T
E
L
K
O
M
NIKA
T
elec
o
mm
un
ica
t
io
n Co
m
pu
t
i
ng
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
23
,
No
.
6
,
Dec
em
b
er
20
25
,
p
p
.
1
5
9
0
~1
599
I
SS
N:
1
6
9
3
-
6
9
3
0
,
DOI
: 1
0
.
1
2
9
2
8
/
T
E
L
KOM
NI
K
A
.
v
23
i
6
.
27090
1590
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
Co
m
pa
ra
tive
perf
o
r
m
a
nce ana
ly
sis
o
f
co
nv
o
lutiona
l neural
netw
o
rk
-
a
rchitec
t
ures o
n cof
fee
-
be
a
n roa
st clas
sifica
tion
I
rf
a
n Asf
y
F
a
k
hry
Ant
o
1
,
J
o
ny
Wina
ry
o
Wibo
w
o
2
,
Aris
M
un
a
nd
a
r
2
,
T
a
ufik
I
bn
u
Sa
li
m
2
1
R
e
se
a
r
c
h
C
e
n
t
e
r
f
o
r
D
a
t
a
a
n
d
I
n
f
o
r
m
a
t
i
o
n
S
c
i
e
n
c
e
,
N
a
t
i
o
n
a
l
R
e
se
a
r
c
h
a
n
d
I
n
n
o
v
a
t
i
o
n
,
B
a
n
d
u
n
g
,
I
n
d
o
n
e
si
a
2
R
e
se
a
r
c
h
C
e
n
t
e
r
f
o
r
S
mart
M
e
c
h
a
t
r
o
n
i
c
s,
N
a
t
i
o
n
a
l
R
e
se
a
r
c
h
a
n
d
I
n
n
o
v
a
t
i
o
n
,
B
a
n
d
u
n
g
,
I
n
d
o
n
e
si
a
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Mar
25
,
2
0
2
5
R
ev
i
s
ed
Sep
12
,
2
0
2
5
A
cc
ep
ted
Oct
19
,
2
0
2
5
T
h
e
c
las
si
f
ica
ti
o
n
o
f
c
o
ff
e
e
b
e
a
n
r
o
a
st
lev
e
ls
u
si
n
g
A
g
tro
n
sta
n
d
a
rd
s
h
a
s
e
v
o
lv
e
d
f
ro
m
trad
it
io
n
a
l
su
b
jec
ti
v
e
m
e
th
o
d
s
to
tec
h
n
o
lo
g
y
-
d
riv
e
n
a
p
p
r
o
a
c
h
e
s
e
m
p
lo
y
in
g
a
d
v
a
n
c
e
d
a
rti
f
icia
l
in
telli
g
e
n
c
e
.
Re
c
e
n
t
a
d
v
a
n
c
e
m
e
n
ts
in
c
o
m
p
u
ter
v
isio
n
h
a
v
e
d
e
m
o
n
stra
ted
th
e
c
a
p
a
b
il
it
y
o
f
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s
(CNN
s)
in
p
r
o
v
id
in
g
o
b
jec
ti
v
e
a
n
d
c
o
n
siste
n
t
r
o
a
st
lev
e
l
c
las
sif
ic
a
ti
o
n
c
o
m
p
a
re
d
to
h
u
m
a
n
v
isu
a
l
a
ss
e
ss
m
e
n
t,
w
h
ich
is
p
ro
n
e
to
v
a
riab
il
it
y
a
n
d
su
b
jec
ti
v
it
y
.
T
h
is
re
se
a
rc
h
p
re
se
n
ts
a
p
e
rf
o
rm
a
n
c
e
a
n
a
l
y
sis
o
f
fiv
e
CNN
a
rc
h
it
e
c
tu
re
s
(A
l
e
x
Ne
t,
Re
sN
e
t,
M
o
b
i
leN
e
t,
V
G
GN
e
t,
a
n
d
De
n
se
Ne
t)
f
o
r
c
las
si
fy
in
g
c
o
ff
e
e
b
e
a
n
s
in
to
e
ig
h
t
d
isti
n
c
t
A
g
tro
n
ro
a
st
le
v
e
ls.
T
h
e
c
o
m
p
re
h
e
n
siv
e
m
e
th
o
d
o
lo
g
y
e
n
c
o
m
p
a
ss
e
s f
o
u
r
p
h
a
se
s:
i
)
d
a
ta
a
c
q
u
isit
io
n
,
ii
)
im
a
g
e
p
re
p
ro
c
e
ss
in
g
,
iii
)
m
o
d
e
l
train
in
g
a
n
d
v
a
li
d
a
ti
o
n
,
a
n
d
iv
)
e
v
a
lu
a
ti
o
n
m
e
tri
c
.
Du
rin
g
train
in
g
-
v
a
li
d
a
ti
o
n
,
De
n
se
Ne
t
o
u
tp
e
rf
o
rm
e
d
o
th
e
r
m
o
d
e
ls,
a
c
h
iev
in
g
9
9
.
7
0
2
%
t
ra
in
i
n
g
a
c
c
u
ra
c
y
a
n
d
7
7
.
6
8
%
v
a
li
d
a
ti
o
n
a
c
c
u
ra
c
y
.
In
th
e
tes
ti
n
g
e
v
a
lu
a
ti
o
n
,
De
n
se
Ne
t
a
lso
led
w
it
h
a
n
a
v
e
ra
g
e
tes
ti
n
g
a
c
c
u
ra
c
y
o
f
9
3
.
8
%
,
f
o
ll
o
w
e
d
b
y
Re
s
Ne
t
a
t
9
2
.
6
%
,
V
G
G
Ne
t
a
n
d
A
lex
N
e
t
b
o
th
a
t
9
2
.
4
%
,
a
n
d
M
o
b
il
e
Ne
t
a
t
8
9
.
7
%
.
T
h
e
re
s
u
lt
s
sh
o
w
th
a
t
th
e
De
n
se
Ne
t
sh
o
w
s
p
ro
m
ise
in
c
las
sify
in
g
A
g
tro
n
c
o
ff
e
e
-
b
e
a
n
ro
a
st cla
s
sif
ica
ti
o
n
.
K
ey
w
o
r
d
s
:
Ag
tr
o
n
le
v
el
C
las
s
i
f
icatio
n
C
o
f
f
ee
-
b
ea
n
r
o
ast
C
o
n
v
o
lu
tio
n
al
n
e
u
r
al
n
e
t
w
o
r
k
s
P
er
f
o
r
m
a
n
ce
an
al
y
s
is
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
:
I
r
f
an
A
s
f
y
Fa
k
h
r
y
A
n
to
R
esear
ch
C
en
ter
f
o
r
Data
an
d
I
n
f
o
r
m
a
tio
n
Scie
n
ce
,
Natio
n
al
R
esear
ch
a
n
d
I
n
n
o
v
atio
n
B
an
d
u
n
g
,
4
0
1
3
5
,
I
n
d
o
n
esia
E
m
ail:
ir
f
a0
1
3
@
b
r
in
.
g
o
.
id
1.
I
NT
RO
D
UCT
I
O
N
R
o
asted
co
f
f
ee
b
ea
n
s
u
n
d
er
g
o
a
d
etailed
class
if
ica
tio
n
p
r
o
ce
s
s
to
m
ain
tain
co
n
s
i
s
ten
c
y
an
d
q
u
alit
y
i
n
th
e
f
i
n
al
p
r
o
d
u
ct
[
1
]
.
A
w
ell
-
k
n
o
w
n
s
tan
d
ar
d
f
o
r
t
h
i
s
clas
s
if
icatio
n
i
s
t
h
e
Ag
tr
o
n
s
y
s
te
m
[
2
]
.
T
h
e
s
p
ec
ialt
y
co
f
f
ee
a
s
s
o
ciatio
n
(
S
C
A
)
,
an
i
n
ter
n
at
io
n
a
l
o
r
g
a
n
izatio
n
i
n
t
h
e
co
f
f
ee
i
n
d
u
s
tr
y
,
p
la
y
s
a
k
e
y
r
o
le
in
es
tab
lis
h
in
g
m
et
h
o
d
o
lo
g
ies f
o
r
ev
al
u
ati
n
g
an
d
class
if
y
i
n
g
r
o
asted
co
f
f
ee
b
ea
n
s
.
Fo
r
m
a
n
y
y
ea
r
s
,
th
e
SC
A
h
as u
ti
lized
to
o
ls
lik
e
th
e
A
g
tr
o
n
s
y
s
te
m
to
ass
es
s
co
f
f
ee
q
u
alit
y
b
y
as
s
i
g
n
i
n
g
s
co
r
es
to
s
a
m
p
les
[
3
]
,
[
4
]
.
T
h
e
A
g
tr
o
n
class
i
f
icat
io
n
p
r
o
v
id
es
an
o
b
j
ec
tiv
e,
s
tan
d
ar
d
ized
m
ea
s
u
r
e
o
f
r
o
ast
le
v
els,
e
n
s
u
r
in
g
u
n
i
f
o
r
m
it
y
a
n
d
q
u
alit
y
i
n
co
f
f
ee
p
r
o
d
u
ctio
n
,
w
h
ich
i
s
cr
u
cial
f
o
r
b
o
th
co
m
m
er
cial
an
d
s
p
ec
ialt
y
co
f
f
ee
m
ar
k
ets.
A
d
d
itio
n
all
y
,
th
e
Ag
tr
o
n
s
y
s
te
m
h
as
b
ee
n
i
n
s
tr
u
m
en
ta
l
i
n
co
f
f
ee
r
esear
c
h
,
p
ar
ticu
lar
l
y
i
n
s
tu
d
ie
s
i
n
v
e
s
ti
g
ati
n
g
h
o
w
r
o
ast
le
v
els
in
f
l
u
e
n
ce
ch
e
m
ical
co
m
p
o
s
itio
n
[
5
]
.
R
esear
ch
er
s
h
av
e
e
m
p
lo
y
ed
A
g
tr
o
n
m
ea
s
u
r
e
m
e
n
t
s
to
an
al
y
ze
th
e
im
p
ac
t
o
f
d
if
f
er
en
t
r
o
ast
lev
el
s
o
n
co
m
p
o
u
n
d
s
li
k
e
ch
lo
r
o
g
e
n
ic
ac
id
s
a
n
d
ca
f
f
ein
e,
w
h
ic
h
a
f
f
ec
t
b
o
th
f
la
v
o
r
an
d
h
ea
lt
h
b
e
n
ef
i
ts
[
6
]
.
T
h
ese
f
in
d
i
n
g
s
h
a
v
e
g
u
id
ed
co
m
m
er
cial
r
o
asti
n
g
p
r
ac
tices
an
d
d
ee
p
en
ed
th
e
u
n
d
er
s
ta
n
d
i
n
g
o
f
co
f
f
ee
s
c
ien
ce
,
e
m
p
h
a
s
izi
n
g
t
h
e
i
m
p
o
r
ta
n
ce
o
f
p
r
ec
is
e
r
o
ast p
r
o
f
ilin
g
i
n
ac
h
iev
i
n
g
d
esire
d
f
lav
o
r
s
.
T
h
e
A
g
tr
o
n
clas
s
i
f
icatio
n
r
elies
o
n
ad
v
an
ce
d
tech
n
o
lo
g
y
to
m
ea
s
u
r
e
r
o
ast
lev
els
[
7
]
.
R
ec
en
t
tech
n
o
lo
g
ical
d
ev
elo
p
m
e
n
ts
h
av
e
i
m
p
r
o
v
ed
th
e
ac
cu
r
ac
y
an
d
u
s
ab
ilit
y
o
f
th
e
A
g
tr
o
n
s
y
s
te
m
.
