODU
C
T
I
ON
T
he
domain
of
c
omput
e
r
vis
ion,
pa
r
ti
c
ular
ly
in
mac
hine
ins
pe
c
ti
on
ha
s
e
xpe
r
ienc
e
d
s
igni
f
ica
nt
a
dva
nc
e
ments
with
the
a
dve
nt
of
de
e
p
lea
r
ning
a
lgor
it
hms
[
1
]
–
[
4
]
.
I
n
the
c
ontext
o
f
p
r
int
e
d
c
ir
c
u
it
boa
r
d
(
P
C
B
)
manuf
a
c
tur
ing
,
a
utom
a
ted
opt
ica
l
ins
pe
c
ti
o
n
(
AO
I
)
s
ys
tem
ha
s
be
c
ome
a
wide
ly
a
c
c
e
pted
s
ta
nda
r
d
f
o
r
vis
ion
-
ba
s
e
d
ins
pe
c
ti
on
f
ull
y
in
tegr
a
ted
int
o
the
manuf
a
c
tur
ing
li
ne
s
[
5]
,
[
6]
.
How
e
ve
r
,
the
vis
ion
s
ys
tem
c
onti
nue
s
to
f
a
c
e
c
ha
ll
e
nge
s
in
a
c
hieving
a
c
c
ur
a
te
ins
pe
c
ti
ons
in
r
e
a
l
-
wo
r
ld
indus
tr
ial
s
e
tt
ings
due
to
the
highl
y
s
pe
c
ular
na
tur
e
o
f
s
older
jo
int
s
ur
f
a
c
e
s
[
7
]
,
[
8]
.
T
o
a
dd
r
e
s
s
th
e
s
e
c
o
nc
e
r
ns
,
r
e
s
e
a
r
c
he
r
s
ha
v
e
pr
o
pos
e
d
n
um
e
r
o
us
in
no
va
ti
ve
m
a
c
h
i
ne
le
a
r
n
in
g
a
pp
r
oa
c
he
s
[
9
]
–
[
1
2
]
.
A
m
on
g
t
he
m
,
c
on
vo
lu
t
io
na
l
n
e
u
r
a
l
ne
tw
or
ks
(
C
NN
)
a
r
e
h
ig
h
ly
r
e
c
o
m
men
de
d
f
o
r
s
o
ld
e
r
j
o
in
t
de
f
e
c
t
d
e
te
c
t
io
n
tas
ks
du
e
to
t
he
i
r
e
f
f
ic
ie
nt
s
p
a
t
i
a
l
p
r
o
c
e
s
s
in
g
c
a
p
a
b
il
i
ti
e
s
[
1
3
]
–
[
21
]
.
De
s
pi
te
t
he
p
r
o
m
is
i
ng
d
e
v
e
l
op
me
nts
,
th
e
i
nd
us
t
r
ia
l
a
pp
l
ica
ti
on
o
f
C
NN
-
b
a
s
e
d
a
lg
o
r
i
th
ms
f
o
r
s
ol
de
r
jo
in
t
qua
l
it
y
i
ns
p
e
c
ti
on
i
n
A
OI
s
ys
t
e
ms
r
e
ma
ins
l
im
i
ted
.
T
he
p
r
i
ma
r
y
h
u
r
d
le
is
d
ue
t
o
de
f
e
c
t
iv
e
s
a
mp
les
a
r
e
s
i
gn
i
f
i
c
a
nt
ly
f
e
we
r
th
a
n
goo
d
one
s
,
l
e
a
di
ng
to
a
n
im
ba
la
nc
e
d
d
a
t
a
s
e
t
,
s
i
gn
i
f
i
c
a
n
t
ly
a
f
f
e
c
t
t
he
c
las
s
i
f
i
c
a
ti
on
a
c
c
u
r
a
c
y
[
22
]
–
[
2
5
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
A
c
c
e
ler
ati
ng
s
older
joi
nt
c
las
s
if
ication
us
ing
ge
ne
r
ati
v
e
ar
ti
fi
c
ial
int
e
ll
igenc
e
for
…
(
T
e
ng
Y
e
ow
Ong
)
4383
Nume
r
ous
s
tudi
e
s
ha
ve
be
e
n
pr
opos
e
d
to
mi
ti
ga
t
e
thi
s
is
s
ue
of
im
ba
lanc
e
d
da
tas
e
t
due
to
li
mi
ted
de
f
e
c
ti
ve
s
a
mpl
e
s
[
26]
–
[
28]
.
How
e
ve
r
,
r
is
k
of
bi
a
s
a
nd
ove
r
f
it
ti
ng
c
a
us
e
d
by
im
ba
lanc
e
d
da
ta
s
e
t
r
e
mains
unr
e
s
olved.
Ge
ne
r
a
ti
ve
a
r
ti
f
icia
l
int
e
ll
igenc
e
(
Ge
n
AI
)
ha
s
r
e
c
e
ntl
y
e
mer
ge
d
a
s
powe
r
f
ul
tool
a
c
r
os
s
va
r
ious
domains
f
or
c
ontent
c
r
e
a
ti
on.
How
e
ve
r
,
i
ts
a
ppli
c
a
ti
on
in
c
r
e
a
ti
ng
s
ynthetic
da
tas
e
ts
f
or
AO
I
model
tr
a
ini
ng
r
e
mains
unde
r
e
xplor
e
d.
T
e
c
hniques
s
uc
h
a
s
dif
f
u
s
ion
models
of
f
e
r
va
s
t
potential
f
o
r
a
ugmenting
tr
a
ini
ng
da
tas
e
t,
pa
r
ti
c
ular
ly
the
s
c
a
r
c
e
ly
a
va
il
a
ble
de
f
e
c
ti
ve
da
ta
[
29]
–
[
31]
.
T
his
s
tudy
a
im
s
to
tac
kle
the
c
ha
ll
e
nge
of
da
ta
im
ba
lanc
e
by
pr
opos
ing
a
nove
l
a
ppr
oa
c
h
ins
pir
e
d
by
the
dif
f
us
ion
pr
oc
e
s
s
us
e
d
in
Ge
n
AI
f
or
de
f
e
c
ti
ve
da
ta
a
ugmenta
ti
on.
T
he
method
invo
lves
pr
ogr
e
s
s
ively
a
dding
nois
e
to
or
igi
na
l
im
a
ge
s
(
f
or
wa
r
d
dif
f
us
io
n)
a
nd
then
de
nois
ing
them
(
r
e
ve
r
s
e
dif
f
us
ion)
to
ge
ne
r
a
te
s
ynthetic
im
a
ge
s
that
c
lo
s
e
ly
matc
h
the
de
s
ir
e
d
d
a
ta
dis
tr
ibut
ion.
T
his
a
ppr
oa
c
h
a
c
c
e
ler
a
tes
tr
a
ini
ng
pr
oc
e
s
s
without
s
a
c
r
if
icing
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y,
f
a
c
il
it
a
ti
ng
r
a
pid
model
de
ploym
e
nt
in
the
indus
tr
ial
a
li
g
ne
d
with
the
pr
a
c
ti
c
a
l
r
e
quir
e
ments
.
I
n
indus
tr
ial
P
C
B
a
s
s
e
mbl
y
manuf
a
c
tur
ing,
e
ns
ur
i
ng
the
e
f
f
e
c
ti
ve
ne
s
s
of
AO
I
f
or
s
older
joi
n
t
qua
li
ty
is
c
r
uc
ial
dur
ing
mas
s
pr
oduc
ti
on.
W
hil
e
C
NN
-
ba
s
e
d
a
lgor
it
hms
of
f
e
r
a
dva
nc
e
d
de
f
e
c
t
de
tec
ti
on
c
a
pa
bil
it
ies
,
their
de
ve
lopm
e
nt
r
e
quir
e
s
e
xtens
ive
a
nd
ba
lan
c
e
d
da
tas
e
ts
a
c
r
os
s
va
r
ious
de
f
e
c
t
c
a
tegor
ies
to
pe
r
f
or
m
r
e
li
a
bly.
Dur
ing
e
a
r
ly
AO
I
p
r
ogr
a
m
de
ve
lopm
e
nt
pha
s
e
,
de
f
e
c
ti
ve
s
a
mpl
e
s
a
r
e
of
ten
li
mi
ted
a
s
c
ompar
e
d
to
good
one
s
,
r
e
s
ult
ing
in
a
n
im
ba
lanc
e
d
da
tas
e
t.
T
r
a
ini
ng
C
NN
models
on
s
uc
h
da
tas
e
t
c
ould
c
ompr
omi
s
e
ge
ne
r
a
li
z
a
ti
on
a
nd
incr
e
a
s
e
the
r
is
ks
of
bias
a
nd
o
ve
r
f
it
ti
ng.
