I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
,
p
p
.
2796
~
2
8
0
8
I
SS
N:
2088
-
8
7
0
8
,
DOI
:
1
0
.
1
1
5
9
1
/ijece.
v
15
i
3
.
pp
2
7
9
6
-
2
8
0
8
2796
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
O
bje
ct
dete
ction i
n print
ed ci
rcuit
bo
a
rd quali
ty co
n
trol:
co
mpa
ring
alg
o
ri
thms fas
te
r
regi
o
n
-
ba
sed co
nv
o
luti
o
na
l neural
networks
a
nd
Y
O
LOv8
J
a
j
a
K
us
t
ij
a
1
,
Dik
i F
a
hriza
l
2
,
M
uh
a
m
a
d Na
s
ir
3
,
Andi
A
d
ria
ns
y
a
h
4
,
M
uh
a
m
m
a
d H
us
ni M
utt
a
qin
2
1
P
r
o
g
r
a
m
S
t
u
d
y
o
f
El
e
c
t
r
i
c
a
l
E
n
g
i
n
e
e
r
i
n
g
E
d
u
c
a
t
i
o
n
,
F
a
c
u
l
t
y
o
f
T
e
c
h
n
o
l
o
g
y
a
n
d
V
o
c
a
t
i
o
n
a
l
E
d
u
c
a
t
i
o
n
,
U
n
i
v
e
r
s
i
t
a
s
P
e
n
d
i
d
i
k
a
n
I
n
d
o
n
e
si
a
,
B
a
n
d
u
n
g
,
I
n
d
o
n
e
si
a
2
M
a
s
t
e
r
o
f
E
l
e
c
t
r
i
c
a
l
E
n
g
i
n
e
e
r
i
n
g
,
S
c
h
o
o
l
o
f
El
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
a
n
d
I
n
f
o
r
mat
i
c
s
,
I
n
st
i
t
u
t
Te
k
n
o
l
o
g
i
B
a
n
d
u
n
g
,
B
a
n
d
u
n
g
,
I
n
d
o
n
e
s
i
a
3
R
e
s
e
a
r
c
h
C
e
n
t
e
r
f
o
r
En
v
i
r
o
n
me
n
t
a
n
d
C
l
e
a
n
Te
c
h
n
o
l
o
g
y
,
B
R
I
N
,
B
a
n
d
u
n
g
,
I
n
d
o
n
e
si
a
4
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
,
U
n
i
v
e
r
si
t
a
s
M
e
r
c
u
B
u
a
n
a
,
Ja
k
a
r
t
a
,
I
n
d
o
n
e
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 1
,
2
0
2
4
R
ev
is
ed
Dec
2
9
,
2
0
2
4
Acc
ep
ted
J
an
1
6
,
2
0
2
5
Alo
n
g
wit
h
th
e
d
e
v
e
lo
p
m
e
n
t
o
f
e
lec
tro
n
ic
tec
h
n
o
lo
g
y
,
t
h
e
i
n
te
g
ra
ti
o
n
o
f
n
u
m
e
ro
u
s
c
o
m
p
o
n
e
n
ts
o
n
p
ri
n
ted
c
ircu
it
b
o
a
rd
(
P
CB)
b
o
a
rd
s
h
a
s
re
su
lt
e
d
i
n
in
c
re
a
sin
g
ly
c
o
m
p
lex
a
n
d
i
n
tri
c
a
t
e
lay
o
u
ts.
S
m
a
ll
d
e
fe
c
ts
i
n
trac
e
s
c
a
n
lea
d
to
fa
il
u
re
s
in
e
lec
tro
n
ic
fu
n
c
ti
o
n
s,
m
a
k
in
g
t
h
e
in
s
p
e
c
ti
o
n
o
f
P
C
B
su
rfa
c
e
lay
o
u
ts
a
c
rit
ica
l
p
r
o
c
e
ss
in
q
u
a
li
ty
c
o
n
tr
o
l.
G
iv
e
n
t
h
e
l
imitatio
n
s
o
f
m
a
n
u
a
l
in
sp
e
c
ti
o
n
,
wh
ic
h
stru
g
g
les
to
d
e
tec
t
su
c
h
d
e
fe
c
ts
d
u
e
to
t
h
e
ir
siz
e
a
n
d
c
o
m
p
lex
it
y
,
t
h
e
re
is a g
ro
win
g
n
e
e
d
fo
r
a
P
CB i
n
sp
e
c
ti
o
n
sy
ste
m
th
a
t
u
ti
li
z
e
s
a
u
to
m
a
ted
o
p
ti
c
a
l
i
n
sp
e
c
ti
o
n
(A
OI)
b
a
se
d
o
n
d
e
e
p
lea
rn
i
n
g
d
e
te
c
ti
o
n
.
Th
is
re
se
a
rc
h
d
e
v
e
lo
p
s
a
n
d
c
o
m
p
a
r
e
s
two
d
e
e
p
lea
rn
i
n
g
a
lg
o
r
it
h
m
s,
fa
ste
r
re
g
io
n
-
b
a
se
d
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
r
a
l
n
e
two
rk
s
(R
-
CNN
)
a
n
d
YO
LOv
8
,
t
o
id
e
n
ti
f
y
t
h
e
m
o
st
e
ffe
c
ti
v
e
a
lg
o
ri
th
m
fo
r
d
e
tec
ti
n
g
d
e
fe
c
ts
o
n
P
C
B
lay
o
u
ts
.
Th
e
fin
d
in
g
s
o
f
th
is
stu
d
y
in
d
ica
te
th
a
t
th
e
YO
LO
v
8
a
l
g
o
rit
h
m
o
u
tp
e
rfo
rm
s
fa
ste
r
R
-
CNN
,
with
th
e
YO
LOv
8
x
v
a
rian
t
e
m
e
rg
in
g
a
s
th
e
b
e
st
m
o
d
e
l
fo
r
d
e
fe
c
t
d
e
tec
ti
o
n
.
Th
e
YO
LOv
8
x
m
o
d
e
l
a
c
h
iev
e
d
p
e
rf
o
rm
a
n
c
e
sc
o
re
s
o
f
0
.
9
6
2
(m
AP@5
0
),
0
.
5
0
3
(m
AP
@5
0
:
9
5
),
0
.
9
5
3
(P
re
c
isio
n
)
,
0
.
9
4
5
(Re
c
a
ll
),
a
n
d
0
.
9
4
9
(F
1
-
sc
o
re
).
Th
e
se
re
su
lt
s
p
ro
v
i
d
e
a
stro
n
g
f
o
u
n
d
a
ti
o
n
f
o
r
fu
r
th
e
r
re
se
a
rc
h
in
to
th
e
a
p
p
li
c
a
ti
o
n
o
f
AO
I
fo
r
P
CB
d
e
fe
c
t
d
e
tec
ti
o
n
a
n
d
o
t
h
e
r
q
u
a
li
t
y
c
o
n
tr
o
l
p
ro
c
e
ss
e
s
in
m
a
n
u
fa
c
tu
rin
g
,
u
sin
g
o
p
ti
m
ize
d
d
e
e
p
lea
rn
in
g
m
o
d
e
ls.
K
ey
w
o
r
d
s
:
Au
to
m
ated
o
p
tical
in
s
p
ec
tio
n
C
o
m
p
ar
is
o
n
o
f
al
g
o
r
ith
m
Fas
ter
r
eg
io
n
-
b
ased
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
Prin
ted
cir
cu
it b
o
a
r
d
d
e
f
ec
t
d
etec
tio
n
YOL
Ov
8
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
:
J
aja
Ku
s
tija
Pro
g
r
am
Stu
d
y
o
f
E
lectr
ical
E
n
g
in
ee
r
in
g
E
d
u
ca
tio
n
,
Facu
lty
o
f
T
ec
h
n
o
lo
g
y
an
d
Vo
ca
tio
n
a
l E
d
u
ca
tio
n
,
Un
iv
er
s
itas
Pen
d
id
ik
an
I
n
d
o
n
esia
4
0
1
5
4
,
B
an
d
u
n
g
,
I
n
d
o
n
esia
E
m
ail:
jaja.
k
u
s
tija@
u
p
i.e
d
u
1.
I
NT
RO
D
UCT
I
O
N
Prin
ted
cir
cu
it
b
o
a
r
d
s
(
PC
B
s
)
ar
e
ess
en
tial
co
m
p
o
n
en
ts
in
m
o
d
e
r
n
elec
tr
o
n
ics
p
r
o
v
id
in
g
th
e
f
o
u
n
d
atio
n
f
o
r
elec
tr
ical
co
n
n
ec
tio
n
s
th
r
o
u
g
h
co
n
d
u
ctiv
e
c
h
an
n
els,
p
ad
s
,
an
d
s
o
ld
er
jo
i
n
ts
.
Ho
wev
er
,
PC
B
s
u
r
f
ac
es
o
f
ten
s
u
f
f
er
f
r
o
m
d
ef
ec
ts
s
u
ch
as
m
is
s
in
g
h
o
les,
m
o
u
s
e
b
ites
,
o
p
en
cir
cu
its
,
s
h
o
r
ts
,
s
p
u
r
s
,
an
d
s
p
u
r
io
u
s
co
p
p
er
[
1
]
,
[
2
]
.
T
h
e
s
e
f
laws
ca
n
lead
to
elec
tr
o
n
ic
d
ev
ice
m
alf
u
n
ctio
n
s
a
n
d
r
ed
u
ce
d
e
f
f
icien
cy
,
is
s
u
es
th
at
ar
e
f
u
r
th
er
ex
ac
e
r
b
ated
b
y
th
e
g
r
o
win
g
co
m
p
lex
ity
o
f
PC
B
d
esig
n
s
d
r
iv
e
n
b
y
th
e
p
o
r
tab
le
elec
tr
o
n
ics
in
d
u
s
tr
y
[
3
]
.
T
r
ad
i
tio
n
al
m
an
u
al
in
s
p
ec
tio
n
m
et
h
o
d
s
,
wh
ile
in
itially
co
s
t
-
ef
f
e
ctiv
e,
h
av
e
b
ec
o
m
e
in
ad
eq
u
ate
d
u
e
to
th
eir
s
u
b
j
ec
tiv
ity
,
in
co
n
s
is
ten
cy
,
an
d
s
u
s
ce
p
tib
ilit
y
to
in
s
p
ec
to
r
f
ati
g
u
e,
m
a
k
in
g
th
em
u
n
s
u
itab
le
f
o
r
m
o
d
er
n
h
ig
h
-
v
o
lu
m
e
p
r
o
d
u
ctio
n
en
v
ir
o
n
m
en
ts
[
4
]
,
[
5
]
.
T
o
ad
d
r
ess
th
ese
lim
itatio
n
s
,
au
to
m
ated
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
Ob
ject
d
etec
tio
n
in
p
r
in
ted
ci
r
cu
it b
o
a
r
d
q
u
a
lity c
o
n
tr
o
l:
c
o
mp
a
r
in
g
a
lg
o
r
ith
ms
…
(
Ja
ja
K
u
s
tija
)
2797
o
p
tical
in
s
p
ec
tio
n
(
AOI
)
s
y
s
tem
s
h
av
e
em
er
g
ed
as
a
t
ec
h
n
o
lo
g
ical
leap
in
q
u
ality
co
n
tr
o
l
f
o
r
PC
B
m
an
u
f
ac
tu
r
in
g
.