Mo
d
er
n
s
p
ec
tr
o
p
h
o
to
m
eter
s
an
d
s
o
f
t
w
ar
e
en
h
a
n
c
e
m
en
t
s
h
a
v
e
s
i
m
p
li
f
ied
r
o
ast
p
r
o
f
ile
an
al
y
s
is
,
m
a
k
i
n
g
it
m
o
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
C
o
mp
a
r
a
tive
p
erfo
r
ma
n
ce
a
n
a
lysi
s
o
f c
o
n
vo
lu
tio
n
a
l n
eu
r
a
l
…
(
I
r
fa
n
A
s
fy
F
a
kh
r
y
A
n
to
)
1591
ac
ce
s
s
ib
le
f
o
r
b
o
th
s
m
all
-
s
ca
le
an
d
lar
g
e
-
s
ca
le
co
f
f
ee
p
r
o
d
u
ce
r
s
[
8
]
.
C
o
n
v
o
lu
tio
n
al
n
e
u
r
al
n
et
w
o
r
k
(
C
NN)
h
a
s
e
m
er
g
ed
a
s
a
p
o
w
er
f
u
l
to
o
l
f
o
r
au
to
m
at
in
g
t
h
e
cla
s
s
i
f
icat
io
n
o
f
co
f
f
ee
b
ea
n
r
o
ast
l
e
v
el
s
,
o
f
f
er
in
g
a
m
ea
n
s
to
r
ef
in
e
an
d
s
ta
n
d
ar
d
ize
th
e
p
r
o
ce
s
s
[
9
]
.
C
NN’
s
ar
ch
itect
u
r
e,
w
h
ic
h
d
r
a
w
s
i
n
s
p
ir
atio
n
f
r
o
m
th
e
h
u
m
a
n
v
is
u
al
s
y
s
te
m
,
is
p
ar
ticu
lar
l
y
w
e
ll
-
s
u
i
ted
to
th
e
v
is
u
al
an
al
y
s
i
s
r
eq
u
ir
ed
f
o
r
class
if
y
i
n
g
r
o
asted
co
f
f
ee
b
ea
n
s
,
an
d
th
e
y
h
av
e
c
o
n
s
eq
u
e
n
tl
y
s
h
o
w
n
co
n
s
id
er
ab
le
p
r
o
m
is
e
i
n
th
i
s
ap
p
licatio
n
[
1
0
]
.
B
y
tr
ain
i
n
g
o
n
lar
g
e
d
atasets
o
f
lab
eled
i
m
a
g
es,
C
NN
s
ca
n
lear
n
to
d
is
tin
g
u
i
s
h
b
et
w
ee
n
d
if
f
er
e
n
t r
o
ast le
v
els
b
ased
o
n
t
h
e
v
i
s
u
a
l c
h
ar
ac
ter
is
tic
s
o
f
th
e
b
ea
n
s
,
ef
f
ec
ti
v
el
y
a
u
to
m
a
tin
g
a
p
r
o
c
ess
th
at
tr
ad
itio
n
all
y
r
eq
u
ir
ed
m
a
n
u
al
in
p
u
t
[
1
1
]
,
[
1
2
]
.
T
h
is
n
o
t
o
n
l
y
ac
ce
ler
ates c
lass
i
f
icat
io
n
b
u
t a
ls
o
m
i
n
i
m
izes
h
u
m
a
n
er
r
o
r
an
d
v
ar
iab
ilit
y
,
en
h
a
n
cin
g
o
v
er
al
l e
f
f
icien
c
y
.
R
ec
en
t
r
e
s
ea
r
ch
h
as
i
n
cr
ea
s
i
n
g
l
y
f
o
c
u
s
ed
o
n
C
NN
-
b
ased
m
et
h
o
d
s
f
o
r
au
to
m
ated
cl
as
s
i
f
icatio
n
o
f
co
f
f
ee
b
ea
n
r
o
ast
lev
el
s
,
o
f
f
e
r
in
g
m
o
r
e
o
b
j
ec
tiv
e
an
d
co
n
s
is
ten
t
ev
al
u
atio
n
co
m
p
ar
ed
to
tr
ad
itio
n
al
h
u
m
a
n
v
is
u
al
as
s
es
s
m
en
t.
I
n
o
n
e
n
o
ta
b
le
s
tu
d
y
[
1
3
]
,
p
r
o
p
o
s
ed
a
cu
s
to
m
C
N
N
ar
ch
itect
u
r
e
d
e
m
o
n
s
tr
ated
ex
ce
p
tio
n
a
l
p
er
f
o
r
m
a
n
ce
in
ca
te
g
o
r
izin
g
co
f
f
ee
b
ea
n
s
i
n
to
f
o
u
r
d
is
ti
n
c
t
r
o
ast
lev
els
(
lig
h
t,
m
ed
i
u
m
,
m
ed
i
u
m
-
d
ar
k
,
an
d
d
ar
k
)
,
ac
h
iev
in
g
9
7
.
5
%
cla
s
s
i
f
icatio
n
ac
cu
r
ac
y
t
h
r
o
u
g
h
o
p
ti
m
ized
h
y
p
er
p
ar
a
m
eter
tu
n
i
n
g
an
d
d
ata
au
g
m
e
n
tatio
n
tech
n
iq
u
es.
An
o
th
er
in
v
e
s
ti
g
atio
n
o
n
[
1
4
]
u
s
ed
R
esNet5
0
ar
ch
itect
u
r
e
im
p
le
m
e
n
ted
o
n
an
An
d
r
o
id
p
latf
o
r
m
,
w
h
ic
h
s
u
cc
ess
f
u
l
l
y
id
en
ti
f
ied
d
if
f
er
e
n
t
t
y
p
es
o
f
r
o
asted
co
f
f
ee
b
ea
n
s
with
8
3
.
3
%
av
er
a
g
e
ac
cu
r
ac
y
,
s
h
o
w
ca
s
in
g
t
h
e
p
o
t
en
tial
f
o
r
p
r
ac
tical,
o
n
-
t
h
e
-
g
o
q
u
alit
y
as
s
es
s
m
e
n
t
in
p
r
o
d
u
ctio
n
f
ac
ilit
ie
s
a
n
d
co
f
f
ee
s
h
o
p
s
.
T
h
e
s
tu
d
y
s
p
ec
i
f
icall
y
e
m
p
h
asized
th
e
m
o
d
el
’
s
p
r
o
f
icien
t
u
tili
za
tio
n
o
f
m
e
m
o
r
y
an
d
its
ca
p
ac
it
y
f
o
r
r
ea
l
-
ti
m
e
an
al
y
s
is
,
attr
ib
u
t
es
th
at
r
en
d
er
it
a
v
iab
le
ca
n
d
id
ate
f
o
r
i
m
p
le
m
e
n
tat
io
n
o
n
r
eso
u
r
ce
-
co
n
s
tr
ai
n
ed
ed
g
e
d
ev
ices.
R
esear
c
h
o
n
[
1
5
]
co
n
d
u
cted
a
s
y
s
te
m
at
ic
p
er
f
o
r
m
an
ce
a
n
al
y
s
i
s
i
n
cl
u
d
in
g
Sq
u
ee
ze
Net,
Sh
u
f
f
le
Net,
Mo
b
ileNetV2
,
an
d
NA
S
Net
Mo
b
ile
in
cla
s
s
i
f
y
in
g
f
o
u
r
r
o
ast
class
e
s
(
g
r
ee
n
,
lig
h
t,
m
ed
i
u
m
,
an
d
d
ar
k
)
.
Si
m
i
lar
l
y
,
w
o
r
k
b
y
Ha
s
s
an
[
1
6
]
ex
p
lo
r
ed
m
u
ltip
le
C
NN
ar
ch
itect
u
r
es
f
o
r
f
o
u
r
-
lev
e
l
r
o
ast
clas
s
if
ic
atio
n
f
r
o
m
an
o
n
li
n
e
co
f
f
ee
-
b
ea
n
d
at
aset,
to
d
eter
m
in
e
th
e
ac
cu
r
ac
y
o
f
p
r
e
-
tr
ai
n
ed
m
o
d
els
in
id
en
tify
i
n
g
f
o
u
r
co
f
f
ee
-
b
ea
n
class
e
s
th
r
o
u
g
h
ad
v
an
ce
d
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es.
C
u
r
r
en
t
r
esear
ch
tr
aj
ec
to
r
ies
p
o
s
itio
n
C
NN
ar
c
h
itect
u
r
es
as
a
s
u
b
s
tan
tia
l
ad
v
a
n
ce
m
e
n
t
w
it
h
in
co
f
f
ee
q
u
alit
y
ass
e
s
s
m
e
n
t.
O
n
g
o
i
n
g
r
esear
ch
a
n
d
d
ev
elo
p
m
e
n
t
s
u
g
g
est
t
h
ese
C
NN
-
b
ased
m
eth
o
d
o
lo
g
ies
ar
e
p
o
is
ed
to
ass
u
m
e
a
cr
itical
f
u
n
ctio
n
in
f
u
tu
r
e
q
u
a
lit
y
co
n
tr
o
l
an
d
r
o
asti
n
g
p
r
o
ce
s
s
es.
Ho
w
e
v
er
,
an
id
en
ti
f
ied
li
m
itatio
n
i
n
th
e
e
x
tan
t
liter
at
u
r
e
is
i
ts
co
n
s
t
r
ain
ed
s
co
p
e
co
n
ce
r
n
i
n
g
th
e
s
p
ec
if
ic
ap
p
licatio
n
o
f
C
NNs f
o
r
class
if
y
i
n
g
co
f
f
ee
b
ea
n
r
o
ast
lev
els.
U
n
li
k
e
p
r
ev
io
u
s
r
esear
ch
,
w
h
ic
h
t
y
p
ical
l
y
ca
teg
o
r
izes
r
o
asti
n
g
le
v
els
i
n
t
o
o
n
l
y
f
o
u
r
class
e
s
(
g
r
ee
n
,
li
g
h
t,
m
ed
iu
m
,
a
n
d
d
ar
k
)
,
th
is
s
tu
d
y
co
n
d
u
c
ts
a
co
m
p
ar
at
iv
e
a
n
al
y
s
i
s
o
f
f
iv
e
C
NN
ar
ch
itect
u
r
es
to
class
i
f
y
ei
g
h
t
Ag
tr
o
n
clas
s
es,
s
u
c
h
as
v
er
y
d
ar
k
,
d
ar
k
,
m
o
d
er
atel
y
d
ar
k
,
m
ed
i
u
m
,
m
ed
iu
m
l
ig
h
t,
m
o
d
er
atel
y
lig
h
t,
li
g
h
t,
an
d
v
er
y
lig
h
t.
T
h
ese
o
f
f
er
a
m
o
r
e
d
etailed
cla
s
s
if
icatio
n
o
f
co
f
f
ee
b
ea
n
f
la
v
o
r
s
.
Fu
r
t
h
er
m
o
r
e,
t
h
i
s
r
esear
ch
cr
ea
tes
its
d
ataset
b
y
e
m
p
lo
y
i
n
g
i
m
a
g
es
a
n
d
a
ca
m
er
a
as
s
en
s
o
r
s
f
o
r
ex
p
er
i
m
en
ta
tio
n
,
w
h
er
ea
s
p
r
ev
io
u
s
s
tu
d
ie
s
o
f
te
n
u
s
ed
d
atase
ts
s
o
u
r
ce
d
f
r
o
m
th
e
in
ter
n
et.
T
h
er
eb
y
,
th
i
s
r
esear
c
h
co
n
tr
ib
u
tes
to
cr
ea
tin
g
a
n
e
w
d
ataset.
Mo
r
eo
v
e
r
,
o
u
r
p
r
io
r
r
esear
ch
[
1
7
]
d
ev
elo
p
ed
a
g
r
ap
h
ical
u
s
er
in
ter
f
ac
e
(
GUI
)
w
ith
i
n
an
e
m
b
ed
d
ed
s
y
s
te
m
f
o
r
co
f
f
ee
b
ea
n
class
if
i
ca
tio
n
,
b
u
t
it
s
ti
ll
u
t
ilized
o
n
e
C
NN
m
o
d
el,
as
t
h
e
p
r
o
p
o
s
ed
s
y
s
te
m
co
u
ld
r
ep
lace
th
e
tr
ai
n
i
n
g
o
u
tp
u
t
f
ile.