T
o
mee
t
the
s
tr
ingent
t
ime
-
to
-
mar
ke
t
d
e
mands
,
AO
I
pr
ogr
a
ms
of
ten
de
ve
loped
us
ing
r
ule
-
ba
s
e
d
a
ppr
oa
c
he
s
,
whic
h
do
not
r
e
quir
e
lar
ge
a
nd
ba
lanc
e
d
tr
a
ini
ng
da
tas
e
ts
but
s
a
c
r
if
ice
the
a
dva
nc
e
d
lea
r
nin
g
c
a
pa
bil
it
ies
of
de
e
p
lea
r
ning
a
lgo
r
it
hms
.
T
his
s
tudy
a
im
s
to
a
ddr
e
s
s
the
ke
y
c
ha
l
lenge
pos
e
d
by
the
s
c
a
r
c
it
y
of
de
f
e
c
ti
ve
s
a
mpl
e
da
ta.
T
his
r
e
s
e
a
r
c
h
int
e
nts
to
a
dd
r
e
s
s
the
f
oll
owing
ga
ps
:
‒
L
im
it
e
d
a
va
il
a
bil
it
y
of
de
f
e
c
t
da
ta
.
I
n
r
e
a
l
-
wor
ld
i
ndus
tr
ial
e
nvir
onments
,
pa
r
ti
c
ular
ly
in
P
C
B
a
s
s
e
mbl
y,
de
f
e
c
ti
ve
s
older
joi
nt
im
a
ge
s
a
r
e
s
c
a
r
c
e
.
De
e
p
le
a
r
ning
models
r
e
ly
on
lar
ge
,
we
ll
-
ba
lanc
e
d
da
tas
e
ts
,
whic
h
a
r
e
of
ten
una
va
il
a
ble,
lea
ding
to
bias
a
nd
ov
e
r
f
it
ti
ng.
‒
R
e
li
a
nc
e
on
r
ule
-
ba
s
e
d
a
ppr
oa
c
he
s
.
Due
to
the
li
mi
ted
a
va
il
a
bil
it
y
of
de
f
e
c
t
da
ta
,
AO
I
s
ys
tems
in
manuf
a
c
tur
ing
o
f
ten
de
pe
nd
on
tr
a
dit
ional
r
ule
-
ba
s
e
d
methods
ins
tea
d
of
mo
r
e
a
dva
nc
e
d
de
e
p
lea
r
ni
ng
-
ba
s
e
d
a
ppr
oa
c
he
s
.
T
he
s
e
r
ule
-
ba
s
e
d
s
y
s
tems
lac
k
f
lexibil
it
y
a
nd
r
obus
tnes
s
.
‒
Unde
r
e
xplor
e
d
a
ppli
c
a
ti
on
of
Ge
n
AI
f
or
da
ta
a
ugmenta
ti
on
.
Alt
hough
Ge
n
A
I
,
pa
r
ti
c
ula
r
ly
dif
f
us
ion
models
,
ha
ve
pr
ove
n
e
f
f
e
c
ti
ve
f
or
da
ta
a
ugmenta
ti
on
a
c
r
os
s
va
r
ious
domains
,
their
a
ppli
c
a
ti
ons
in
s
older
joi
nt
ins
pe
c
ti
on
r
e
mains
lar
ge
ly
une
xplor
e
d.
T
his
s
tudy
a
im
s
to
a
ddr
e
s
s
thi
s
g
a
p
by
de
mons
tr
a
ti
ng
the
potential
of
ge
ne
r
a
ti
ve
AI
in
c
r
e
a
ti
ng
s
ynthetic
de
f
e
c
t
im
a
ge
s
to
e
nha
nc
e
de
e
p
lea
r
ning
mo
de
l
pe
r
f
or
manc
e
.
2.
M
E
T
HO
D
T
he
pr
opos
e
d
tec
hnique
is
e
va
luate
d
th
r
ough
f
iv
e
s
e
que
nti
a
l
s
tage
s
,
a
s
de
picte
d
in
F
igur
e
1.
T
he
pr
oc
e
s
s
be
gins
with
da
ta
pr
e
pa
r
a
ti
on,
f
oll
owe
d
by
model
tr
a
ini
ng,
da
ta
a
ugmenta
ti
on
a
nd
da
ta
s
ynthe
ti
z
a
ti
on,
then
c
onc
ludes
with
pe
r
f
or
manc
e
e
va
luation.
E
a
c
h
of
thes
e
s
tage
s
will
be
de
s
c
r
ibed
in
de
tail
in
the
s
ubs
e
que
n
t
s
e
c
ti
ons
.
F
igur
e
1
.
S
tep
-
by
-
s
tep
e
va
luation
of
the
p
r
opos
e
d
t
e
c
hnique
2.
1.
Dat
as
e
t
p
r
e
p
ar
at
ion
(
s
t
e
p
1)
I
n
o
r
de
r
to
e
ns
ur
e
t
r
a
ns
pa
r
e
nc
y
a
nd
r
e
pr
oduc
ibi
li
t
y,
thi
s
s
tudy
uti
li
z
e
s
a
n
ope
n
-
s
our
c
e
publi
c
da
tas
e
t
obtaine
d
f
r
om
the
Ka
ggle
r
e
pos
it
or
y
a
t
htt
ps
:/
/
ww
w.
ka
ggle.
c
om/
da
tas
e
ts
.
I
n
thi
s
pa
pe
r
,
the
do
wnloa
de
d
da
t
a
s
e
t
is
r
e
f
e
r
r
e
d
a
s
“
Da
ta_O
r
igi
n_1”
.
I
t
c
ontains
323
i
mage
s
of
good
s
older
lea
ds
a
nd
30
im
a
ge
s
of
de
f
e
c
t
s
older
lea
ds
.
T
he
s
e
im
a
ge
s
a
r
e
s
e
gmente
d
r
e
gion
-
of
-
int
e
r
e
s
t
f
oc
us
on
the
s
older
joi
nt
lea
ds
of
s
ur
f
a
c
e
-
mount
de
vice
s
,
typi
c
a
ll
y
f
ound
in
pa
c
ka
ge
s
li
ke
s
mall
outl
ine
int
e
gr
a
ted
c
ir
c
uit
(
S
OI
C
)
,
s
mall
outl
ine
pa
c
ka
ge
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
5:
Oc
tober
2025
:
43
82
-
4389
4384
(
S
OP)
,
a
nd
qua
d
f
lat
pa
c
ka
ge
(
QFP
)
.
F
igur
e
2
pr
e
s
e
nts
s
a
mpl
e
im
a
ge
s
f
r
om
“
Da
ta_O
r
igi
n_1,
”
while
F
igur
e
3
pr
ovides
a
c
los
e
r
view
of
the
s
older
joi
nts
on
the
c
omponent
lea
ds
.
F
igur
e
2
.
E
xa
mpl
e
s
o
f
good
a
nd
de
f
e
c
t
i
mage
s
f
r
o
m
“
Da
ta_O
r
igi
n_1”
da
tas
e
t
F
igur
e
3
.
C
los
e
-
up
view
of
s
older
jo
int
s
a
t
lea
ds
T
o
c
ons
tr
uc
t
the
tes
ti
ng
da
tas
e
t,
r
e
f
e
r
r
e
d
a
s
“
Da
ta_T
e
s
t
”
,
thr
e
e
good
im
a
ge
s
a
nd
thr
e
e
de
f
e
c
t
im
a
ge
s
we
r
e
r
a
ndoml
y
s
a
mpl
e
d
f
r
om
the
“
Da
ta_O
r
igi
n_1
”
da
tas
e
t
downloa
de
d
f
r
om
the
r
e
pos
it
or
y.
T
he
r
e
maining
im
a
ge
s
,
c
ompr
is
ing
320
good
im
a
ge
s
a
nd
27
de
f
e
c
t
im
a
ge
s
,
a
r
e
c
a
ll
e
d
“
Da
ta_O
r
igi
n_2”
.
T
h
e
da
tas
e
t
pa
r
ti
ti
oning
i
s
i
ll
us
tr
a
ted
in
F
igur
e
4
.
F
igur
e
4
.
Dow
nloade
d
da
tas
e
t
is
divi
de
d
in
to
tr
a
in
i
ng
a
nd
tes
ti
ng
da
tas
e
ts
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
A
c
c
e
ler
ati
ng
s
older
joi
nt
c
las
s
if
ication
us
ing
ge
ne
r
ati
v
e
ar
ti
fi
c
ial
int
e
ll
igenc
e
for
…
(
T
e
ng
Y
e
ow
Ong
)
4385
2.
2.
CNN
m
od
e
l
t
r
ai
n
in
g
(
s
t
e
p
2)
Give
n
that
the
“
Da
ta_O
r
igi
n_2
”
da
tas
e
t
is
a
c
oll
e
c
ti
on
of
gr
a
ys
c
a
le
im
a
ge
s
with
dim
e
ns
ions
of
179×
38
pixels
,
a
s
im
ple
C
NN
-
ba
s
e
d
a
lgor
it
hm
w
it
h
a
li
ne
a
r
s
tr
uc
tur
e
wa
s
c
hos
e
n.
