T
E
L
K
O
M
N
I
K
A
T
elec
o
m
m
un
ica
t
io
n,
Co
m
pu
t
ing
,
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
19
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
1
,
p
p
.
2
4
4
~
2
5
1
I
SS
N:
1
6
9
3
-
6
9
3
0
,
ac
cr
ed
ited
First Gr
ad
e
b
y
Kem
en
r
is
tek
d
i
k
ti,
Dec
r
ee
No
: 2
1
/E/KPT
/2
0
1
8
DOI
:
1
0
.
1
2
9
2
8
/TE
L
KOM
NI
K
A.
v
1
9
i1
.
1
6
2
3
2
244
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//jo
u
r
n
a
l.u
a
d
.
a
c.
id
/in
d
ex
.
p
h
p
/TELK
OM
N
I
K
A
Apply
ing
conv
o
lutiona
l neural ne
t
wo
rks for
limi
ted
-
memo
ry
a
pplica
tion
Xua
n
-
K
ien D
a
ng
1
,
H
uy
nh
-
N
hu
T
ruo
ng
2
,
Viet
-
Chi
nh
Ng
u
y
en
3
,
T
hi
-
Duy
en
-
An
h P
ha
m
4
1,
3,
4
Ho
Ch
i
M
in
h
Cit
y
U
n
iv
e
rsit
y
o
f
Tran
sp
o
rt,
Vie
tn
a
m
2
Ba
Ria
-
Vu
n
g
Tau
Co
ll
e
g
e
o
f
Tec
h
n
o
lo
g
y
,
Vie
tn
a
m
Ar
t
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ap
r
2
,
2
0
2
0
R
ev
is
ed
J
u
n
1
2
,
2
0
2
0
Acc
ep
ted
Au
g
2
9
,
2
0
2
0
Cu
rre
n
tl
y
,
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(CNN
)
a
re
c
o
n
sid
e
re
d
a
s
th
e
m
o
st
e
ffe
c
ti
v
e
to
o
l
in
ima
g
e
d
iag
n
o
sis a
n
d
p
ro
c
e
ss
in
g
tec
h
n
i
q
u
e
s.
In
th
is
p
a
p
e
r,
w
e
stu
d
ied
a
n
d
a
p
p
li
e
d
th
e
m
o
d
ifi
e
d
S
S
DLi
te_
M
o
b
il
e
Ne
tV2
a
n
d
p
ro
p
o
se
d
a
so
lu
ti
o
n
to
a
lwa
y
s
m
a
in
tain
t
h
e
b
o
u
n
d
a
ry
o
f
t
h
e
t
o
tal
m
e
m
o
ry
c
a
p
a
c
it
y
in
th
e
fo
ll
o
wi
n
g
r
o
b
u
st
b
o
u
n
d
a
n
d
a
p
p
li
e
d
o
n
t
h
e
b
rid
g
e
n
a
v
i
g
a
ti
o
n
a
l
wa
tch
&
a
larm
sy
ste
m
(
BNWAS
)
.
Th
e
h
a
rd
wa
re
wa
s
d
e
sig
n
e
d
b
a
se
d
o
n
ra
sp
b
e
rr
y
Pi
-
3
,
a
n
e
m
b
e
d
d
e
d
si
n
g
le
b
o
a
rd
c
o
m
p
u
ter
with
C
P
U
sm
a
rtp
h
o
n
e
lev
e
l,
li
m
it
e
d
RAM
with
o
u
t
CUD
A
G
P
U.
E
x
p
e
rime
n
tal
re
su
lt
s
sh
o
we
d
t
h
a
t
th
e
d
e
e
p
lea
rn
i
n
g
m
o
d
e
l
o
n
a
n
e
m
b
e
d
d
e
d
sin
g
le
b
o
a
rd
c
o
m
p
u
ter b
rin
g
s
u
s
h
ig
h
e
ffe
c
ti
v
e
n
e
ss
in
a
p
p
li
c
a
ti
o
n
.
K
ey
w
o
r
d
s
:
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
I
m
ag
e
p
r
o
ce
s
s
in
g
L
im
ited
h
ar
d
war
e
d
ev
ices
Ma
r
itime
ap
p
licatio
n
Ob
ject
class
if
icatio
n
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
:
Xu
an
-
Kien
Dan
g
Gr
ad
u
ate
Sch
o
o
l
Ho
C
h
i M
in
h
C
ity
Un
iv
er
s
ity
o
f
T
r
an
s
p
o
r
t
Nu
m
b
er
2
,
Vo
Oan
h
Stre
et,
W
ar
d
2
5
,
B
in
h
T
h
a
n
h
Dis
tr
ict,
Ho
C
h
i M
in
h
C
ity
,
Viet
n
am
E
m
ail:
d
an
g
x
u
an
k
ien
@
h
cm
u
tr
an
s
.
ed
u
.
v
n
1.
I
NT
RO
D
UCT
I
O
N
Ob
ject
d
etec
tio
n
is
a
co
m
p
u
ter
tech
n
o
lo
g
y
r
elate
d
to
co
m
p
u
t
er
v
is
io
n
an
d
im
ag
e
p
r
o
ce
s
s
in
g
th
at
d
ea
ls
with
a
co
m
b
in
atio
n
o
f
o
b
ject
class
if
icatio
n
an
d
o
b
ject
p
o
s
it
io
n
in
g
.
T
h
e
ad
v
en
t
o
f
m
o
d
er
n
ad
v
an
ce
s
in
d
ee
p
lear
n
in
g
[
1
-
3
]
h
as led
to
s
ig
n
if
ican
t a
d
v
an
ce
s
in
o
b
ject
d
etec
t
io
n
.
Mo
s
t r
ec
en
t r
esear
ch
f
o
cu
s
ed
o
n
d
esig
n
in
g
a
co
m
p
lex
n
etwo
r
k
f
o
r
o
b
ject
d
e
tectio
n
b
ased
o
n
n
eu
r
al
n
etwo
r
k
to
en
h
an
ce
ac
c
u
r
ac
y
,
s
u
ch
as
s
in
g
le
s
h
o
t
d
etec
to
r
(
SS
D)
[
4
]
an
d
f
aster
R
-
C
NN
[
5
]
.
Ma
n
y
r
esear
ch
er
s
ar
e
d
ev
o
ted
to
d
ev
elo
p
in
g
a
co
m
p
u
ter
tech
n
o
lo
g
y
a
nd
d
ee
p
lear
n
i
n
g
in
th
e
m
o
d
er
n
life
f
o
r
ít
o
u
ts
tan
d
in
g
ad
v
a
n
ta
g
es.
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
et
wo
r
k
s
(
C
NNs
)
ap
p
lied
o
n
th
e
d
ataset
o
f
im
ag
e
d
ata
(
esp
ec
ially
lu
n
g
X
-
r
ay
)
[
3
]
f
o
r
class
if
icatio
n
o
f
p
n
eu
m
o
n
ia
d
i
s
ea
s
e
an
d
th
e
r
esu
lt
was
o
b
tain
ed
an
ac
cu
r
ac
y
r
ate
of
9
7
%.
T
h
e
Alex
Net’
s
d
ee
p
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
u
s
ed
as
a
p
r
e
-
tr
ai
n
ed
n
eu
r
al
n
etwo
r
k
with
1
0
0
0
ca
teg
o
r
ies
f
o
r
im
ag
e
class
if
icatio
n
[
6
]
to
d
etec
t
a
n
d
g
eo
t
ag
ad
v
er
tis
em
en
t
b
illb
o
ar
d
in
r
ea
l
-
tim
e
co
n
d
itio
n
,
an
d
e
x
p
er
im
e
n
tal
r
esu
lts
ac
h
i
ev
ed
9
2
.
7
%
tr
ain
in
g
ac
cu
r
ac
y
f
o
r
ad
v
er
tis
em
en
t
b
illb
o
ar
d
d
etec
tio
n
.
B
y
u
s
in
g
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
s
,
Z
.
R
u
s
tam
,
et
a
l.,
[
7
]
p
r
o
p
o
s
ed
th
e
m
eth
o
d
to
ass
is
t
d
o
ct
o
r
s
in
p
r
o
v
id
i
n
g
th
e
ap
p
r
o
p
r
iate
b
elief
s
an
d
p
r
ed
ict
io
n
s
to
p
atien
ts
,
th
e
r
esu
lts
s
h
o
wed
th
e
ca
p
ab
ilit
y
o
f
C
NNs
m
eth
o
d
to
ac
cu
r
ately
id
en
tify
th
e
p
atien
t'
s
X
-
r
ay
t
est
im
ag
es.
Acc
o
r
d
in
g
to
th
e
r
esu
lts
p
u
b
l
is
h
ed
in
[
8
]
,
th
e
C
NNs
m
o
d
el
u
s
es
6
4
x
6
4
in
p
u
t
s
h
ap
e,
0
.
0
0
0
1
lea
r
n
in
g
r
ate,
3
x
3
f
ilter
s
ize,
e
p
o
ch
1
0
0
co
u
n
t,
d
ata
tr
ain
i
n
g
1
6
0
,
an
d
d
ata
test
in
g
4
0
,
th
e
ac
cu
r
ac
y
lev
el
o
f
t
r
ain
i
n
g
an
d
test
in
g
in
class
if
icatio
n
o
f
g
o
lek
p
u
p
p
et
im
ag
e
attain
e
d
1
0
0
%
ac
cu
r
ac
y
.
T
h
is
is
an
id
ea
l
r
esu
lt
th
at
d
e
m
o
n
s
tr
ates
th
e
ef
f
ec
tiv
en
ess
o
f
u
s
in
g
C
NNs
m
eth
o
d
i
n
o
b
je
ct
class
if
icatio
n
.
An
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
ommun
C
o
m
p
u
t E
l Co
n
tr
o
l
A
p
p
lyin
g
c
o
n
vo
lu
tio
n
a
l n
e
u
r
a
l
n
etw
o
r
ks fo
r
limited
-
mem
o
r
y
a
p
p
lica
tio
n
(
X
u
a
n
-
K
ien
Da
n
g
)
245
ap
p
licatio
n
o
f
tr
an
s
f
er
lear
n
in
g
b
y
u
s
in
g
C
NNs
m
eth
o
d
b
as
ed
o
n
t
h
e
in
ce
p
tio
n
-
v
3
ar
ch
ite
ctu
r
al
m
o
d
el
[
9
]
f
o
r
ea
r
ly
d
etec
tio
n
o
f
ter
r
y
’
s
n
ail.
T
h
e
ac
cu
r
ac
y
o
b
tain
ed
with
tr
ain
in
g
d
ata
9
0
%,
p
r
ec
is
io
n
a
n
d
m
em
o
r
y
,
ea
c
h
o
f
wh
ich
is
wo
r
th
9
5
.
2
4
%,
1
0
0
%,
an
d
9
0
.
9
1
%.
Sp
ec
ially
,
we
in
tr
o
d
u
ce
y
o
u
o
n
ly
l
o
o
k
o
n
ce
(
YO
LO
)
,
a
u
n
if
ied
m
o
d
el
f
o
r
o
b
ject
d
etec
tio
n
.
T
h
e
YOL
O
m
o
d
el
[
1
0
]
is
s
im
p
le
to
co
n
s
tr
u
ct
an
d
ca
n
b
e
tr
ai
n
ed
d
ir
ec
tly
o
n
f
u
ll
im
ag
es.
Un
lik
e
class
if
ier
-
b
ase
d
ap
p
r
o
ac
h
es,
f
ast
YOL
O
i
s
t
h
e
f
astes
t
g
en
er
al
-
p
u
r
p
o
s
e
o
b
ject
d
etec
to
r
in
th
e
liter
atu
r
e
an
d
YOL
O
p
u
s
h
es
th
e
s
tate
-
of
-
th
e
-
ar
t in
r
ea
l
-
tim
e
o
b
ject
d
etec
tio
n
,
to
d
o
s
o
YOL
O
g
en
er
alize
s
well
to
n
ew
d
o
m
ain
s
m
ak
in
g
it
id
e
al
,
f
ast,
r
o
b
u
s
t
o
b
ject
d
etec
tio
n
f
o
r
ap
p
licatio
n
s
th
at
r
ely
o
n
.