AOI
s
y
s
tem
s
an
aly
ze
h
ig
h
-
r
eso
lu
tio
n
PC
B
im
ag
es
u
s
in
g
alg
o
r
ith
m
s
to
id
e
n
tify
d
ef
ec
ts
b
ased
o
n
p
r
ed
ef
in
e
d
c
r
iter
ia,
in
clu
d
i
n
g
r
e
f
er
en
ce
-
b
ased
c
o
m
p
ar
is
o
n
s
,
co
m
p
lian
ce
ch
ec
k
s
,
o
r
h
y
b
r
id
ap
p
r
o
ac
h
es
[
6
]
.
T
h
e
in
teg
r
atio
n
o
f
d
ee
p
lear
n
in
g
tech
n
iq
u
es
h
as
f
u
r
th
er
r
ev
o
lu
tio
n
ized
AOI
ca
p
ab
ilit
ie
s
,
en
ab
lin
g
h
ig
h
er
ac
cu
r
ac
y
a
n
d
r
ea
l
-
tim
e
a
n
aly
s
is
.
Desp
ite
th
ese
ad
v
an
ce
m
e
n
ts
,
ch
allen
g
es
r
em
ain
in
v
alid
atin
g
m
o
d
els
with
r
ea
l
-
wo
r
ld
d
ata
a
n
d
a
d
d
r
ess
in
g
lim
itatio
n
s
in
h
an
d
li
n
g
u
n
tr
a
in
ed
d
atasets
[
7
]
.
T
h
e
ad
v
en
t
o
f
d
ee
p
lear
n
in
g
tech
n
iq
u
es
h
as
r
e
v
o
lu
tio
n
ized
co
m
p
u
ter
v
is
io
n
,
e
n
h
a
n
cin
g
th
e
ca
p
ab
ilit
ies
o
f
AOI
s
y
s
tem
s
.
R
ec
en
t
s
tu
d
ies,
s
u
ch
as
th
o
s
e
u
tili
zin
g
th
e
Sk
ip
-
C
o
n
n
ec
ted
C
o
n
v
o
lu
tio
n
al
Au
to
en
co
d
e
r
m
o
d
el,
h
av
e
ac
h
iev
ed
d
e
f
ec
t
d
etec
tio
n
ac
cu
r
ac
ies
ex
ce
ed
in
g
9
0
%.
H
o
wev
er
,
th
ese
m
o
d
els
s
till
f
ac
e
ch
allen
g
es,
in
clu
d
in
g
th
e
n
ee
d
f
o
r
r
ea
l
-
wo
r
ld
v
alid
atio
n
an
d
lim
itatio
n
s
in
h
an
d
lin
g
u
n
tr
ain
ed
d
ata
[
8
]
.
Ad
d
itio
n
ally
,
s
em
i
-
s
u
p
e
r
v
is
ed
lear
n
in
g
(
SS
L
)
m
o
d
els
th
at
lev
er
ag
e
b
o
th
lab
eled
a
n
d
u
n
lab
eled
d
ata
with
d
if
f
er
en
t
au
g
m
en
tatio
n
s
h
av
e
s
h
o
wn
im
p
r
o
v
e
d
p
er
f
o
r
m
a
n
ce
o
v
er
p
u
r
ely
Su
p
er
v
is
ed
L
ea
r
n
in
g
m
o
d
els,
with
er
r
o
r
i
n
cr
ea
s
es
o
f
less
th
an
0
.
5
%
[
9
]
.
Desp
ite
d
if
f
er
e
n
ce
s
i
n
d
atasets
an
d
o
b
ject
class
if
icatio
n
,
th
is
r
esear
ch
p
r
o
v
id
es
v
alu
a
b
le
in
s
ig
h
t
th
at
in
f
o
r
m
th
e
d
ev
elo
p
m
en
t
o
f
m
o
r
e
ad
v
an
ce
d
alg
o
r
ith
m
ic
m
o
d
els,
s
u
ch
as
th
o
s
e
ex
p
lo
r
ed
in
th
is
r
esear
ch
.
T
h
i
s
r
es
e
a
r
c
h
e
x
a
m
i
n
es
t
w
o
a
d
v
a
n
c
e
d
d
e
e
p
l
ea
r
n
i
n
g
a
l
g
o
r
i
t
h
m
,
f
a
s
te
r
R
-
C
N
N
a
n
d
YO
L
O
v
8
,
f
o
r
PC
B
d
e
f
e
c
t
d
e
t
e
c
ti
o
n
.
F
a
s
t
e
r
R
-
C
NN
e
m
p
l
o
y
s
a
r
e
g
i
o
n
p
r
o
p
o
s
a
l
n
e
t
w
o
r
k
(
R
PN
)
f
o
r
a
c
c
u
r
a
t
e
o
b
je
c
t
r
e
c
o
g
n
i
t
i
o
n
[
1
0
]
,
w
h
i
l
e
Y
O
L
O
v
8
,
t
h
e
l
at
es
t
it
er
a
t
i
o
n
i
n
t
h
e
Y
OL
O
s
e
r
i
es
,
ex
c
e
l
s
i
n
r
e
a
l
-
t
i
m
e
d
et
e
c
ti
o
n
wi
t
h
h
i
g
h
s
p
e
e
d
a
n
d
p
r
e
c
i
s
i
o
n
[
1
1
]
.
A
b
i
b
l
i
o
m
e
t
r
ic
a
n
a
l
y
s
is
o
f
1
0
5
j
o
u
r
n
a
l
a
r
t
i
c
le
s
i
n
d
e
x
e
d
b
y
G
o
o
g
l
e
S
c
h
o
l
a
r
a
n
d
S
c
o
p
u
s
f
u
r
t
h
e
r
h
i
g
h
l
i
g
h
t
s
t
h
e
i
m
p
o
r
ta
n
c
e
o
f
th
e
s
e
a
l
g
o
r
i
t
h
m
i
n
a
d
v
a
n
c
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
l
i
c
a
ti
o
n
s
f
o
r
PC
B
d
e
f
e
c
t
d
e
t
e
ct
i
o
n
[
1
2
]
.
A
s
s
h
o
w
n
i
n
Fi
g
u
r
e
1
,
“
d
e
e
p
l
e
a
r
n
i
n
g
,
”
“
Y
O
L
O
v
8
,
”
an
d
“
f
a
s
t
e
r
R
-
C
N
N
”
a
r
e
c
e
n
t
r
a
l
n
o
d
e
s
i
n
t
h
is
f
ie
l
d
,
d
e
m
o
n
s
t
r
a
t
i
n
g
t
h
ei
r
p
i
v
o
t
a
l
c
o
n
t
r
i
b
u
t
i
o
n
s
t
o
r
ea
l
-
t
i
m
e
a
p
p
l
i
c
at
io
n
s
a
n
d
e
f
f
i
c
i
e
n
t
i
m
a
g
e
p
r
o
c
e
s
s
i
n
g
.
T
h
is
r
esear
ch
aim
s
to
an
aly
z
e
an
d
co
m
p
ar
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
f
aster
R
-
C
NN
an
d
YOL
Ov
8
in
d
etec
tin
g
v
ar
i
o
u
s
ty
p
es
o
f
PC
B
d
ef
ec
ts
,
s
u
ch
as
m
is
s
in
g
h
o
les,
m
o
u
s
e
b
ites
,
o
p
en
cir
cu
its
,
s
h
o
r
ts
,
s
p
u
r
s
,
an
d
s
p
u
r
io
u
s
co
p
p
e
r
.
B
y
d
o
in
g
s
o
,
th
is
r
esear
ch
co
n
tr
ib
u
tes
to
th
e
d
ev
elo
p
m
e
n
t
o
f
ef
f
icien
t
an
d
ef
f
ec
tiv
e
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
f
o
r
PC
B
d
ef
ec
t
id
en
tific
atio
n
,
o
f
f
er
in
g
s
o
lu
tio
n
s
to
th
e
in
cr
ea
s
in
g
co
m
p
lex
ity
o
f
v
is
u
ally
u
n
d
etec
tab
le
f
a
u
lts
.
Fu
r
th
er
m
o
r
e,
it
p
r
o
v
id
es
v
alu
ab
le
in
s
i
g
h
ts
in
to
th
e
a
p
p
licatio
n
o
f
c
o
m
p
u
ter
v
is
io
n
an
d
d
ee
p
lear
n
i
n
g
in
in
d
u
s
tr
ial
s
ettin
g
s
,
p
av
in
g
th
e
wa
y
f
o
r
m
o
r
e
ad
v
an
ce
d
d
etec
tio
n
s
y
s
tem
s
ca
p
ab
le
o
f
h
an
d
lin
g
d
iv
er
s
e
PC
B
d
ef
ec
ts
.
Fig
u
r
e
1
.
An
al
y
s
is
n
etwo
r
k
v
i
s
u
aliza
tio
n
VOSv
iewe
r
2.
O
VE
RVI
E
W
O
F
S
E
L
E
C
T
E
D
AL
G
O
RI
T
H
M
2
.
1
.
F
a
s
t
er
R
-
CNN
a
lg
o
rit
hm
On
e
m
o
d
el
o
r
m
eth
o
d
t
h
at
c
an
b
e
u
s
ed
t
o
d
etec
t
o
b
jects
in
an
im
ag
e
is
th
e
f
aster
r
eg
io
n
-
b
ased
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
o
r
wh
at
is
k
n
o
wn
as
f
aster
R
-
C
NN.
T
h
is
m
o
d
el
is
a
d
ev
elo
p
m
en
t
o
f
th
e
p
r
ev
io
u
s
alg
o
r
ith
m
,
n
am
ely
f
ast
R
-
C
NN.
T
h
e
g
o
al
o
f
f
ast
er
R
-
C
NN
is
to
in
cr
ea
s
e
th
e
ac
cu
r
ac
y
o
f
o
b
ject
d
etec
tio
n
b
y
r
ed
u
cin
g
th
e
n
u
m
b
er
o
f
p
r
o
p
o
s
al
r
eg
io
n
s
g
en
er
ated
[
1
3
]
.
T
o
ac
h
iev
e
th
is
g
o
al,
th
is
alg
o
r
ith
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
:
2
7
9
6
-
2
8
0
8
2798
r
ep
lace
s
th
e
s
elec
tiv
e
s
ea
r
ch
m
eth
o
d
in
th
e
f
ast
R
-
C
NN
m
o
d
el
with
a
r
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
.
T
h
e
f
aster
R
-
C
NN
m
o
d
el
is
g
en
er
ally
s
im
ilar
to
th
e
f
ast
R
-
C
NN
m
o
d
el
b
u
t
o
n
ly
ch
an
g
es
p
ar
t
o
f
t
h
e
p
r
o
p
o
s
al
r
eg
io
n
[
1
4
]
.
R
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
(
R
PN)
wo
r
k
s
th
r
o
u
g
h
th
e
f
o
ll
o
win
g
s
tep
s
:
a.
Usi
n
g
a
3
×
3
co
n
v
o
lu
tio
n
al
lay
er
with
p
ad
d
in
g
1
to
ch
a
n
g
e
th
e
C
NN
o
u
tp
u
t
in
to
a
n
ew
o
u
tp
u
t
with
a
n
u
m
b
er
o
f
c
h
an
n
els.
b.
Nex
t,
s
ev
er
al
lab
eled
b
o
x
es
ar
e
cr
ea
ted
f
o
r
ea
ch
p
ix
el
in
th
e
f
ea
tu
r
e
m
ap
with
d
if
f
e
r
en
t
s
ca
les
an
d
asp
ec
t
r
atio
s
.
c.
Fo
r
ea
ch
la
b
eled
b
o
x
,
b
in
a
r
y
class
p
r
ed
ictio
n
(
o
b
ject
o
r
b
ac
k
g
r
o
u
n
d
)
is
p
er
f
o
r
m
e
d
,
a
n
d
th
e
f
ea
tu
r
e
v
ec
to
r
is
p
lace
d
at
th
e
ce
n
ter
o
f
ea
ch
lab
eled
b
o
x
.
d.