B
u
ild
in
g
o
n
t
h
i
s
,
t
h
is
s
t
u
d
y
a
i
m
s
to
c
o
n
d
u
ct
a
m
o
r
e
i
n
-
d
ep
th
a
n
al
y
s
is
o
f
m
u
lt
ip
le
C
N
N
-
b
ased
m
o
d
els
f
o
r
class
i
f
y
i
n
g
Ag
tr
o
n
co
lo
r
lev
els
in
co
f
f
ee
b
ea
n
s
.
T
o
ac
h
iev
e
a
h
ig
h
l
y
ac
cu
r
ate
s
y
s
te
m
,
t
h
i
s
p
ap
er
p
r
esen
ts
a
co
m
p
ar
ativ
e
p
er
f
o
r
m
a
n
ce
an
al
y
s
is
o
f
f
i
v
e
C
NN
ar
ch
itect
u
r
es,
s
u
c
h
as
A
le
x
Net,
R
e
s
Net,
Mo
b
ileNet,
VGGN
et,
an
d
Den
s
eNe
t,
to
clas
s
if
y
ei
g
h
t
Ag
tr
o
n
class
es
a
n
d
id
en
ti
f
y
t
h
e
b
est
-
p
er
f
o
r
m
i
n
g
m
o
d
e
l
f
o
r
r
o
asted
c
o
f
f
ee
b
ea
n
class
i
f
icatio
n
.
T
h
e
ev
alu
atio
n
m
etr
ic
s
in
clu
d
e
a
co
n
f
u
s
io
n
m
atr
ix
u
s
ed
f
o
r
ea
ch
o
f
th
e
f
i
v
e
ar
ch
itect
u
r
es.
2.
RE
L
A
T
E
D
S
T
UDI
E
S
2
.
1
.
Ag
t
ro
n c
la
s
s
if
ica
t
io
n
T
h
e
Ag
tr
o
n
cla
s
s
i
f
icat
io
n
s
y
s
te
m
h
as
u
n
d
er
g
o
n
e
s
i
g
n
i
f
i
ca
n
t
ad
v
a
n
ce
m
e
n
ts
a
n
d
f
o
u
n
d
d
iv
er
s
e
ap
p
licatio
n
s
in
r
ec
en
t
y
ea
r
s
,
m
ir
r
o
r
in
g
th
e
ev
o
l
v
i
n
g
la
n
d
s
ca
p
e
o
f
co
f
f
ee
tech
n
o
lo
g
y
[
1
8
]
.
T
h
is
s
y
s
te
m
co
n
s
tit
u
tes a
w
id
el
y
e
m
p
lo
y
ed
m
et
h
o
d
f
o
r
ev
alu
ati
n
g
co
f
f
ee
b
ea
n
r
o
ast lev
els.
I
t e
m
p
lo
y
s
a
s
p
ec
tr
o
p
h
o
to
m
eter
to
m
ea
s
u
r
e
co
lo
r
an
d
ass
ig
n
a
n
u
m
er
ical
v
a
lu
e
o
n
th
e
Ag
tr
o
n
s
ca
le,
w
h
ic
h
r
an
g
es f
r
o
m
li
g
h
t to
d
a
r
k
r
o
ast
[
1
9
]
.
T
h
e
s
ca
le
p
lay
s
a
cr
u
cial
r
o
le
in
m
ai
n
tai
n
i
n
g
co
n
s
is
te
n
c
y
i
n
co
f
f
ee
r
o
asti
n
g
b
y
o
f
f
er
i
n
g
an
o
b
j
ec
tiv
e
m
ea
s
u
r
e
o
f
r
o
ast
d
eg
r
ee
,
w
h
ich
d
ir
ec
tl
y
in
f
lu
e
n
ce
s
f
lav
o
r
p
r
o
f
iles
a
n
d
co
n
s
u
m
er
p
r
ef
er
e
n
ce
s
[
2
0
]
.
I
n
th
is
s
t
u
d
y
,
it is
u
s
e
th
e
A
g
tr
o
n
r
o
ast
co
lo
r
s
tan
d
ar
d
is
u
s
ed
f
o
r
r
o
ast
ed
co
f
f
ee
b
ea
n
s
w
it
h
eig
h
t
co
lo
r
lev
els
co
n
s
is
tin
g
o
f
R
-
2
5
,
R
-
3
5
,
R
-
4
5
,
R
-
5
5
,
R
-
6
5
,
R
-
7
5
,
R
-
8
5
,
an
d
R
-
95.
2
.
1
.
CNN
a
rc
hite
ct
ures
C
NNs
h
a
v
e
b
ec
o
m
e
a
p
r
ev
ale
n
t
to
o
l
f
o
r
clas
s
i
f
icatio
n
,
a
u
til
it
y
lar
g
el
y
a
ttrib
u
tab
le
to
t
h
eir
ab
ilit
y
to
p
er
f
o
r
m
a
u
to
n
o
m
o
u
s
f
ea
tu
r
e
ex
tr
ac
tio
n
d
ir
ec
tl
y
f
r
o
m
r
a
w
d
ata.
T
h
is
ap
p
r
o
ac
h
w
o
u
ld
a
d
d
r
ess
th
e
g
r
o
w
i
n
g
d
em
a
n
d
f
o
r
m
u
lti
m
o
d
al
class
if
icatio
n
s
y
s
te
m
s
i
n
r
ea
l
-
w
o
r
ld
ap
p
licatio
n
s
[
1
8
]
,
[
2
0
]
.
On
g
o
i
n
g
r
e
s
ea
r
ch
i
n
t
h
ese
ar
ea
s
w
i
ll
f
u
r
t
h
er
en
h
an
ce
th
e
p
o
ten
tial
o
f
C
NNs
f
o
r
clas
s
if
i
ca
tio
n
ta
s
k
s
,
s
u
ch
a
s
d
eter
m
i
n
in
g
co
f
f
ee
-
b
ea
n
Ag
tr
o
n
clas
s
es
.
Alex
Ne
t
w
a
s
th
e
f
ir
s
t
d
ee
p
C
NN
to
ef
f
ec
t
iv
el
y
u
tili
ze
GP
U
ac
ce
ler
atio
n
f
o
r
tr
ain
in
g
.
T
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
23
,
No
.
6
,
Dec
em
b
er
20
25
:
1
5
9
0
-
1
599
1592
ac
co
m
p
li
s
h
m
e
n
t
s
o
f
th
i
s
n
et
wo
r
k
w
er
e
in
s
tr
u
m
en
tal
in
estab
lis
h
in
g
th
e
v
iab
ilit
y
o
f
d
ee
p
lear
n
in
g
f
o
r
co
m
p
lex
i
m
a
g
e
clas
s
i
f
icatio
n
,
t
h
er
eb
y
c
atal
y
zi
n
g
th
e
b
r
o
ad
in
te
g
r
atio
n
o
f
C
NN
in
to
co
m
p
u
ter
v
is
io
n
[
2
1
]
.
Fo
r
ex
am
p
le,
a
s
tu
d
y
b
y
[
2
2
]
in
v
esti
g
ated
co
m
b
in
i
n
g
A
lex
Net
w
ith
e
n
s
e
m
b
le
g
en
etic
al
g
o
r
ith
m
s
to
b
o
o
s
t
im
ag
e
cla
s
s
i
f
icatio
n
p
er
f
o
r
m
a
n
ce
.
T
h
is
h
y
b
r
id
m
et
h
o
d
s
o
u
g
h
t
to
o
p
ti
m
ize
f
ea
tu
r
e
s
elec
tio
n
a
n
d
en
h
a
n
ce
clas
s
if
icatio
n
ac
cu
r
ac
y
,
illu
s
tr
atin
g
A
le
x
Net
’
s
ad
ap
tab
ilit
y
w
h
e
n
in
te
g
r
ated
w
it
h
o
th
er
m
ac
h
i
n
e
lear
n
i
n
g
tech
n
iq
u
e
s
.
C
o
llecti
v
el
y
,
t
h
i
s
r
esear
ch
h
i
g
h
li
g
h
ts
A
le
x
Net
’
s
e
n
d
u
r
i
n
g
in
f
l
u
en
ce
a
n
d
f
le
x
ib
ilit
y
in
i
m
a
g
e
cla
s
s
i
f
i
ca
tio
n
,
w
h
i
le
also
ac
k
n
o
w
led
g
in
g
it
s
p
iv
o
tal
r
o
le
in
th
e
b
r
o
ad
er
d
ev
elo
p
m
e
n
t o
f
C
N
N
ar
ch
itect
u
r
es.
VGGN
et
is
a
C
NN
ar
ch
i
tectu
r
e
th
at
m
ad
e
s
i
g
n
i
f
ica
n
t
s
tr
id
es in
th
e
p
ar
t
o
f
co
m
p
u
ter
v
i
s
io
n
[
2
3
]
.
Ov
er
th
e
p
r
ec
ed
in
g
f
i
v
e
-
y
ea
r
p
er
io
d
,
VGGN
et
h
a
s
b
ee
n
e
x
te
n
s
i
v
e
l
y
u
tili
ze
d
an
d
m
o
d
if
ied
ac
r
o
s
s
d
iv
er
s
e
d
o
m
ai
n
s
,
w
it
h
p
ar
tic
u
lar
p
r
o
m
i
n
en
ce
in
th
e
f
ield
s
o
f
m
ed
ical
i
m
a
g
i
n
g
an
d
o
b
j
ec
t
d
etec
tio
n
[
2
4
]
.
T
h
is
r
esear
ch
u
t
ilized
VGGNet
,
as
it
h
as
n
o
t
o
n
l
y
d
r
iv
en
ad
v
a
n
ce
m
en
t
s
i
n
i
m
a
g
e
c
lass
i
f
icatio
n
b
u
t
also
co
n
tr
ib
u
t
ed
s
ig
n
i
f
ica
n
tl
y
to
p
r
o
g
r
ess
in
o
b
j
ec
t
d
etec
tio
n
an
d
tr
an
s
f
er
lear
n
in
g
[
2
5
]
.
R
es
Net
h
as
co
n
s
t
itu
ted
a
s
e
m
in
a
l
co
n
tr
ib
u
tio
n
to
th
e
f
ield
o
f
d
ee
p
lear
n
i
n
g
[
2
6
]
.
I
ts
f
o
u
n
d
a
tio
n
a
l
ele
m
en
t
is
t
h
e
r
esid
u
al
b
lo
ck
[
2
7
]
,
an
ar
ch
itect
u
r
al
m
o
d
u
le
co
m
p
r
is
in
g
m
u
ltip
le
co
n
v
o
lu
t
i
o
n
al
an
d
b
atch
n
o
r
m
aliza
tio
n
l
a
y
er
s
,
in
ter
s
p
er
s
ed
w
ith
ac
tiv
a
tio
n
f
u
n
ctio
n
s
.
T
h
is
ar
ch
itect
u
r
e
p
r
io
r
itizes
a
lig
h
t
w
ei
g
h
t
d
esi
g
n
an
d
m
i
n
i
m
al
co
m
p
u
tatio
n
al
o
v
er
h
ea
d
,
th
er
eb
y
r
ep
r
ese
n
ti
n
g
a
s
ig
n
i
f
ica
n
t
ad
v
a
n
ce
m
e
n
t
i
n
t
h
e
d
ev
elo
p
m
e
n
t
o
f
ef
f
icie
n
t
d
ee
p
lear
n
in
g
m
o
d
els
[
2
8
]
,
[
2
9
]
.
R
ec
en
t
r
esear
ch
ef
f
o
r
t
s
h
av
e
f
o
cu
s
ed
o
n
en
h
an
cin
g
th
e
o
p
er
atio
n
al
ef
f
ici
en
c
y
an
d
f
u
n
ctio
n
a
l
ad
ap
tab
ilit
y
o
f
Mo
b
ileNet
ar
ch
itect
u
r
es.