A
P
ython
s
c
r
ip
t,
us
ing
Ope
nC
V
a
nd
T
e
ns
or
F
low
l
ibr
a
r
ies
,
wa
s
d
e
ve
loped
f
or
im
a
ge
loading
a
nd
model
t
r
a
ini
ng.
T
he
“
Da
ta_O
r
igi
n_2”
da
tas
e
t
wa
s
us
e
d
f
or
t
r
a
ini
n
g.
A
f
ter
10
e
poc
hs
,
the
model
a
c
hieve
d
a
tes
t
a
c
c
ur
a
c
y
of
98.
57%
with
a
tes
t
los
s
of
2.
18
%
.
2.
3.
Dat
as
e
t
a
u
gm
e
n
t
a
t
ion
(
s
t
e
p
3)
M
a
ny
tec
hniques
we
r
e
r
e
po
r
ted
s
uit
a
ble
f
or
da
ta
a
ugmenta
ti
on
[
32]
–
[
34]
.
C
omm
on
a
ugmenta
ti
on
methods
a
r
e
r
otation,
s
c
a
li
ng,
f
li
pping
,
a
nd
c
r
opp
i
ng.
How
e
ve
r
,
a
s
“
Da
ta_O
r
igi
n_2
”
da
tas
e
t
c
ons
is
ts
of
only
s
e
gmente
d
im
a
ge
s
,
he
nc
e
the
mos
t
r
e
leva
nt
im
a
ge
a
ugmenta
ti
on
will
be
pixel
va
lue
modi
f
ica
ti
on.
T
o
a
ddr
e
s
s
the
im
ba
lanc
e
in
“
Da
ta_O
r
igi
n_2”
(
320
good
i
mage
s
ve
r
s
us
27
de
f
e
c
t
im
a
ge
s
)
,
da
ta
a
ugmenta
ti
on
wa
s
pe
r
f
or
med
e
xc
lus
ively
on
the
de
f
e
c
t
im
a
ge
s
.
F
ive
de
f
e
c
t
im
a
ge
s
we
r
e
r
a
ndoml
y
s
a
mpl
e
d
a
nd
their
s
ynthes
ize
d
c
ounter
pa
r
ts
we
r
e
ge
ne
r
a
ted
p
r
ogr
e
s
s
ivel
y
ba
s
e
d
on
a
di
f
f
us
ion
model.
T
he
s
ynthes
ize
d
im
a
ge
f
r
om
e
a
c
h
a
ugmenta
ti
on
loop
is
loade
d
to
the
tr
a
i
ne
d
C
NN
model
f
r
om
s
tep
2
to
e
va
luate
the
c
las
s
if
ica
ti
on
r
e
s
ult
s
.
S
ynthes
ize
d
im
a
ge
s
with
a
c
on
f
ident
leve
l
unde
r
95%
we
r
e
dis
c
a
r
de
d
.
F
igu
r
e
5
s
hows
or
ig
in
a
l
a
nd
s
ynthes
ize
d
im
a
ge
s
a
f
ter
a
ugmenta
ti
on
pr
oc
e
s
s
.
T
hr
ough
e
xpe
r
im
e
ntation,
it
wa
s
f
ound
that
a
pplyi
ng
Ga
us
s
ian
nois
e
with
a
mea
n
of
0
a
nd
a
s
tanda
r
d
d
e
viation
up
to
40
pr
e
s
e
r
ve
d
ke
y
i
mage
f
e
a
tur
e
s
,
a
c
hieving
c
onf
idenc
e
leve
ls
a
bove
95%
.
E
a
c
h
de
f
e
c
t
im
a
ge
wa
s
s
ynthes
ize
d
f
ive
ti
mes
t
o
ge
ne
r
a
te
in
a
to
tal
of
25
s
ynthes
ize
d
im
a
ge
s
.
T
o
pr
e
s
e
r
ve
the
c
ons
is
tenc
y
of
r
a
ti
o
be
twe
e
n
g
ood
a
nd
de
f
e
c
t
im
a
ge
s
,
thr
e
e
s
ynthes
ize
d
im
a
ge
s
we
r
e
de
li
be
r
a
tely
e
xc
luded
.
T
he
f
inal
s
e
t
of
22
s
ynthes
ize
d
im
a
ge
s
,
a
long
with
the
or
igi
na
l
5
de
f
e
c
t
a
nd
3
20
good
im
a
ge
s
f
r
om
“
Da
ta_O
r
igi
n_2”
,
c
ons
ti
tut
e
d
a
ne
w
da
tas
e
t
labe
ll
e
d
“
Da
ta_Syn_A”
.
T
he
c
ompos
it
ion
of
thi
s
da
tas
e
t
is
s
umm
a
r
ize
d
in
F
igur
e
6
.
F
igur
e
5
.
Or
igi
na
l
im
a
ge
is
a
ugmente
d
to
f
ive
s
ynthes
ize
d
im
a
ge
s
F
igur
e
6
.
F
or
mi
ng
a
ne
w
da
tas
e
t
f
r
om
or
igi
na
l
a
nd
s
ynthes
ize
d
im
a
ge
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
5:
Oc
tober
2025
:
43
82
-
4389
4386
2.
4.
S
yn
t
h
e
s
ize
d
d
at
as
e
t
t
r
ain
in
g
(
s
t
e
p
4)
I
n
thi
s
s
tep,
t
he
C
NN
model
wa
s
tr
a
ined
us
ing
th
e
“
Da
ta_Syn_A
”
da
tas
e
t
.
T
o
e
ns
ur
e
r
e
pe
a
tabili
ty
a
nd
r
e
pr
oduc
ibi
li
ty
,
the
s
a
me
tr
a
ini
ng
pr
oc
e
dur
e
de
s
c
r
ib
ed
in
s
tep
2
wa
s
a
ppli
e
d
.
Upon
c
omp
leti
on
of
10
e
poc
hs
,
the
model
a
c
hiev
e
d
a
tes
t
a
c
c
ur
a
c
y
of
1
00%
with
a
c
or
r
e
s
ponding
tes
t
los
s
of
1.
60
%
.
2.
5.
P
e
r
f
or
m
an
c
e
e
valu
at
io
n
(
s
t
e
p
5)
T
o
c
ompar
e
the
c
las
s
if
ica
ti
on
pe
r
f
or
manc
e
,
b
oth
models
tr
a
ined
on
“
Da
ta_O
r
ig
in_2”
a
nd
“
Da
ta_Syn_A”
we
r
e
tes
ted
us
ing
a
s
ynthe
s
iz
e
d
tes
ti
ng
da
tas
e
t,
“
Da
ta_T
e
s
t_S
yn,
”
c
r
e
a
ted
thr
ough
a
ugmenta
ti
on
of
the
o
r
igi
na
l
“
Da
ta_T
e
s
t”
.
T
he
a
ugmente
d
“
Da
ta_T
e
s
t_S
yn”
da
tas
e
t
c
o
mpr
is
e
d
15
s
ynthes
ize
d
im
a
ge
s
e
a
c
h
f
or
the
good
a
nd
de
f
e
c
t
c
las
s
e
s
,
in
a
ddit
ion
to
the
or
igi
na
l
thr
e
e
good
a
nd
thr
e
e
de
f
e
c
t
im
a
ge
s
.
F
igur
e
7
il
lus
tr
a
tes
thi
s
a
ugmenta
ti
on
pr
oc
e
s
s
.
F
igur
e
7
.
Da
ta
a
ugmenta
ti
on
o
f
the
tes
ti
ng
da
tas
e
t
T
he
C
NN
model
ope
r
a
tes
unde
r
a
two
-
c
las
s
c
las
s
if
ica
ti
on
f
r
a
mew
or
k
,
c
a
tegor
izing
pr
e
dictions
a
s
pos
it
ive
(
P
)
or
ne
ga
ti
ve
(
N)
.
C
las
s
if
ica
ti
on
outcome
s
include
:
i
)
tr
ue
pos
it
ive
(
T
P
)
the
model
c
or
r
e
c
tl
y
pr
e
dicts
P
f
or
a
n
a
c
tual
P
s
a
mpl
e
;
ii
)
f
a
ls
e
pos
it
i
ve
(
F
P
)
the
model
incor
r
e
c
tl
y
pr
e
dicts
P
f
o
r
a
n
a
c
tual
N
s
a
mpl
e
;
i
ii
)
tr
ue
n
e
g
a
t
iv
e
(
T
N)
the
mo
de
l
c
o
r
r
e
c
t
ly
p
r
e
d
ic
ts
N
f
o
r
a
n
a
c
t
ua
l
N
s
a
m
p
le
;
a
nd
iv
)
f
a
ls
e
ne
ga
t
ive
(
F
N
)
the
m
od
e
l
i
nc
o
r
r
e
c
t
ly
p
r
e
di
c
ts
N
f
or
a
n
a
c
tua
l
P
s
a
m
ple
.