Ho
wev
er
,
all
o
f
th
e
alg
o
r
ith
m
s
r
eq
u
ir
e
a
la
r
g
e
am
o
u
n
t
o
f
r
eso
u
r
ce
s
o
f
th
e
s
y
s
te
m
,
an
d
to
p
u
t
th
em
o
n
lim
ited
h
ar
d
war
e
d
e
v
ices
n
ee
d
s
to
b
e
s
tr
ea
m
lin
ed
an
d
c
o
m
p
iled
in
to
lim
ited
h
ar
d
war
e
.
R
elate
d
to
en
s
u
r
e
th
e
m
ar
itime
s
af
ety
,
th
e
m
ain
o
b
jectiv
e
co
n
s
titu
tes
th
e
f
o
llo
win
g
two
task
s
as
f
o
llo
w:
th
e
f
ir
s
t
is
en
s
u
r
in
g
th
e
s
af
e
ty
o
f
life
an
d
p
r
o
p
er
ty
at
s
ea
f
r
o
m
th
e
g
eo
g
r
ap
h
ic
an
d
o
p
er
atio
n
al
h
az
a
r
d
s
(
u
n
d
er
wate
r
o
b
s
tacle
s
,
co
llis
i
o
n
,
h
ar
m
s
an
d
d
am
ag
es
ca
u
s
e
d
b
y
th
e
u
n
f
av
o
r
ab
le
wea
th
er
co
n
d
itio
n
s
)
an
d
th
e
s
ec
o
n
d
is
en
s
u
r
in
g
th
e
s
af
ety
o
f
s
h
ip
co
n
tr
o
l
th
r
o
u
g
h
o
u
t
t
h
e
jo
u
r
n
ey
b
y
t
h
e
s
ailer
,
if
d
u
r
in
g
a
n
em
e
r
g
en
c
y
s
itu
atio
n
,
a
n
a
v
ig
atio
n
al
o
f
f
icer
is
n
o
t
ca
p
ab
le
o
f
h
an
d
lin
g
th
at
s
itu
atio
n
,
i
t
ca
n
lead
to
m
ar
i
tim
e
co
llis
io
n
.
Fo
r
th
e
f
ir
s
t
task
,
th
er
e
ar
e
m
an
y
s
tu
d
ies
to
im
p
r
o
v
e,
u
p
g
r
ad
e
cu
r
r
en
t
s
y
s
tem
s
th
at
h
av
e
s
h
o
r
tc
o
m
in
g
s
in
r
e
g
ar
d
to
av
ailab
ili
ty
,
in
teg
r
ity
,
m
o
n
ito
r
in
g
an
d
s
y
s
tem
life
e
x
p
ec
tan
c
y
as
th
e
g
l
o
b
al
n
av
ig
atio
n
s
atellite
s
y
s
tem
[
1
1
]
an
d
th
e
r
eg
io
n
al
s
atellite
au
g
m
en
ta
tio
n
s
y
s
tem
f
o
r
m
ar
itime
ap
p
li
ca
tio
n
s
[
1
2
]
,
o
r
th
e
d
esig
n
o
f
s
atellite
co
n
s
tellatio
n
f
o
r
I
n
d
o
n
esian
m
ar
itime
s
u
r
v
eillan
ce
u
s
in
g
th
e
AI
S
d
at
a
ac
q
u
is
itio
n
b
y
L
APAN
-
A
2
an
d
L
APAN
-
A3
s
atellite
s
[
1
3
]
with
t
h
e
eig
h
t
s
atellit
es
in
an
eq
u
ato
r
ial
o
r
b
it
f
o
r
n
ea
r
r
ea
l
-
tim
e
A
I
S
m
o
n
ito
r
in
g
in
I
n
d
o
n
esia
an
d
th
e
o
th
er
eq
u
ato
r
ial
r
eg
i
o
n
m
ak
e
a
b
etter
g
lo
b
al
m
ar
itime
awa
r
en
ess
an
d
en
s
u
r
in
g
t
h
e
m
ar
itime
s
af
ety
.
T
h
e
s
ec
o
n
d
task
,
to
d
esig
n
a
n
d
m
an
u
f
ac
tu
r
e
s
y
s
tem
s
s
er
v
in
g
s
h
ip
s
to
en
s
u
r
e
s
a
f
ety
in
s
h
ip
o
p
er
atio
n
p
r
o
ce
s
s
b
y
u
s
in
g
n
ew
co
m
p
u
ter
tec
h
n
o
n
o
g
ies as n
eu
r
al
n
etwo
r
k
,
f
u
zz
y
-
n
eu
r
al,
o
r
g
e
n
etic
alg
o
r
ith
m
.
I
n
th
is
p
ap
er
,
we
ai
m
ed
t
o
a
p
p
ly
th
e
m
o
d
if
ied
SS
DL
ite_
Mo
b
ileNetV2
b
o
u
n
d
e
d
C
NN
al
g
o
r
ith
m
to
b
r
id
g
e
n
av
ig
atio
n
al
watc
h
&
al
ar
m
s
y
s
tem
(
B
N
W
AS)
,
e
x
ten
s
iv
e
ex
p
e
r
im
en
ts
s
h
o
wed
t
h
at
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
ca
n
ac
h
iev
e
t
h
e
s
tate
-
of
-
th
e
-
ar
t
r
esu
lts
co
m
p
ar
e
d
with
th
e
b
est
cu
r
r
en
t
m
eth
o
d
b
ased
o
n
h
an
d
c
r
af
te
d
f
ea
tu
r
es
[
1
4
]
an
d
th
r
ee
o
th
e
r
r
elate
d
C
NN
b
ased
m
eth
o
d
s
[
1
5
-
1
7
]
an
d
o
u
r
p
r
e
v
io
u
s
wo
r
k
[
1
8
]
f
o
r
im
a
g
e
an
aly
s
is
.
Mo
r
eo
v
er
,
we
h
a
v
e
v
alid
ated
th
e
r
atio
n
ality
a
n
d
r
o
b
u
s
tn
ess
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
with
m
o
r
e
s
u
p
p
lem
en
tar
y
r
esu
lts
.
T
h
e
in
v
er
ted
r
esid
u
al
b
o
ttle
n
ec
k
lay
er
s
allo
w
a
p
ar
ticu
lar
ly
m
em
o
r
y
-
e
f
f
icien
t
im
p
lem
en
tatio
n
wh
ich
is
v
e
r
y
im
p
o
r
ta
n
t
f
o
r
m
o
b
ile
ap
p
l
icatio
n
s
.
A
s
tan
d
ar
d
ef
f
icien
t
im
p
lem
en
tatio
n
o
f
in
f
er
en
ce
t
h
at
wer
e
u
s
ed
f
o
r
i
n
s
tan
ce
T
en
s
o
r
Flo
w
[
1
9
]
o
r
C
af
f
e
[
20
]
b
u
ilt
a
d
ir
ec
ted
ac
y
clic
co
m
p
u
te
h
y
p
er
g
r
ap
h
G
.
W
ith
a
s
m
all
h
a
r
d
wa
r
e
s
y
s
tem
,
we
u
s
ed
th
e
SS
D
L
i
te
Mo
b
ileNetV2
s
tr
u
ctu
r
e
b
ec
au
s
e
it wa
s
f
a
s
t a
n
d
ac
cu
r
ate.
No
t
o
n
ly
wer
e
th
e
r
eq
u
ir
em
en
ts
f
o
r
im
ag
e
p
r
o
ce
s
s
in
g
,
o
b
ject
d
etec
tio
n
an
d
cla
s
s
if
icatio
n
m
et,
th
e
s
y
s
tem
also
ab
o
d
e
b
y
I
MO
[
21
,
22
]
,
I
E
C
[
23
]
an
d
[
24
,
25
]
r
eg
u
latio
n
s
wh
ich
co
u
ld
b
e
test
ed
an
d
d
ir
ec
tly
o
p
er
ated
o
n
b
o
a
r
d
.
W
e
ca
r
ef
u
lly
d
esig
n
ed
a
n
ew
C
NN
b
ased
m
eth
o
d
f
o
r
d
etec
ti
n
g
v
ar
io
u
s
t
y
p
ical
im
ag
e
-
p
r
o
ce
s
s
in
g
o
p
er
atio
n
s
,
th
e
m
ain
co
n
tr
ib
u
tio
n
s
o
f
th
is
p
ap
er
ar
e
g
iv
en
as f
o
llo
w:
−
W
e
f
ir
s
t
co
n
v
er
ted
th
e
in
p
u
t
im
ag
e
in
to
r
esid
u
als
to
s
u
p
p
r
ess
th
e
in
f
lu
en
ce
o
f
im
ag
e
co
n
te
n
ts
,
an
d
th
en
u
s
ed
a
co
n
v
o
lu
tio
n
al
lay
er
to
in
cr
ea
s
e
th
e
ch
an
n
el
n
u
m
b
e
r
.
−
W
e
em
p
lo
y
ed
s
ix
s
im
ilar
lay
er
g
r
o
u
p
s
to
o
b
tain
th
e
h
i
g
h
-
lev
e
l f
ea
tu
r
es o
f
th
e
in
p
u
t im
a
g
e.
−
Fin
all
y
,
we
ap
p
lied
t
h
e
r
esu
ltin
g
f
ea
tu
r
es
in
t
o
th
e
f
u
ll
co
n
n
ec
t
lay
er
f
o
r
class
if
icatio
n
o
f
th
e
s
y
s
tem
,
we
p
r
o
p
o
s
ed
a
s
o
lu
tio
n
to
alwa
y
s
m
ain
tain
th
e
b
o
u
n
d
ar
y
o
f
th
e
t
o
tal
m
em
o
r
y
ca
p
ac
ity
in
th
e
f
o
llo
win
g
r
o
b
u
s
t
b
o
u
n
d
an
d
ap
p
lied
o
n
t
h
e
B
NW
AS.
T
h
e
r
est
o
f
th
e
p
ap
e
r
is
o
r
g
an
i
ze
d
as
f
o
llo
ws
;
s
ec
tio
n
2
s
h
o
ws
s
o
m
e
r
elate
d
wo
r
k
s
an
d
p
r
o
p
o
s
ed
th
e
m
eth
o
d
r
e
d
u
cin
g
m
e
m
o
r
y
wh
ile
en
s
u
r
in
g
im
ag
e
q
u
ality
f
o
r
o
b
ject
d
etec
tio
n
a
n
d
s
ec
tio
n
3
d
escr
ib
es
th
e
s
tr
u
ctu
r
e
o
f
th
e
p
r
o
p
o
s
ed
B
NW
AS
b
ased
o
n
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
,
p
r
esen
ts
th
e
ex
p
er
im
en
tal
r
esu
lts
an
d
d
is
cu
s
s
io
n
s
.
Fin
ally
,
th
e
c
o
n
clu
d
in
g
r
em
a
r
k
s
ar
e
g
i
v
en
i
n
s
ec
t
io
n
4.
2.
CNNs
B
A
SE
D
SS
D
L
I
T
E
-
M
O
B
I
L
E
N
E
T
M
E
T
H
O
D
F
O
R
O
B
J
E
CT
DE
T
E
C
T
I
O
N
W
I
T
H
L
I
M
I
T
E
D
-
M
E
M
O
RY
C
NN
m
o
d
els
ar
e
h
ig
h
ly
ac
cu
r
ate,
b
u
t
th
ey
all
h
av
e
a
co
m
m
o
n
d
r
awb
ac
k
t
h
at
is
th
ey
ar
e
n
o
t
s
u
itab
le
f
o
r
m
o
b
ile
ap
p
licatio
n
s
o
r
e
m
b
ed
d
e
d
s
y
s
tem
s
with
lo
w
p
o
wer
co
m
p
u
tin
g
.