T
h
i
s
h
as
b
ee
n
p
u
r
s
u
ed
th
r
o
u
g
h
t
h
e
i
n
co
r
p
o
r
ati
o
n
o
f
o
p
ti
m
izatio
n
tech
n
iq
u
e
s
s
u
c
h
as
n
u
m
er
ical
q
u
an
tizat
io
n
an
d
th
eir
in
teg
r
atio
n
w
it
h
s
p
ec
ialized
ed
g
e
ar
tif
icial
in
tel
lig
e
n
ce
f
r
a
m
e
w
o
r
k
s
,
in
cl
u
d
in
g
T
en
s
o
r
Flo
w
L
ite
an
d
P
y
T
o
r
ch
Mo
b
ile
[
3
0
]
,
[
3
1
]
.
T
h
ese
ad
v
an
ce
m
e
n
t
s
h
av
e
estab
li
s
h
ed
Mo
b
ileNet
as
an
o
th
er
ch
o
ice
f
o
r
r
ea
l
-
ti
m
e
ap
p
licati
o
n
s
,
i
n
clu
d
i
n
g
o
b
j
ec
t
d
etec
tio
n
,
f
ac
ial
r
ec
o
g
n
itio
n
,
an
d
a
u
g
m
e
n
ted
r
ea
lit
y
o
n
e
m
b
ed
d
ed
o
r
m
o
b
ile
d
e
v
ices
[
3
2
]
.
Den
s
eNe
t
is
a
C
NN
ar
c
h
itect
u
r
e
d
is
ti
n
g
u
is
h
ed
b
y
i
ts
d
en
s
e
co
n
n
ec
ti
v
it
y
p
atter
n
,
w
h
er
e
ea
ch
la
y
er
is
co
n
n
ec
ted
to
ev
er
y
o
th
er
la
y
er
in
a
f
ee
d
-
f
o
r
w
ar
d
m
an
n
er
.
T
h
is
d
esig
n
en
co
u
r
ag
e
s
f
ea
t
u
r
e
r
eu
s
e,
r
ed
u
ce
s
t
h
e
n
u
m
b
er
o
f
p
ar
a
m
e
ter
s
,
a
n
d
alle
v
iates
t
h
e
v
a
n
i
s
h
in
g
g
r
ad
ien
t
p
r
o
b
lem
,
m
a
k
in
g
Den
s
eNe
t
h
i
g
h
l
y
e
f
f
ic
ien
t
f
o
r
d
ee
p
lear
n
in
g
tas
k
s
[
3
3
]
.
R
esear
ch
o
n
Den
s
eNe
t
h
as
ce
n
ter
ed
o
n
im
p
r
o
v
i
n
g
it
s
s
ca
lab
ilit
y
a
n
d
ad
ap
tab
ilit
y
f
o
r
ap
p
licatio
n
s
s
u
ch
as
au
to
n
o
m
o
u
s
d
r
iv
in
g
,
m
ed
ical
i
m
ag
e
class
if
icatio
n
,
an
d
n
atu
r
al
la
n
g
u
a
g
e
p
r
o
ce
s
s
in
g
[
3
4
]
,
[
3
5
]
.
Fu
r
th
er
m
o
r
e,
Den
s
e
Net
r
em
a
in
s
a
p
o
w
er
f
u
l
a
n
d
v
er
s
atile
ar
ch
itect
u
r
e,
f
o
s
ter
i
n
g
i
n
n
o
v
atio
n
in
b
o
th
ac
ad
em
ic
r
esear
ch
a
n
d
in
d
u
s
tr
ial
ap
p
licatio
n
s
[
3
6
]
.
3.
M
E
T
H
O
D
T
h
e
co
m
p
ar
ativ
e
p
er
f
o
r
m
an
ce
an
al
y
s
is
in
t
h
i
s
r
esear
ch
w
a
s
ex
ec
u
ted
ac
co
r
d
in
g
to
a
s
tr
u
ct
u
r
ed
,
m
u
lti
-
s
tag
e
m
e
th
o
d
o
lo
g
y
,
o
u
tli
n
ed
i
n
Fi
g
u
r
e
1
.
T
h
e
p
r
o
ce
s
s
b
eg
a
n
w
ith
t
h
e
co
llectio
n
o
f
a
co
f
f
ee
-
b
ea
n
r
o
ast
d
ataset,
ca
teg
o
r
ized
in
to
eig
h
t
clas
s
es
b
ased
o
n
A
g
tr
o
n
clas
s
es
.
T
h
e
in
itial
p
h
a
s
e
in
v
o
lv
ed
p
r
ep
r
o
ce
s
s
i
n
g
t
h
e
d
ataset
to
en
s
u
r
e
th
e
i
m
a
g
e
s
w
er
e
s
ta
n
d
ar
d
ized
an
d
r
ea
d
y
f
o
r
a
n
al
y
s
is
.
Fo
ll
o
w
i
n
g
t
h
is
,
t
h
e
p
er
f
o
r
m
an
ce
o
f
s
e
v
er
al
p
r
o
m
i
n
en
t
C
NN
ar
ch
itect
u
r
es
n
a
m
e
l
y
A
le
x
Net,
R
e
s
Net,
Mo
b
ileNet,
VGGN
et,
an
d
Den
s
e
Net
w
as
e
v
alu
a
ted
.
E
ac
h
m
o
d
el
u
n
d
er
w
e
n
t
a
co
m
p
r
eh
en
s
i
v
e
tr
ain
in
g
an
d
v
a
lid
atio
n
p
r
o
ce
s
s
to
g
en
er
ate
a
clas
s
if
ier
as
t
h
e
o
u
tp
u
t.
T
h
ese
cla
s
s
i
f
ier
s
w
er
e
th
e
n
r
i
g
o
r
o
u
s
l
y
test
ed
u
s
i
n
g
e
v
al
u
at
io
n
m
etr
ic
s
co
n
s
i
s
t
o
f
t
h
e
co
n
f
u
s
io
n
m
atr
ix
a
n
d
ac
cu
r
ac
y
s
co
r
e,
to
ass
ess
th
e
ir
ef
f
ec
ti
v
en
e
s
s
i
n
class
i
f
y
in
g
th
e
co
f
f
ee
-
b
ea
n
r
o
ast
i
m
ag
e
s
.
T
h
is
s
y
s
te
m
atic
m
et
h
o
d
o
lo
g
y
f
ac
ilit
ated
a
co
m
p
r
e
h
e
n
s
i
v
e
e
v
alu
at
io
n
o
f
th
e
m
o
d
els
’
ca
p
ab
ilit
ies,
y
ield
i
n
g
cr
it
ical
i
n
s
i
g
h
ts
r
eg
ar
d
in
g
t
h
eir
p
er
f
o
r
m
a
n
ce
a
n
d
ap
p
r
o
p
r
iaten
ess
f
o
r
th
e
in
te
n
d
ed
ap
p
licatio
n
.
Fig
u
r
e
1.
R
esear
ch
m
et
h
o
d
3
.
1
.
Da
t
a
c
o
llect
io
n
T
h
is
r
esear
ch
is
co
llec
ted
th
e
d
ataset
o
u
r
s
el
v
es
f
o
r
th
e
e
x
p
er
i
m
en
t.
T
h
e
m
ater
ials
u
s
ed
d
u
r
in
g
th
e
d
ataset
co
llectio
n
ar
e
s
h
o
w
n
in
Fig
u
r
e
2
.
A
co
m
p
u
ter
,
a
m
in
i b
o
x
p
h
o
to
s
tu
d
io
,
a
B
r
i
o
4
K
W
eb
C
a
m
er
a,
an
d
its
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
C
o
mp
a
r
a
tive
p
erfo
r
ma
n
ce
a
n
a
lysi
s
o
f c
o
n
vo
lu
tio
n
a
l n
eu
r
a
l
…
(
I
r
fa
n
A
s
fy
F
a
kh
r
y
A
n
to
)
1593
s
o
f
t
w
ar
e
w
er
e
u
s
ed
,
as
s
h
o
wn
in
Fi
g
u
r
e
2
(
a)
.
T
o
r
ed
u
ce
t
h
e
i
m
p
ac
t
o
f
en
v
ir
o
n
m
e
n
tal
f
ac
to
r
s
d
u
r
in
g
i
m
ag
e
ca
p
tu
r
e,
th
e
s
a
m
e
p
ar
a
m
eter
s
s
u
ch
as b
r
ig
h
tn
e
s
s
,
co
n
tr
ast,
ex
p
o
s
u
r
e,
an
d
o
th
er
s
ettin
g
s
w
er
e
m
ai
n
tai
n
ed
f
o
r
all
i
m
a
g
es,
as
s
ee
n
in
Fi
g
u
r
e
2
(
b
)
.
T
h
e
m
ater
ials
co
n
s
is
t
o
f
r
o
as
ted
co
f
f
ee
b
ea
n
s
w
it
h
v
ar
io
u
s
co
lo
r
lev
els
an
d
th
e
Ag
tr
o
n
r
o
ast
co
lo
r
s
tan
d
a
r
d
,
as
s
h
o
w
n
i
n
Fi
g
u
r
e
2
(
c)
.
T
h
e
im
a
g
e
w
as
ta
k
en
u
s
i
n
g
l
ig
h
ti
n
g
co
n
d
itio
n
s
u
s
in
g
li
g
h
t
f
r
o
m
th
e
L
E
D
o
n
th
e
m
i
n
ip
h
o
t
o
b
o
x
an
d
th
e
f
o
llo
w
i
n
g
ca
m
er
a
s
ettin
g
s
:
i)
b
r
ig
h
tn
e
s
s
s
et
to
1
1
5
,
ii)
co
n
tr
ast s
et
to
1
2
5
,
iii)
s
h
ar
p
n
es
s
o
r
s
at
u
r
a
tio
n
s
e
t
to
1
1
0
,
iv
)
n
o
is
e
r
ed
u
c
tio
n
s
e
t
to
1
1
0
,
an
d
v
)
c
h
r
o
m
a
k
e
y
ef
f
ec
t
in
te
n
s
it
y
s
et
to
5
5
.
C
o
f
f
ee
b
ea
n
s
w
er
e
ch
o
s
en
to
p
r
ec
is
e
Ag
tr
o
n
s
p
ec
if
icatio
n
s
,
an
d
i
m
ag
e
s
wer
e
ca
p
tu
r
ed
u
n
d
er
co
n
tr
o
lled
lig
h
ti
n
g
co
n
d
itio
n
s
an
d
s
etti
n
g
s
to
m
i
n
i
m
ize
v
ar
ia
b
ilit
y
.
E
ac
h
r
o
asted
co
f
f
ee
b
ea
n
w
a
s
m
atc
h
ed
w
it
h
th
e
Ag
tr
o
n
r
o
ast
co
lo
r
s
tan
d
ar
d
f
o
r
th
e
lab
elin
g
p
r
o
ce
s
s
an
d
th
en
p
lace
d
in
to
th
e
p
h
o
to
b
o
x
s
tu
d
io
f
o
r
i
m
a
g
e
ca
p
tu
r
e.
T
h
e
ca
p
tu
r
ed
im
ag
e
s
w
er
e
o
r
g
an
ized
in
to
f
o
ld
er
s
o
n
th
e
co
m
p
u
ter
co
r
r
esp
o
n
d
in
g
to
th
e
r
esp
ec
tiv
e
Ag
tr
o
n
r
o
ast
co
lo
r
class
es.
Af
ter
all
im
a
g
es
w
er
e
ca
p
tu
r
ed
,
d
ata
au
g
m
e
n
tatio
n
th
r
o
u
g
h
i
m
ag
e
tr
an
s
f
o
r
m
atio
n
w
a
s
i
m
p
le
m
e
n
ted
to
e
x
p
an
d
t
h
e
d
ataset.
B
y
ap
p
l
y
i
n
g
v
ar
io
u
s
i
m
a
g
e
tr
a
n
s
f
o
r
m
atio
n
s
,
ab
le
t
o
s
i
m
u
late
d
i
f
f
er
en
t
v
ie
w
i
n
g
co
n
d
itio
n
s
a
n
d
p
er
s
p
ec
tiv
es,
w
h
ich
h
elp
s
i
m
p
r
o
v
e
th
e
g
en
er
a
l
izab
ilit
y
o
f
th
e
m
o
d
els
tr
ain
ed
o
n
t
h
is
d
ataset.