P
e
r
f
or
manc
e
wa
s
e
va
luate
d
us
ing
a
c
c
ur
a
c
y,
pr
e
c
is
ion
a
nd
r
e
c
a
ll
de
f
ined
a
s
f
oll
ows
:
‒
Ac
c
ur
a
c
y:
pr
opor
ti
on
o
f
c
o
r
r
e
c
tl
y
p
r
e
dicte
d
s
a
mpl
e
s
a
mong
a
ll
s
a
mpl
e
s
.
=
(
+
)
(
+
+
+
)
(
1)
‒
P
r
e
c
is
ion:
pr
opor
ti
on
of
TP
pr
e
dictions
a
mong
a
ll
pos
it
ive
pr
e
dictions
.
=
(
+
)
(
2)
‒
R
e
c
a
ll
:
pr
opor
ti
on
o
f
TP
p
r
e
dictions
a
mong
a
ll
a
c
tual
pos
it
ives
.
=
(
+
)
(
3)
3.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
T
a
bl
e
1
pr
e
s
e
nts
th
e
pe
r
f
or
ma
nc
e
o
f
C
N
N
m
ode
ls
t
r
a
i
ne
d
o
n
d
if
f
e
r
e
n
t
d
a
t
a
s
e
ts
.
A
lt
ho
ug
h
t
he
r
e
s
ul
ts
r
e
ve
a
l
o
nl
y
m
in
o
r
d
if
f
e
r
e
n
c
e
s
o
ve
r
a
ll
,
th
e
mo
de
l
t
r
a
i
ne
d
on
“
Da
ta
_S
yn
_A
”
de
mo
ns
t
r
a
tes
a
no
t
ice
a
bl
e
b
ias
t
o
wa
r
ds
t
he
g
oo
d
c
l
a
s
s
.
S
pe
c
i
f
i
c
a
ll
y
,
2
ou
t
o
f
1
8
de
f
e
c
t
im
a
ge
s
w
e
r
e
m
is
c
las
s
i
f
ie
d
a
s
g
oo
d
,
wh
il
e
a
l
l
1
8
g
oo
d
i
m
a
ge
s
we
r
e
c
o
r
r
e
c
tl
y
c
las
s
i
f
ie
d
w
it
h
a
ve
r
y
hi
gh
c
o
n
f
i
de
nc
e
le
ve
l
.
T
h
is
b
ias
i
s
l
ike
l
y
a
t
t
r
i
bu
te
d
t
o
t
he
d
a
t
a
s
e
t
’
s
i
m
ba
lan
c
e
,
w
he
r
e
t
he
r
a
ti
o
o
f
go
od
to
de
f
e
c
t
im
a
ge
s
i
s
1
2:
1
.
T
o
ve
r
i
f
y
whe
ther
the
is
s
ue
is
c
a
us
e
d
by
da
tas
e
t
i
mbala
nc
e
,
de
f
e
c
t
i
mage
s
in
“
Da
ta_O
r
igi
n_2
”
we
r
e
s
ynthes
ize
d
f
ive
ti
mes
us
ing
the
pr
oc
e
dur
e
s
im
il
a
r
to
that
de
s
c
r
ibed
in
s
tep
3.
T
his
p
r
oc
e
s
s
r
e
s
ult
e
d
in
a
n
e
xpa
nde
d
da
tas
e
t
r
e
f
e
r
r
e
d
to
a
s
“
Da
ta_Syn_B
”
.
F
igur
e
8
il
lus
tr
a
tes
the
d
a
tas
e
t
e
xpa
ns
ion
pr
oc
e
s
s
i
n
de
tail.
F
oll
owing
the
da
ta
a
ugmenta
ti
on,
the
r
a
ti
o
of
good
to
de
f
e
c
t
im
a
ge
s
wa
s
a
djus
ted
to
2:1.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
A
c
c
e
ler
ati
ng
s
older
joi
nt
c
las
s
if
ication
us
ing
ge
ne
r
ati
v
e
ar
ti
fi
c
ial
int
e
ll
igenc
e
for
…
(
T
e
ng
Y
e
ow
Ong
)
4387
T
a
ble
1
.
C
ompar
ing
pe
r
f
or
manc
e
of
“
Da
ta_O
r
igi
n
_2”
a
ga
ins
t
“
Da
ta_Syn_A”
da
tas
e
t
A
c
c
ur
a
c
y
(
%)
P
r
e
c
is
io
n
(%)
R
e
c
a
ll
(%)
M
ode
l
tr
a
in
e
d by “
D
a
ta
_O
r
ig
in
_2”
100
100
100
M
ode
l
tr
a
in
e
d by “
D
a
ta
_S
yn_A”
94.4
100
90
F
igur
e
8
.
Da
ta
a
ugmenta
ti
on
o
f
de
f
e
c
t
im
a
ge
s
to
b
a
lanc
e
the
da
tas
e
t
T
he
“
Da
ta_Syn_B
”
da
tas
e
t
wa
s
then
loade
d
int
o
the
s
a
me
C
NN
model
f
or
tr
a
ini
ng
a
nd
e
va
luate
d
us
ing
the
s
a
me
tes
t
im
a
ge
da
tas
e
t,
“
Da
ta_T
e
s
t_S
yn
”
.
T
he
r
e
s
ult
s
indi
c
a
te
that
a
c
c
ur
a
c
y,
pr
e
c
is
ion
a
nd
r
e
c
a
ll
indi
c
a
tor
s
a
c
hieve
d
ha
ving
the
leve
l
e
quivale
nt
to
thos
e
with
“
Da
ta_O
r
i
gin_2
”
.
I
n
a
ddit
ion
,
the
a
ve
r
a
ge
c
las
s
if
ica
ti
on
c
onf
idenc
e
leve
l
s
howe
d
a
no
table
im
pr
ove
ment
.
T
a
ble
2
pr
ovides
a
s
umm
a
r
y
of
the
pe
r
f
or
manc
e
metr
ics
o
f
“
Da
ta_Syn_B
”
c
ompar
e
d
t
o
thos
e
of
“
Da
ta_O
r
igi
n_2
”
.
T
a
ble
2
.
C
ompar
is
on
of
“
Da
ta_O
r
igi
n_2”
a
ga
ins
t
“
Da
t
a
_S
yn_B
”
da
tas
e
t
s
A
c
c
ur
a
c
y
(%)
P
r
e
c
is
io
n
(%)
R
e
c
a
ll
(%)
A
ve
r
a
ge
c
onf
id
e
nc
e
l
e
ve
l
(%)
M
ode
l
tr
a
in
e
d by “
D
a
ta
_O
r
ig
in
_2”
100
100
100
96.33
M
ode
l
tr
a
in
e
d by “
D
a
ta
_S
yn_B
”
100
100
100
99.99
4.
CONC
L
USI
ON
T
his
s
tudy
p
r
e
s
e
nts
a
n
innovative
f
r
a
mew
or
k
lev
e
r
a
ging
ge
ne
r
a
ti
ve
AI
,
s
pe
c
if
ica
ll
y
the
di
f
f
us
ion
model,
to
a
dd
r
e
s
s
the
inhe
r
e
nt
c
ha
ll
e
nge
of
li
mi
t
e
d
de
f
e
c
t
im
a
ge
da
ta
in
AOI
s
ys
tems
.
B
y
a
ugmenting
the
da
tas
e
t
with
s
ynthe
s
ize
d
im
a
ge
s
,
the
pr
opos
e
d
a
pp
r
oa
c
h
e
nha
nc
e
s
the
pe
r
f
or
manc
e
of
C
NN
-
b
a
s
e
d
c
l
a
s
s
if
ier
s
,
yielding
a
mor
e
ba
lanc
e
d
a
nd
r
e
p
r
e
s
e
ntative
tr
a
ini
ng
s
e
t.
T
he
di
f
f
us
ion
model
e
na
bles
pr
ogr
e
s
s
ive
im
a
ge
s
ynthes
is
thr
ough
the
a
ddit
ion
a
nd
r
e
moval
o
f
Ga
us
s
ian
nois
e
,
ther
e
by
e
nr
iching
the
da
tas
e
t
while
mai
ntaining
c
ha
r
a
c
ter
is
ti
c
s
to
the
or
igi
na
l
de
f
e
c
t
di
s
tr
ibut
ion.
T
his
pr
oc
e
s
s
e
ns
ur
e
s
the
pr
e
s
e
r
va
ti
on
o
f
c
r
it
ica
l
vis
ua
l
f
e
a
tur
e
s
,
whic
h
is
e
s
s
e
nti
a
l
f
or
a
c
hieving
r
obus
t
c
las
s
if
ica
ti
on
pe
r
f
or
manc
e
.