I
n
liter
atu
r
e
r
ev
iew,
th
e
a
u
th
o
r
s
in
[
2
6
]
in
tr
o
d
u
ce
r
eso
u
r
ce
-
f
r
u
g
al
q
u
a
n
tized
co
n
v
o
lu
ti
o
n
al
n
eu
r
al
n
etwo
r
k
s
to
r
ed
u
ce
th
eir
s
ize
with
o
u
t
ad
v
er
s
ely
af
f
ec
tin
g
th
e
class
if
icatio
n
ca
p
ab
ilit
y
f
o
r
s
eg
m
e
n
tin
g
h
y
p
er
s
p
ec
tr
al
s
atellite
im
ag
es,
esp
ec
ially
f
o
cu
s
in
g
o
n
th
e
m
em
o
r
y
s
av
i
n
g
s
o
f
q
u
a
n
tized
C
NNs.
Mo
r
eo
v
er
,
a
n
a
p
p
r
o
ac
h
u
s
in
g
o
b
ject
class
clu
s
ter
in
g
to
lo
wer
b
it
p
r
ec
is
io
n
b
e
y
o
n
d
q
u
an
tizatio
n
lim
its
p
r
o
p
o
s
ed
b
y
Pra
teeth
Nay
ak
,
et
a
l
.
[
2
7
]
u
s
e
d
3
s
ch
e
m
es,
wh
ich
ar
e
u
n
if
o
r
m
-
ASYMM
,
u
n
if
o
r
m
-
SYMM
,
an
d
p
o
wer
-
of
-
2
.
T
h
e
r
esu
lt
is
all
o
f
q
u
a
n
tizatio
n
s
ch
em
e
ac
h
iev
e
d
n
ea
r
o
r
ig
in
al
m
o
d
el
ac
cu
r
ac
y
f
o
r
ev
er
y
te
s
ted
m
o
d
el.
I
f
y
o
u
wa
n
t
to
d
ev
elo
p
th
ese
m
o
d
els
f
o
r
r
ea
l
-
tim
e
a
p
p
lica
tio
n
s
,
y
o
u
n
ee
d
an
ex
tr
e
m
ely
p
o
wer
f
u
l
co
n
f
ig
u
r
atio
n
(
GPU/C
PU)
f
o
r
em
b
ed
d
ed
s
y
s
tem
s
(
r
asp
b
e
r
r
y
Pi,
n
an
o
PC
)
o
r
ap
p
licatio
n
s
r
u
n
n
in
g
o
n
s
m
ar
tp
h
o
n
es.
T
h
er
e
f
o
r
e,
we
n
ee
d
to
b
u
ild
a
m
o
d
el
lik
e
SS
D
L
ite
-
Mo
b
ileNet
h
y
b
r
id
.
T
h
e
m
ain
f
ac
to
r
will
h
elp
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
1
:
2
4
4
-
2
5
1
246
SS
D
L
ite
-
Mo
b
ile
Net
ac
h
iev
e
h
ig
h
ac
cu
r
ac
y
wh
ile
lo
w
co
m
p
u
tatio
n
tim
e
lies
in
th
e
h
y
b
r
id
s
tr
u
ctu
r
e
f
r
o
m
SS
D
an
d
M
o
b
ileNet
s
tr
u
ctu
r
e.
SS
D
(
s
in
g
le
s
h
o
t
m
u
lti
b
o
x
d
etec
t
o
r
)
is
an
o
b
ject
d
etec
to
r
(
Fig
u
r
e
1
)
t
h
at
p
er
f
o
r
m
s
two
m
ain
s
tep
s
:
ex
t
r
ac
t
f
ea
tu
r
e
m
ap
s
o
f
f
ea
tu
r
es
(
f
ea
tu
r
e
m
ap
s
)
an
d
ap
p
ly
c
o
n
v
o
lu
tio
n
f
ilter
s
(
co
n
v
o
lu
tio
n
f
ilter
s
)
to
d
etec
t o
b
jects.
Fig
u
r
e
1
.
Stru
ctu
r
e
o
f
s
in
g
le
s
h
o
t m
u
lti b
o
x
d
etec
to
r
u
s
ed
to
d
etec
t a
n
av
ig
atio
n
al
o
f
f
ice
r
T
h
e
lo
s
s
f
u
n
ctio
n
[
3
]
:
(
,
,
,
)
=
1
(
(
,
)
+
(
,
,
)
)
(
1
)
T
h
e
lo
s
s
f
u
n
ctio
n
co
n
s
is
ts
o
f
t
wo
ter
m
s
:
an
d
wh
er
e
N
is
th
e
m
atch
ed
d
ef
au
lt
b
o
x
es.
Ma
tch
ed
d
ef
au
lt
b
o
x
es:
(
,
,
)
=
∑
∑
∈
{
,
,
,
ℎ
}
∈
ℎ
1
(
−
̂
)
(
2
)
wh
er
e
̂
=
(
−
)
/
,
̂
=
(
−
)
/
ℎ
,
̂
=
(
)
an
d
̂
ℎ
=
(
)
;
is
th
e
lo
ca
lizatio
n
lo
s
s
wh
ich
is
th
e
s
m
o
o
th
lo
s
s
b
etwe
en
th
e
p
r
e
d
icted
b
o
x
a
n
d
th
e
g
r
o
u
n
d
-
tr
u
th
b
o
x
p
ar
am
eter
s
.
T
h
is
lo
s
s
f
u
n
ctio
n
is
s
im
ilar
to
th
e
o
n
e
in
Fas
ter
R
-
C
NN.
is
t
h
e
co
n
f
id
en
c
e
lo
s
s
wh
ich
is
th
e
s
o
f
tm
ax
lo
s
s
o
v
er
m
u
ltip
le
class
es
co
n
f
id
en
ce
s
(
c)
.
(
α
is
s
et
to
1
b
y
cr
o
s
s
v
alid
atio
n
)
.
(
,
)
=
−
∑
(
̂
)
−
∑
(
̂
0
)
∈
∈
(
3
)
wh
er
e
:
̂
=
(
)
∑
(
)
;
=
{
1
,
0
}
is
an
in
d
icato
r
f
o
r
m
atch
in
g
i
-
th
d
ef
au
lt b
o
x
to
th
e
j
-
th
g
r
o
u
n
d
t
r
u
th
b
o
x
o
f
ca
teg
o
r
y
P.
I
f
m
d
ef
au
lt m
a
p
s
ar
e
u
s
ed
f
o
r
p
r
e
d
ictio
n
,
we
s
u
g
g
est th
e
f
o
r
m
th
e
s
ca
le
o
f
t
h
e
d
ef
au
lt
b
o
x
es
f
o
r
ea
ch
f
ea
tu
r
e
m
ap
is
co
m
p
u
ted
as:
m
a
x
m
i
n
m
i
n
(
1
)
,
[
1
,
]
1
k
ss
s
s
k
k
m
m
−
=
+
−
−
(
4)
B
ased
o
n
[
2
4
]
,
we
s
et
p
ar
am
et
er
is
0
.
2
an
d
is
0
.
9
(
s
k
is
0
.
1
,
0
.
2
,
0
.
3
7
5
,
0
.
5
5
,
0
.
7
2
5
.
0
.
9
m
e
an
s
3
0
,
6
0
,
1
1
2
.
5
,
1
6
5
,
2
1
7
.
5
,
2
7
0
p
i
x
els in
p
u
t im
ag
e
(
3
0
0
x
3
0
0
)
)
.
T
h
e
s
tr
u
ctu
r
e
co
n
tain
s
a
co
m
p
letely
o
r
ig
in
al
c
o
n
v
o
lu
tio
n
lay
er
with
3
2
f
ilt
er
s
an
d
1
9
lay
er
s
o
f
b
o
ttlen
ec
k
.
Mo
b
ileNetV2
d
etailed
s
tr
u
ctu
r
e
is
d
escr
ib
ed
b
y
M.
San
d
ler
[
2
5
]
.
T
h
e
in
v
er
ted
r
esid
u
al
b
o
ttlen
ec
k
lay
er
s
allo
w
th
e
s
y
s
tem
to
h
a
v
e
a
p
a
r
ticu
lar
ly
e
f
f
icien
t
m
e
m
o
r
y
,
wh
ich
is
v
e
r
y
im
p
o
r
tan
t
f
o
r
a
p
p
licatio
n
s
.
A
s
tan
d
ar
d
ef
f
icien
t
im
p
lem
en
tat
io
n
o
f
in
f
er
en
ce
is
u
s
ed
in
T
e
n
s
o
r
Flo
w
[
1
9
]
o
r
C
af
f
e
[
2
0
]
.
T
h
e
co
m
p
u
tatio
n
is
s
ch
ed
u
led
to
m
in
im
ize
t
h
e
to
ta
l n
u
m
b
e
r
o
f
ten
s
o
r
s
th
at
n
ee
d
s
to
b
e
s
to
r
e
d
in
m
em
o
r
y
.
I
n
m
o
s
t g
en
er
al
ca
s
es,
it
s
ea
r
ch
es o
v
er
all
p
lau
s
ib
le
co
m
p
u
tatio
n
o
r
d
er
s
Σ
(
G)
a
n
d
p
ic
k
s
th
e
m
in
im
u
m
o
n
e.
(
)
=
∈
∑
(
)
∈
1
.
.
[
∑
|
|
∈
(
,
,
)
]
+
(
)
(
5
)
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
ommun
C
o
m
p
u
t E
l Co
n
tr
o
l
A
p
p
lyin
g
c
o
n
vo
lu
tio
n
a
l n
e
u
r
a
l
n
etw
o
r
ks fo
r
limited
-
mem
o
r
y
a
p
p
lica
tio
n
(
X
u
a
n
-
K
ien
Da
n
g
)
247
w
h
e
r
e
:
R
(
i
,
π
,
G
)
i
s
t
h
e
l
i
s
t
o
f
i
n
t
e
r
m
e
d
i
a
t
e
t
e
n
s
o
r
s
t
h
a
t
a
r
e
c
o
n
n
e
c
t
e
d
t
o
a
n
y
o
f
π
i
.
.
.
π
n
n
o
d
e
s
,
|
A
|
r
e
p
r
e
s
e
n
t
s
t
h
e
s
i
z
e
o
f
t
h
e
t
e
n
s
o
r
A
,
a
n
d
s
i
z
e
(
π
i
)
i
s
t
h
e
t
o
t
a
l
a
m
o
u
n
t
o
f
m
e
m
o
r
y
n
e
e
d
e
d
f
o
r
i
n
t
e
r
n
a
l
s
t
o
r
a
g
e
d
u
r
i
n
g
o
p
e
r
a
t
i
o
n
i
.
Fo
r
g
r
ap
h
s
th
at
h
av
e
o
n
l
y
tr
iv
ial
p
ar
allel
s
tr
u
ctu
r
e
(
s
u
ch
as
r
esid
u
al
c
o
n
n
ec
tio
n
)
,
th
er
e
is
o
n
ly
o
n
e
n
o
n
tr
iv
ial
f
ea
s
ib
l
e
co
m
p
u
tatio
n
o
r
d
e
r
,
an
d
th
u
s
t
h
e
to
tal
am
o
u
n
t
an
d
a
b
o
u
n
d
o
n
th
e
m
e
m
o
r
y
M(
G)
n
ee
d
e
d
f
o
r
i
n
f
er
en
ce
o
n
co
m
p
u
te
g
r
ap
h
G
ca
n
b
e
s
im
p
lifie
d
:
(
)
=
∈
[
∑
|
|
+
∑
|
|
+
|
|
∈
∈
]
(
6
)
Fo
llo
win
g
[
2
5
]
,
th
e
am
o
u
n
t
o
f
m
em
o
r
y
is
s
im
p
ly
th
e
m
ax
i
m
u
m
to
tal
s
ize
o
f
co
m
b
in
e
d
in
p
u
ts
an
d
o
u
tp
u
ts
ac
r
o
s
s
all
o
p
er
atio
n
s
.