Af
ter
th
e
n
o
n
-
m
ax
im
u
m
s
u
p
p
r
ess
io
n
p
r
o
ce
s
s
,
th
e
b
o
u
n
d
i
n
g
b
o
x
p
r
e
d
icted
as
an
o
b
ject
is
u
s
ed
as
a
p
r
o
p
o
s
al
r
eg
io
n
.
T
h
e
class
if
icatio
n
lo
s
s
an
d
r
e
g
r
ess
io
n
lo
s
s
ar
e
co
m
p
u
ted
as
(
1
)
,
(
2
)
:
(
,
∗
)
=
−
l
og
[
∗
+
(
1
−
∗
)
(
1
−
)
]
(
1
)
(
,
∗
)
=
ℎ
1
(
−
∗
)
=
{
0
.
5
(
−
∗
)
2
×
1
2
|
−
∗
|
<
1
2
|
−
∗
|
−
0
.
5
ℎ
(
2
)
T
h
e
p
o
s
itio
n
v
ec
t
o
r
,
r
ep
r
ese
n
ts
th
e
p
r
o
jecte
d
o
f
f
s
et
o
f
th
e
an
ch
o
r
in
t
h
e
r
e
g
io
n
p
r
o
p
o
s
al
n
etwo
r
k
.
T
h
e
p
r
ac
tical
o
f
f
s
et,
∗
is
a
v
ec
to
r
o
f
th
e
s
am
e
d
im
en
s
io
n
s
as
an
d
is
co
m
p
ar
ed
with
th
e
g
r
o
u
n
d
t
r
u
th
.
T
h
e
t
o
tal
lo
s
s
f
u
n
ctio
n
is
co
m
p
u
ted
with
an
d
,
as
(
7
)
s
h
o
wn
[
1
5
]
.
(
,
)
=
1
∑
(
,
∗
)
+
1
∑
∗
(
,
∗
)
(
3
)
I
t
is
im
p
o
r
tan
t
to
n
o
te
t
h
at
th
e
r
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
is
tr
ain
ed
s
im
u
ltan
eo
u
s
ly
with
o
th
er
co
m
p
o
n
en
ts
o
f
th
e
f
aster
R
-
C
NN
alg
o
r
ith
m
m
o
d
el.
T
h
e
r
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
was
d
ev
elo
p
ed
to
p
r
o
d
u
ce
h
ig
h
q
u
ality
r
eg
io
n
p
r
o
p
o
s
als,
s
o
th
at
it
ca
n
d
etec
t
o
b
jects
with
a
s
m
aller
am
o
u
n
t
o
f
d
ata
.
T
h
e
ar
ch
itectu
r
e
o
f
th
e
f
aster
R
-
C
NN
alg
o
r
ith
m
is
illu
s
tr
ated
in
Fig
u
r
e
2
.
Fig
u
r
e
2
.
Ar
c
h
itectu
r
e
f
aster
R
-
C
NN
a
lg
o
r
ith
m
2
.
2
.
YO
L
O
v
8
a
lg
o
rit
h
m
YOL
Ov
8
is
a
s
ig
n
if
ican
t
b
r
ea
k
th
r
o
u
g
h
in
th
e
YOL
O
s
er
ies,
n
o
ted
f
o
r
its
r
ea
l
-
tim
e
o
b
ject
d
etec
tio
n
ca
p
ab
ilit
ies.
T
h
e
m
o
d
el
ar
c
h
itectu
r
e
is
co
m
p
o
s
ed
o
f
f
o
u
r
p
r
im
ar
y
co
m
p
o
n
e
n
ts
:
th
e
in
p
u
t
m
o
d
u
le,
b
ac
k
b
o
n
e
f
ea
tu
r
e
e
x
tr
ac
tio
n
n
etwo
r
k
,
n
ec
k
n
etwo
r
k
,
a
n
d
d
etec
ti
o
n
m
o
d
u
le.
T
h
ese
co
m
p
o
n
en
ts
wo
r
k
to
g
eth
e
r
s
ea
m
less
ly
to
p
r
o
ce
s
s
im
ag
es
,
ex
tr
ac
t
f
ea
tu
r
es,
an
d
f
o
r
ec
ast
o
b
ject
class
es
an
d
b
o
u
n
d
in
g
b
o
x
es
with
g
r
ea
t
ef
f
icien
cy
,
m
a
k
in
g
YOL
Ov
8
a
s
o
lid
s
o
lu
tio
n
f
o
r
m
an
y
o
b
ject
r
ec
o
g
n
itio
n
wo
r
k
lo
ad
s
[
1
6
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Ob
ject
d
etec
tio
n
in
p
r
in
ted
ci
r
cu
it b
o
a
r
d
q
u
a
lity c
o
n
tr
o
l:
c
o
mp
a
r
in
g
a
lg
o
r
ith
ms
…
(
Ja
ja
K
u
s
tija
)
2799
T
h
e
in
p
u
t
m
o
d
u
le
o
f
YOL
Ov
8
co
v
er
s
cr
u
cial
f
u
n
ctio
n
ality
s
u
ch
as
p
ictu
r
e
in
p
u
t,
d
ata
au
g
m
en
tatio
n
,
an
d
ad
ap
tiv
e
an
ch
o
r
b
o
x
co
m
p
u
tatio
n
s
.
T
h
e
b
ac
k
b
o
n
e
f
e
atu
r
e
ex
tr
ac
tio
n
n
etwo
r
k
lev
er
ag
es
s
tr
u
ctu
r
es
lik
e
C
o
n
v
+Bn
+SiL
U
(
C
B
L
)
,
C
S
PLa
y
er
_
2
C
o
n
v
(
c
2
F),
an
d
s
p
atial
p
y
r
am
id
p
o
o
lin
g
-
f
ast
(
SP
PF
)
to
ex
tr
ac
t
m
ea
n
in
g
f
u
l
f
ea
tu
r
es
f
r
o
m
im
ag
es.
T
h
ese
s
tr
u
ctu
r
es
en
h
an
ce
th
e
m
o
d
el’
s
ab
ilit
y
to
r
ec
o
g
n
ize
an
d
class
if
y
o
b
jects
ac
cu
r
ately
.
Me
an
wh
il
e,
th
e
n
ec
k
n
etwo
r
k
em
p
lo
y
s
th
e
p
ath
ag
g
r
eg
atio
n
n
etwo
r
k
(
PAN)
s
tr
u
ctu
r
e
to
f
u
s
e
o
b
ject
in
f
o
r
m
atio
n
ac
r
o
s
s
d
if
f
er
en
t
s
ca
les.
T
h
is
ap
p
r
o
a
ch
m
ak
es
YOL
Ov
8
h
ig
h
ly
ef
f
ec
tiv
e
in
d
etec
tin
g
o
b
jects
o
f
v
a
r
y
in
g
s
izes
b
y
en
s
u
r
in
g
th
at
f
ea
t
u
r
e
m
a
p
s
f
r
o
m
m
u
ltip
le
lay
er
s
ar
e
u
tili
ze
d
o
p
t
im
ally
.
Fin
ally
,
th
e
d
etec
tio
n
m
o
d
u
le
co
m
b
in
es
c
lass
if
icatio
n
an
d
r
eg
r
ess
io
n
t
ask
s
b
y
u
s
in
g
a
d
v
an
ce
d
lo
s
s
f
u
n
ctio
n
s
s
u
ch
as
v
ar
if
o
ca
l lo
s
s
(
VFL)
f
o
r
class
i
f
icatio
n
an
d
d
is
tr
ib
u
tio
n
f
o
ca
l
lo
s
s
(
DFL)
with
co
m
p
lete
in
ter
s
ec
tio
n
o
v
er
u
n
io
n
(
C
I
o
U)
lo
s
s
f
o
r
r
e
g
r
ess
io
n
,
r
es
u
ltin
g
in
p
r
ec
is
e
an
d
r
eliab
le
o
b
ject
d
etec
tio
n
[
1
7
]
.
I
n
ad
d
itio
n
to
its
ar
c
h
itectu
r
al
ad
v
a
n
tag
es,
YOL
Ov
8
o
f
f
er
s
f
lex
ib
ilit
y
in
its
co
n
f
i
g
u
r
atio
n
s
,
allo
win
g
u
s
er
s
to
ad
ju
s
t
p
ar
a
m
eter
s
s
u
ch
as
wid
th
,
d
ep
th
,
an
d
r
atio
to
m
ee
t
s
p
ec
if
ic
p
e
r
f
o
r
m
an
ce
an
d
c
o
m
p
u
tatio
n
al
r
eq
u
ir
em
e
n
ts
.
T
h
e
m
o
d
el
is
av
ailab
le
in
f
iv
e
co
n
f
ig
u
r
atio
n
s
,
in
clu
d
in
g
YOL
Ov
8
n
,
YOL
Ov
8
s
,
YOL
Ov
8
m
,
YOL
Ov
8
l,
an
d
YOL
Ov
8
x
ea
ch
ca
ter
in
g
to
d
if
f
er
e
n
t
ap
p
li
ca
tio
n
n
ee
d
s
[
1
8
]
.
E
v
alu
atio
n
s
h
av
e
s
h
o
wn
th
at
YOL
Ov
8
m
ac
h
iev
es
th
e
h
ig
h
est
m
AP
at
8
7
.
7
2
%,
w
h
ile
YOL
Ov
8
s
p
r
o
v
i
d
es
a
b
alan
ce
b
etwe
en
p
er
f
o
r
m
an
c
e
an
d
ef
f
icie
n
cy
with
a
s
m
aller
f
ile
s
ize
an
d
f
ewe
r
lay
er
s
.
T
h
ese
ch
ar
ac
ter
is
tics
m
ak
e
YOL
Ov
8
s
u
itab
le
f
o
r
a
wid
e
r
an
g
e
o
f
s
ce
n
ar
io
s
,
f
r
o
m
r
eso
u
r
ce
-
c
o
n
s
tr
ain
ed
en
v
ir
o
n
m
en
ts
to
h
ig
h
-
p
er
f
o
r
m
a
n
ce
ap
p
licatio
n
s
.
T
h
e
o
v
er
all
ar
ch
itectu
r
e
o
f
th
e
YOL
Ov
8
m
o
d
el,
in
clu
d
i
n
g
its
b
a
ck
b
o
n
e
,
n
ec
k
,
an
d
h
ea
d
c
o
m
p
o
n
en
ts
,
is
illu
s
tr
ated
in
Fig
u
r
e
3
.
Fig
u
r
e
3
.
Ar
c
h
itectu
r
e
YOL
Ov
8
a
lg
o
r
ith
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
:
2
7
9
6
-
2
8
0
8
2800
2
.
3
.
Select
ing
h
y
perpa
ra
m
et
er
T
o
tr
ain
th
e
d
ef
ec
t
r
ec
o
g
n
itio
n
s
y
s
tem
o
n
PC
B
s
,
we
ap
p
lied
two
d
ataset
ap
p
r
o
ac
h
es:
n
o
n
-
a
u
g
m
en
ted
an
d
a
u
g
m
en
te
d
d
atasets
.
A
k
ey
s
tag
e
in
t
h
e
m
o
d
el
c
o
n
s
tr
u
ctio
n
p
r
o
ce
s
s
b
ef
o
r
e
tr
ai
n
in
g
is
s
ettin
g
th
e
h
y
p
er
p
ar
am
eter
s
,
as
th
ese
ca
n
co
n
s
id
er
ab
ly
af
f
ec
t
th
e
p
ac
e
o
f
tr
ain
in
g
c
o
n
v
e
r
g
en
ce
an
d
th
e
q
u
ality
o
f
th
e
r
esu
ltin
g
m
o
d
el.