T
he
e
xpe
r
im
e
ntal
outcome
s
va
li
da
te
the
e
f
f
ica
c
y
of
thi
s
ge
n
e
r
a
ti
v
e
a
ppr
oa
c
h,
de
mons
tr
a
ti
ng
c
las
s
if
ica
ti
on
r
e
s
ult
s
that
a
r
e
c
ompar
a
ble
to
models
tr
a
ined
on
r
e
a
l
-
wor
ld
de
f
e
c
t
da
ta.
T
he
methodology
s
uppor
ts
r
a
pid
da
tas
e
t
pr
e
pa
r
a
ti
on,
f
a
c
il
it
a
ti
ng
the
de
ploym
e
nt
of
C
NN
-
ba
s
e
d
model
s
in
indus
tr
ial
s
e
tt
ings
whe
r
e
r
e
duc
e
d
time
-
to
-
ma
r
ke
t
is
a
s
tr
a
tegic
pr
ior
it
y.
M
or
e
ove
r
,
it
unde
r
s
c
or
e
s
the
c
ompl
e
menta
r
y
r
ole
of
ge
ne
r
a
ti
ve
lea
r
ning
in
a
ugmenting
tr
a
dit
ional
dis
c
r
im
inative
methods
,
e
s
pe
c
ially
dur
ing
the
e
a
r
ly
pha
s
e
s
of
mas
s
pr
oduc
ti
on
whe
n
de
f
e
c
t
s
a
mpl
e
s
a
r
e
s
c
a
r
c
e
.
T
his
s
tu
dy
us
e
d
a
manua
l
a
n
d
it
e
r
a
ti
ve
a
ppr
oa
c
h
f
or
im
a
ge
s
ynthes
is
,
whic
h
i
nhe
r
e
ntl
y
li
mi
ted
the
e
xplor
a
ti
on
o
f
diver
s
e
nois
e
models
a
nd
a
ugmenta
ti
on
tec
hniques
.
T
he
r
e
is
a
n
e
e
d
f
or
c
ompr
e
he
ns
ive
s
of
twa
r
e
tool
s
c
a
pa
ble
of
s
uppor
ti
ng
va
r
ious
nois
e
modeling
tec
hniq
ue
s
,
a
ut
omating
s
im
ulation
pr
oc
e
s
s
e
s
a
nd
opti
mi
z
ing
the
s
e
lec
ti
on
of
nois
e
types
a
nd
int
e
ns
it
ies
.
De
ve
lopi
ng
s
uc
h
a
dva
nc
e
d
s
of
twa
r
e
s
olut
ion
in
the
f
utur
e
will
be
c
r
uc
ial
to
s
tr
e
a
ml
ine
a
nd
e
nha
nc
e
da
tas
e
t
a
ugmenta
ti
on.
M
e
a
nwhile,
thi
s
pa
pe
r
f
oc
us
e
s
e
xc
lus
i
ve
ly
on
a
s
ingl
e
type
of
s
older
ing
de
f
e
c
t.
I
n
p
r
a
c
ti
c
a
l,
a
wide
r
s
pe
c
tr
um
of
s
older
ing
-
r
e
late
d
a
nomalies
,
s
uc
h
a
s
s
older
br
idgi
n
g,
nonwe
tt
ing
,
e
xc
e
s
s
ive
s
older
,
a
nd
dis
tur
be
d
s
old
e
r
joi
nts
a
r
e
c
omm
only
e
nc
ounter
e
d
but
not
a
ddr
e
s
s
e
d
i
n
the
pr
e
s
e
nt
s
tudy.
F
u
tur
e
r
e
s
e
a
r
c
h
s
hould
e
xtend
the
f
r
a
mew
or
k
to
include
thes
e
a
ddit
ional
de
f
e
c
t
typ
e
s
,
ther
e
by
im
p
r
oving
the
ge
ne
r
a
li
z
a
bil
it
y
a
nd
i
ndus
tr
ial
r
e
leva
nc
e
of
the
f
indi
ngs
.
L
a
s
tl
y,
s
c
a
li
ng
the
da
t
a
s
e
t
s
ize
is
r
e
c
omm
e
nde
d
to
r
e
inf
o
r
c
e
the
r
obus
t
ne
s
s
a
nd
r
e
li
a
bil
it
y
of
the
model
pe
r
f
o
r
manc
e
s
.
F
UN
DI
NG
I
NF
ORM
AT
I
ON
T
his
r
e
s
e
a
r
c
h
is
s
uppor
ted
by
W
a
wa
s
a
n
Ope
n
Uni
ve
r
s
it
y
f
or
the
C
e
ntr
e
f
or
R
e
s
e
a
r
c
h
a
nd
I
nnova
ti
on
I
nc
e
nti
ve
Gr
a
nt
S
c
he
me
with
P
r
ojec
t
C
ode
:
W
OU
/
C
e
R
I
/2025(
0079)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
5:
Oc
tober
2025
:
43
82
-
4389
4388
AU
T
HO
R
CONT
RI
B
U
T
I
ONS
S
T
AT
E
M
E
N
T
T
his
jour
na
l
us
e
s
the
C
ontr
ibut
o
r
R
oles
T
a
xo
nomy
(
C
R
e
diT
)
to
r
e
c
ognize
indi
vidual
a
uthor
c
ontr
ibut
ions
,
r
e
duc
e
a
utho
r
s
hip
dis
putes
,
a
nd
f
a
c
il
it
a
te
c
oll
a
bor
a
ti
on.
Nam
e
of
Au
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
T
e
ng
Ye
ow
Ong
✓
✓
✓
✓
✓
✓
✓
✓
✓
P
ing
C
how
T
e
oh
✓
✓
✓
✓
✓
✓
Koon
T
a
tt
T
a
n
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
li
z
a
ti
on
M
:
M
e
th
odol
ogy
So
:
So
f
twa
r
e
Va
:
Va
li
da
ti
on
Fo
:
Fo
r
ma
l
a
na
ly
s
i
s
I
:
I
nve
s
ti
ga
ti
on
R
:
R
e
s
our
c
e
s
D
:
D
a
ta
C
ur
a
ti
on
O
:
W
r
it
in
g
-
O
r
ig
in
a
l
D
r
a
f
t
E
:
W
r
it
in
g
-
R
e
vi
e
w
&
E
di
ti
ng
Vi
:
Vi
s
ua
li
z
a
ti
on
Su
:
Su
pe
r
vi
s
io
n
P
:
P
r
oj
e
c
t
a
dmi
ni
s
tr
a
ti
on
Fu
:
Fu
ndi
ng
a
c
qui
s
it
io
n
CONF
L
I
CT
OF
I
NT
E
RE
S
T
S
T
AT
E
M
E
N
T
T
he
a
uthor
s
de
c
lar
e
that
ther
e
is
no
c
onf
li
c
t
of
int
e
r
e
s
t
r
e
ga
r
ding
the
publi
c
a
ti
on
o
f
thi
s
manus
c
r
ipt
.
DA
T
A
AV
AI
L
A
B
I
L
I
T
Y
T
he
da
ta
that
s
uppor
t
the
f
indi
ngs
of
thi
s
s
tudy
a
r
e
ope
nly
a
va
il
a
ble
da
tas
e
t
in
the
Ka
ggle
r
e
po
s
it
or
y
a
t
htt
ps
:/
/www
.
ka
ggle.
c
om/
da
tas
e
ts
unde
r
the
na
m
e
“
lea
d
-
legs
-
on
-
c
hips
e
t”
.
RE
F
E
RE
NC
E
S
[
1]
A
.
P
r
a
ka
s
h
a
nd
S
.
C
h
a
uha
n,
“
A
c
ompr
e
he
n
s
iv
e
s
ur
ve
y
of
tr
e
n
di
ng
to
ol
s
a
nd
te
c
hni
que
s
in
de
e
p
le
a
r
ni
ng,”
in
2023
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on Dis
r
upt
i
v
e
T
e
c
hnol
ogi
e
s
, I
E
E
E
, M
a
y 2023, pp.
289
–
292
, doi
:
10.1109/I
C
D
T
57929.2023.10151083.
[
2]
N
.
H
üt
te
n,
M
.
A
.
G
ome
s
,
F
.
H
öl
ke
n,
K
.
A
nd
r
ic
e
vi
c
,
R
.
M
e
ye
s
,
a
nd
T
.
M
e
is
e
n,
“
D
e
e
p
le
a
r
ni
ng
f
or
a
ut
oma
te
d
vi
s
ua
l
in
s
pe
c
ti
on
in
ma
nuf
a
c
tu
r
in
g
a
nd
ma
in
te
na
nc
e
:
a
s
ur
ve
y
of
ope
n
-
a
c
c
e
s
s
pa
pe
r
s
,”
A
ppl
ie
d
Sy
s
te
m
I
nnov
at
io
n
,
vo
l.
7,
no.
1,
J
a
n.
2024,
doi
:
10.3390/a
s
i7
010011.
[
3]
P
.