I
t
m
ea
n
s
we
r
ec
o
g
n
ize
th
at
if
we
tr
ea
t
a
b
o
ttlen
ec
k
r
esid
u
al
b
lo
ck
as
a
s
in
g
le
o
p
er
atio
n
(
an
d
tr
ea
t
in
n
er
c
o
n
v
o
lu
tio
n
as
a
d
is
p
o
s
ab
le
ten
s
o
r
)
,
th
e
to
tal
am
o
u
n
t
o
f
m
e
m
o
r
y
w
o
u
ld
b
e
d
o
m
in
ated
b
y
th
e
s
ize
o
f
b
o
ttlen
ec
k
ten
s
o
r
s
,
r
ath
er
th
an
th
e
s
ize
o
f
ten
s
o
r
s
th
at
ar
e
in
ter
n
al
to
b
o
ttlen
ec
k
(
an
d
m
u
ch
lar
g
er
)
.
I
n
a
T
en
s
o
r
Flo
w
g
r
ap
h
,
ea
c
h
n
o
d
e
h
as
ze
r
o
o
r
m
o
r
e
in
p
u
ts
an
d
ze
r
o
o
r
m
o
r
e
o
u
tp
u
ts
,
an
d
r
e
p
r
esen
ts
th
e
in
s
tan
tiatio
n
o
f
an
o
p
e
r
atio
n
.
Valu
e
s
th
at
f
lo
w
alo
n
g
n
o
r
m
a
l
ed
g
es
in
th
e
g
r
a
p
h
(
f
r
o
m
o
u
tp
u
ts
to
in
p
u
ts
)
ar
e
ten
s
o
r
s
,
ar
b
itra
r
y
d
im
e
n
s
io
n
a
lity
ar
r
ay
s
wh
er
e
th
e
u
n
d
e
r
ly
in
g
elem
e
n
t
ty
p
e
is
s
p
ec
if
ied
o
r
in
f
er
r
ed
a
t
g
r
ap
h
-
co
n
s
tr
u
ctio
n
tim
e.
Fo
r
s
m
all
ap
p
licatio
n
s
,
r
ed
u
cin
g
m
em
o
r
y
wh
ile
en
s
u
r
i
n
g
im
ag
e
q
u
ality
is
g
r
ea
t.
Ho
wev
er
,
wh
e
n
we
a
b
u
s
e
th
i
s
,
it
ca
n
ea
s
ily
lead
to
in
s
tab
ilit
y
in
im
ag
e
p
r
o
ce
s
s
in
g
,
s
u
ch
as
r
ed
u
ci
n
g
im
a
g
e
q
u
ality
,
wh
ic
h
r
elate
s
t
o
th
e
m
ar
g
in
al
lim
it
o
f
to
tal
m
em
o
r
y
ca
p
ac
ity
.
I
n
t
h
is
p
ap
er
,
we
p
r
o
p
o
s
ed
a
s
o
lu
tio
n
to
alwa
y
s
m
ain
t
ain
th
e
b
o
u
n
d
ar
y
o
f
th
e
to
tal
m
em
o
r
y
ca
p
ac
ity
in
th
e
f
o
llo
win
g
r
o
b
u
s
t
b
o
u
n
d
o
f
OP
as
(
7
)
as
f
o
llo
ws:
(
)
=
∈
[
∑
|
|
+
∑
|
|
+
‖
‖
∞
∈
∈
]
(
7
)
Similar
with
(
)
=
∈
[
∑
|
|
+
∑
|
|
∈
∈
]
+
‖
‖
∞
(
8
)
T
h
en
,
f
o
r
h
y
b
r
id
SS
D
an
d
Mo
b
ileNetV2
,
we
r
ep
lace
d
all
r
eg
u
lar
co
n
v
o
lu
tio
n
s
with
s
ep
ar
ab
le
co
n
v
o
l
u
tio
n
s
in
th
e
SS
D
n
etwo
r
k
'
s
p
r
ed
ictiv
e
class
e
s
[
2
]
to
r
ed
u
ce
th
e
n
u
m
b
er
o
f
p
ar
am
eter
s
an
d
h
e
lp
th
e
m
o
d
el
d
ec
r
ea
s
e
th
e
am
o
u
n
t
o
f
to
tal
m
em
o
r
y
ca
p
ac
ity
as
s
h
o
wed
in
(
8
)
b
u
t
s
till
m
ain
tain
th
e
b
o
u
n
d
ar
y
o
f
co
m
p
u
tin
g
s
tep
s
.
I
n
p
a
r
ticu
lar
,
t
h
e
o
u
tp
u
t
is
lab
eled
with
th
e
o
b
ject
an
d
t
h
e
co
n
f
id
en
ce
lev
el
is
in
p
er
ce
n
tag
e
ter
m
s
.
I
n
th
e
ex
p
er
im
en
ts
o
f
th
is
p
ap
e
r
,
th
e
im
p
r
o
v
e
d
SS
D
-
Mo
b
ile
Net
V
2
m
eth
o
d
also
s
h
o
wed
h
i
g
h
er
ef
f
i
cien
cy
th
an
th
e
m
eth
o
d
o
f
[
2
5
]
esp
ec
ially
wh
e
n
ap
p
lied
t
o
th
e
B
NW
AS
.
3.
AP
P
L
YING
C
NNs
T
O
DE
SI
G
N
T
H
E
B
RIDG
E
NA
VIGA
T
I
O
NAL
WA
T
CH
A
ND
AL
A
RM
SYST
E
M
3
.
1
.
B
NWAS
des
ig
n ba
s
e
d o
n
re
g
ula
t
io
ns
o
f
I
M
O
M
SC.
1
2
8
(
7
5
)
I
n
r
ec
en
t
y
ea
r
s
,
it
is
k
n
o
wn
t
h
at
s
h
ip
s
u
s
u
ally
p
er
f
o
r
m
u
n
d
er
th
e
c
o
m
p
lex
ity
an
d
v
u
ln
e
r
ab
ilit
y
o
f
en
v
ir
o
n
m
en
t,
s
o
th
at
th
e
ch
alle
n
g
e
o
f
s
h
i
p
d
e
v
elo
p
m
en
t
r
em
a
in
s
an
p
r
o
b
lem
o
f
s
ig
n
i
f
ican
t a
d
v
an
ce
m
e
n
ts
f
r
o
m
r
esear
ch
er
s
.
T
h
ey
h
av
e
b
ee
n
p
aid
at
ten
tio
n
to
s
tu
d
y
o
f
s
h
ip
[
2
7
-
3
0
]
to
m
ee
t
th
e
I
MO
s
tan
d
ar
d
s
.
R
ec
en
tly
,
th
e
au
th
o
r
s
[
1
8
]
h
av
e
s
tu
d
ied
an
d
ap
p
lied
th
e
m
o
d
if
ied
SS
DL
ite_
Mo
b
ileNetV2
h
y
b
r
i
d
alg
o
r
i
th
m
to
B
N
W
AS
b
y
u
s
in
g
th
e
h
ar
d
war
e
b
ased
o
n
r
asp
b
er
r
y
Pi
-
3
to
m
ee
t
t
h
e
r
e
q
u
ir
em
en
ts
o
f
I
MO
MSC
.
1
2
8
(
7
5
)
a
n
d
SOLA
S
C
h
ap
ter
V,
R
eg
.
1
9
MSC
.
2
8
2
(
8
6
)
[
2
3
]
r
ev
is
ed
o
n
J
u
n
e
5
,
2
0
0
9
[
2
0
]
v
alid
f
o
r
s
h
ip
s
class
if
ied
b
y
s
ize:
−
J
u
ly
2
0
1
1
: n
ew
v
ess
els in
ex
c
ess
o
f
1
5
0
to
n
n
es.
−
J
u
ly
2
0
1
1
: a
ll p
ass
en
g
er
v
ess
els.
−
J
u
ly
2
0
1
2
: a
ll v
ess
els in
ex
ce
s
s
o
f
3
,
0
0
0
to
n
n
es.
−
J
u
ly
2
0
1
3
: a
ll v
ess
els b
etwe
en
5
0
0
a
n
d
3
,
0
0
0
to
n
n
es.
−
J
u
ly
2
0
1
4
: a
ll v
e
s
s
els b
etwe
en
1
5
0
a
n
d
5
0
0
to
n
n
es.
B
NW
AS
is
a
m
o
n
ito
r
in
g
an
d
Alar
m
s
y
s
tem
wh
ich
n
o
tifie
s
o
th
er
o
f
f
icer
s
o
r
ca
p
tain
s
if
th
e
o
f
f
icer
o
n
watc
h
(
OOW)
d
o
es
n
o
t
r
esp
o
n
d
o
r
h
e/sh
e
is
in
ca
p
ab
le
o
f
p
er
f
o
r
m
in
g
th
e
watc
h
d
u
ties
ef
f
icien
tly
wh
ich
ca
n
lead
to
m
ar
itime
a
cc
id
e
n
ts
.
T
h
e
s
y
s
tem
m
o
n
ito
r
s
th
e
aw
ar
en
ess
o
f
th
e
o
f
f
icer
o
f
th
e
watc
h
(
OOW)
an
d
au
to
m
atica
lly
aler
ts
th
e
Ma
s
ter
o
r
an
o
th
e
r
q
u
alif
ied
OOW
if
f
o
r
an
y
r
ea
s
o
n
t
h
e
OOW
b
ec
o
m
es
in
ca
p
ab
le
o
f
p
er
f
o
r
m
in
g
d
u
ties
.
T
h
is
is
ac
h
i
ev
ed
th
r
o
u
g
h
a
m
i
x
o
f
alar
m
s
an
d
in
d
ica
tio
n
s
wh
ich
aler
t
b
a
ck
u
p
OOWs
as
wel
l
as
th
e
Ma
s
ter
.
B
NW
A
S
war
n
in
g
s
ar
e
g
iv
en
in
t
h
e
ca
s
e
o
f
in
ca
p
ac
ity
o
f
th
e
watc
h
k
e
ep
in
g
o
f
f
icer
d
u
e
to
ac
cid
en
ts
,
s
ick
n
ess
o
r
in
th
e
ev
en
t
o
f
a
s
ec
u
r
ity
b
r
ea
ch
,
e.
g
.
p
ir
ac
y
an
d
/o
r
h
ijack
in
g
.
Un
l
ess
d
ec
id
ed
b
y
th
e
Ma
s
ter
o
n
ly
,
th
e
B
NW
AS sh
al
l r
em
ain
o
p
e
r
atio
n
al
at
all
tim
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
1
:
2
4
4
-
2
5
1
248
Ou
tp
u
ts
o
f
th
e
s
y
s
tem
s
h
o
u
l
d
b
e
av
ailab
le
f
o
r
co
n
n
ec
tio
n
o
f
ad
d
itio
n
al
b
r
id
g
e
v
is
u
al
in
d
icatio
n
s
,
au
d
ib
le
alar
m
s
an
d
r
e
m
o
te
au
d
ib
le
alar
m
s
as
in
[
9
]
.
T
h
e
ap
p
li
ed
to
th
e
ac
tu
al
s
y
s
tem
d
esig
n
in
Fig
u
r
e
2
(
a)
an
d
d
esig
n
in
g
d
iag
r
am
is
s
h
o
wn
in
Fig
u
r
e
2
(
b
)
.