T
ec
h
n
iq
u
es
s
u
c
h
as
v
er
tical
-
h
o
r
izo
n
ta
l
f
lip
p
i
n
g
,
r
o
tatio
n
,
zo
o
m
i
n
g
in
a
n
d
o
u
t,
f
ea
tu
r
e
w
i
s
e
tr
an
s
f
o
r
m
atio
n
,
w
id
th
-
h
eig
h
t
s
h
i
f
ti
n
g
,
a
n
d
r
an
d
o
m
cr
o
p
p
in
g
w
er
e
ap
p
lied
to
in
tr
o
d
u
ce
g
r
ea
ter
v
ar
iatio
n
i
n
th
e
i
m
a
g
e
s
.
T
he
d
ataset
laid
th
e
g
r
o
u
n
d
w
o
r
k
f
o
r
tr
ain
in
g
an
d
ev
alu
a
tin
g
f
o
r
ea
ch
A
lex
Net,
R
esNet,
Mo
b
ileNet,
VGGN
et,
an
d
De
n
s
eNe
t.
(
a)
(
b
)
(
c)
Fig
u
r
e
2
.
Ma
ter
ials
f
o
r
d
ataset
co
llectio
n
; (
a)
to
o
ls
f
o
r
d
ata
c
o
llectio
n
,
(
b
)
s
o
f
t
w
ar
e
s
e
tti
n
g
s
,
an
d
(
c)
A
g
tr
o
n
r
o
ast co
lo
r
an
d
co
f
f
ee
-
b
ea
n
r
o
asted
3
.
2
.
M
o
del t
ra
ini
ng
a
nd
v
a
li
da
t
io
n
T
h
e
tr
ain
in
g
s
tep
w
a
s
ca
r
ef
u
ll
y
m
o
n
i
to
r
ed
to
en
s
u
r
e
th
at
th
e
m
o
d
els
w
er
e
lear
n
in
g
ef
f
ec
t
iv
el
y
.
T
h
e
tr
ain
i
n
g
p
r
o
ce
s
s
f
o
r
all
f
iv
e
m
o
d
els
A
lex
Net,
R
esNet,
Mo
b
il
eNe
t
,
VGGN
et,
an
d
Den
s
eNe
t
w
as
co
n
d
u
cted
in
p
ar
allel
to
en
s
u
r
e
co
n
s
i
s
te
n
c
y
an
d
f
air
n
e
s
s
i
n
th
e
co
m
p
ar
i
s
o
n
.
E
ac
h
m
o
d
el
w
a
s
tr
ain
ed
f
o
r
5
0
ep
o
ch
s
,
a
d
u
r
atio
n
d
eter
m
in
ed
to
b
e
s
u
f
f
icie
n
t
f
o
r
th
e
m
o
d
els
to
lear
n
m
ea
n
in
g
f
u
l
p
atter
n
s
f
r
o
m
t
h
e
d
ataset
w
it
h
o
u
t
e
x
ce
s
s
iv
e
co
m
p
u
tatio
n
al
o
v
er
h
ea
d
.
T
h
e
en
v
ir
o
n
m
e
n
t
f
o
r
tr
ain
in
g
w
as
s
et
u
p
u
s
i
n
g
t
h
e
An
ac
o
n
d
a
d
is
tr
ib
u
tio
n
o
n
t
h
e
h
i
g
h
-
p
er
f
o
r
m
a
n
ce
co
m
p
u
ti
n
g
(
HP
C
)
Ma
h
e
m
er
u
B
R
I
N
f
ac
ilit
y
,
w
h
ich
p
r
o
v
id
ed
th
e
n
ec
e
s
s
ar
y
co
m
p
u
tatio
n
al
r
eso
u
r
ce
s
f
o
r
h
a
n
d
lin
g
lar
g
e
-
s
ca
le
m
ac
h
in
e
lear
n
i
n
g
tas
k
s
.
H
y
p
er
p
ar
a
m
eter
t
u
n
in
g
p
la
y
ed
a
cr
u
cial
r
o
le
i
n
o
p
tim
izin
g
t
h
e
m
o
d
el
s
’
p
er
f
o
r
m
an
ce
.
T
h
is
s
tu
d
y
e
m
p
lo
y
ed
th
e
A
d
a
m
o
p
ti
m
izer
o
n
ac
co
u
n
t
o
f
it
s
ad
ap
tiv
e
lear
n
in
g
r
ate
ca
p
ab
ilit
ies,
w
h
ich
ar
e
i
n
s
tr
u
m
en
tal
i
n
attai
n
in
g
s
u
p
er
io
r
co
n
v
er
g
e
n
ce
s
p
ee
d
o
v
er
s
tan
d
ar
d
g
r
ad
ien
t
d
escen
t.
T
h
e
lear
n
in
g
r
ate
o
f
0
.
0
0
5
w
a
s
s
elec
ted
af
ter
s
ev
er
al
tr
ials
to
b
alan
ce
th
e
s
p
ee
d
o
f
co
n
v
er
g
en
ce
w
it
h
t
h
e
s
tab
ilit
y
o
f
tr
ai
n
i
n
g
.
S
i
m
ilar
l
y
,
th
e
b
atc
h
s
ize
o
f
64
w
as
ch
o
s
e
n
to
en
s
u
r
e
ef
f
icie
n
t
u
s
e
o
f
co
m
p
u
ta
tio
n
a
l
r
eso
u
r
ce
s
w
h
i
le
m
ai
n
tai
n
i
n
g
a
s
u
f
f
icie
n
t
n
u
m
b
er
o
f
s
a
m
p
les
f
o
r
g
r
ad
ien
t
u
p
d
ates.
W
eig
h
t
m
o
m
e
n
t
u
m
an
d
d
ec
a
y
w
er
e
s
e
t
to
0
.
9
an
d
0
.
0
0
5
,
to
r
eg
u
lar
ize
th
e
m
o
d
els
a
n
d
p
r
ev
en
t
o
v
er
f
itti
n
g
w
h
ile
e
n
s
u
r
i
n
g
s
m
o
o
th
u
p
d
ates
to
th
e
m
o
d
el
p
ar
a
m
eter
s
.
3
.
3
.
E
v
a
lua
t
i
o
n
m
et
ric
E
v
alu
a
tio
n
m
etr
ic
s
co
n
s
tit
u
te
a
f
u
n
d
a
m
en
tal
co
m
p
o
n
e
n
t in
t
h
e
as
s
es
s
m
e
n
t
o
f
m
ac
h
in
e
lear
n
in
g
m
o
d
el
p
er
f
o
r
m
a
n
ce
,
as
th
e
y
s
u
p
p
l
y
q
u
an
t
itati
v
e
ev
id
en
ce
o
f
a
m
o
d
el
’
s
ca
p
ac
it
y
to
g
e
n
er
alize
to
u
n
s
ee
n
d
ata.
I
n
th
e
ev
alu
a
tio
n
o
f
th
e
C
N
N
ar
ch
ite
ctu
r
e,
th
e
p
r
im
ar
y
m
etr
ic
s
e
m
p
lo
y
ed
w
er
e
th
e
co
n
f
u
s
io
n
m
atr
i
x
an
d
class
i
f
icatio
n
ac
cu
r
ac
y
.
T
h
e
co
n
f
u
s
io
n
m
atr
i
x
s
er
v
es
as
a
d
iag
n
o
s
t
ic
to
o
l
th
at
d
elin
ea
tes
m
o
d
el
p
er
f
o
r
m
a
n
ce
b
y
tab
u
lati
n
g
t
h
e
in
cid
en
ce
o
f
tr
u
e
an
d
f
al
s
e
class
i
f
icatio
n
s
ac
r
o
s
s
all
ca
teg
o
r
ies,
th
er
eb
y
p
r
o
v
id
in
g
a
d
eta
iled
b
r
ea
k
d
o
w
n
o
f
p
r
ed
ictiv
e
o
u
tco
m
e
s
[
3
7
]
.
A
co
n
f
u
s
io
n
m
atr
ix
p
r
o
v
id
es
a
s
tr
u
ct
u
r
ed
tab
u
lar
o
v
er
v
i
e
w
f
o
r
ev
al
u
ati
n
g
class
i
f
icatio
n
m
o
d
el
p
er
f
o
r
m
a
n
ce
,
q
u
an
ti
f
y
i
n
g
p
r
ed
ictiv
e
o
u
tco
m
e
s
t
h
r
o
u
g
h
t
h
e
ca
te
g
o
r
ies
o
f
tr
u
e
p
o
s
i
tiv
e
s
(
T
P
)
,
tr
u
e
n
eg
ati
v
es
(
T
N)
,
f
als
e
p
o
s
itiv
e
s
(
FP
)
,
an
d
f
al
s
e
n
eg
ativ
es
(
FN)
.
Dis
p
ar
itie
s
i
n
m
o
d
el
p
er
f
o
r
m
a
n
ce
ca
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
23
,
No
.
6
,
Dec
em
b
er
20
25
:
1
5
9
0
-
1
599
1594
b
e
f
u
r
th
er
el
u
cid
ated
th
r
o
u
g
h
t
h
e
d
er
iv
atio
n
o
f
k
e
y
m
etr
ic
s
in
clu
d
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
—
co
m
p
u
ted
f
r
o
m
t
h
e
co
n
s
tit
u
e
n
t
v
alu
e
s
o
f
t
h
e
co
n
f
u
s
io
n
m
atr
i
x
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
4
.
1
.
Da
t
a
s
et
co
llect
i
o
n
T
h
e
r
esu
lts
o
f
t
h
e
d
ataset
co
llectio
n
p
r
o
ce
s
s
ar
e
s
u
m
m
ar
ize
d
in
Fig
u
r
e
3
.
A
ls
o
,
Fi
g
u
r
e
4
p
r
o
v
id
es
v
is
u
al
ex
a
m
p
les
o
f
i
m
a
g
e
s
f
r
o
m
ea
c
h
class
,
o
f
f
er
in
g
a
clea
r
r
ep
r
esen
tatio
n
o
f
th
e
d
ataset
’
s
d
iv
er
s
it
y
a
n
d
q
u
alit
y
.
T
h
is
d
ataset
co
n
s
is
t
s
o
f
i
m
a
g
es
o
f
co
f
f
ee
b
ea
n
s
r
o
asted
at
eig
h
t
d
is
ti
n
ct
Ag
tr
o
n
r
o
ast
cla
s
s
es,
r
ep
r
esen
tin
g
a
r
an
g
e
o
f
r
o
ast
p
r
o
f
iles
f
r
o
m
v
er
y
d
ar
k
to
v
er
y
li
g
h
t.
I
n
to
tal,
2
,
5
2
2
im
ag
e
s
o
f
co
f
f
ee
b
ea
n
s
w
er
e
g
at
h
er
ed
an
d
ca
teg
o
r
ized
in
to
ei
g
h
t c
las
s
es,
ea
ch
co
r
r
esp
o
n
d
in
g
to
a
s
p
ec
i
f
ic
A
g
tr
o
n
clas
s
.
T
h
ese
clas
s
es
ar
e
lab
eled
as R
2
5
as
s
h
o
w
n
as
Fi
g
u
r
e
4
(
a)
,
R
3
5
,
as
s
h
o
w
n
i
n
Fig
u
r
e
4
(
b
)
,
R
4
5
,
as
s
h
o
w
n
in
Fig
u
r
e
4
(
c)
,
R
5
5
,
as
s
h
o
w
n
in
Fig
u
r
e
4
(
d
)
,
R
6
5
as
s
h
o
w
n
i
n
Fig
u
r
e
4
(
e)
,
R
7
5
as
s
h
o
w
n
in
F
ig
u
r
e
4
(
f
)
,
R
8
5
,
as
s
h
o
w
n
in
Fi
g
u
r
e
4
(
g
)
,
an
d
R
9
5
,
as
s
h
o
w
n
in
Fi
g
u
r
e
4
(
h
)
.