T
ab
le
1
d
is
p
lay
s
th
e
s
elec
ted
h
y
p
er
p
ar
a
m
eter
s
u
tili
ze
d
in
tr
ai
n
in
g
t
h
e
f
aster
R
-
C
NN
an
d
YOL
Ov
8
alg
o
r
ith
m
s
in
t
h
is
r
esear
ch
.
I
n
ad
d
itio
n
to
th
e
s
elec
ted
h
y
p
er
p
a
r
am
eter
s
,
th
is
r
esear
ch
u
tili
ze
d
p
r
e
-
tr
ai
n
ed
weig
h
ts
f
o
r
m
o
d
el
in
itializatio
n
,
wh
ich
im
p
r
o
v
e
d
b
o
th
tr
ain
in
g
ef
f
icien
c
y
a
n
d
ac
cu
r
ac
y
i
n
d
e
f
ec
t
d
etec
tio
n
.
T
h
e
f
aster
R
-
C
NN
m
o
d
els
em
p
lo
y
ed
R
esNet5
0
FP
N
b
ac
k
b
o
n
es,
w
h
ile
YOL
Ov
8
in
clu
d
ed
f
iv
e
c
o
n
f
ig
u
r
at
io
n
s
r
an
g
i
n
g
f
r
o
m
YOL
Ov
8
n
(
6
.
2
MB)
to
YOL
Ov
8
x
(
1
3
1
MB).
Am
o
n
g
t
h
ese,
YOL
Ov
8
x
d
e
m
o
n
s
tr
ated
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
d
u
e
to
its
h
ig
h
er
m
o
d
el
ca
p
ac
i
ty
an
d
o
p
tim
ize
d
weig
h
t
co
n
f
ig
u
r
atio
n
s
.
B
y
s
elec
tin
g
r
elev
a
n
t
h
y
p
er
p
ar
am
eter
s
an
d
ap
p
ly
in
g
p
r
e
-
tr
ai
n
ed
weig
h
ts
,
we
s
tr
iv
e
to
im
p
r
o
v
e
b
o
th
th
e
tr
ain
in
g
s
p
ee
d
an
d
th
e
ac
c
u
r
ac
y
o
f
th
e
m
o
d
el,
en
s
u
r
in
g
th
e
h
ig
h
est p
o
s
s
ib
le
p
er
f
o
r
m
an
ce
in
d
ef
ec
t d
etec
tio
n
o
n
PC
B
lay
o
u
ts
.
T
ab
le
1
.
Settin
g
h
y
p
er
p
ar
am
et
er
alg
o
r
ith
m
H
y
p
e
r
p
a
r
a
me
t
e
r
F
a
st
e
r
R
-
C
N
N
Y
O
LO
v
8
I
n
p
u
t
i
m
a
g
e
s
i
z
e
6
4
0
×
6
4
0
(
p
x
)
6
4
0
×
6
4
0
(
p
x
)
Ep
o
c
h
s
50
1
0
0
Le
a
r
n
i
n
g
r
a
t
e
0
.
0
0
1
0
.
0
1
B
a
t
c
h
s
i
z
e
8
16
W
o
r
k
e
r
s
4
8
D
e
v
i
c
e
C
u
d
a
C
u
d
a
W
e
i
g
h
t
s
d
e
c
a
y
N
u
l
l
0
.
0
0
0
5
3.
RE
S
E
ARCH
M
E
T
H
O
D
3
.
1
.
T
he
re
s
ea
rc
h sta
g
e
T
h
is
r
esear
ch
was
u
n
d
e
r
tak
e
n
th
r
o
u
g
h
a
s
er
ies
o
f
s
tag
es
m
ea
n
t
to
test
th
e
e
f
f
icac
y
o
f
th
e
f
aster
R
-
C
NN
an
d
YOL
Ov
8
alg
o
r
it
h
m
s
in
f
in
d
in
g
d
ef
ec
ts
in
PC
B
lay
o
u
ts
.
T
h
e
r
esear
c
h
p
r
o
ce
s
s
is
s
u
m
m
ar
ized
in
th
e
f
lo
wch
ar
t
s
h
o
wn
in
Fig
u
r
e
4
.
T
h
e
ap
p
r
o
ac
h
b
eg
i
n
s
with
a
d
etailed
liter
atu
r
e
r
esear
c
h
,
f
o
c
u
s
in
g
o
n
b
o
th
d
ef
ec
t
d
etec
tio
n
u
tili
zin
g
f
aste
r
R
-
C
NN
an
d
YOL
Ov
8
alg
o
r
i
th
m
s
an
d
g
en
e
r
ic
d
e
f
ec
t
id
e
n
tific
atio
n
tech
n
iq
u
es
f
o
r
PC
B
lay
o
u
ts
.
T
h
is
f
o
u
n
d
atio
n
al
r
ev
iew
was
cr
itical
f
o
r
co
m
p
r
eh
en
d
in
g
t
h
e
co
r
e
t
h
eo
r
ies
an
d
r
elate
d
ad
v
an
ce
m
e
n
ts
,
d
ir
ec
tin
g
th
e
e
n
s
u
in
g
ex
p
er
im
en
tal
d
esig
n
[
1
9
]
.
T
h
e
d
ataset
f
o
r
th
is
r
esear
ch
,
s
o
u
r
ce
d
f
r
o
m
Hu
a
n
g
et
a
l.
[
2
0
]
,
c
o
n
s
is
t
o
f
1
0
PC
B
s
with
6
9
3
im
a
g
es,
r
ep
r
esen
tin
g
s
ix
d
ef
ec
t
ty
p
es,
in
clu
d
in
g
m
is
s
in
g
h
o
le,
m
o
u
s
e
b
ite,
s
h
o
r
t,
o
p
en
cir
cu
it,
s
p
u
r
,
an
d
s
p
u
r
io
u
s
co
p
p
er
.
I
m
ag
es
wer
e
m
an
u
all
y
lab
eled
u
s
in
g
R
o
b
o
f
lo
w
to
o
ls
,
g
en
er
atin
g
.
x
m
l
f
iles
f
o
r
f
as
ter
R
-
C
NN
an
d
.
tx
t
f
iles
f
o
r
YOL
Ov
8
.
T
h
e
lab
el
ed
d
ata
was
d
iv
id
ed
in
to
tr
ai
n
in
g
(
8
5
%),
v
alid
atio
n
(
1
0
%)
,
an
d
test
in
g
(
5
%)
s
u
b
s
ets.
Data
au
g
m
en
tatio
n
te
ch
n
iq
u
es,
s
u
c
h
as
90
-
d
e
g
r
ee
r
o
tatio
n
s
,
s
m
all
-
an
g
le
r
o
tatio
n
s
(
-
15
o
to
+
1
5
o
)
,
an
d
h
o
r
izo
n
tal/v
e
r
tical
s
h
ea
r
s
,
wer
e
ap
p
lied
to
en
h
a
n
ce
d
ataset
d
iv
er
s
ity
an
d
r
ed
u
ce
o
v
er
f
itti
n
g
.
Sep
ar
ate
t
r
ain
in
g
s
ess
io
n
s
wer
e
co
n
d
u
cted
f
o
r
Fas
ter
R
-
C
NN
an
d
YOL
Ov
8
u
s
in
g
au
g
m
e
n
ted
an
d
n
o
n
-
a
u
g
m
en
ted
d
atasets
.
Mo
d
el
p
er
f
o
r
m
an
ce
was
ev
alu
ated
u
s
in
g
a
co
n
f
u
s
io
n
m
atr
i
x
,
f
o
cu
s
in
g
o
n
k
e
y
m
etr
ics
lik
e
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e
,
I
o
U,
an
d
m
AP.
T
h
e
s
e
ev
alu
atio
n
s
,
co
m
b
in
ed
with
tr
ain
in
g
tim
es
an
d
d
etec
tio
n
s
p
ee
d
s
,
p
r
o
v
id
ed
a
th
o
r
o
u
g
h
ass
ess
m
en
t o
f
ea
ch
a
lg
o
r
ith
m
'
s
ca
p
ab
ilit
ies f
o
r
PC
B
d
ef
ec
t
d
etec
tio
n
[
2
1
]
,
[
2
2
]
.
Fig
u
r
e
4
.
T
h
e
p
r
o
ce
d
u
r
e
r
esea
r
ch
3
.
2
.
Da
t
a
s
et
c
o
llect
io
n
I
n
th
is
r
esear
c
h
,
th
e
d
ataset
f
o
r
d
ef
ec
t
d
etec
tio
n
in
PC
B
lay
o
u
ts
co
m
p
r
is
es
6
9
3
im
a
g
es
ca
p
tu
r
ed
with
a
1
6
-
m
eg
ap
i
x
el
HD
in
d
u
s
tr
ial
ca
m
er
a
eq
u
i
p
p
ed
with
a
co
m
p
lem
en
tar
y
m
etal
-
o
x
id
e
-
s
em
ic
o
n
d
u
ct
o
r
(
C
MO
S
)
s
en
s
o
r
.
T
h
e
o
r
ig
in
al
r
eso
lu
tio
n
o
f
4
6
0
8
×3
4
5
6
p
ix
els
was
r
e
s
ized
to
6
4
0
×6
4
0
p
ix
els
to
s
tan
d
ar
d
ize
i
n
p
u
t
s
ize
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
Ob
ject
d
etec
tio
n
in
p
r
in
ted
ci
r
cu
it b
o
a
r
d
q
u
a
lity c
o
n
tr
o
l:
c
o
mp
a
r
in
g
a
lg
o
r
ith
ms
…
(
Ja
ja
K
u
s
tija
)
2801
f
o
r
ea
ch
m
eth
o
d
[
2
0
]
.
T
h
is
r
eso
lu
tio
n
was
ch
o
s
en
b
ased
o
n
p
r
elim
in
ar
y
ex
p
e
r
im
en
ts
,
b
alan
cin
g
co
m
p
u
tatio
n
al
ef
f
icien
c
y
with
th
e
p
r
eser
v
atio
n
o
f
ess
en
tial
f
ea
tu
r
es
f
o
r
ac
cu
r
ate
d
e
f
ec
t
d
et
ec
tio
n
.
T
h
e
d
ataset
in
clu
d
es
s
ix
d
ef
ec
t
t
y
p
es,
as
s
u
m
m
ar
ized
i
n
T
ab
le
2
,
t
h
e
im
ag
es
ar
e
d
is
tr
ib
u
ted
ev
en
l
y
ac
r
o
s
s
d
ef
ec
t
class
es,
en
co
m
p
ass
in
g
2
,
5
9
3
d
ef
ec
t
p
o
in
ts
.
T
h
is
b
alan
ce
d
d
is
tr
ib
u
tio
n
is
cr
u
cial
f
o
r
tr
ain
in
g
a
r
o
b
u
s
t
m
o
d
el
ca
p
ab
le
o
f
g
en
er
alizin
g
e
f
f
ec
tiv
ely
ac
r
o
s
s
all
d
ef
ec
t ty
p
es.
T
ab
le
2
.