T
h
e
co
n
n
ec
te
d
co
m
p
u
te
r
wo
r
k
s
in
tan
d
em
with
r
asp
b
er
r
y
Pi
-
3
(
p
lay
s
th
e
r
o
le
o
f
th
e
ce
n
tr
al
p
r
o
ce
s
s
in
g
b
o
a
r
d
o
n
Fig
u
r
e
2
(
b
)
to
c
o
llect
in
p
u
t
a
n
d
o
u
tp
u
t
d
ata
o
f
th
e
test
in
g
p
r
o
ce
s
s
.
Har
d
war
e
is
d
esig
n
ed
to
p
er
f
o
r
m
alar
m
f
u
n
ctio
n
s
.
T
o
co
m
p
a
r
e
th
e
e
f
f
ec
tiv
en
ess
o
f
th
e
s
o
lu
tio
n
with
o
th
e
r
ap
p
licatio
n
s
b
ased
o
n
h
ar
d
war
e
an
d
p
r
ac
tical
co
n
d
itio
n
s
in
th
e
b
r
id
g
e
o
f
th
e
Saig
o
n
Millen
n
iu
m
Sh
ip
,
we
d
ep
lo
y
ed
f
o
u
r
s
o
lu
tio
n
s
to
g
et
r
esu
lts
.
I
n
th
is
wo
r
k
,
we
f
o
cu
s
ed
o
n
tw
o
f
ac
to
r
s
,
in
clu
d
in
g
p
r
o
ce
s
s
in
g
s
p
ee
d
an
d
o
u
tp
u
t
r
eliab
ilit
y
to
ap
p
ly
o
b
j
ec
t
d
etec
to
r
s
o
n
th
e
d
esig
n
ed
s
y
s
tem
b
y
u
s
in
g
th
e
m
o
d
if
ied
SS
DL
ite_
Mo
b
ileNetV2
b
o
u
n
d
ed
C
NN
alg
o
r
ith
m
.
220
V
A
C
24
V
D
C
C
A
M
E
R
A
B
U
T
T
O
N
S
C
O
M
P
U
T
E
R
M
O
T
I
O
N
S
E
N
S
O
R
H
I
S
T
O
R
Y
D
A
T
A
A
L
A
R
M
D
E
V
I
C
E
S
T
O
U
C
H
S
C
R
E
E
N
M
A
IN
C
O
N
T
R
O
L
S
Y
S
T
E
M
(
RA
S
PI
3
B
+
U
S
IN
G
T
E
N
S
O
R
F
L
O
W
A
PI
)
C
O
N
T
R
O
L
VI
A
I
N
T
E
R
N
E
T
(
a
)
(
b
)
F
i
g
u
r
e
2
.
T
h
e
d
e
s
i
g
n
e
d
B
N
W
A
S
G
T
S
.
V
1
s
y
s
t
e
m
t
e
s
t
e
d
o
n
H
C
M
C
i
t
y
U
n
i
v
e
r
s
i
t
y
o
f
T
r
a
n
s
p
o
r
t
;
(
a
)
B
N
W
A
S
-
G
T
S
.
V
1
s
y
s
t
e
m
t
e
s
t
ed
o
n
H
C
M
C
i
t
y
U
n
i
v
e
r
s
i
t
y
o
f
T
r
a
n
s
p
o
r
t
,
a
n
d
(
b
)
S
t
r
u
c
t
u
r
e
o
f
d
e
s
i
g
n
e
d
B
N
W
A
S
3
.
2
.
T
esting
t
he
des
ig
ned B
NWAS
o
n Sa
ig
o
n M
il
lenni
u
m
Vess
el
in Sa
ig
o
n Riv
er
T
h
e
im
ag
e
h
as
b
ee
n
r
ec
o
r
d
ed
f
r
o
m
Saig
o
n
Millen
n
iu
m
s
h
ip
at
So
n
Hai
Sh
ip
y
ar
d
,
Ho
C
h
i
Min
h
C
ity
,
Vietn
am
.
T
h
is
im
a
g
e
is
ca
p
tu
r
ed
th
r
o
u
g
h
th
e
l
o
g
itech
C
2
7
0
ca
m
er
a
an
d
p
r
o
ce
s
s
ed
b
y
h
y
b
r
id
n
etwo
r
k
-
b
ased
o
b
ject
id
en
t
if
icatio
n
alg
o
r
ith
m
s
SS
D
-
Mo
b
ile
Net
V2
.
T
h
e
o
u
tp
u
t
is
th
e
p
r
o
ce
s
s
ed
im
ag
e
e
x
tr
ac
tin
g
th
e
d
etec
ted
o
b
ject
f
r
am
e
a
n
d
th
e
r
eliab
ilit
y
ca
lcu
lated
as
a
p
er
ce
n
tag
e.
W
ith
th
e
tech
n
iq
u
e
u
s
ed
i
n
th
i
s
p
ap
er
,
th
e
s
y
s
tem
ca
n
id
en
tif
y
m
a
n
y
o
f
f
icer
s
in
th
e
b
r
id
g
e
an
d
th
e
m
ax
im
u
m
n
u
m
b
er
o
f
ac
ce
s
s
es
to
th
e
d
etec
tio
n
f
r
am
e
is
2
0
pe
o
p
le
at
a
tim
e.
W
h
en
id
en
tif
y
in
g
o
f
f
icer
s
in
th
e
b
r
id
g
e
,
th
e
s
y
s
tem
allo
ws
cu
s
to
m
ized
f
u
n
ctio
n
s
v
ia
th
e
to
u
ch
s
cr
ee
n
o
r
p
u
s
h
-
b
u
tto
n
o
n
th
e
b
r
id
g
e.
T
esti
n
g
th
e
d
esig
n
ed
B
NW
AS
o
n
Saig
o
n
Millen
n
iu
m
V
ess
el
in
Saig
o
n
R
iv
er
as b
elo
ws:
−
C
ase
1
:
if
th
e
s
y
s
tem
d
eter
m
in
es
th
at
th
er
e
is
n
o
o
f
f
icer
in
th
e
b
r
id
g
e,
a
tim
er
will
b
e
tu
r
n
ed
o
n
an
d
th
e
co
u
n
td
o
wn
tim
e
will
wait
f
o
r
th
e
o
f
f
icer
to
ap
p
ea
r
.
Du
r
in
g
th
e
ac
tiv
e
tim
er
p
er
io
d
,
th
e
f
u
n
ctio
n
o
f
s
witch
m
o
d
es
an
d
c
o
u
n
td
o
wn
tim
er
ar
e
d
is
ab
led
.
I
f
d
u
r
i
n
g
th
e
co
u
n
td
o
wn
,
th
er
e
is
an
o
f
f
ic
er
in
th
e
b
r
i
d
g
e
(
n
o
p
h
y
s
ical
im
p
ac
t
is
n
ee
d
ed
o
n
th
e
s
y
s
tem
)
,
th
e
tim
er
is
r
eset
an
d
th
e
s
y
s
tem
r
etu
r
n
s
to
its
n
o
r
m
al
s
tate
,
o
f
f
icer
s
ca
n
o
p
er
ate
an
d
u
s
e
th
e
s
y
s
tem
f
u
n
ctio
n
k
ey
s
.
−
C
ase
2
:
if
n
o
o
f
f
icer
r
etu
r
n
s
a
n
d
th
e
tim
er
h
as
co
u
n
ted
to
z
er
o
(
tim
eo
u
t)
,
a
f
lash
war
n
in
g
s
ig
n
al
will
b
e
ac
tiv
ated
in
th
e
b
r
id
g
e;
th
is
s
ta
g
e
is
ca
lled
th
e
p
r
im
ar
y
alar
m
s
tag
e.
T
h
is
s
ig
n
al
ca
n
b
e
s
ee
n
an
y
wh
er
e
in
th
e
b
r
id
g
e
an
d
in
ac
c
o
r
d
a
n
ce
with
I
MO
s
tan
d
a
r
d
s
.
On
th
e
d
is
p
la
y
s
cr
ee
n
,
th
e
alar
m
lev
el
wil
l
ap
p
ea
r
,
an
d
all
s
y
s
tem
p
ar
am
eter
s
will
b
e
s
av
ed
to
t
h
e
h
is
to
r
y
f
ile,
th
en
a
n
ex
t
tim
er
is
s
tar
ted
t
o
m
o
v
e
to
th
e
n
e
x
t
alar
m
s
tag
e.
Su
b
s
eq
u
en
t
alar
m
test
s
ar
e
test
ed
an
d
th
e
f
in
al
r
esu
lt
s
ar
e
co
n
s
is
ten
t
with
I
MO
r
e
q
u
ir
em
en
ts
.
No
t
o
n
ly
d
id
th
e
s
y
s
tem
r
ec
o
g
n
ize
th
e
o
f
f
icer
p
r
esen
ce
in
th
e
b
r
i
d
g
e,
it
also
an
aly
ze
d
th
e
o
f
f
ic
er
s
'
ac
tio
n
s
an
d
is
s
u
ed
war
n
in
g
s
wh
en
th
ey
f
o
u
n
d
o
f
f
icer
s
s
tan
d
in
g
s
till
f
o
r
to
o
lo
n
g
o
r
s
leep
in
g
wh
ile
o
n
d
u
ty
.
I
n
ex
p
er
im
en
t,
t
h
e
test
d
etec
ted
an
o
f
f
icer
wh
o
s
at
in
s
ilen
ce
f
o
r
to
o
lo
n
g
o
r
s
h
o
wed
s
ig
n
s
o
f
d
r
o
wsi
n
ess
as in
Fig
u
r
e
3
.
T
h
e
test
was
r
ec
o
r
d
ed
wh
en
w
e
ask
ed
an
o
f
f
icer
t
o
s
it
s
ilen
tl
y
o
n
th
e
d
r
iv
er
'
s
s
ea
t
(
at
lea
s
t
2
0
s
ec
o
n
d
s
)
to
s
ee
wh
eth
er
th
e
o
f
f
icer
s
tan
d
s
s
till
f
o
r
to
o
lo
n
g
o
r
h
as
a
d
r
o
wsi
n
ess
.
At
th
e
s
am
e
tim
e,
an
u
n
d
er
g
r
o
u
n
d
r
u
n
n
in
g
co
u
n
ter
will
an
aly
ze
th
e
r
elativ
e
p
o
s
itio
n
o
f
th
e
o
f
f
i
ce
r
an
d
g
iv
e
a
r
elativ
e
er
r
o
r
.
B
ased
o
n
th
e
r
esu
lts
o
f
ea
ch
f
r
am
e
an
aly
s
is
,
af
ter
2
0
s
ec
o
n
d
s
,
if
th
e
r
elativ
e
p
o
s
iti
o
n
er
r
o
r
d
o
es n
o
t e
x
ce
ed
1
0
%
,
th
e
p
r
im
ar
y
alar
m
is
s
et
an
d
th
e
n
ex
t a
lar
m
tim
e
r
will st
ar
t c
o
u
n
tin
g
d
o
wn
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
ommun
C
o
m
p
u
t E
l Co
n
tr
o
l
A
p
p
lyin
g
c
o
n
vo
lu
tio
n
a
l n
e
u
r
a
l
n
etw
o
r
ks fo
r
limited
-
mem
o
r
y
a
p
p
lica
tio
n
(
X
u
a
n
-
K
ien
Da
n
g
)
249
(
a
)
(
b
)
Fig
u
r
e
3
.
T
esti
n
g
t
h
e
d
esig
n
e
d
B
NW
A
S o
n
Saig
o
n
Millen
n
i
u
m
Vess
el
in
Saig
o
n
R
iv
er
;
(
a)
test
in
g
n
o
ala
r
m
s
tag
e,
an
d
(
b
)
test
in
g
alar
m
s
tag
es
3
.
2
.