K
lc
o,
D
.
K
oni
a
r
,
L
.
H
a
r
ga
s
,
K
.
P
.
D
im
ova
,
a
nd
M
.
C
hna
pko,
“
Q
ua
li
ty
in
s
pe
c
ti
on
of
s
pe
c
if
ic
e
le
c
tr
oni
c
boa
r
ds
by
de
e
p
ne
ur
a
l
ne
twor
ks
,”
Sc
ie
nt
if
ic
R
e
por
ts
, vol
. 13, no. 1, Nov. 202
3, doi:
10
.1038/s
4159
8
-
023
-
47958
-
0.
[
4]
M
.
D
ol
a
n
d
A
. G
e
e
t
h
a
,
“
A
l
e
a
r
ni
ng
tr
a
n
s
i
ti
on f
r
o
m m
a
c
hi
n
e
l
e
a
r
ni
ng
to
d
e
e
p l
e
a
r
n
in
g:
a
s
ur
v
e
y
,”
in
20
21
I
nt
e
r
nat
io
na
l
C
o
nf
e
r
e
n
c
e
on
E
m
e
r
gi
ng
T
e
c
hn
iq
u
e
s
i
n
C
om
pu
ta
ti
on
al
I
n
te
ll
i
g
e
n
c
e
,
I
E
E
E
,
A
u
g.
2
02
1,
p
p.
8
9
–
94
,
doi
:
10
.1
10
9/
I
C
E
T
C
I
51
97
3.
2
0
21
.9
57
40
66
.
[
5]
A
.
A
.
R
. M
.
A
.
E
ba
yye
h
a
nd
A
.
M
ou
s
a
vi
,
“
A
r
e
vi
e
w
a
nd a
na
ly
s
is
of
a
ut
oma
ti
c
opt
ic
a
l
in
s
pe
c
ti
on a
nd
qua
li
ty
moni
to
r
in
g
me
th
ods
in
e
le
c
tr
oni
c
s
i
ndus
tr
y,”
I
E
E
E
A
c
c
e
s
s
, vol
. 8, pp. 183192
–
183271, 2020, doi:
10.1109/AC
C
E
S
S
.2020.3029127.
[
6]
M
.
M
ic
ha
l
s
ka
,
“
O
ve
r
vi
e
w
of
A
O
I
u
s
e
in
s
ur
f
a
c
e
-
mount
te
c
hno
lo
gy
c
ont
r
ol
,”
I
nf
or
m
at
y
k
a,
A
ut
om
at
y
k
a,
P
om
ia
r
y
w
G
o
s
podar
c
e
i
O
c
hr
oni
e
Śr
odo
w
is
k
a
, vol
. 10, no. 4, pp. 61
–
64, 2020.
[
7]
W
.
D
a
i,
A
.
M
uj
e
e
b,
M
.
E
r
dt
,
a
nd
A
.
S
our
in
,
“
S
ol
de
r
in
g
de
f
e
c
t
de
te
c
ti
on
in
a
ut
oma
t
ic
opt
ic
a
l
in
s
pe
c
ti
on,”
A
dv
anc
e
d
E
ngi
ne
e
r
in
g
I
nf
or
m
at
ic
s
, vol
. 43, J
a
n. 2020, doi:
10.1016/j
.a
e
i.
2019.101004.
[
8]
V
.
R
e
s
ha
d
a
t
a
nd
R
.
A
.
J
.
W
.
K
a
pt
e
ij
ns
,
“
I
mpr
ovi
ng
th
e
pe
r
f
or
ma
nc
e
of
a
ut
oma
te
d
opt
ic
a
l
in
s
pe
c
ti
on
(
AOI
)
us
in
g
ma
c
h
in
e
le
a
r
ni
ng
c
la
s
s
if
ie
r
s
,”
i
n
2021
I
nt
e
r
nat
io
nal
C
onf
e
r
e
n
c
e
on
D
at
a
and
Sof
tw
ar
e
E
ngi
ne
e
r
in
g
,
I
E
E
E
,
N
ov.
2021,
pp.
1
–
5
,
doi
:
10.1109/I
C
oD
S
E
53690.2021.9648445.
[
9]
I.
-
C
.
C
he
n,
R
.
-
C
.
H
w
a
ng,
a
nd
H
.
-
C
.
H
ua
ng,
“
P
C
B
de
f
e
c
t
de
te
c
ti
on
ba
s
e
d
on
de
e
p
le
a
r
ni
ng
a
lg
or
it
hm,”
P
r
oc
e
s
s
e
s
,
vol
.
11,
n
o.
3,
M
a
r
. 2023, doi:
10.3390/pr
11030775.
[
10]
G
.
L
a
ks
hmi
,
V
.
U
.
S
a
nka
r
,
a
nd
Y
.
S
.
S
a
nka
r
,
“
A
s
ur
ve
y
of
P
C
B
de
f
e
c
t
d
e
te
c
ti
on
a
lg
or
it
hms
,”
J
our
nal
of
E
le
c
tr
oni
c
T
e
s
t
in
g
,
vol
. 39, no. 5
–
6, pp. 541
–
554, De
c
. 2023, doi:
10.1007/s
10836
-
023
-
06091
-
6.
[
11]
Z
.
Z
h
a
ng,
W
.
Z
ha
ng,
D
.
Z
hu,
Y
.
X
u,
a
nd
C
.
Z
hou,
“
P
r
in
te
d
c
ir
c
ui
t
boa
r
d
s
ol
de
r
jo
in
t
qua
li
ty
in
s
pe
c
ti
on
ba
s
e
d
on
li
ght
w
e
i
ght
c
la
s
s
if
ic
a
ti
on ne
twor
k,”
I
E
T
C
y
be
r
-
Sy
s
te
m
s
and R
obot
ic
s
, vol
.
5, no. 4, De
c
. 2023, doi:
10.1049/cs
y2.12101.
[
12]
F
.
U
lg
e
r
,
S
.
E
.
Y
uks
e
l,
a
nd
A
.
Y
il
ma
z
,
“
A
noma
ly
de
te
c
ti
on
f
or
s
ol
de
r
jo
in
ts
us
in
g
β
-
va
e
,”
I
E
E
E
T
r
ans
ac
ti
ons
on
C
om
pon
e
nt
s
,
P
ac
k
agi
ng and M
anuf
ac
tu
r
in
g T
e
c
hnol
ogy
, vol
. 11, no. 12, pp. 2214
–
2221, De
c
. 2021, doi:
10.1109/T
C
P
M
T
.2021.3121265.
[
13]
M
.
B
.
A
kht
a
r
,
“
T
he
u
s
e
of
a
c
onvolu
ti
ona
l
ne
ur
a
l
n
e
twor
k
in
de
te
c
ti
ng
s
ol
de
r
in
g
f
a
ul
t
s
f
r
om
a
pr
in
te
d
c
ir
c
ui
t
boa
r
d
a
s
s
e
mb
ly
,”
H
ig
hT
e
c
h and I
nnov
at
io
n J
our
nal
, vol
. 3, no. 1, pp. 1
–
14, M
a
r
.
2022, doi:
10.28991/HI
J
-
2022
-
03
-
01
-
01.
[
14]
Y
.
T
ia
n,
“
A
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
im
a
g
e
r
e
c
ogni
ti
on
me
th
o
d
b
a
s
e
d
on
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
a
lg
or
it
hm,”
I
E
E
E
A
c
c
e
s
s
,
vol
. 8, pp. 125731
–
125744, 2020, doi:
10.1109/AC
C
E
S
S
.2020.
3006097.
[
15]
Y.
-
G
.
K
im
a
nd
T
.
-
H
.
P
a
r
k,
“
S
M
T
a
s
s
e
mbl
y
in
s
pe
c
ti
on
u
s
in
g
dua
l
-
s
tr
e
a
m
c
onvolut
io
na
l
ne
twor
ks
a
nd
two
s
ol
de
r
r
e
gi
ons
,”
A
ppl
ie
d Sc
ie
nc
e
s
, vol
. 10, no. 13, J
ul
. 2020, doi:
10.3390/app10
134598.
[
16]
B
.
C
a
i,
“
F
ul
ly
c
onne
c
te
d
c
onvolut
io
na
l
ne
ur
a
l
ne
two
r
k
in
P
C
B
s
ol
de
r
in
g
poi
nt
in
s
pe
c
ti
on,”
J
our
nal
of
C
om
put
e
r
and
C
om
m
uni
c
at
io
ns
, vol
. 10, no. 12, pp. 62
–
70, 2022, doi:
10.423
6/
jc
c
.2022.1012005.
[
17]
Q
.
Z
hu
a
nd
X
.
Z
u,
“
F
ul
ly
c
onvolut
io
na
l
ne
ur
a
l
n
e
twor
k
s
tr
uc
tu
r
e
a
nd
it
s
lo
s
s
f
unc
ti
on
f
or
im
a
ge
c
la
s
s
if
ic
a
ti
on,”
I
E
E
E
A
c
c
e
s
s
,
vol
. 10, pp. 35541
–
35549, 2022, doi:
10.1109/AC
C
E
S
S
.2022.3
163849.