Dis
tr
ib
u
tio
n
o
f
n
u
m
b
er
o
f
im
a
g
es a
n
d
d
e
f
ec
t p
o
i
n
ts
o
f
HR
I
PC
B
d
atase
t
R
e
f
e
r
e
n
c
e
P
C
B
A
d
j
u
st
e
d
s
i
z
e
D
e
f
e
c
t
t
y
p
e
M
i
ss
i
n
g
h
o
l
e
M
o
u
s
e
b
i
t
e
O
p
e
n
c
i
r
c
u
i
t
S
h
o
r
t
S
p
u
r
S
p
u
r
i
o
u
s
c
o
p
p
e
r
I
mag
e
1
6
4
0
×
6
4
0
20
20
20
20
20
20
I
mag
e
4
6
4
0
×
6
4
0
20
20
20
20
20
20
I
mag
e
5
6
4
0
×
6
4
0
10
10
10
10
10
10
I
mag
e
6
6
4
0
×
6
4
0
10
10
10
10
10
10
I
mag
e
7
6
4
0
×
6
4
0
10
10
10
10
10
10
I
mag
e
8
6
4
0
×
6
4
0
10
10
10
10
10
10
I
mag
e
9
6
4
0
×
6
4
0
10
10
10
10
10
10
I
mag
e
1
0
6
4
0
×
6
4
0
5
5
6
6
5
6
I
mag
e
1
1
6
4
0
×
6
4
0
10
10
10
10
10
10
I
mag
e
1
2
6
4
0
×
6
4
0
10
10
10
10
10
10
N
u
mb
e
r
i
m
a
g
e
(
a
mo
u
n
t
o
f
d
e
f
e
c
t
)
1
1
5
(
4
9
7
)
1
1
5
(
4
9
2
)
1
1
6
(
4
8
2
)
1
1
6
(
4
9
1
)
1
1
5
(
4
8
8
)
1
1
6
(
5
0
3
)
6
9
3
(
2
5
9
3
)
3
.
3
.
O
v
er
a
ll
s
y
s
t
em
c
o
nfig
ura
t
io
n
T
h
e
p
ip
elin
e
o
f
th
e
PC
B
d
ef
ec
t
d
etec
tio
n
s
y
s
tem
p
r
o
p
o
s
ed
in
th
is
r
esear
ch
is
illu
s
tr
ated
i
n
Fig
u
r
e
5
.
T
h
e
p
r
o
ce
s
s
b
eg
in
s
with
p
r
e
-
p
r
o
ce
s
s
in
g
,
wh
er
e
th
e
PC
B
d
ataset
is
s
tan
d
ar
d
ized
u
s
in
g
a
n
au
to
-
o
r
ien
tatio
n
tech
n
iq
u
e
to
alig
n
all
im
ag
es
ac
co
r
d
in
g
to
d
e
f
in
ed
s
tan
d
a
r
d
s
[
2
3
]
.
T
h
e
im
ag
es
ar
e
th
en
r
esized
to
6
4
0
×
6
4
0
p
ix
els,
b
alan
cin
g
c
o
m
p
u
tatio
n
al
ef
f
icien
c
y
with
th
e
r
eso
lu
tio
n
r
eq
u
ir
e
d
of
ac
c
u
r
ate
d
e
f
ec
t
d
etec
tio
n
.
Data
au
g
m
en
tatio
n
tech
n
iq
u
es,
s
u
c
h
as
r
o
tatio
n
s
an
d
s
h
ea
r
in
g
,
a
r
e
ap
p
lied
to
en
h
a
n
ce
d
ataset
d
iv
er
s
ity
,
s
im
u
latin
g
r
ea
l
-
wo
r
ld
v
a
r
iatio
n
s
in
PC
B
m
an
u
f
ac
tu
r
in
g
a
n
d
in
s
p
ec
tio
n
[
2
4
]
.
Fo
llo
win
g
p
r
e
-
p
r
o
ce
s
s
in
g
a
n
d
au
g
m
en
tatio
n
,
th
e
d
ataset
is
u
s
ed
to
tr
ai
n
two
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
,
f
aster
R
-
C
NN
an
d
YOL
Ov
8
.
B
o
th
alg
o
r
ith
m
s
ar
e
tr
ain
ed
o
n
a
u
g
m
en
te
d
a
n
d
n
o
n
-
au
g
m
en
ted
d
atasets
to
ev
alu
ate
th
eir
p
er
f
o
r
m
an
ce
u
n
d
er
d
if
f
e
r
en
t
co
n
d
itio
n
s
.
T
h
is
ap
p
r
o
ac
h
p
r
o
v
id
e
s
in
s
ig
h
ts
in
to
ea
ch
m
o
d
el'
s
s
tr
en
g
th
s
an
d
lim
itatio
n
s
in
h
an
d
lin
g
d
ata
v
ar
iab
ili
ty
.
On
ce
tr
ain
ed
,
th
e
m
o
d
els
p
r
ed
ict
d
ef
ec
ts
in
PC
B
lay
o
u
ts
,
g
e
n
er
atin
g
a
d
e
f
ec
t
d
etec
tio
n
m
ap
.
T
h
is
m
a
p
h
ig
h
lig
h
ts
d
etec
ted
d
ef
ec
ts
b
ased
o
n
p
r
e
d
ef
in
ed
th
r
esh
o
ld
s
,
clea
r
ly
in
d
icatin
g
wh
eth
er
a
PC
B
is
d
ef
ec
tiv
e
o
r
n
o
r
m
al.
Su
ch
o
u
tp
u
ts
ar
e
c
r
itical
f
o
r
p
r
ac
tical
ap
p
licatio
n
s
,
en
s
u
r
in
g
q
u
ic
k
a
n
d
r
eliab
le
d
e
f
ec
t id
en
tific
atio
n
in
r
ea
l
-
wo
r
ld
s
ce
n
ar
io
s
[
2
5
]
.
Fig
u
r
e
5
.
Ov
e
r
all
s
y
s
tem
co
n
f
i
g
u
r
atio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
:
2
7
9
6
-
2
8
0
8
2802
3
.
4
.
P
re
-
pro
ce
s
s
ing
da
t
a
s
et
Data
p
r
ep
ar
atio
n
was
u
n
d
er
ta
k
en
with
two
p
r
im
ar
y
o
b
jecti
v
es,
to
c
o
r
r
ec
t
im
a
g
e
o
r
ien
tat
io
n
an
d
to
n
o
r
m
alize
im
ag
e
d
im
en
s
io
n
.
T
h
e
“
A
u
to
-
Ori
en
t
”
f
ea
tu
r
e
was u
tili
ze
d
to
au
to
m
atica
lly
f
ix
th
e
o
r
ien
tatio
n
b
ased
o
n
E
XI
F
m
etad
ata,
en
s
u
r
in
g
u
n
if
o
r
m
v
is
u
al
alig
n
m
en
t
a
cr
o
s
s
all
p
h
o
to
s
.
Su
b
s
eq
u
e
n
tly
,
all
p
h
o
to
g
r
ap
h
s
wer
e
d
o
wn
s
ized
to
6
4
0
×
6
4
0
p
ix
els
u
s
in
g
th
e
“
R
esiz
e
I
ma
g
e
”
t
o
o
l,
s
tan
d
ar
d
izin
g
th
eir
p
r
o
p
o
r
ti
o
n
s
f
o
r
th
e
t
r
ain
in
g
p
r
o
ce
s
s
.
T
h
ese
p
r
e
p
ar
atio
n
s
te
p
s
ar
e
cr
itical
f
o
r
e
n
s
u
r
in
g
th
a
t
th
e
d
ataset
is
co
n
s
is
ten
t
an
d
s
u
itab
le
f
o
r
m
o
d
el
tr
ain
in
g
.
Pro
p
er
o
r
ien
tatio
n
a
n
d
s
ize
u
n
if
o
r
m
ity
ar
e
cr
itical
f
o
r
d
eliv
er
in
g
ac
cu
r
ate
an
d
d
ep
en
d
a
b
le
tr
ain
in
g
o
u
tco
m
es
[
2
6
]
.
Step
1:
Au
to
m
atica
lly
c
o
r
r
ec
t
im
ag
e
o
r
ien
tatio
n
,
t
h
e
A
u
to
-
Ori
en
t
f
u
n
ctio
n
id
e
n
tifie
s
a
n
d
c
o
r
r
ec
ts
im
a
g
e
o
r
ien
tatio
n
b
ased
o
n
E
XI
F
m
etad
ata
to
en
s
u
r
e
v
is
u
al
co
n
s
is
ten
cy
.
T
h
e
im
ag
e
is
r
o
tated
as
r
eq
u
ir
ed
b
ased
o
n
th
e
r
ec
o
r
d
ed
o
r
ien
tat
io
n
o
f
1
8
0
o
,
2
7
0
o
,
an
d
9
0
o
.
Step
2:
No
r
m
alize
im
ag
e
d
im
en
s
io
n
s
,
all
im
ag
es
in
th
e
d
ataset
ar
e
r
esized
to
6
4
0
×
6
4
0
p
ix
els
u
s
in
g
th
e
f
u
n
ctio
n
R
esiz
e
I
ma
g
e
,
to
en
s
u
r
e
s
ize
u
n
if
o
r
m
ity
b
ef
o
r
e
tr
ain
i
n
g
th
e
n
eu
r
al
n
etwo
r
k
m
o
d
el.
3
.
5
.
Da
t
a
s
et
a
ug
m
ent
a
t
io
n
T
r
ain
in
g
n
e
u
r
al
n
etwo
r
k
m
o
d
els
o
f
ten
r
eq
u
ir
es
lar
g
e
-
s
ca
le
d
atasets
d
u
e
to
th
e
lar
g
e
co
m
p
lex
ity
o
f
m
o
d
el
p
ar
a
m
eter
s
[
2
7
]
.
Ho
we
v
er
,
p
r
o
d
u
ct
d
ef
ec
ts
d
u
r
in
g
th
e
m
an
u
f
ac
tu
r
in
g
p
r
o
ce
s
s
ten
d
to
b
e
r
ar
e
an
d
t
h
e
v
ar
iety
o
f
d
ef
ec
t
ty
p
es
m
ay
c
h
an
g
e
d
u
r
in
g
m
ass
p
r
o
d
u
cti
o
n
.
I
n
th
is
r
esear
ch
,
th
e
u
s
e
o
f
u
n
-
au
g
m
e
n
ted
an
d
au
g
m
en
ted
d
atasets
is
tak
en
in
to
co
n
s
id
er
atio
n
to
d
eter
m
in
e
t
h
e
lev
el
o
f
p
er
f
o
r
m
a
n
ce
an
d
in
f
lu
en
ce
with
th
ese
two
m
eth
o
d
s
.
W
e
a
p
p
ly
a
u
g
m
en
tatio
n
tech
n
iq
u
es,
9
0
o
r
o
t
ate
(
clo
ck
wis
e,
co
u
n
ter
-
clo
c
k
wis
e,
u
p
s
id
e
d
o
wn
)
,
r
o
tatio
n
(
b
etwe
en
-
1
5
°
an
d
+1
5
°),
an
d
s
h
ea
r
(
±
1
0
°
h
o
r
iz
o
n
tal
,
±
1
0
°
v
er
tical
)
.
C
lo
ck
w
is
e,
ea
ch
im
ag
e
is
r
o
tated
9
0
o
cl
o
ck
wis
e
to
s
im
u
late
a
ch
an
g
e
in
o
r
ien
tatio
n
.
C
o
u
n
ter
-
cl
o
ck
wis
e
,
a
s
im
ilar
p
r
o
ce
s
s
is
p
er
f
o
r
m
ed
b
u
t
in
th
e
o
p
p
o
s
ite
d
ir
ec
tio
n
t
o
ad
d
o
r
ien
tatio
n
v
ar
iatio
n
.
U
p
s
id
e
d
o
wn
,
th
e
im
ag
e
is
r
o
tated
b
y
1
8
0
o
to
f
u
lly
cr
ea
te
th
e
b
est
im
ag
e
c
o
n
d
itio
n
s
,
ad
d
in
g
m
o
r
e
o
r
ien
tatio
n
v
a
r
iatio
n
s
.