Su
mm
a
ry
ex
perim
ent
a
l
re
s
ults
Hig
h
ly
co
n
f
i
g
u
r
a
b
le
m
o
d
els
r
u
n
n
in
g
o
n
T
I
T
AN
X
GPUs
p
r
o
d
u
ce
d
p
r
o
ce
s
s
in
g
s
p
ee
d
s
b
etwe
en
1
7
an
d
3
7
f
r
am
es
p
e
r
s
ec
o
n
d
.
Ho
wev
er
,
wh
en
ex
p
er
im
en
tin
g
o
n
C
OC
O
d
ata
s
et
s
an
d
m
AP
ca
lcu
latio
n
s
o
n
all
o
b
ject
class
es,
th
e
r
esu
lts
o
n
ly
r
ea
ch
ed
2
1
-
2
8
%.
E
x
p
e
r
im
en
tal
r
esu
l
ts
o
n
p
r
o
ce
s
s
in
g
s
p
ee
d
o
n
s
y
s
tem
wer
e
test
ed
d
ir
ec
tly
o
n
th
e
b
r
id
g
e
with
n
o
r
m
al
wo
r
k
i
n
g
co
n
d
itio
n
s
an
d
th
e
r
esu
lts
wer
e
s
h
o
wn
h
ig
h
p
er
f
o
r
m
an
ce
f
r
o
m
76
-
9
7
% a
s
in
T
ab
le
1
.
T
ab
le
1
.
T
esti
n
g
p
er
f
o
r
m
an
ce
r
esu
lts
o
f
4
m
o
d
els in
ex
p
e
r
im
en
tal
M
o
d
e
l
n
a
m
e
Te
st
o
n
G
P
U
TI
TA
N
X
Te
st
o
n
R
a
s
p
b
e
r
r
y
P
i
3
B
+
S
p
e
e
d
(
ms)
C
O
C
O
(
mA
P
)
S
p
e
e
d
(
F
P
S
)
R
e
a
l
t
i
m
e
o
n
b
r
i
d
g
e
(
mA
P
)
ssd
_
mo
b
i
l
e
n
e
t
_
v
1
_
c
o
c
o
30
21
1
.
0
5
76
ssd
_
mo
b
i
l
e
n
e
t
_
v
2
_
c
o
c
o
31
22
0
.
8
3
94
ssd
l
i
t
e
_
m
o
b
i
l
e
n
e
t
_
v
2
_
c
o
c
o
27
22
1
.
0
8
86
f
a
st
e
r
_
r
c
n
n
_
i
n
c
e
p
t
i
o
n
_
v
2
_
c
o
c
o
58
28
0
.
0
8
97
T
h
is
im
p
r
ess
iv
e
r
esu
lt
is
ac
h
i
ev
ed
wh
e
n
in
s
tallin
g
th
e
ca
m
er
a
in
t
h
e
b
r
id
g
e
in
a
co
n
v
en
i
en
t
p
o
s
itio
n
wh
ile
th
e
h
ar
d
war
e
is
a
m
o
b
ile
d
ev
ice
with
o
n
ly
AR
M
C
PU
an
d
n
o
in
te
g
r
ated
GPU.
T
h
e
h
ig
h
est
p
r
o
ce
s
s
in
g
s
p
ee
d
is
o
n
ly
ap
p
r
o
x
im
ately
1
FP
S.
T
h
e
d
is
cu
s
s
o
f
th
e
ex
p
er
im
en
tal
r
esu
lts
f
o
cu
s
m
o
r
e
d
et
ail
in
T
ab
le
1
.
T
h
e
r
esu
lt
s
h
o
wed
th
at
4
m
o
d
els
test
ed
o
n
o
u
r
h
ar
d
war
e
(
r
asp
b
er
r
y
Pi
3
B
+)
u
s
in
g
o
u
r
m
eth
o
d
b
etter
th
an
GPU
T
I
T
AN
X
h
ar
d
war
e
(
d
if
f
e
r
en
ce
h
ar
d
war
e)
ab
o
u
t
s
p
ee
d
(
m
s
)
an
d
m
AP.
So
th
at,
th
e
FP
S
s
p
ee
d
o
f
th
e
te
s
t
m
eth
o
d
s
is
in
d
icate
d
in
Fig
u
r
e
4
an
d
t
h
is
is
a
g
o
o
d
r
esp
o
n
s
e
r
ate
f
o
r
a
m
o
n
ito
r
in
g
s
y
s
tem
.
T
h
e
o
u
tp
u
t
r
eliab
ilit
y
is
h
ig
h
e
s
t
wh
en
test
ed
with
th
e
f
aster
R
C
N
N
d
etec
to
r
,
h
o
wev
e
r
with
0
.
0
8
FP
S
(
ab
o
u
t
1
2
.
5
s
ec
o
n
d
s
to
p
r
o
ce
s
s
a
f
r
am
e)
it
i
s
n
o
t
p
o
s
s
ib
le
t
o
m
e
et
o
n
a
m
o
n
ito
r
in
g
s
y
s
tem
.
Ob
ject
d
etec
to
r
s
b
ased
o
n
th
e
SS
D_
Mo
b
ileNet
s
tr
u
ctu
r
e
(
in
b
r
o
wn
co
lo
r
)
p
r
o
d
u
ce
h
i
g
h
ly
r
eliab
le
r
esu
lts
an
d
m
ee
t
p
r
o
ce
s
s
in
g
s
p
ee
d
r
eq
u
ir
em
en
ts
.
Me
a
n
wh
il
e,
th
e
r
esu
lt
o
f
SS
D_
Mo
b
ileNe
tV1
(
y
ello
w)
an
d
SS
D_
Mo
b
ile
N
etV2
(
g
r
ee
n
)
s
ets
ar
e
alm
o
s
t
eq
u
iv
alen
t,
b
u
t
th
e
l
o
ad
tim
e
o
f
th
e
m
o
d
el
is
s
lo
w
d
u
e
to
lar
g
e
ca
p
ac
ity
an
d
ac
tu
al
o
u
tp
u
t.
T
h
er
e
ar
e
s
till
ce
r
tain
d
ev
iatio
n
s
.
T
h
u
s
,
th
e
im
p
r
o
v
ed
SS
DL
ite_
Mo
b
ileNetV2
s
o
lu
tio
n
g
iv
es
g
o
o
d
r
esu
lts
r
elatin
g
to
q
u
ality
,
p
r
o
ce
s
s
in
g
s
p
ee
d
,
f
ast
m
o
d
el
lo
a
d
tim
e
(
s
tab
le
r
u
n
n
in
g
o
n
r
asp
b
er
r
y
PI
-
3
)
an
d
h
as
h
i
g
h
er
ac
cu
r
ac
y
th
an
th
e
o
th
er
s
o
lu
tio
n
s
.
(
a
)
(
b
)
Fig
u
r
e
4
.
T
h
e
FP
S sp
ee
d
o
f
th
e
test
m
eth
o
d
s
; (
a)
co
m
p
ar
e
p
r
o
ce
s
s
in
g
s
p
ee
d
o
f
o
b
ject
d
etec
to
r
s
o
n
B
NW
AS
h
ar
d
war
e
a
n
d
(
b
)
co
m
p
ar
e
th
e
o
u
tp
u
t
r
eliab
ilit
y
o
f
o
b
ject
d
et
ec
to
r
s
o
n
B
NW
AS h
ar
d
war
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
1
:
2
4
4
-
2
5
1
250
4.
CO
NCLU
SI
O
N
I
n
th
is
p
ap
er
,
we
s
tu
d
ie
d
an
d
ap
p
lied
th
e
m
o
d
if
ied
SS
DL
ite_
Mo
b
ileNetV2
b
o
u
n
d
ed
C
NN
alg
o
r
ith
m
to
B
NW
AS
-
GT
S.V1
.
T
h
e
h
ar
d
war
e
was
d
esig
n
ed
b
ased
o
n
r
a
s
p
b
er
r
y
Pi
-
3
,
a
n
e
m
b
ed
d
ed
s
in
g
le
b
o
a
r
d
co
m
p
u
ter
with
C
PU
s
m
ar
tp
h
o
n
e
lev
el,
lim
ited
R
AM
with
o
u
t C
UDA
G
PU.
E
x
p
er
im
en
tal
r
esu
lts
o
n
p
r
o
ce
s
s
in
g
s
p
ee
d
o
n
B
NW
AS
-
GT
S.V1
wer
e
test
ed
d
ir
ec
tly
o
n
t
h
e
b
r
id
g
e
with
n
o
r
m
al
wo
r
k
in
g
co
n
d
itio
n
s
.
T
h
is
im
p
r
ess
iv
e
r
esu
lt
was
ac
h
iev
ed
wh
e
n
i
n
s
tallin
g
th
e
ca
m
e
r
a
in
th
e
b
r
id
g
e
i
n
a
co
n
v
en
ien
t
p
o
s
itio
n
w
h
ile
t
h
e
h
ar
d
war
e
u
s
ed
a
m
o
b
ile
d
ev
ice.
T
h
e
im
p
r
o
v
ed
SS
D
-
Mo
b
ile
Net
V2
b
ased
o
n
b
o
u
n
d
ed
C
NN
alg
o
r
ith
m
also
s
h
o
wed
h
ig
h
er
ef
f
icien
cy
esp
ec
ially
wh
e
n
ap
p
lied
to
th
e
B
NW
AS
.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
is
s
tu
d
y
was
s
u
p
p
o
r
ted
b
y
t
h
e
Ap
p
lied
B
asic
R
esear
ch
Pro
g
r
am
o
f
Min
is
tr
y
o
f
T
r
an
s
p
o
r
t
o
f
Vietn
am
DT
1
8
4
0
8
(
2
0
1
8
)
,
DT
2
0
3
0
3
6
(
2
0
2
0
)
an
d
th
e
Star
t u
p
an
d
I
n
n
o
v
atio
n
C
en
ter
o
f
B
a
R
ia
-
Vu
n
g
T
au
,
Viet
Nam
.
RE
F
E
R
E
NC
E
S
[1
]
A.
Kriz
h
e
v
s
k
y
,
e
t
a
l.
,
“
Im
a
g
e
n
e
t
c
las
sifica
ti
o
n
with
d
e
e
p
c
o
n
v
o
l
u
t
io
n
a
l
n
e
u
ra
l
n
e
two
rk
s
,
”
Co
mm
u
n
ica
ti
o
n
s
o
f
t
h
e
ACM
,
v
o
l
.
6
0
,
n
o
.
6
,
p
p
.
1
0
9
7
-
1
1
0
5
,
2
0
1
2
.
[2
]
W.
Li
u
,
e
t
a
l.
,
“
S
sd
:
S
in
g
le sh
o
t
m
u
lt
ib
o
x
d
e
tec
to
r
,
”
Le
c
tu
re
N
o
te
in
Co
m
p
u
ter
Vi
si
o
n
-
ECCV
,
p
p
.
2
1
-
3
7
,
2
0
1
6
.
[
3
]
Z
u
h
e
r
m
a
n
R
u
s
t
a
m
,
e
t
a
l
.
,
“
P
u
l
m
o
n
a
r
y
r
o
n
t
g
e
n
c
l
a
s
s
i
f
i
c
a
t
i
o
n
t
o
d
e
t
e
c
t
p
n
e
u
m
o
n
i
a
d
i
s
e
a
s
e
u
s
i
n
g
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
,
”.
T
E
L
K
O
M
N
I
K
A
T
e
l
e
c
o
m
m
u
n
i
c
a
t
i
o
n
C
o
m
p
u
t
i
n
g
E
l
e
c
t
r
o
n
i
c
s
a
n
d
C
o
n
t
r
o
l
,
v
o
l
.
1
8
,
n
o
.
3
,
p
p
.
1
5
2
2
-
1528,
2020.
[4
]
Ch
e
n
g
c
h
e
n
g
Nin
g
,
e
t
a
l
.