E
ac
h
clas
s
i
s
s
to
r
ed
in
a
s
ep
ar
ate
f
o
ld
er
n
a
m
ed
f
o
r
its
r
e
s
p
ec
tiv
e
Ag
tr
o
n
clas
s
.
Fu
r
t
h
er
m
o
r
e,
th
e
d
ataset
w
a
s
p
ar
titi
o
n
ed
in
to
tr
ain
i
n
g
,
v
al
id
atio
n
,
an
d
test
i
n
g
s
u
b
s
ets
i
n
ac
co
r
d
an
ce
w
i
th
estab
lis
h
ed
m
ac
h
i
n
e
lear
n
i
n
g
p
r
o
to
co
ls
,
u
tili
zin
g
a
n
8
0
(
tr
ain
in
g
)
,
10
(
v
alid
atio
n
)
,
a
n
d
10
(
test
in
g
)
r
atio
.
A
d
h
er
en
ce
to
th
is
co
n
v
e
n
tio
n
al
f
r
a
m
e
w
o
r
k
en
s
u
r
es
th
e
d
at
aset
p
r
o
v
id
es
a
r
o
b
u
s
t
f
o
u
n
d
a
tio
n
f
o
r
d
ev
elo
p
in
g
m
o
d
el
s
p
r
o
f
icien
t in
t
h
e
ac
cu
r
ate
cl
ass
i
f
icatio
n
o
f
co
f
f
ee
b
ea
n
r
o
ast lev
el
s
.
Fig
u
r
e
3
.
Data
s
et
co
m
p
o
s
itio
n
(
a)
(
b
)
(
c)
(
d
)
(
e)
(
f
)
(
g
)
(
h
)
Fig
u
r
e
4
.
Sa
m
p
le
i
m
ag
e
s
f
r
o
m
th
e
d
ataset
f
o
r
ea
ch
cla
s
s
o
f
A
g
tr
o
n
clas
s
; (
a)
R
2
5
,
(
b
)
R
3
5
,
(
c)
R
4
5
,
(
d
)
R
5
5
,
(
e)
R
6
5
,
(
f
)
R
7
5
,
(
g
)
R
8
5
,
an
d
(
h
)
R
9
5
4
.
2
.
T
ra
ini
ng
a
nd
v
a
lid
a
t
ing
Af
ter
th
e
tr
ai
n
i
n
g
s
tep
o
n
HP
C
Ma
h
a
m
er
u
,
t
h
e
tr
ain
i
n
g
o
u
t
p
u
t
f
ile
s
f
o
r
ea
ch
m
o
d
el
w
er
e
o
b
tain
ed
in
th
e
f
o
r
m
o
f
“
.
p
th
”
an
d
“
.
p
k
l
”
f
o
r
m
a
ts
.
T
h
ese
f
iles
co
n
tai
n
e
d
th
e
tr
ain
ed
w
e
ig
h
ts
,
co
n
f
i
g
u
r
atio
n
s
,
an
d
o
th
er
r
elev
an
t
d
ata
n
ec
e
s
s
ar
y
f
o
r
f
u
r
t
h
er
an
al
y
s
is
.
T
h
e
tr
ai
n
i
n
g
r
esu
l
t
f
ile
s
w
er
e
th
e
n
a
n
al
y
ze
d
u
s
in
g
J
u
p
y
ter
No
teb
o
o
k
f
o
r
ev
alu
ati
o
n
,
w
h
er
e
p
er
f
o
r
m
an
ce
m
etr
ics,
in
c
lu
d
i
n
g
ac
c
u
r
ac
y
,
lo
s
s
,
an
d
co
n
f
u
s
io
n
m
atr
ic
es,
w
er
e
co
m
p
u
ted
an
d
v
i
s
u
al
ized
.
Me
tr
ics
s
u
ch
as
lo
s
s
a
n
d
ac
cu
r
ac
y
w
er
e
lo
g
g
ed
,
p
r
o
v
id
in
g
i
n
s
ig
h
t
s
i
n
to
th
e
m
o
d
els
’
b
eh
a
v
io
r
o
v
er
ep
o
ch
s
.
Fi
g
u
r
e
5
d
ep
ict
t
h
e
tr
ai
n
in
g
p
er
f
o
r
m
a
n
ce
m
etr
i
cs,
in
c
lu
d
i
n
g
ac
cu
r
ac
y
a
n
d
lo
s
s
,
f
o
r
b
o
th
tr
ain
i
n
g
an
d
v
alid
atio
n
p
h
ase
s
ac
r
o
s
s
all
ep
o
ch
s
f
o
r
th
e
f
iv
e
C
NN
m
o
d
els.
T
h
e
r
esu
lts
d
e
m
o
n
s
tr
ate
a
p
o
s
itiv
e
co
r
r
elatio
n
b
et
w
ee
n
ep
o
ch
co
u
n
t
a
n
d
m
o
d
el
p
er
f
o
r
m
a
n
ce
,
ch
ar
ac
ter
i
ze
d
b
y
in
cr
ea
s
in
g
ac
c
u
r
ac
y
an
d
a
c
o
n
co
m
ita
n
t
r
ed
u
ctio
n
in
lo
s
s
f
o
r
b
o
th
tr
ain
in
g
a
n
d
v
alid
atio
n
d
atasets
.
D
en
s
eNe
t
ac
h
ie
v
ed
th
e
h
ig
h
es
t
t
r
ain
in
g
ac
c
u
r
ac
y
at
9
9
.
7
0
2
%
an
d
th
e
h
ig
h
es
t
v
ali
d
atio
n
ac
cu
r
ac
y
at
7
7
.
6
8
%.
T
h
e
r
e
m
ai
n
i
n
g
m
o
d
el
s
’
tr
ai
n
i
n
g
ac
cu
r
ac
ies
w
er
e
as
f
o
llo
w
s
:
R
es
Net
at
9
7
.
0
7
%,
M
o
b
ileNet
at
9
5
.
5
4
4
%,
VGGN
et
at
8
9
.
8
5
1
%
an
d
A
lex
Net
at
8
5
.
3
4
6
%,
w
h
ic
h
w
a
s
th
e
lo
w
e
s
t.
T
h
e
b
est
-
p
er
f
o
r
m
i
n
g
m
o
d
el
d
is
p
la
y
ed
ex
ce
p
tio
n
al
co
n
v
er
g
e
n
ce
an
d
r
ea
ch
ed
p
ea
k
ac
cu
r
ac
y
w
it
h
m
i
n
i
m
al
tr
ai
n
i
n
g
a
n
d
v
alid
atio
n
lo
s
s
es.
I
t
ac
h
ie
v
ed
f
a
s
ter
co
n
v
er
g
e
n
ce
,
k
ep
t
th
e
d
is
p
ar
it
y
b
e
t
w
ee
n
tr
ai
n
i
n
g
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
C
o
mp
a
r
a
tive
p
erfo
r
ma
n
ce
a
n
a
lysi
s
o
f c
o
n
vo
lu
tio
n
a
l n
eu
r
a
l
…
(
I
r
fa
n
A
s
fy
F
a
kh
r
y
A
n
to
)
1595
v
alid
atio
n
lo
s
s
to
a
m
in
i
m
u
m
,
an
d
co
n
s
is
ten
t
l
y
d
eliv
er
ed
s
u
p
er
io
r
ac
cu
r
ac
y
co
m
p
ar
ed
to
th
e
o
th
er
m
o
d
els.
R
esNet
al
s
o
s
h
o
w
ed
s
tr
o
n
g
p
er
f
o
r
m
a
n
ce
,
ac
h
ie
v
i
n
g
n
ea
r
-
p
er
f
ec
t
tr
ain
i
n
g
ac
c
u
r
ac
y
,
i
n
d
icatin
g
its
ab
ilit
y
to
lear
n
co
m
p
le
x
p
atter
n
s
e
f
f
ec
tiv
e
l
y
.
Mo
b
ileNet,
w
h
ile
s
l
ig
h
tl
y
less
ac
cu
r
ate,
d
e
m
o
n
s
tr
ated
its
ef
f
icie
n
c
y
a
n
d
s
u
itab
il
it
y
f
o
r
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
en
v
ir
o
n
m
e
n
t
s
d
u
e
to
its
lig
h
t
w
e
ig
h
t
ar
ch
itect
u
r
e.
A
le
x
Net,
d
esp
ite
b
ein
g
th
e
o
ld
est
ar
ch
itect
u
r
e
a
m
o
n
g
t
h
e
test
ed
m
o
d
el
s
,
s
till
ac
h
iev
ed
r
ea
s
o
n
ab
le
ac
cu
r
ac
y
,
s
h
o
w
ca
s
in
g
it
s
f
o
u
n
d
atio
n
al
r
o
le
i
n
t
h
e
ev
o
l
u
t
io
n
o
f
C
NNs.
Up
o
n
co
m
p
leti
n
g
t
h
e
tr
ain
i
n
g
p
r
o
ce
s
s
,
th
e
m
o
d
els
w
er
e
e
v
al
u
ated
u
s
i
n
g
a
s
ep
ar
ate
test
d
ataset
t
o
m
ea
s
u
r
e
t
h
eir
g
e
n
er
aliza
tio
n
ca
p
ab
ilit
ies.
P
er
f
o
r
m
a
n
ce
m
etr
ics
d
escr
ib
ed
i
n
s
ec
tio
n
4
.
3
,
to
p
r
o
v
id
e
a
co
m
p
r
eh
en
s
iv
e
as
s
ess
m
e
n
t o
f
ea
ch
m
o
d
el
’
s
e
f
f
ec
tiv
e
n
es
s
.
T
h
e
r
es
u
lts
w
er
e
co
m
p
ar
ed
ac
r
o
s
s
th
e
f
iv
e
C
NN
ar
c
h
itect
u
r
es
to
id
en
tify
t
h
e
b
est
-
p
er
f
o
r
m
i
n
g
m
o
d
el
f
o
r
th
e
co
f
f
ee
b
ea
n
r
o
ast
class
if
icatio
n
task
.
T
h
i
s
s
y
s
te
m
at
ic
a
p
p
r
o
ac
h
to
tr
ai
n
i
n
g
an
d
e
v
al
u
atio
n
e
n
s
u
r
ed
th
at
t
h
e
m
o
d
el
s
w
er
e
r
ig
o
r
o
u
s
l
y
te
s
ted
an
d
o
p
tim
ized
,
p
r
o
v
id
in
g
r
eliab
le
i
n
s
i
g
h
ts
f
o
r
th
e
s
t
u
d
y
.
Fig
u
r
e
5
.
Mo
d
els an
d
v
alid
ati
n
g
ac
c
u
r
ac
y
a
n
d
lo
s
s
4
.
3
.
E
v
a
lua
t
i
o
n
m
a
t
rix
W
ith
in
t
h
e
d
o
m
a
in
o
f
i
m
ag
e
class
if
icatio
n
,
s
e
v
er
al
ev
alu
a
tio
n
m
etr
ics
ar
e
ex
te
n
s
i
v
el
y
e
m
p
lo
y
ed
,
n
o
tab
l
y
t
h
e
co
n
f
u
s
io
n
m
atr
i
x
(
C
M)
an
d
c
lass
if
icatio
n
ac
c
u
r
ac
y
.
T
h
e
e
v
al
u
atio
n
m
etr
ic
s
an
d
co
m
p
ar
ativ
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
i
s
s
t
u
d
y
ar
e
i
llu
s
tr
ated
in
Fi
g
u
r
e
6
.
Sp
e
cif
ic
all
y
,
th
e
C
M
co
r
r
esp
o
n
d
s
to
Alex
Net
F
ig
u
r
e
6
(
a)
,
VGGN
et
Fi
g
u
r
e
6
(
b
)
,
R
esNet
Fig
u
r
e
6
(
c)
,
Mo
b
ileNet
Fig
u
r
e
6
(
d
)
,
an
d
Den
s
eNe
t
Fi
g
u
r
e
6
(
e)
.