T
h
en
,
r
an
d
o
m
r
o
tatio
n
b
etwe
en
-
15
o
a
n
d
+1
5
o
to
s
im
u
late
p
o
s
itio
n
al
im
p
er
f
ec
tio
n
s
th
at
m
ay
o
cc
u
r
d
u
r
in
g
t
h
e
p
r
o
d
u
ctio
n
o
r
i
n
s
p
ec
tio
n
p
r
o
ce
s
s
.
Fu
r
th
er
m
o
r
e
,
s
h
ea
r
au
g
m
e
n
tatio
n
is
ca
r
r
ied
o
u
t
to
s
im
u
late
th
e
ef
f
ec
ts
o
f
p
r
ess
u
r
e
o
r
ten
s
io
n
o
n
th
e
PC
B
wh
ich
ca
n
af
f
ec
t th
e
s
h
ap
e
o
r
r
elativ
e
p
o
s
itio
n
o
f
c
o
m
p
o
n
en
t
s
an
d
d
ef
ec
ts
.
3
.
6
.
Da
t
a
s
et
s
pli
t
t
ing
Fo
llo
win
g
p
r
e
p
r
o
ce
s
s
in
g
a
n
d
d
ata
au
g
m
e
n
tatio
n
,
th
e
d
ataset
was
s
ep
ar
ated
in
to
tr
ain
in
g
,
v
alid
atio
n
,
an
d
test
in
g
s
ets.
T
h
e
s
p
lit
was
ac
co
m
p
lis
h
ed
p
r
o
p
o
r
tio
n
ally
,
8
5
%
f
o
r
tr
ain
in
g
,
1
0
%
f
o
r
v
ali
d
atio
n
,
an
d
5
%
f
o
r
test
in
g
.
T
ab
le
3
s
h
o
wn
th
e
d
is
tr
ib
u
tio
n
o
f
p
h
o
to
s
an
d
d
ef
e
cts
ac
r
o
s
s
th
ese
s
u
b
s
et
s
f
o
r
b
o
th
au
g
m
en
te
d
an
d
non
-
a
u
g
m
en
te
d
d
ataset.
T
h
e
p
r
o
p
o
r
ti
o
n
8
5
/1
0
/5
s
p
lit
is
a
g
e
n
er
ally
estab
lis
h
ed
m
eth
o
d
in
m
ac
h
in
e
lear
n
in
g
,
g
iv
in
g
ad
e
q
u
ate
d
ata
f
o
r
tr
ain
i
n
g
wh
ile
g
u
ar
a
n
teein
g
th
at
th
e
v
alid
atio
n
an
d
test
in
g
s
ets
ar
e
r
ep
r
esen
tativ
e
o
f
th
e
wh
o
le
d
ataset.
T
h
is
d
iv
id
e
s
u
p
p
o
r
ts
r
o
b
u
s
t m
o
d
el
e
v
alu
at
io
n
an
d
r
ed
u
ce
s
th
e
p
o
s
s
ib
ilit
y
o
f
o
v
er
f
itti
n
g
.
T
ab
le
3
.
Sp
litt
in
g
n
o
n
-
a
u
g
m
e
n
ted
d
ataset
an
d
au
g
m
en
ted
d
at
aset
N
o
n
-
a
u
g
me
n
t
a
t
i
o
n
A
u
g
m
e
n
t
a
t
i
o
n
N
u
mb
e
r
o
f
i
m
a
g
e
s
N
u
mb
e
r
o
f
d
e
f
e
c
t
s
N
u
mb
e
r
o
f
i
m
a
g
e
s
N
u
mb
e
r
o
f
d
e
f
e
c
t
s
Tr
a
i
n
5
8
9
2
1
4
8
1
7
6
7
7
5
2
3
V
a
l
i
d
69
2
9
3
69
2
9
3
Te
st
35
1
5
2
35
1
5
2
To
t
a
l
6
9
3
2
5
9
3
1
8
7
1
7
9
6
8
3
.
7
.
P
er
f
o
rma
nce
e
v
a
lua
t
io
n
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
d
ef
e
ct
d
etec
tio
n
m
o
d
els
was
ev
alu
ated
u
s
in
g
a
c
o
n
f
u
s
io
n
m
atr
ix
,
wh
ich
p
r
o
v
id
es
in
s
ig
h
ts
in
to
th
e
m
o
d
els’
p
r
ed
ictio
n
s
ac
r
o
s
s
s
ix
d
ef
ec
t
class
e
s
,
in
clu
d
in
g
m
is
s
i
n
g
h
o
le,
m
o
u
s
e
b
ite,
s
h
o
r
t,
o
p
e
n
cir
cu
it,
s
p
u
r
,
s
p
u
r
io
u
s
co
p
p
e
r
.
T
h
e
c
o
n
f
u
s
io
n
m
atr
ix
is
a
tab
le
with
f
o
u
r
c
ells
th
at
s
h
o
w
th
e
n
u
m
b
er
o
f
ac
cu
r
ate
a
n
d
e
r
r
o
n
e
o
u
s
g
u
ess
es
f
o
r
ea
ch
item
ty
p
e
.
T
h
e
co
n
f
u
s
io
n
m
atr
ix
h
as
ce
l
ls
f
o
r
tr
u
e
p
o
s
itiv
e
(
T
P),
tr
u
e
n
eg
ativ
e
(
T
N)
,
f
alse p
o
s
i
tiv
e
(
FP
)
,
an
d
f
alse n
eg
ati
v
e
(
FN)
.
K
e
y
m
e
t
r
i
c
s
s
u
c
h
a
s
m
e
a
n
a
v
e
r
a
g
e
p
r
e
c
i
s
i
o
n
(
m
A
P
)
,
p
r
e
c
i
s
i
o
n
,
r
e
c
a
l
l
,
a
n
d
F
1
-
s
c
o
r
e
w
e
r
e
d
e
r
i
v
e
d
f
r
o
m
t
h
e
c
o
n
f
u
s
i
o
n
m
at
r
i
x
.
P
r
e
c
is
i
o
n
m
e
a
s
u
r
es
t
h
e
p
r
o
p
o
r
t
i
o
n
o
f
c
o
r
r
e
c
t
l
y
i
d
e
n
t
i
f
i
e
d
p
o
s
i
ti
v
e
i
n
s
t
a
n
c
es
,
w
h
i
l
e
r
e
c
a
ll
e
v
a
l
u
a
t
es
t
h
e
m
o
d
e
l
’
s
a
b
i
l
it
y
t
o
d
e
t
e
c
t
a
ll
r
e
l
e
v
a
n
t
p
o
s
i
ti
v
e
i
n
s
t
a
n
c
es
.
T
h
e
F
1
-
s
c
o
r
e
c
o
m
b
i
n
e
s
p
r
e
c
is
i
o
n
a
n
d
r
e
c
a
l
l
t
o
p
r
o
v
i
d
e
a
b
al
a
n
c
e
d
a
s
s
e
s
s
m
e
n
t
o
f
t
h
e
m
o
d
el
’
s
p
e
r
f
o
r
m
a
n
c
e
.
A
d
d
i
t
i
o
n
a
ll
y
,
m
AP
u
t
i
li
z
e
s
t
h
e
i
n
t
e
r
s
e
c
ti
o
n
o
v
e
r
u
n
i
o
n
(
I
o
U
)
t
o
e
v
a
l
u
a
t
e
t
h
e
s
i
m
il
a
r
i
t
y
b
et
w
e
e
n
p
r
e
d
i
ct
e
d
a
n
d
g
r
o
u
n
d
t
r
u
t
h
b
o
u
n
d
i
n
g
b
o
x
e
s
[
2
8
]
.
@
=
1
∑
=
1
(
4
)
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
Ob
ject
d
etec
tio
n
in
p
r
in
ted
ci
r
cu
it b
o
a
r
d
q
u
a
lity c
o
n
tr
o
l:
c
o
mp
a
r
in
g
a
lg
o
r
ith
ms
…
(
Ja
ja
K
u
s
tija
)
2803
=
+
(
5
)
=
+
(
6
)
1
−
=
2
∗
(
∗
)
(
+
)
(
7
)
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
4
.
1
.
P
er
f
o
rm
re
s
ult
wit
h no
n
-
a
ug
m
ent
a
t
io
n da
t
a
s
et
T
h
e
o
b
jectiv
e
o
f
th
is
in
v
es
tig
atio
n
is
to
ascer
tain
t
h
e
o
p
tim
al
m
o
d
els
an
d
p
r
o
ce
d
u
r
es
f
o
r
un
-
au
g
m
en
ted
o
r
u
n
m
o
d
if
ied
d
atasets
.
T
ab
le
4
s
h
o
w
as
s
u
m
m
ar
ized
d
ep
icts
th
e
p
e
r
f
o
r
m
a
n
ce
an
al
y
s
is
o
u
tco
m
es
o
f
b
o
th
alg
o
r
ith
m
s
o
n
v
ar
i
o
u
s
m
o
d
el
v
ar
iatio
n
s
u
tili
zin
g
an
u
n
alter
e
d
d
ataset
m
eth
o
d
o
l
o
g
y
.
T
h
e
ex
am
in
ed
m
etr
ics
co
n
s
is
t
o
f
m
AP@
5
0
,
m
AP@
5
0
:9
5
,
p
r
ec
is
io
n
,
r
ec
all
,
a
n
d
F1
-
Sco
r
e
f
o
r
ea
ch
s
etu
p
o
f
th
e
m
o
d
el.
T
ab
le
4
.
B
est
p
er
f
o
r
m
m
o
d
el
with
n
o
n
-
a
u
g
m
en
tatio
n
d
atase
t
M
o
d
e
l
V
a
r
i
a
n
mA
P
@
5
0
mA
P
@
5
0
:
9
5
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
S
c
o
r
e
F
a
st
e
r
R
-
C
N
N
Re
sN
e
t
5
0
FP
N
0
.
7
5
5
0
.
3
0
2
0
.
2
1
6
0
.
3
2
6
0
.
2
5
7
Re
sN
e
t
5
0
FP
N
v
2
0
.
7
6
3
0
.
3
2
5
0
.
2
4
3
0
.
3
4
6
0
.
2
8
3
Y
O
LO
v
8
n
0
.
8
7
3
0
.
4
1
8
0
.
8
8
4
0
.
8
4
8
0
.
8
6
6
v
8
s
0
.
9
2
4
0
.
4
6
1
0
.
9
5
1
0
.
8
9
4
0
.
9
2
1
v
8
m
0
.
9
3
7
0
.
4
8
6
0
.
9
5
4
0
.
9
0
1
0
.
9
2
7
v
8
l
0
.
9
6
3
0
.
5
1
8
0
.
9
4
6
0
.
9
2
5
0
.
9
3
5
v
8
x
0
.
9
4
9
0
.
4
9
1
0
.
9
5
6
0
.
9
2
0
0
.
9
3
8
Fas
ter
R
-
C
NN
with
th
e
R
e
s
Net5
0
FP
N
v
2
b
ac
k
b
o
n
e
ac
h
iev
ed
a
m
AP@
5
0
o
f
0
.
7
6
3
an
d
a
m
AP@
5
0
:9
5
o
f
0
.
3
2
5
,
m
ak
in
g
it
th
e
b
est
-
p
er
f
o
r
m
in
g
v
ar
ia
n
t
am
o
n
g
Fas
ter
R
-
C
NN
m
o
d
els.
I
n
co
n
tr
ast,
th
e
YOL
Ov
8
m
o
d
els
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
e
d
f
aster
R
-
C
NN,
with
YOL
Ov
8
l
ac
h
iev
in
g
th
e
h
ig
h
est
m
AP@
5
0
o
f
0
.