,
“
In
c
e
p
ti
o
n
S
i
n
g
le
S
h
o
t
M
u
l
ti
b
o
x
De
tec
to
r
fo
r
Ob
jec
t
De
tec
ti
o
n
,
”
IEE
E
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
M
u
lt
ime
d
i
a
&
Ex
p
o
W
o
rk
sh
o
p
s (ICM
EW
),
p
p
.
5
4
9
-
5
5
4
,
2
0
1
7
.
[5
]
S
.
Re
n
,
e
t
a
l.
,
“
F
a
ste
r
r
-
c
n
n
:
T
o
-
w
a
rd
s
re
a
l
-
ti
m
e
o
b
jec
t
d
e
tec
ti
o
n
wit
h
re
g
i
o
n
p
r
o
p
o
sa
l
n
e
two
r
k
s
,
”
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
P
a
tt
e
rn
An
a
lys
is a
n
d
M
a
c
h
in
e
In
telli
g
e
n
c
e
,
v
o
l.
3
9
,
n
o
.
6
,
p
p
.
1
1
3
7
-
1
1
4
9
,
2
0
1
7
.
[6
]
R.
Ra
h
m
a
t,
e
t
a
l.
,
“
Ad
v
e
rt
ise
m
e
n
t
b
il
lb
o
a
rd
d
e
tec
ti
o
n
a
n
d
g
e
o
tag
g
i
n
g
sy
ste
m
with
in
d
u
c
ti
v
e
tran
sfe
r
lea
rn
in
g
i
n
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
,
”
T
EL
KOM
NIKA
T
e
lec
o
mm
u
n
ica
t
i
o
n
Co
m
p
u
t
in
g
El
e
c
tro
n
ics
a
n
d
Co
n
tro
l
,
v
o
l.
1
7
,
n
o
.
5
,
p
p
.
2
6
5
9
-
2
6
6
6
,
2
0
1
9
.
[
7
]
Z
.
R
u
s
t
a
m
,
e
t
a
l
.
,
“
P
u
l
m
o
n
a
r
y
r
o
n
t
g
e
n
c
l
a
s
s
i
f
i
c
a
t
i
o
n
t
o
d
e
t
e
c
t
p
n
e
u
m
o
n
i
a
d
i
s
e
a
s
e
u
s
i
n
g
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
,
”
T
E
L
K
O
M
N
I
K
A
T
e
l
e
c
o
m
m
u
n
i
c
a
t
i
o
n
C
o
m
p
u
t
i
n
g
E
l
e
c
t
r
o
n
i
c
s
a
n
d
C
o
n
t
r
o
l
,
v
o
l
.
1
8
,
n
o
.
3
,
p
p
.
1
5
2
2
-
1528,
2
0
2
0
.
[8
]
Tu
ti
P
.
,
e
t
a
l.
,
“
Im
a
g
e
c
las
sific
a
ti
o
n
o
f
g
o
lek
p
u
p
p
e
t
ima
g
e
s
u
sin
g
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
a
lg
o
r
it
h
m
,
”
In
ter
n
a
t
io
n
a
l
J
o
u
rn
a
l
o
f
A
d
v
a
n
c
e
s,
v
o
l
.
1
1
,
p
p
.
3
4
-
4
5
,
2
0
1
9
.
[9
]
M
u
h
a
m
a
d
.
Y,
e
t
a
l.
,
“
Ap
p
li
c
a
ti
o
n
o
f
tran
sfe
r
lea
rn
in
g
u
sin
g
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
m
e
th
o
d
f
o
r
e
a
rly
d
e
tec
ti
o
n
o
f
terry
’s
n
a
il
,
”
J
o
u
r
n
a
l
o
f
Ph
y
sic
s: Co
n
fer
e
n
c
e
S
e
rie
s,
v
o
l.
1
2
0
1
,
p
p
.
1
-
9
,
2
0
1
9
.
[1
0
]
Jo
se
p
h
Re
d
m
o
n
,
e
t
a
l.
,
“
Yo
u
O
n
ly
Lo
o
k
On
c
e
:
U
n
ifi
e
d
,
Re
a
l
-
Ti
m
e
Ob
jec
t
De
tec
ti
o
n
,
”
IEE
E
Co
n
fer
e
n
c
e
o
n
C
o
mp
u
ter
Vi
sio
n
a
n
d
P
a
tt
e
rn
Rec
o
g
n
it
i
o
n
,
pp
.
7
7
9
-
7
8
8
,
2
0
1
5
.
[1
1
]
Dim
o
v
S
to
jce
Ilce
v
,
“
Arc
h
it
e
c
tu
re
o
f
th
e
g
l
o
b
a
l
n
a
v
ig
a
ti
o
n
sa
telli
te
sy
ste
m
fo
r
m
a
rit
ime
a
p
p
li
c
a
ti
o
n
s
,
”
T
EL
KOM
NIKA
T
e
lec
o
mm
u
n
ic
a
ti
o
n
C
o
mp
u
ti
n
g
El
e
c
tro
n
ics
a
n
d
Co
n
tro
l
,
v
o
l
.
1
8
,
n
o
.
3
,
p
p
.
1
6
0
0
-
1
6
0
9
,
J
u
n
e
2
0
2
0
.
[1
2
]
Dim
o
v
S
t
o
jce
Ilce
v
,
“
Arc
h
it
e
c
t
u
re
o
f
t
h
e
re
g
i
o
n
a
l
sa
telli
te
a
u
g
m
e
n
tatio
n
sy
ste
m
fo
r
m
a
rit
im
e
a
p
p
li
c
a
ti
o
n
s
,
”
T
EL
KOM
NIKA
T
e
lec
o
mm
u
n
ic
a
ti
o
n
C
o
mp
u
ti
n
g
El
e
c
tro
n
ics
a
n
d
Co
n
tro
l
,
v
o
l
.
1
8
,
n
o
.
3
,
p
p
.
1
6
1
0
-
1
6
2
0
,
J
u
n
e
2
0
2
0
.
[1
3
]
M.
M
u
k
h
a
y
a
d
i
,
e
t
a
l.
,
“
De
sig
n
i
n
g
a
c
o
n
ste
ll
a
ti
o
n
fo
r
AIS
m
issi
o
n
b
a
se
d
o
n
d
a
ta
a
c
q
u
isit
io
n
o
f
LAP
AN
-
A2
a
n
d
LAP
AN
-
A3
sa
telli
te
,
”
T
EL
KOM
NIKA
T
e
lec
o
mm
u
n
ica
ti
o
n
C
o
mp
u
ti
n
g
El
e
c
tro
n
ics
a
n
d
C
o
n
tr
o
l
,
vol
.
1
7
,
no
.
4
,
p
p
.
1
7
7
4
-
1
7
8
4
,
Au
g
u
st
2
0
1
9
.
[1
4
]
Xia
n
-
Ba
o
,
e
t
a
l.
,
“
S
o
l
a
r
Ce
ll
s
S
u
rfa
c
e
De
f
e
c
ts
De
t
e
c
ti
o
n
Ba
se
d
o
n
De
e
p
Lea
rn
in
g
,
”
P
a
tt
e
rn
Rec
o
g
n
it
.
Arti
f.
I
n
tell
.
v
o
l.
2
7
,
p
p
.
5
1
7
-
5
2
3
,
2
0
1
4
.
[1
5
]
S
imo
n
y
a
n
K
.
,
e
t
a
l.
,
.
“
Ve
ry
De
e
p
Co
n
v
o
lu
t
io
n
a
l
Ne
two
rk
s
f
o
r
La
rg
e
-
S
c
a
le
Im
a
g
e
Re
c
o
g
n
it
i
o
n
,
”
C
o
mp
u
ter
V
isio
n
a
n
d
Pa
tt
e
rn
Rec
o
g
n
it
io
n
,
p
p
.
1
4
0
9
-
1
5
5
6
,
2
0
1
4
.
[1
6
]
Ho
n
g
S
.
,
e
t
a
l.
,
“
Li
g
h
twe
ig
h
t
De
e
p
Ne
u
ra
l
Ne
two
rk
s
f
o
r
Re
a
l
-
ti
m
e
Ob
jec
t
De
tec
ti
o
n
,
”
T
h
e
1
st
In
ter
n
a
ti
o
n
a
l
W
o
rk
sh
o
p
o
n
Ef
fi
c
ien
t
M
e
th
o
d
s
f
o
r De
e
p
Ne
u
ra
l
Ne
tw
o
rk
s
,
p
p
.
1
-
7
,
2
0
1
6
.
[1
7
]
Hu
n
t
M
.
A
.
,
e
t
a
l.
,
“
Op
ti
m
izin
g
a
u
t
o
m
a
ti
c
d
e
fe
c
t
c
las
sifica
ti
o
n
fe
a
tu
re
a
n
d
c
las
sifier
p
e
rfo
rm
a
n
c
e
fo
r
p
o
st
,
”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
2
0
0
0
IEE
E/
S
E
M
I
Ad
v
a
n
c
e
d
S
e
mic
o
n
d
u
c
t
o
r
M
a
n
u
fa
c
tu
ri
n
g
Co
n
fer
e
n
c
e
a
n
d
W
o
rk
sh
o
p
,
p
p
.
1
1
6
-
1
2
3
,
2
0
0
0
.
[1
8
]
Da
n
g
Xu
a
n
Kie
n
,
e
t
a
l
.
,
"
A
p
p
l
y
in
g
Hy
b
rid
C
o
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
s
fo
r
Im
a
g
e
P
ro
c
e
s
sin
g
t
o
Brid
g
e
Na
v
ig
a
ti
o
n
a
l
Watc
h
&
Ala
rm
S
y
s
tem
,
"
J
o
u
rn
a
l
o
f
T
ra
n
sp
o
rt
a
ti
o
n
S
c
ien
c
e
a
n
d
T
e
c
h
n
o
lo
g
y
,
p
p
.
47
-
5
3
,
no
.
3
2
,
2
0
1
9
.
[1
9
]
M
.
A
b
a
d
i,
e
t
a
l
.
,
“
Te
n
so
rF
l
o
w:
Larg
e
-
sc
a
le
m
a
c
h
in
e
lea
rn
in
g
o
n
h
e
tero
g
e
n
e
o
u
s
s
y
s
tem
s
,
”
a
r
Xi
v
:1
6
0
3
.
0
4
4
6
,
p
p
.
1
-
1
9
,
2
0
1
5
.
[2
0
]
Ya
n
g
q
i
n
g
Jia
,
e
t
a
l
.
,
“
Ca
ffe
:
Co
n
v
o
l
u
ti
o
n
a
l
a
rc
h
it
e
c
tu
re
f
o
r
fa
st f
e
a
tu
re
e
m
b
e
d
d
i
n
g
,
”
Pr
o
c
e
e
d
in
g
s
o
f
th
e
2
2
n
d
AC
M
in
ter
n
a
t
io
n
a
l
c
o
n
fer
e
n
c
e
o
n
M
u
lt
i
me
d
ia
,
p
p
.
6
7
5
-
6
7
8
,
2
0
1
4
.
[2
1
]
IM
O M
S
C.
1
2
8
(7
5
),
“
P
e
rfo
rm
a
n
c
e
S
tan
d
a
rd
s
f
o
r
a
Bri
d
g
e
Na
v
i
g
a
ti
o
n
a
l
Wat
c
h
Ala
rm
S
y
ste
m
(BNW
AS)
,
”
2
0
0
2
.
[2
2
]
IM
O A.1
0
2
1
(2
6
),
“
Co
d
e
o
n
Ale
rt
s a
n
d
I
n
d
ica
to
rs
,
”
2
0
0
9
.