B
ased
o
n
th
e
e
m
p
ir
ical
r
esu
lts
,
t
h
e
test
i
n
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
w
er
e
q
u
a
n
ti
f
ied
f
o
r
ea
ch
m
o
d
el
an
d
in
d
iv
id
u
al
class
,
as
co
m
p
r
eh
en
s
i
v
el
y
d
etailed
in
Fi
g
u
r
e
6
(
f
)
.
T
h
e
Den
s
eNe
t
m
o
d
el
em
er
g
ed
as
th
e
to
p
-
p
er
f
o
r
m
in
g
ar
ch
itect
u
r
e
af
ter
av
er
ag
i
n
g
th
e
ex
p
er
i
m
e
n
tal
r
es
u
lts
,
ac
h
iev
in
g
an
i
m
p
r
es
s
i
v
e
av
er
ag
e
f
o
r
ac
cu
r
ac
y
o
f
9
3
.
8
%,
p
r
ec
is
io
n
o
f
7
6
%,
r
ec
all
o
f
7
5
,
4
%
an
d
f
1
-
s
co
r
e
o
f
0
.
7
5
5
.
R
esNet
f
o
llo
w
ed
clo
s
el
y
w
it
h
an
ac
cu
r
ac
y
o
f
9
2
.
6
%,
w
h
ile
VGGN
et
an
d
A
le
x
Net
b
o
th
ac
h
ie
v
ed
id
en
ti
ca
l
ac
cu
r
ac
ies
o
f
9
2
.
4
%.
Mo
b
ileNet
r
ec
o
r
d
ed
th
e
lo
w
es
t
ac
cu
r
ac
y
at
8
9
.
7
%.
W
h
en
e
v
al
u
ati
n
g
th
e
a
v
er
ag
e
d
ac
cu
r
ac
y
s
co
r
es
ac
r
o
s
s
all
lo
ca
tio
n
s
,
De
n
s
eNe
t
co
n
s
is
ten
tl
y
o
u
tp
er
f
o
r
m
ed
th
e
o
th
er
m
o
d
els,
w
h
er
ea
s
Mo
b
ileNet
s
tr
u
g
g
led
to
d
eliv
er
co
m
p
etitiv
e
r
es
u
lts
in
t
h
is
class
i
f
icatio
n
task
.
No
tab
l
y
,
Den
s
eNe
t
’
s
tr
ain
i
n
g
a
n
d
test
i
n
g
ac
cu
r
ac
y
m
etr
ic
s
alig
n
ed
p
er
f
ec
tl
y
,
r
ein
f
o
r
cin
g
Den
s
eNe
t
’
s
s
u
p
e
r
io
r
p
er
f
o
r
m
an
ce
ac
r
o
s
s
t
h
e
b
o
ar
d
.
Den
s
eNe
t
en
co
u
r
a
g
es
f
ea
t
u
r
e
r
eu
s
e
an
d
s
u
b
s
tan
tia
ll
y
i
m
p
r
o
v
es
p
ar
a
m
eter
ef
f
icie
n
c
y
,
m
ak
in
g
it
m
o
r
e
e
f
f
ec
tiv
e
in
t
ask
s
r
eq
u
ir
in
g
f
i
n
e
-
g
r
ai
n
ed
v
i
s
u
al
d
is
t
in
ctio
n
s
.
I
n
o
th
er
ca
s
es,
w
h
ile
De
n
s
eNe
t
’
s
ar
ch
itect
u
r
e
en
h
a
n
ce
s
f
ea
tu
r
e
r
eu
s
e
an
d
i
m
p
r
o
v
e
s
p
ar
a
m
eter
ef
f
icien
c
y
,
it
is
also
s
u
s
ce
p
tib
le
to
o
v
er
f
itti
n
g
.
T
r
a
d
itio
n
al
d
r
o
p
o
u
t
m
et
h
o
d
s
ca
n
in
ter
f
er
e
w
it
h
th
e
d
en
s
e
f
ea
t
u
r
e
p
r
o
p
ag
atio
n
in
Den
s
eNe
t,
r
ed
u
cin
g
th
eir
r
eg
u
lar
izatio
n
ef
f
ec
ti
v
en
e
s
s
a
n
d
u
n
in
te
n
tio
n
al
l
y
h
ar
m
i
n
g
g
e
n
er
ali
za
t
io
n
[
3
3
]
,
[
3
8
]
.
A
clo
s
er
ex
a
m
i
n
atio
n
o
f
cla
s
s
-
s
p
ec
if
ic
p
er
f
o
r
m
an
ce
r
ev
ea
led
t
h
at
A
lex
Net
ac
h
iev
ed
p
er
f
ec
t
ac
cu
r
ac
y
1
0
0
%
i
n
t
h
e
R
2
5
class
,
m
ak
i
n
g
it
th
e
b
est
p
er
f
o
r
m
er
f
o
r
th
at
p
ar
ticu
lar
c
lass
.
VGGN
et
an
d
R
e
s
Net
f
o
l
lo
w
ed
w
ith
9
9
.
2
%
ac
cu
r
ac
y
,
w
h
ile
De
n
s
eN
e
t
ac
h
iev
ed
9
6
.
8
%.
Mo
b
ileNet
lag
g
ed
s
i
g
n
i
f
ica
n
tl
y
b
eh
i
n
d
,
r
ec
o
r
d
in
g
o
n
l
y
8
8
.
0
4
8
%
ac
cu
r
ac
y
i
n
t
h
e
s
a
m
e
clas
s
.
T
h
e
o
b
s
er
v
ed
v
ar
iatio
n
in
p
er
f
o
r
m
an
ce
ac
r
o
s
s
clas
s
es
i
n
d
icate
s
th
at
ce
r
tai
n
ar
ch
itect
u
r
es
m
a
y
b
e
m
o
r
e
e
f
f
ec
ti
v
e
f
o
r
p
ar
ticu
lar
d
ata
t
y
p
es,
u
n
d
er
s
co
r
in
g
th
e
n
ec
es
s
it
y
o
f
alig
n
i
n
g
m
o
d
el
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
23
,
No
.
6
,
Dec
em
b
er
20
25
:
1
5
9
0
-
1
599
1596
s
elec
tio
n
w
it
h
t
h
e
d
is
ti
n
ct
c
h
ar
ac
ter
is
tic
s
o
f
th
e
d
ata
s
et.
A
d
d
itio
n
al
l
y
,
co
m
b
in
i
n
g
m
u
l
tip
le
p
r
ep
r
o
ce
s
s
in
g
m
et
h
o
d
s
o
r
c
o
n
ca
ten
ati
n
g
f
ea
t
u
r
e
v
ec
to
r
s
d
er
iv
ed
f
r
o
m
d
if
f
e
r
en
t
ap
p
r
o
ac
h
es
co
u
ld
en
h
an
c
e
th
e
o
v
er
all
q
u
alit
y
o
f
th
e
in
p
u
t
d
ata.
E
x
p
lo
r
in
g
alter
n
ati
v
e
n
eu
r
al
n
et
w
o
r
k
ar
ch
itectu
r
es,
s
u
c
h
as
tr
an
s
f
o
r
m
er
s
o
r
h
y
b
r
id
m
o
d
els
,
co
u
ld
also
y
ield
b
etter
r
esu
lts
.
Fu
r
th
er
m
o
r
e,
th
e
lo
w
p
er
f
o
r
m
an
ce
o
f
ce
r
tain
m
o
d
el
s
in
s
p
ec
if
ic
class
e
s
m
a
y
in
d
icate
a
n
ee
d
f
o
r
m
et
h
o
d
o
lo
g
ical
ad
j
u
s
t
m
e
n
ts
.
A
d
d
itio
n
all
y
,
ex
p
lo
r
in
g
t
h
e
in
teg
r
atio
n
o
f
u
n
s
u
p
er
v
i
s
ed
o
r
s
elf
-
s
u
p
er
v
i
s
ed
lear
n
i
n
g
tec
h
n
iq
u
e
s
m
i
g
h
t
r
ed
u
ce
th
e
r
elian
ce
o
n
lar
g
e
lab
eled
d
atasets
.
Fu
tu
r
e
d
ev
elo
p
m
e
n
ts
i
n
d
ee
p
lear
n
in
g
tec
h
n
iq
u
es,
i
n
cl
u
d
in
g
r
e
f
in
ed
at
ten
t
io
n
m
ec
h
a
n
is
m
s
an
d
ad
ap
ti
v
e
lear
n
in
g
r
ates,
ar
e
an
t
icip
ated
to
co
n
s
id
er
ab
ly
e
x
p
an
d
t
h
e
f
r
o
n
tier
s
o
f
ac
h
ie
v
ab
le
ac
cu
r
ac
y
i
n
co
f
f
ee
-
b
ea
n
r
o
ast cla
s
s
if
ica
ti
o
n
task
s
.
(
a)
(
b
)
(
c)
(
d
)
(
e)
(
f
)
Fig
u
r
e
6
.
C
o
n
f
u
s
io
n
m
atr
ix
a
n
d
co
m
p
ar
ativ
e
p
er
f
o
r
m
an
ce
; (
a)
A
le
x
Net,
(
b
)
VGGN
et,
(
c)
R
estNet,
(
d
)
Mo
b
ileNet,
(
e)
Den
s
eNe
t
,
an
d
(
f
)
test
in
g
ac
c
u
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
5.
CO
NCLU
SI
O
N
R
ec
en
t
ad
v
a
n
ce
m
en
t
s
in
C
N
N
ar
ch
itectu
r
es
h
a
v
e
s
i
g
n
if
ica
n
tl
y
e
n
h
a
n
ce
d
c
o
f
f
ee
-
b
ea
n
r
o
asti
n
g
lev
e
l
class
i
f
icatio
n
,
p
ar
ticu
lar
l
y
i
n
p
r
ed
ictin
g
A
g
tr
o
n
class
e
s
.
T
h
is
s
tu
d
y
p
r
o
v
id
es
a
co
m
p
r
eh
e
n
s
i
v
e
p
er
f
o
r
m
a
n
ce
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
C
o
mp
a
r
a
tive
p
erfo
r
ma
n
ce
a
n
a
lysi
s
o
f c
o
n
vo
lu
tio
n
a
l n
eu
r
a
l
…
(
I
r
fa
n
A
s
fy
F
a
kh
r
y
A
n
to
)
1597
co
m
p
ar
is
o
n
o
f
s
ev
er
al
C
NN
m
o
d
els
A
le
x
Net,
R
esNet,
Mo
b
ileNet,
VGGN
et,
an
d
D
en
s
eNe
t
t
h
r
o
u
g
h
a
s
tr
u
ct
u
r
ed
ev
al
u
atio
n
p
r
o
ce
s
s
i
n
v
o
l
v
i
n
g
d
ata
co
llectio
n
,
p
r
ep
r
o
ce
s
s
in
g
,
tr
ai
n
i
n
g
,
v
alid
ati
o
n
,
an
d
ev
al
u
atio
n
m
atr
i
x
.
T
h
is
r
esear
ch
u
tili
ze
d
co
s
t
-
e
f
f
ec
tiv
e
i
m
a
g
e
-
b
ased
d
ata
co
llectio
n
w
it
h
a
ca
m
er
a.
Du
r
in
g
tr
ain
i
n
g
an
d
v
alid
atio
n
,
De
n
s
eNe
t
o
u
tp
er
f
o
r
m
ed
o
th
er
m
o
d
els,
ac
h
iev
i
n
g
9
9
.
7
0
2
%
tr
a
in
in
g
ac
c
u
r
ac
y
an
d
7
7
.
6
8
%
v
alid
atio
n
ac
cu
r
ac
y
.
I
n
t
h
e
f
i
n
al
ev
al
u
ati
o
n
,
Den
s
e
Net
also
led
w
it
h
an
av
er
ag
e
te
s
ti
n
g
ac
c
u
r
ac
y
o
f
9
3
.