9
6
3
an
d
m
AP@
5
0
:9
5
o
f
0
.
5
1
8
.
T
h
ese
r
esu
lts
em
p
h
asize
YOL
Ov
8
l
s
u
p
er
io
r
a
b
ilit
y
to
d
etec
t
PC
B
d
ef
ec
ts
,
s
h
o
wca
s
in
g
h
ig
h
e
r
ac
cu
r
ac
y
an
d
r
o
b
u
s
tn
ess
ac
r
o
s
s
all
m
etr
ics.
T
h
e
f
in
d
in
g
s
alig
n
with
p
r
io
r
r
esear
ch
,
s
u
ch
as
Yu
s
r
o
et
a
l.
[
2
9
]
,
wh
ic
h
h
ig
h
lig
h
ted
Y
OL
O
-
b
ased
m
o
d
els
as
th
e
p
r
ef
er
r
ed
ch
o
i
ce
f
o
r
r
ea
l
-
tim
e
o
b
ject
d
etec
tio
n
d
u
e
to
th
eir
s
p
ee
d
an
d
p
r
ec
is
io
n
.
T
h
is
s
tu
d
y
ex
ten
d
s
th
o
s
e
f
in
d
in
g
s
,
d
e
m
o
n
s
tr
atin
g
YOL
Ov
8
l
n
o
tab
le
ad
v
an
tag
es
in
g
en
er
aliza
tio
n
,
p
ar
ticu
lar
l
y
f
o
r
d
etec
tin
g
s
m
all
d
ef
ec
ts
.
T
h
e
r
esu
lts
u
n
d
er
s
co
r
e
YOL
Ov
8
l
ef
f
icien
cy
a
n
d
s
u
itab
ilit
y
f
o
r
PC
B
s
u
r
f
ac
e
f
law
d
etec
tio
n
task
s
,
esp
ec
ially
in
s
ce
n
ar
io
s
r
eq
u
i
r
in
g
h
ig
h
p
r
ec
is
io
n
an
d
s
p
ee
d
.
4
.
2
.
P
er
f
o
rm
re
s
ult
wit
h a
ug
m
ent
a
t
io
n da
t
a
s
et
Af
ter
an
aly
zin
g
th
e
p
e
r
f
o
r
m
a
n
ce
o
n
th
e
d
ataset
with
o
u
t
a
u
g
m
en
tatio
n
,
it
is
cr
u
cial
to
ex
am
in
e
h
o
w
th
e
two
alg
o
r
ith
m
s
r
esp
o
n
d
t
o
th
e
co
n
d
itio
n
s
o
f
an
ex
p
a
n
d
ed
d
ataset.
T
h
is
co
m
p
ar
is
o
n
o
f
f
e
r
s
v
ital
in
s
ig
h
ts
in
to
th
e
r
esil
ien
ce
an
d
ad
ap
tab
ilit
y
o
f
ea
c
h
alg
o
r
ith
m
u
n
d
er
i
n
cr
ea
s
in
g
ly
c
o
m
p
licate
d
an
d
v
ar
ie
d
d
ata
cir
cu
m
s
tan
ce
s
.
T
ab
le
5
s
h
o
w
th
e
p
er
f
o
r
m
a
n
ce
o
f
b
o
th
alg
o
r
ith
m
s
wh
en
ass
ess
ed
o
n
th
e
au
g
m
en
ted
d
ataset,
u
s
in
g
th
e
s
am
e
p
er
f
o
r
m
a
n
ce
a
s
s
es
s
m
en
t m
etr
ics as in
th
e
n
o
n
-
au
g
m
e
n
ted
d
ataset
r
esear
ch
.
T
ab
le
5
.
B
est
p
er
f
o
r
m
m
o
d
el
with
au
g
m
en
tatio
n
d
ataset
M
o
d
e
l
V
a
r
i
a
n
mA
P
@
5
0
mA
P
@
5
0
:
9
5
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
S
c
o
r
e
F
a
st
e
r
R
-
C
N
N
Re
sN
e
t
5
0
FP
N
0
.
7
5
4
0
.
2
7
5
0
.
2
8
6
0
.
3
7
2
0
.
3
2
3
Re
sN
e
t
5
0
FP
N
v
2
0
.
8
2
6
0
.
3
3
8
0
.
3
1
6
0
.
4
1
1
0
.
3
5
7
Y
O
LO
v
8
n
0
.
9
1
4
0
.
4
3
4
0
.
9
4
8
0
.
8
4
1
0
.
8
9
2
v
8
s
0
.
9
2
7
0
.
4
8
8
0
.
9
3
9
0
.
9
0
6
0
.
9
2
2
v
8
m
0
.
9
4
9
0
.
4
8
1
0
.
9
4
9
0
.
9
1
6
0
.
9
3
2
v
8
l
0
.
9
5
1
0
.
4
8
0
0
.
9
5
4
0
.
9
0
9
0
.
9
3
1
v
8
x
0
.
9
6
2
0
.
5
0
3
0
.
9
5
3
0
.
9
4
5
0
.
9
4
9
On
th
e
au
g
m
en
ted
d
ataset,
f
aster
R
-
C
NN
with
th
e
R
esNet5
0
FP
N
v
2
b
ac
k
b
o
n
e
ac
h
iev
e
d
its
h
ig
h
est
m
AP@
5
0
s
co
r
e
o
f
0
.
8
2
6
an
d
m
AP@
5
0
:9
5
o
f
0
.
3
3
8
,
in
d
icat
in
g
an
im
p
r
o
v
em
en
t
co
m
p
a
r
e
d
to
its
p
er
f
o
r
m
an
ce
o
n
n
o
n
-
au
g
m
e
n
ted
d
ata.
T
h
is
d
em
o
n
s
tr
ates
th
at
au
g
m
en
tatio
n
en
h
an
ce
s
its
ab
ilit
y
to
d
et
ec
t
d
ef
ec
ts
o
n
PC
B
s
u
r
f
ac
es.
Me
an
wh
ile,
YOL
Ov
8
x
s
h
o
we
d
th
e
b
est
o
v
er
all
p
er
f
o
r
m
a
n
ce
,
ac
h
iev
in
g
a
m
AP@
5
0
o
f
0
.
9
6
2
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
3
,
J
u
n
e
20
25
:
2
7
9
6
-
2
8
0
8
2804
m
AP@
5
0
:9
5
o
f
0
.
5
0
3
,
f
u
r
th
er
s
o
lid
if
y
in
g
its
p
o
s
itio
n
as
th
e
m
o
s
t
ef
f
ec
tiv
e
m
o
d
el
f
o
r
f
au
lt
d
etec
tio
n
o
n
PC
B
lay
o
u
ts
.
T
h
e
YOL
Ov
8
x
m
o
d
el
o
u
t
p
er
f
o
r
m
ed
f
aster
R
-
C
NN
ac
r
o
s
s
all
m
etr
ics,
s
h
o
wca
s
in
g
s
u
p
er
io
r
r
o
b
u
s
tn
ess
an
d
v
er
s
atility
wh
en
tr
ain
ed
o
n
h
eter
o
g
en
eo
u
s
d
ata.
T
h
e
s
ig
n
if
ican
t
im
p
r
o
v
em
e
n
t
in
m
AP@
5
0
:9
5
h
ig
h
lig
h
ts
YOL
Ov
8
x
a
b
ilit
y
t
o
g
e
n
er
alize
b
etter
,
p
a
r
ticu
lar
l
y
f
o
r
d
etec
tin
g
s
m
aller
d
ef
ec
ts
.
T
h
ese
f
in
d
in
g
s
ar
e
co
n
s
is
ten
t
with
p
r
ev
io
u
s
s
tu
d
ies,
s
u
ch
as
Yao
et
a
l.
[
3
0
]
,
wh
ich
em
p
h
asized
th
e
im
p
o
r
tan
ce
o
f
au
g
m
en
tatio
n
in
im
p
r
o
v
in
g
th
e
p
er
f
o
r
m
a
n
ce
o
f
d
ee
p
lear
n
in
g
m
o
d
els f
o
r
o
b
ject
d
etec
tio
n
.
4
.
3
.
Ana
ly
s
is
co
m
pa
riso
n a
lg
o
rit
hm
T
h
is
s
ec
tio
n
co
m
p
ar
es
th
e
f
aster
R
-
C
NN
an
d
YOL
Ov
8
al
g
o
r
ith
m
s
,
f
o
c
u
s
in
g
o
n
th
eir
s
p
ee
d
an
d
ac
cu
r
ac
y
in
d
etec
tin
g
PC
B
d
ef
ec
ts
.
Fas
ter
R
-
C
NN
em
p
lo
y
s
a
two
-
s
tag
e
ap
p
r
o
ac
h
with
a
R
PN,
wh
ich
is
ef
f
ec
tiv
e
in
id
en
tify
in
g
s
m
all
an
d
c
o
m
p
le
x
o
b
jects
b
u
t
r
e
s
u
lts
in
s
lo
wer
p
er
f
o
r
m
an
ce
.
T
r
ain
in
g
tim
es
f
o
r
Fas
ter
R
-
C
NN
ar
e
r
ec
o
r
d
ed
at
1
.
8
5
s
ec
o
n
d
s
p
er
iter
ati
o
n
an
d
4
0
8
.
8
5
s
ec
o
n
d
s
p
er
ep
o
ch
d
u
e
to
its
co
m
p
u
tatio
n
ally
in
te
n
s
iv
e
ar
ch
itectu
r
e.
I
n
co
n
tr
ast,
YOL
Ov
8
u
tili
ze
s
a
s
in
g
le
-
s
h
o
t
d
etec
tio
n
ap
p
r
o
ac
h
,
p
r
ed
ictin
g
b
o
u
n
d
i
n
g
b
o
x
es
an
d
o
b
ject
p
r
o
b
ab
ilit
ies
d
ir
ec
tly
,
s
ig
n
if
ican
tly
r
ed
u
cin
g
tr
ain
in
g
tim
es
to
1
.
1
3
s
ec
o
n
d
s
p
er
iter
atio
n
a
n
d
1
1
3
s
ec
o
n
d
s
p
er
ep
o
c
h
.
T
h
is
h
ig
h
lig
h
ts
YOL
Ov
8
ef
f
icien
c
y
an
d
s
u
itab
ilit
y
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
s
.
T
h
e
im
p
ac
t
o
f
d
ataset
au
g
m
e
n
tatio
n
o
n
m
o
d
el
p
e
r
f
o
r
m
an
ce
i
s
ev
id
en
t,
as
s
ee
n
in
Fig
u
r
e
6
.
W
h
ile
th
e
non
-
a
u
g
m
en
te
d
YOL
Ov
8
l
ac
h
iev
es
a
p
r
ec
is
io
n
o
f
0
.
9
4
6
,
r
ec
all
o
f
0
.
9
2
5
,
an
d
F1
-
s
c
o
r
e
o
f
0
.
9
3
5
,
th
e
au
g
m
en
ted
YOL
Ov
8
x
s
u
r
p
ass
es
it
with
m
etr
ics
o
f
0
.
9
5
3
,
0
.
9
4
5
,
an
d
0
.
9
4
9
,
r
esp
ec
tiv
ely
.
T
h
is
d
em
o
n
s
tr
ates
th
e
s
ig
n
if
ican
ce
o
f
a
u
g
m
en
tat
io
n
in
im
p
r
o
v
in
g
m
o
d
el
p
er
f
o
r
m
an
ce
.
C
o
n
s
is
ten
t
with
Yao
et
a
l.