[2
3
]
IEC
6
2
6
1
6
,
“
M
a
rit
ime
Na
v
ig
a
ti
o
n
a
n
d
Ra
d
i
o
Co
m
m
u
n
ica
ti
o
n
Eq
u
ip
m
e
n
t
a
n
d
S
y
ste
m
s
-
Brid
g
e
Na
v
i
g
a
ti
o
n
a
l
Watc
h
Ala
rm
S
y
ste
m
(BNWAS
)
,”
2
0
1
0
.
[2
4
]
1
9
IM
O
M
S
C.
2
8
2
(8
6
),
“
Ch
a
p
ter
V,
Re
g
,
”
2
0
0
9
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
ommun
C
o
m
p
u
t E
l Co
n
tr
o
l
A
p
p
lyin
g
c
o
n
vo
lu
tio
n
a
l n
e
u
r
a
l
n
etw
o
r
ks fo
r
limited
-
mem
o
r
y
a
p
p
lica
tio
n
(
X
u
a
n
-
K
ien
Da
n
g
)
251
[2
5
]
M
a
rk
S
a
n
d
ler,
e
t
a
l
.
,
“
M
o
b
i
leN
e
tV2
:
I
n
v
e
rted
Re
sid
u
a
ls
a
n
d
Li
n
e
a
r
Bo
tt
len
e
c
k
s
,
”
Pr
o
c
.
I
EE
E
Co
n
fer
e
n
c
e
on
Co
mp
u
ter
V
isio
n
a
n
d
Pa
t
ter
n
Rec
o
g
n
it
io
n
,
p
p
.
4
5
1
0
-
4
5
2
0
,
2
0
1
8
.
[2
6
]
Ja
k
u
b
Na
lep
a
a
,
e
t
a
l.
,
“
To
wa
rd
s
re
so
u
rc
e
-
fru
g
a
l
d
e
e
p
c
o
n
v
o
l
u
t
io
n
a
l
n
e
u
ra
l
n
e
two
r
k
s
f
o
r
h
y
p
e
rsp
e
c
tral
ima
g
e
se
g
m
e
n
tatio
n
,
”
M
icr
o
p
r
o
c
e
ss
o
rs
a
n
d
M
icr
o
sy
ste
ms
,
v
ol
.
73
,
p
p
.
1
-
1
4
,
2
0
2
0
.
[
2
7
]
P
r
a
t
e
e
t
h
N
a
y
a
k
,
e
t
a
l
.
,
”
B
i
t
E
f
f
i
c
i
e
n
t
Q
u
a
n
t
i
z
a
t
i
o
n
f
o
r
D
e
e
p
N
e
u
r
a
l
N
e
t
w
o
r
k
s
”
E
M
C
2
-
N
e
u
r
I
P
S
W
o
r
k
s
h
o
p
,
p
p
.
1
-
5
,
2
0
1
9
.
[2
8
]
Xu
a
n
Kie
n
Da
n
g
,
e
t
a
l
.
,
“
An
a
l
y
z
in
g
th
e
se
a
we
a
th
e
r
e
ffe
c
ts
to
t
h
e
sh
ip
m
a
n
e
u
v
e
rin
g
i
n
Vie
t
Na
m
se
a
fro
m
Bin
h
Th
u
a
n
p
r
o
v
i
n
c
e
to
Ca
M
a
u
p
r
o
v
i
n
c
e
b
a
se
d
o
n
F
u
z
z
y
c
o
n
tr
o
l
m
e
th
o
d
,
”
T
EL
KO
M
NIKA
T
e
lec
o
mm
u
n
ica
t
io
n
Co
mp
u
t
in
g
El
e
c
tro
n
ics
a
n
d
C
o
n
tr
o
l
,
vol
.
1
6
,
no
.
2
,
p
p
.
5
3
3
-
5
4
3
,
2
0
1
8
.
[2
9
]
Vie
t
Du
n
g
Do
,
e
t
a
l.
,
"
Op
ti
m
a
l
c
o
n
tr
o
l
f
o
r
d
y
n
a
m
ic
p
o
siti
o
n
in
g
s
y
ste
m
b
a
se
d
o
n
F
u
z
z
y
-
P
S
O
a
d
v
a
n
c
e
d
tec
h
n
ica
l
,
"
T
EL
KOM
NIKA
T
e
lec
o
mm
u
n
ic
a
ti
o
n
C
o
mp
u
ti
n
g
El
e
c
tro
n
ics
a
n
d
Co
n
tro
l
,
v
o
l
.
1
6
,
n
o.
6
,
p
p
.
2
9
9
9
-
3
0
0
7
,
2
0
1
8
.
[3
0
]
Vie
t
Du
n
g
Do
,
e
t
a
l.
,
"
Th
e
fu
z
z
y
p
a
rti
c
le
sw
a
rm
o
p
ti
m
iza
ti
o
n
a
lg
o
rit
h
m
d
e
si
g
n
fo
r
d
y
n
a
m
ic
p
o
si
ti
o
n
in
g
s
y
ste
m
u
n
d
e
r
u
n
e
x
p
e
c
ted
imp
a
c
ts
,
"
J
o
u
rn
a
l
o
f
M
e
c
h
a
n
ica
l
E
n
g
in
e
e
rin
g
a
n
d
S
c
ien
c
e
s (JM
ES
),
v
o
l.
1
3
,
p
p
.
5
4
0
7
-
5
4
2
3
,
2
0
1
9
.
B
I
O
G
RAP
H
I
E
S O
F
A
UT
H
O
RS
Xua
n
-
K
ien
D
a
n
g
re
c
e
iv
e
d
P
h
.
D.
d
e
g
r
e
e
in
Co
n
tro
l
S
c
ien
c
e
a
n
d
En
g
in
e
e
ri
n
g
,
Hu
a
z
h
o
n
g
Un
iv
e
rsity
o
f
S
c
ien
c
e
a
n
d
Tec
h
n
o
lo
g
y
i
n
J
u
n
e
2
0
1
2
.
He
is
se
rv
in
g
a
s
th
e
Dire
c
to
r
o
f
G
ra
d
u
a
te
S
c
h
o
o
l,
Ho
Ch
i
M
i
n
h
Cit
y
U
n
iv
e
rsity
o
f
Tran
sp
o
rt,
Vie
tn
a
m
.
He
h
a
s
b
e
e
n
a
wa
rd
e
d
t
h
e
Be
st
P
a
p
e
r
A
wa
rd
in
t
h
e
4
th
C
o
n
f
e
re
n
c
e
o
f
S
c
ien
c
e
a
n
d
Tec
h
n
o
lo
g
y
,
H
o
Ch
i
M
in
h
Cit
y
Un
iv
e
rsity
o
f
Tran
sp
o
rt
(2
0
1
8
),
t
h
e
P
re
sid
e
n
t
P
rize
fo
r
Aw
a
rd
Wi
n
n
e
r
o
f
Th
e
E
x
c
e
ll
e
n
t
P
a
p
e
r
o
f
t
h
e
1
7
th
As
ia M
a
rit
ime
&
F
ish
e
ries
Un
iv
e
rsiti
e
s F
o
ru
m
(
2
0
1
8
)
.
His c
u
rre
n
t
re
se
a
rc
h
in
tere
s
ts f
o
c
u
s o
n
Co
n
tr
o
l
T
h
e
o
ry
,
Au
t
o
m
a
ti
o
n
,
M
a
rit
ime
Tec
h
n
o
l
o
g
y
,
Un
d
e
rwa
ter
Ve
h
icle
s,
Op
ti
m
a
l
a
n
d
Ro
b
u
st
C
o
n
tr
o
l,
a
n
d
Ne
two
rk
e
d
Co
n
tro
l
S
y
ste
m
.
He
h
a
s
b
e
e
n
se
rv
in
g
a
s
a
n
As
so
c
iatio
n
e
x
e
c
u
ti
v
e
c
o
m
m
it
tee
m
e
m
b
e
r
o
f
Vie
tn
a
m
Au
to
m
a
ti
o
n
As
so
c
iatio
n
(VA
A).
H
u
y
n
h
-
Nhu
Tr
u
o
n
g
re
c
e
iv
e
d
M
a
ste
r'
s
d
e
g
re
e
in
Au
to
m
a
ti
o
n
,
Ho
Ch
i
M
i
n
h
Cit
y
Un
i
v
e
rsity
o
f
Tran
s
p
o
rt
,
Vie
t
Na
m
,
2
0
1
1
.
S
h
e
wa
s
re
c
o
g
n
ize
d
a
s
a
P
h
.
D
.
st
u
d
e
n
t
o
f
Au
t
o
m
a
ti
o
n
a
n
d
Co
n
tr
o
l
E
n
g
i
n
e
e
rin
g
,
Ho
C
h
i
M
i
n
h
Cit
y
Un
iv
e
rsity
o
f
Tra
n
sp
o
r
t,
Vie
tn
a
m
,
2
0
1
9
.
S
h
e
is
t
h
e
P
rin
c
ip
a
l
o
f
Ba
Ria
-
Vu
n
g
Tau
C
o
ll
e
g
e
o
f
Tec
h
n
o
lo
g
y
,
Vie
t
n
a
m
.
M
rs.
Nh
u
’s
c
u
rre
n
t
re
se
a
rc
h
fo
c
u
se
s o
n
th
e
a
re
a
s o
f
C
o
n
tr
o
l
T
h
e
o
ry
,
Au
t
o
m
a
ti
o
n
,
a
n
d
Un
d
e
rwa
ter Ro
b
o
ti
c
.
Vie
t
-
Chi
n
h
Ng
u
y
e
n
re
c
e
iv
e
d
Ba
c
h
e
lo
r'
s
d
e
g
re
e
in
El
e
c
tri
c
a
l
a
n
d
El
e
c
tro
n
ic
E
n
g
i
n
e
e
rin
g
in
Ho
Ch
i
M
i
n
h
Cit
y
Un
i
v
e
rsity
o
f
T
e
c
h
n
o
l
o
g
y
,
Vie
t
Na
m
.
He
is
stu
d
y
in
g
t
h
e
M
a
ste
r’s
d
e
g
re
e
in
Au
to
m
a
ti
o
n
a
n
d
Co
n
tro
l
En
g
i
n
e
e
rin
g
,
Ho
C
h
i
M
i
n
h
Cit
y
U
n
iv
e
rsit
y
o
f
Tran
sp
o
rt,
Vie
t
Na
m
.
His c
u
rre
n
t
re
se
a
rc
h
in
tere
sts fo
c
u
s o
n
Co
n
tro
l
T
h
e
o
ry
,
Au
t
o
m
a
ti
o
n
a
n
d
De
e
p
Lea
rn
in
g
.
Th
i
-
Duy
e
n
-
Anh
P
h
a
m
n
o
w
is a
Tea
c
h
e
r
o
f
E
n
g
li
sh
a
t
Ho
Ch
i
M
in
h
Un
iv
e
rsity
o
f
Tran
s
p
o
rt
.
S
h
e
g
ra
d
u
a
ted
fro
m
Un
iv
e
rsity
o
f
S
o
c
ial
S
c
ien
c
e
s
a
n
d
Hu
m
a
n
it
ies
in
2
0
1
2
with
a
d
e
g
re
e
in
En
g
li
s
h
L
in
g
u
isti
c
s
a
n
d
Li
tera
tu
re
a
n
d
c
o
m
p
lete
d
h
e
r
M
a
ste
r’s
De
g
re
e
(M
A)
in
Tea
c
h
in
g
En
g
li
s
h
t
o
S
p
e
a
k
e
rs
o
f
Ot
h
e
r
La
n
g
u
a
g
e
s
(TE
S
OL),
Vic
t
o
ria
Un
i
v
e
rsity
,
Au
stra
li
a
,
i
n
2
0
1
6
.
He
r
re
se
a
rc
h
h
a
s fo
c
u
se
d
o
n
En
g
l
ish
fo
r
M
a
rit
ime
a
n
d
M
a
rit
ime
S
a
fe
ty
.
Evaluation Warning : The document was created with Spire.PDF for Python.