[
18]
A
.
K
ha
n,
A
.
S
oha
il
,
U
.
Z
a
hoor
a
,
a
nd
A
.
S
.
Q
ur
e
s
hi
,
“
A
s
ur
ve
y of
th
e
r
e
c
e
nt
a
r
c
hi
te
c
tu
r
e
s
of
de
e
p c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
s
,”
A
r
ti
fi
c
ia
l
I
nt
e
ll
ig
e
nc
e
R
e
v
ie
w
, vol
. 53, no. 8,
pp. 5455
–
5516, 2020, doi:
10.1007/s
10462
-
020
-
09825
-
6.
[
19]
Z
.
L
i
,
F
.
L
i
u
,
W
.
Y
a
n
g
,
S
.
P
e
n
g
,
a
n
d
J
.
Z
h
o
u
,
“
A
s
u
r
v
e
y
o
f
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
:
a
n
a
l
y
s
i
s
,
a
p
p
l
i
c
a
t
i
o
n
s
,
a
n
d
p
r
o
s
p
e
c
t
s
,
”
I
E
E
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
N
e
u
r
a
l
N
e
t
w
o
r
k
s
a
n
d
L
e
a
r
n
i
n
g
S
y
s
t
e
m
s
,
v
o
l
.
3
3
, no. 12, pp. 6999
–
7
0
1
9
,
D
e
c
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
T
N
N
L
S
.
2
0
2
1
.
3
0
8
4
8
2
7
.
[
20]
M
. K
r
ic
he
n, “
C
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
:
a
s
ur
ve
y,”
C
o
m
p
ut
e
r
s
, vol
. 12, no. 8, J
ul
. 2023, doi:
10.3390/computer
s
1208015
1.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
A
c
c
e
ler
ati
ng
s
older
joi
nt
c
las
s
if
ication
us
ing
ge
ne
r
ati
v
e
ar
ti
fi
c
ial
int
e
ll
igenc
e
for
…
(
T
e
ng
Y
e
ow
Ong
)
4389
[
21]
J
.
L
ia
n,
L
.
W
a
ng,
T
.
L
iu
,
X
.
D
in
g,
a
nd
Z
.
Y
u,
“
A
ut
om
a
ti
c
vi
s
ua
l
in
s
pe
c
ti
on
f
or
pr
in
te
d
c
ir
c
ui
t
boa
r
d
vi
a
nove
l
ma
s
k
R
-
C
N
N
in
s
ma
r
t
c
it
y a
ppl
ic
a
ti
ons
,”
Su
s
ta
in
abl
e
E
ne
r
gy
T
e
c
hnol
ogi
e
s
and
A
s
s
e
s
s
m
e
nt
s
, vol
. 44, Apr
. 2021, doi:
10.1016/j
.s
e
ta
.2021.1010
32.
[
22]
A
. I
. M
. S
c
hw
e
bi
g a
nd R
. T
ut
s
c
h, “
C
ompi
la
ti
on of
t
r
a
in
in
g da
ta
s
e
ts
f
or
us
e
of
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
s
uppor
ti
ng a
ut
om
a
ti
c
in
s
pe
c
ti
on
pr
oc
e
s
s
e
s
in
in
dus
tr
y
4.0
ba
s
e
d
e
le
c
tr
oni
c
ma
nuf
a
c
tu
r
in
g,”
J
our
nal
of
Se
ns
or
s
and
Se
ns
o
r
Sy
s
te
m
s
,
vol
.
9,
no.
1,
pp
. 167
–
178, J
ul
. 2020, doi:
10.5194/j
s
s
s
-
9
-
167
-
2020.
[
23]
Y
.
L
iu
,
H
.
W
u,
Y
.
X
u,
X
.
L
iu
,
a
nd
X
.
Y
u,
“
A
ut
oma
ti
c
P
C
B
s
a
mpl
e
ge
ne
r
a
ti
on
a
nd
de
f
e
c
t
de
te
c
ti
on
ba
s
e
d
on
c
ont
r
ol
ne
t
a
nd
s
w
in
tr
a
ns
f
or
me
r
,”
Se
ns
or
s
, vol
. 24, no. 11,
M
a
y 2024, doi:
10.3390/s
24113473.
[
24]
S.
-
H
.
T
s
a
ng,
Z
.
S
uo,
T
.
T
.
-
L
.
C
ha
n,
H
.
-
T
.
N
guye
n,
a
nd
D
.
P
.
-
K
.
L
un,
“
P
C
B
s
ol
de
r
in
g
de
f
e
c
t
in
s
pe
c
ti
on
us
in
g
mul
ti
ta
s
k
le
a
r
ni
ng
unde
r
l
ow
da
ta
r
e
gi
me
s
,”
A
dv
anc
e
d I
nt
e
ll
ig
e
nt
Sy
s
te
m
s
, vol
. 5,
no. 12, De
c
. 2023, doi:
10.1002/ais
y.202300364.
[
25]
R
.
D
in
g,
C
.
Z
ha
ng,
Q
.
Z
hu,
a
nd
H
.
L
iu
,
“
U
nknown
de
f
e
c
t
de
te
c
ti
on
f
or
pr
in
te
d
c
ir
c
ui
t
boa
r
d
ba
s
e
d
on
mu
lt
i‐
s
c
a
le
de
e
p
s
im
il
a
r
it
y
me
a
s
ur
e
me
th
od,”
T
he
J
ou
r
nal
of
E
ngi
ne
e
r
in
g
, vol
. 2020, no. 1
3, pp. 388
–
393, J
ul
. 2020, doi:
10.1049/j
oe
.2019.1188.
[
26]
T.
-
C
.
T
s
a
n,
T
.
-
F.
S
hi
h,
a
nd
C
.
-
S
.
F
uh,
“
T
s
a
nK
it
:
a
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
f
or
s
ol
de
r
ba
ll
he
a
d
-
in
-
pi
ll
ow
de
f
e
c
t
in
s
pe
c
ti
on,”
M
ac
hi
ne
V
is
io
n and A
ppl
ic
at
io
ns
, vol
. 32, no. 3, M
a
y 2021, doi:
10.1007/s
00138
-
021
-
01192
-
8.
[
27]
Y
.
W
a
n,
L
.
G
a
o,
X
.
L
i,
a
nd
Y
.
G
a
o,
“
S
e
mi
-
s
upe
r
vi
s
e
d
de
f
e
c
t
de
te
c
ti
on
me
th
od
w
it
h
da
ta
-
e
xpa
ndi
ng
s
tr
a
te
gy
f
or
P
C
B
qua
li
ty
in
s
pe
c
ti
on,”
Se
n
s
or
s
, vol
. 22, no. 20, O
c
t.
2022, doi:
10.3390/s
22207971.
[
28]
J
.
Z
hou,
G
.
L
i,
R
.
W
a
ng,
R
.
C
h
e
n,
a
nd
S
.
L
uo,
“
A
nov
e
l
c
ont
r
a
s
ti
ve
s
e
lf
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
f
r
a
me
w
or
k
f
or
s
o
lv
in
g
da
ta
im
ba
la
nc
e
i
n s
ol
de
r
j
oi
nt
de
f
e
c
t
de
te
c
ti
on,”
E
nt
r
op
y
, vol
. 25, no. 2, J
a
n. 2023, doi:
10.3390/e25020268
.
[
29]
F.
-
A
.
C
r
oi
to
r
u,
V
.
H
ond
r
u,
R
.
T
.
I
one
s
c
u,
a
nd
M
.
S
ha
h,
“
D
if
f
us
io
n
mode
ls
in
vi
s
io
n:
a
s
ur
ve
y,”
I
E
E
E
T
r
ans
ac
ti
ons
on
P
at
t
e
r
n
A
nal
y
s
is
an
d M
ac
hi
ne
I
nt
e
ll
ig
e
nc
e
, vol
. 45, no. 9, pp. 10850
–
1
0869, S
e
p. 2023, doi:
10.1109/T
P
A
M
I
.2023.3261988.
[
30]
L
.
Y
a
ng
e
t
al
.
,
“
D
if
f
us
io
n
mode
ls
:
a
c
ompr
e
he
ns
iv
e
s
ur
ve
y
of
me
th
ods
a
nd
a
ppl
ic
a
ti
ons
,
”
A
C
M
C
om
put
in
g
Sur
v
e
y
s
,
vol
.
56,
no. 4, pp. 1
–
39, Apr
. 2024, d
oi
:
10.1145/3626235.
[
31]
H
. C
a
o
e
t
al
.