[
3
0
]
,
t
h
ese
f
in
d
in
g
s
h
ig
h
lig
h
t
th
e
au
g
m
en
ted
YOL
Ov
8
x
as
th
e
m
o
s
t
ef
f
ec
tiv
e
co
n
f
ig
u
r
atio
n
,
c
o
m
b
in
in
g
s
u
p
er
io
r
ac
cu
r
ac
y
with
p
r
o
ce
s
s
in
g
ef
f
i
cien
cy
,
p
ar
ticu
lar
ly
f
o
r
d
etec
t
in
g
f
a
u
lts
o
n
PC
B
lay
o
u
ts
.
o
wn
in
Fig
u
r
e
7
,
th
e
au
g
m
en
ted
YOL
Ov
8
x
ac
cu
r
a
tely
id
en
tifie
s
d
ef
ec
ts
s
u
ch
as
s
p
u
r
io
u
s
co
p
p
er
an
d
o
p
en
cir
cu
its
with
h
ig
h
co
n
f
id
en
ce
,
o
u
tp
er
f
o
r
m
in
g
b
o
th
th
e
n
o
n
-
au
g
m
en
ted
YOL
Ov
8
l
an
d
f
aster
R
-
C
NN
(
R
esNet5
0
FP
N
v
2
)
i
n
d
etec
tio
n
ac
cu
r
ac
y
.
T
h
ese
r
e
s
u
lts
em
p
h
asize
th
e
cr
itical
r
o
le
o
f
d
ataset
au
g
m
en
tatio
n
in
en
h
an
ci
n
g
th
e
r
o
b
u
s
tn
ess
an
d
r
eliab
ilit
y
o
f
d
ee
p
lear
n
in
g
m
o
d
els f
o
r
d
e
f
ec
t d
etec
tio
n
.
Fig
u
r
e
6
.
An
al
y
s
is
r
esu
lt c
o
m
p
ar
is
o
n
alg
o
r
ith
m
I
n
co
n
clu
s
io
n
,
YOL
Ov
8
x
d
em
o
n
s
tr
ates
s
u
p
er
io
r
p
er
f
o
r
m
an
c
e
in
b
o
th
s
p
ee
d
an
d
ac
cu
r
ac
y
,
m
ak
in
g
it
th
e
m
o
s
t
r
o
b
u
s
t
an
d
ef
f
icien
t
s
o
lu
tio
n
f
o
r
r
ea
l
-
tim
e
PC
B
d
ef
ec
t
d
etec
tio
n
.
Fu
tu
r
e
r
esea
r
ch
co
u
l
d
f
o
c
u
s
o
n
d
ev
elo
p
in
g
h
y
b
r
id
m
o
d
els
th
at
co
m
b
in
e
t
h
e
p
r
ec
is
io
n
o
f
two
-
s
tag
e
d
etec
to
r
s
lik
e
f
aste
r
R
-
C
NN
with
th
e
co
m
p
u
tatio
n
al
ef
f
icien
c
y
o
f
s
i
n
g
le
-
s
tag
e
d
etec
to
r
s
lik
e
YOL
Ov
8
.
Ad
d
itio
n
ally
,
ap
p
ly
i
n
g
YOL
Ov
8
x
to
d
etec
t
o
th
er
d
ef
ec
t
ty
p
es
o
r
o
p
er
ate
in
d
iv
er
s
e
in
d
u
s
tr
ial
co
n
te
x
ts
wo
u
ld
v
alid
ate
its
v
er
s
atility
f
u
r
th
er
.
E
v
alu
atin
g
th
ese
alg
o
r
ith
m
s
o
n
ed
g
e
d
ev
i
ce
s
o
r
in
d
is
tr
ib
u
ted
co
m
p
u
tin
g
en
v
ir
o
n
m
en
ts
co
u
ld
u
n
lo
ck
n
ew
o
p
p
o
r
tu
n
ities
f
o
r
r
ea
l
-
tim
e
d
ef
ec
t d
etec
tio
n
i
n
in
d
u
s
tr
ies r
eq
u
i
r
in
g
lo
w
-
late
n
cy
p
r
o
ce
s
s
in
g
.
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
Ob
ject
d
etec
tio
n
in
p
r
in
ted
ci
r
cu
it b
o
a
r
d
q
u
a
lity c
o
n
tr
o
l:
c
o
mp
a
r
in
g
a
lg
o
r
ith
ms
…
(
Ja
ja
K
u
s
tija
)
2805
Fig
u
r
e
7
.
Dete
ctio
n
r
esu
lts
o
f
ad
v
an
ce
d
o
b
ject
d
etec
tio
n
alg
o
r
ith
m
s
5.
CO
NCLU
SI
O
N
T
h
is
r
esear
ch
s
u
cc
ess
f
u
lly
d
ev
elo
p
ed
a
r
o
b
u
s
t
d
ef
ec
t
d
etec
tio
n
s
y
s
tem
f
o
r
PC
B
lay
o
u
ts
u
s
in
g
two
ad
v
an
ce
d
d
ee
p
lear
n
in
g
al
g
o
r
i
th
m
s
,
f
aster
R
-
C
NN
an
d
YOL
Ov
8
.
T
h
r
o
u
g
h
ev
alu
atio
n
s
o
n
n
o
n
-
a
u
g
m
en
te
d
an
d
au
g
m
en
ted
d
atasets
,
th
e
s
tu
d
y
h
ig
h
lig
h
te
d
th
eir
r
esp
ec
t
iv
e
s
tr
en
g
th
s
a
n
d
lim
itatio
n
s
.
Fas
ter
R
-
C
NN,
p
ar
ticu
lar
ly
t
h
e
R
esNet5
0
FP
N
v
2
v
ar
ian
t,
s
h
o
wed
im
p
r
o
v
ed
p
e
r
f
o
r
m
an
ce
with
au
g
m
e
n
ted
d
ata
b
u
t
was
lim
ited
b
y
its
s
lo
wer
two
-
s
tag
e
d
etec
tio
n
p
r
o
ce
s
s
,
m
a
k
in
g
it
less
s
u
itab
le
f
o
r
r
ea
l
-
ti
m
e
ap
p
licatio
n
s
.
I
n
co
n
tr
ast,
YOL
Ov
8
,
esp
ec
ially
th
e
YOL
Ov
8
x
v
a
r
ian
t,
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
e
d
f
aster
R
-
C
NN
in
b
o
th
s
p
ee
d
an
d
ac
cu
r
ac
y
.
I
ts
s
in
g
le
-
s
h
o
t
d
etec
tio
n
m
ec
h
an
is
m
en
ab
l
ed
f
aster
co
m
p
u
tatio
n
wh
ile
m
ain
tain
in
g
h
ig
h
ac
cu
r
ac
y
,
p
o
s
itio
n
in
g
YOL
Ov
8
x
as th
e
o
p
tim
al
ch
o
ice
f
o
r
r
e
al
-
tim
e
PC
B
d
ef
ec
t d
etec
tio
n
.
T
h
e
f
i
n
d
in
g
s
u
n
d
er
s
co
r
e
th
e
s
ig
n
if
ican
ce
o
f
d
ata
p
r
e
p
ar
a
tio
n
in
en
h
an
cin
g
d
ee
p
lear
n
in
g
m
o
d
el
r
o
b
u
s
tn
ess
.
YOL
Ov
8
x
s
u
p
er
i
o
r
p
e
r
f
o
r
m
an
ce
m
ak
es
it
a
p
r
ac
tical
s
o
lu
tio
n
f
o
r
im
p
r
o
v
in
g
q
u
ality
co
n
tr
o
l
p
r
o
ce
s
s
es
in
in
d
u
s
tr
ial
in
s
p
ec
tio
n
s
y
s
tem
s
,
o
f
f
er
in
g
en
h
an
ce
d
ac
cu
r
ac
y
an
d
s
p
ee
d
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
s
.
Fu
tu
r
e
r
esear
ch
co
u
ld
f
o
cu
s
o
n
d
ev
elo
p
in
g
h
y
b
r
id
m
o
d
els
th
at
co
m
b
in
e
th
e
p
r
ec
is
io
n
o
f
f
aster
R
-
C
N
N
wit
h
th
e
ef
f
icien
cy
o
f
YOL
Ov
8
to
b
alan
ce
ac
c
u
r
ac
y
a
n
d
c
o
m
p
u
tatio
n
al
co
s
t.
E
x
p
a
n
d
in
g
th
e
a
p
p
licatio
n
o
f
th
ese
m
o
d
els
to
d
i
v
er
s
e
d
ef
ec
t
ty
p
e
s
an
d
in
d
u
s
tr
ial
s
ettin
g
s
wo
u
ld
f
u
r
t
h
er
v
alid
ate
th
eir
v
er
s
atility
.
Ad
d
itio
n
ally
,
d
ep
lo
y
in
g
th
ese
alg
o
r
ith
m
s
o
n
ed
g
e
d
ev
ices
o
r
d
is
tr
ib
u
te
d
s
y
s
tem
s
co
u
ld
u
n
lo
ck
n
ew
o
p
p
o
r
tu
n
ities
f
o
r
r
ea
l
-
tim
e
d
ef
ec
t d
etec
tio
n
i
n
lo
w
-
la
ten
cy
en
v
ir
o
n
m
e
n
ts
.
ACK
NO
WL
E
DG
M
E
N
T
S
W
e
ex
p
r
ess
o
u
r
s
in
ce
r
e
g
r
atitu
d
e
to
all
p
ar
ties
wh
o
co
n
t
r
ib
u
ted
to
th
e
s
u
cc
ess
f
u
l
co
m
p
le
tio
n
o
f
th
is
r
esear
ch
.
Ou
r
d
ee
p
est
a
p
p
r
ec
i
atio
n
g
o
es
to
th
e
De
p
ar
tm
en
t
o
f
E
lectr
ical
E
n
g
in
ee
r
in
g
E
d
u
ca
tio
n
,
Facu
lty
o
f
I
n
d
u
s
tr
ial
T
ec
h
n
o
lo
g
y
an
d
E
d
u
ca
tio
n
,
U
n
iv
er
s
itas
Pen
d
id
ik
an
I
n
d
o
n
esia,
f
o
r
p
r
o
v
i
d
in
g
th
e
n
ec
ess
ar
y
r
eso
u
r
ce
s
an
d
in
s
titu
tio
n
al
s
u
p
p
o
r
t
th
r
o
u
g
h
o
u
t
th
is
s
tu
d
y
.
L
a
s
tly
,
we
ac
k
n
o
wled
g
e
th
e
ef
f
o
r
ts
o
f
th
e
jo
u
r
n
al’
s
ed
ito
r
ial
team
an
d
p
u
b
lis
h
er
s
in
f
ac
ilit
atin
g
th
e
d
is
s
em
in
atio
n
o
f
o
u
r
f
in
d
in
g
s
,
en
s
u
r
in
g
th
i
s
r
esear
ch
s
er
v
es
as
a
v
alu
ab
le
r
e
f
er
en
ce
f
o
r
t
h
e
ac
ad
em
ic
co
m
m
u
n
ity
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
No
ex
ter
n
al
f
u
n
d
in
g
was
in
v
o
lv
ed
in
th
is
r
esear
ch
.
T
h
e
r
esear
ch
was
s
elf
-
f
u
n
d
ed
,
wit
h
f
in
an
cial
s
u
p
p
o
r
t a
llo
ca
ted
f
r
o
m
th
e
p
r
o
f
ess
o
r
ial
h
o
n
o
r
a
r
iu
m
to
f
u
lf
ill
th
e
s
p
ec
ial
d
u
ties
o
f
a
p
r
o
f
ess
o
r
.
Evaluation Warning : The document was created with Spire.PDF for Python.