, “
A
s
ur
ve
y on ge
ne
r
a
ti
ve
di
f
f
us
io
n mode
ls
,”
I
E
E
E
T
r
ans
ac
ti
ons
on K
now
le
dge
and Data E
ngi
ne
e
r
in
g
, vol
. 36, no. 7,
pp. 2814
–
2830, J
ul
. 2024, doi:
10.1109/T
K
D
E
.2024.3361474.
[
32]
L
.
N
a
nni
,
M
.
P
a
c
i,
S
.
B
r
a
hna
m,
a
nd
A
.
L
u
mi
ni
,
“
C
ompa
r
is
o
n
of
di
f
f
e
r
e
nt
im
a
ge
da
t
a
a
ugme
nt
a
ti
on
a
ppr
oa
c
he
s
,”
J
our
nal
of
I
m
agi
ng
, vol
. 7, no. 12, Nov. 202
1, doi:
10.3390/j
im
a
gi
ng7120
254.
[
33]
T
.
K
uma
r
,
R
.
B
r
e
nna
n,
A
.
M
il
e
o,
a
nd
M
.
B
e
nde
c
h
a
c
he
,
“
I
ma
ge
da
ta
a
ugme
nt
a
ti
on
a
ppr
oa
c
he
s
:
a
c
ompr
e
he
ns
iv
e
s
ur
ve
y
a
nd
f
ut
ur
e
di
r
e
c
ti
ons
,”
I
E
E
E
A
c
c
e
s
s
, vol
. 12, pp. 187536
–
187571, 2024, doi:
10.1109/AC
C
E
S
S
.2024.3470122.
[
34]
M
.
X
u,
S
.
Y
oon,
A
.
F
ue
nt
e
s
,
a
nd
D
.
S
.
P
a
r
k,
“
A
c
ompr
e
he
ns
iv
e
s
ur
ve
y
of
im
a
ge
a
ugme
nt
a
ti
on
te
c
hni
que
s
f
or
de
e
p
le
a
r
ni
ng,”
P
at
te
r
n R
e
c
ogni
ti
on
, vol
. 137, M
a
y 2023, doi:
10.1016/j
.pa
tc
og.2023.109347.
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
Teng
Y
e
o
w
Ong
o
b
t
ai
n
ed
h
i
s
D
o
c
t
o
r
o
f
P
h
i
l
o
s
o
p
h
y
i
n
Man
u
fact
u
ri
n
g
S
y
s
t
ems
an
d
A
u
t
o
ma
t
i
o
n
fr
o
m
U
n
i
v
ers
i
t
y
Sci
e
n
ce
o
f
Ma
l
ay
s
i
a
i
n
2
0
1
0
.
H
e
al
s
o
ear
n
ed
h
i
s
Bach
e
l
o
r
o
f
E
n
g
i
n
eeri
n
g
(H
o
n
s
)
fro
m
t
h
e
U
n
i
v
er
s
i
t
y
o
f
Mal
ay
a,
Ma
l
ay
s
i
a
i
n
1
9
8
9
.
Cu
rren
t
l
y
,
h
e
s
erv
es
as
a
Sen
i
o
r
L
ect
u
rer
at
W
aw
a
s
an
O
p
e
n
U
n
i
v
er
s
i
t
y
,
Mal
a
y
s
i
a.
Pri
o
r
t
o
h
i
s
aca
d
emi
c
career,
h
e
accu
mu
l
at
e
d
o
v
er
t
w
o
d
ecad
e
s
o
f
i
n
d
u
s
t
r
y
ex
p
eri
e
n
ce
a
s
a
p
rac
t
i
t
i
o
n
er.
H
i
s
re
s
earch
i
n
t
ere
s
t
s
en
co
m
p
as
s
man
u
fact
u
ri
n
g
p
ro
ce
s
s
e
s
an
d
au
t
o
ma
t
i
o
n
,
ar
t
i
f
i
ci
a
l
n
e
u
ral
n
et
w
o
r
k
s
,
mach
i
n
e
d
eep
l
earn
i
n
g
,
i
mag
e
p
ro
ce
s
s
i
n
g
an
d
cl
a
s
s
i
fi
ca
t
i
o
n
.
H
e
can
b
e
co
n
t
ac
t
ed
at
emai
l
:
t
y
o
n
g
@
w
o
u
.
ed
u
.
m
y
.
P
i
ng
C
ho
w
Teo
h
h
o
l
d
s
a
D
o
ct
o
r
o
f
Ph
i
l
o
s
o
p
h
y
i
n
Man
u
fact
u
ri
n
g
E
n
g
i
n
eer
i
n
g
fro
m
L
o
u
g
h
b
o
r
o
u
g
h
U
n
i
v
ers
i
t
y
,
U
n
i
t
e
d
K
i
n
g
d
o
m
.
H
e
al
s
o
recei
v
ed
h
i
s
Bach
e
l
o
r
o
f
E
n
g
i
n
eeri
n
g
(Mec
h
an
i
cal
)
fr
o
m
t
h
e
U
n
i
v
ers
i
t
y
o
f
Mal
ay
a,
Mal
ay
s
i
a
a
w
ar
d
ed
i
n
1
9
9
1
.
H
e
i
s
cu
rren
t
l
y
an
A
s
s
o
ci
a
t
e
Pro
fe
s
s
o
r
at
t
h
e
Sc
h
o
o
l
o
f
T
ech
n
o
l
o
g
y
an
d
E
n
g
i
n
eer
i
n
g
Sci
en
ce,
W
aw
a
s
an
O
p
e
n
U
n
i
v
ers
i
t
y
,
Mal
ay
s
i
a.
H
i
s
res
earc
h
fo
c
u
s
es
o
n
d
at
a
-
d
ri
v
en
ma
n
u
fac
t
u
r
i
n
g
,
l
ea
n
man
u
fac
t
u
r
i
n
g
,
k
n
o
w
l
ed
g
e
man
ag
eme
n
t
an
d
fai
l
u
re
m
o
d
e
an
d
effect
s
a
n
al
y
s
i
s
(FME
A
).
He
h
as
au
t
h
o
red
n
u
mero
u
s
p
u
b
l
i
cat
i
o
n
s
i
n
acad
em
i
c
j
o
u
rn
a
l
an
d
co
n
fere
n
ce
p
r
o
ceed
i
n
g
s
as
w
el
l
as
h
o
l
d
i
n
g
9
U
S
p
a
t
en
t
s
.
H
e
ca
n
b
e
co
n
t
ac
t
ed
at
ema
i
l
:
p
c
t
e
o
h
@
w
o
u
.
ed
u
.
my
.
Ko
o
n
T
a
tt
Ta
n
earn
ed
h
i
s
P
h
.
D
.
i
n
Mec
h
an
i
cal
a
n
d
Mat
eri
a
l
s
E
n
g
i
n
eeri
n
g
fro
m
U
n
i
v
er
s
i
t
y
K
e
b
an
g
s
aa
n
Mal
ay
s
i
a
.
H
e
cu
rren
t
l
y
s
erv
e
s
as
an
A
s
s
o
ci
a
t
e
Pro
fe
s
s
o
r
an
d
t
h
e
D
ean
o
f
t
h
e
Sch
o
o
l
o
f
T
ec
h
n
o
l
o
g
y
an
d
E
n
g
i
n
eer
i
n
g
Sci
en
ce
at
W
aw
as
a
n
O
p
e
n
U
n
i
v
er
s
i
t
y
,
Mal
ay
s
i
a.
H
i
s
areas
o
f
re
s
earch
e
x
p
er
t
i
s
e
i
n
c
l
u
d
e
p
o
w
d
er
me
t
al
l
u
rg
y
,
met
al
i
n
j
ec
t
i
o
n
m
o
u
l
d
i
n
g
,
met
a
l
fo
ams
,
s
t
at
i
s
t
i
ca
l
p
r
o
ces
s
o
p
t
i
mi
s
at
i
o
n
,
b
i
o
mat
er
i
al
s
,
man
u
fact
u
ri
n
g
p
ro
ce
s
s
e
s
,
an
d
man
u
fac
t
u
r
i
n
g
man
ag
emen
t
.
He
i
s
al
s
o
a
reg
i
s
t
ere
d
Pro
fes
s
i
o
n
a
l
T
ech
n
o
l
o
g
i
s
t
w
i
t
h
Mal
ay
s
i
a
Bo
ard
of
T
ech
n
o
l
o
g
i
s
t
s
(MBO
T
)
an
d
an
act
i
v
e
Co
mm
i
t
t
ee
Memb
er
of
t
h
e
Mal
a
y
s
i
a
Po
w
d
er
Met
al
l
u
r
g
y
an
d
Part
i
cu
l
at
e
Ma
t
eri
a
l
s
A
s
s
o
c
i
at
i
o
n
(MPM2
A
).
H
e
ca
n
b
e
co
n
t
ac
t
ed
at
emai
l
:
s
ean
t
an
@
w
o
u
.
ed
u
.
my
.
Evaluation Warning : The document was created with Spire.PDF for Python.