8
%,
f
o
llo
w
ed
b
y
R
esNet
at
9
2
.
6
%,
VGGN
et
an
d
A
le
x
Net
b
o
th
at
9
2
.
4
%,
an
d
Mo
b
ileNet
at
8
9
.
7
%.
T
h
e
f
i
n
d
in
g
s
r
e
v
ea
l
cr
itica
l
tr
ad
e
-
o
f
f
s
b
et
w
ee
n
co
m
p
u
tatio
n
al
ef
f
icie
n
c
y
a
n
d
ac
cu
r
ac
y
,
w
h
ich
m
u
s
t
b
e
co
n
s
id
er
ed
w
h
e
n
d
ep
lo
y
in
g
co
f
f
ee
-
b
ea
n
class
if
icatio
n
s
y
s
te
m
s
i
n
r
ea
l
-
w
o
r
ld
ap
p
licatio
n
s
.
W
h
ile
De
n
s
e
Net
d
eli
v
er
ed
th
e
h
i
g
h
e
s
t
ac
c
u
r
ac
y
,
lig
h
t
w
ei
g
h
t
m
o
d
els
lik
e
Mo
b
ileNet
co
u
ld
b
e
m
o
r
e
p
r
ac
tic
al.
Fu
t
u
r
e
r
esear
ch
co
u
ld
ex
p
lo
r
e
h
y
b
r
id
ar
ch
itect
u
r
es
th
at
co
m
b
i
n
e
th
e
s
tr
en
g
t
h
s
o
f
d
if
f
er
e
n
t
C
NN
s
w
h
i
le
o
p
ti
m
iz
in
g
co
m
p
u
tatio
n
al
ef
f
icie
n
c
y
.
Nex
t,
u
n
s
u
p
er
v
i
s
ed
lear
n
in
g
tec
h
n
iq
u
es
co
u
ld
r
e
d
u
ce
r
elian
ce
o
n
lar
g
e
lab
el
ed
d
atasets
,
m
ak
i
n
g
th
e
s
y
s
t
e
m
m
o
r
e
s
c
alab
le.
E
m
er
g
i
n
g
ad
v
a
n
ce
m
e
n
t
s
i
n
att
en
tio
n
m
ec
h
a
n
is
m
s
a
n
d
ad
ap
tiv
e
lear
n
i
n
g
r
ate
o
p
ti
m
iza
tio
n
c
o
u
ld
f
u
r
th
er
r
e
f
in
e
Ag
tr
o
n
-
le
v
el
clas
s
i
f
icatio
n
,
p
a
v
in
g
t
h
e
w
a
y
f
o
r
m
o
r
e
p
r
ec
is
e
an
d
au
to
m
a
ted
co
f
f
ee
r
o
asti
n
g
p
r
o
ce
s
s
es.
ACK
NO
WL
E
D
G
M
E
NT
S
T
h
e
au
th
o
r
s
ac
k
n
o
w
led
g
e
th
e
s
u
p
p
o
r
t
o
f
th
e
Natio
n
al
R
esear
c
h
an
d
I
n
n
o
v
atio
n
Ag
en
c
y
(
B
R
I
N)
.
A
ls
o
,
th
e
co
m
p
u
tatio
n
s
p
er
f
o
r
m
ed
i
n
th
i
s
w
o
r
k
u
tili
ze
d
th
e
M
AH
A
ME
R
U
HP
C
f
ac
ilit
y
,
m
a
n
a
g
ed
b
y
B
R
I
N.
F
UNDIN
G
I
NF
O
RM
AT
I
O
N
Au
t
h
o
r
s
s
tate
n
o
f
u
n
d
i
n
g
i
n
v
o
l
v
ed
.
AUTHO
R
CO
NT
RIB
UT
I
O
NS ST
A
T
E
M
E
NT
T
h
is
j
o
u
r
n
al
u
s
e
s
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT
)
to
r
ec
o
g
n
ize
in
d
i
v
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
t
h
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
lla
b
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
I
r
f
an
A
s
f
y
Fa
k
h
r
y
A
n
to
✓
✓
✓
✓
✓
✓
✓
✓
J
o
n
y
W
i
n
ar
y
o
W
ib
o
w
o
✓
✓
✓
✓
✓
A
r
i
s
Mu
n
an
d
ar
✓
✓
✓
✓
✓
T
au
f
ik
I
b
n
u
Sali
m
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
si
s
I
:
I
n
v
e
st
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
si
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
th
o
r
s
d
ec
lar
e
th
at
t
h
e
y
h
av
e
n
o
co
n
f
lict o
f
i
n
ter
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
atasets
u
s
ed
an
d
/o
r
a
n
al
y
ze
d
in
th
is
s
tu
d
y
ar
e
a
v
ailab
le
f
r
o
m
th
e
co
r
r
esp
o
n
d
in
g
au
t
h
o
r
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
A
.
A
.
C
a
t
ã
o
e
t
a
l
.
,
“
C
o
f
f
e
e
-
r
o
a
st
i
n
g
v
a
r
i
a
b
l
e
s
a
sso
c
i
a
t
e
d
w
i
t
h
v
o
l
a
t
i
l
e
o
r
g
a
n
i
c
p
r
o
f
i
l
e
s
a
n
d
se
n
so
r
y
e
v
a
l
u
a
t
i
o
n
u
si
n
g
mu
l
t
i
v
a
r
i
a
t
e
a
n
a
l
y
si
s,”
Ap
p
l
i
e
d
Fo
o
d
R
e
se
a
rc
h
,
v
o
l
.
2
,
n
o
.
2
,
D
e
c
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
a
f
r
e
s.2
0
2
2
.
1
0
0
2
2
3
.
[
2
]
M
.
L
i
n
,
C
.
C
h
e
n
,
a
n
d
J
.
L
u
,
“
I
n
t
e
l
l
i
g
e
n
t
c
l
a
ss
i
f
i
e
r
f
o
r
v
a
r
i
o
u
s
d
e
g
r
e
e
s
o
f
c
o
f
f
e
e
r
o
a
st
s
u
si
n
g
smar
t
m
u
l
t
i
s
p
e
c
t
r
a
l
v
i
si
o
n
sy
st
e
m,”
J
o
u
rn
a
l
o
f
Fi
e
l
d
R
o
b
o
t
i
c
s
,
v
o
l
.
4
1
,
n
o
.
3
,
p
p
.
6
3
9
–
6
5
3
,
M
a
y
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
2
/
r
o
b
.
2
2
2
8
5
.
[
3
]
L
.
A
n
o
k
y
e
-
B
e
mp
a
h
,
I
.
R
.
D
o
n
i
s
-
G
o
n
z
á
l
e
z
,
a
n
d
W
.
D
.
R
i
st
e
n
p
a
r
t
,
“
W
h
a
t
c
o
l
o
r
i
s
y
o
u
r
c
o
f
f
e
e
?
”
A
c
c
e
sse
d
:
O
c
t
.
1
9
,
2
0
2
4
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s:
/
/
s
c
a
.
c
o
f
f
e
e
/
sca
-
n
e
w
s
/
2
5
/
i
ss
u
e
-
2
1
/
w
h
a
t
-
c
o
l
o
r
-
is
-
y
o
u
r
-
c
o
f
f
e
e
[
4
]
F
.
d
e
C
a
r
v
a
l
h
o
P
i
r
e
s,
R
.
G
.
F
.
A
.
P
e
r
e
i
r
a
,
M
.
R
.
B
a
q
u
e
t
a
,
P
.
V
a
l
d
e
r
r
a
ma,
a
n
d
R
.
A
.
d
a
R
o
c
h
a
,
“
C
o
mp
a
r
i
so
n
b
e
t
w
e
e
n
b
i
n
a
r
y
a
n
d
mu
l
t
i
c
l
a
ss
p
a
r
t
i
a
l
l
e
a
st
s
q
u
a
r
e
w
i
t
h
d
i
scri
mi
n
a
n
t
a
n
a
l
y
si
s:
A
c
a
se
st
u
d
y
o
n
c
o
f
f
e
e
c
l
a
ssi
f
i
c
a
t
i
o
n
b
a
se
d
o
n
n
e
a
r
-
i
n
f
r
a
r
e
d
sp
e
c
t
r
a
,
”
i
n
Ad
v
a
n
c
e
s i
n
S
p
e
c
t
r
o
sc
o
p
i
c
A
n
a
l
y
s
i
s
o
f
F
o
o
d
a
n
d
D
ri
n
k
,
I
O
P
P
u
b
l
i
s
h
i
n
g
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
8
8
/
9
7
8
-
0
-
7
5
0
3
-
5
5
7
3
-
5
c
h
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
23
,
No
.
6
,
Dec
em
b
er
20
25
:
1
5
9
0
-
1
599
1598
[
5
]
I
.
S
a
n
t
o
so
,
S
.
A
.
M
u
s
t
a
n
i
r
o
h
,
a
n
d
A
.
C
h
o
i
r
u
n
,
“
M
e
t
h
o
d
s
f
o
r
q
u
a
l
i
t
y
c
o
f
f
e
e
r
o
a
st
i
n
g
d
e
g
r
e
e
e
v
a
l
u
a
t
i
o
n
:
A
l
i
t
e
r
a
t
u
r
e
r
e
v
i
e
w
o
n
r
i
s
k
p
e
r
sp
e
c
t
i
v
e
,
”
I
O
P
C
o
n
f
e
re
n
c
e
S
e
ri
e
s:
Ea
r
t
h
a
n
d
E
n
v
i
r
o
n
m
e
n
t
a
l
S
c
i
e
n
c
e
,
v
o
l
.
9
2
4
,
n
o
.
1
,
N
o
v
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
8
8
/
1
7
5
5
-
1
3
1
5
/
9
2
4
/
1
/
0
1
2
0
5
8
.
[
6
]
S
.
E.
Y
e
a
g
e
r
e
t
a
l
.
,
“
R
o
a
s
t
l
e
v
e
l
a
n
d
b
r
e
w
t
e
mp
e
r
a
t
u
r
e
si
g
n
i
f
i
c
a
n
t
l
y
a
f
f
e
c
t
t
h
e
c
o
l
o
r
o
f
b
r
e
w
e
d
c
o
f
f
e
e
,
”
J
o
u
rn
a
l
o
f
F
o
o
d
S
c
i
e
n
c
e
,
v
o
l
.
8
7
,
n
o
.
4
,
p
p
.
1
8
3
7
–
1
8
5
0
,
A
p
r
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
1
1
/
1
7
5
0
-
3
8
4
1
.
1
6
0
8
9
.
[
7
]
R
.
E
.
B
u
m
b
a
u
g
h
e
t
a
l
.
,
“
A
n
e
l
e
c
t
r
o
c
h
e
mi
c
a
l
a
p
p
r
a
i
s
a
l
o
f
c
o
f
f
e
e
q
u
a
l
i
t
i
e
s,”
Pre
p
r
.
a
rX
i
v
.
2
5
0
1
.
1
4
9
5
0
,
O
c
t
.
2
0
2
5
.
[
8
]
J.
A
l
st
r
u
p
,
M
.
A
.
P
e
t
e
r
se
n
,
F
.
H
.
L
a
r
se
n
,
a
n
d
M
.
M
ü
n
c
h
o
w
,
“
T
h
e
e
f
f
e
c
t
o
f
r
o
a
st
d
e
v
e
l
o
p
me
n
t
t
i
me
mo
d
u
l
a
t
i
o
n
s
o
n
t
h
e
se
n
so
r
y
p
r
o
f
i
l
e
a
n
d
c
h
e
mi
c
a
l
c
o
mp
o
si
t
i
o
n
o
f
t
h
e
c
o
f
f
e
e
b
r
e
w
a
s
me
a
su
r
e
d
b
y
N
M
R
a
n
d
D
H
S
-
GC
–
M
S
,
”
Be
v
e
r
a
g
e
s
,
v
o
l
.
6
,
n
o
.
4
,
D
e
c
.
2
0
2
0
,
d
o
i
:
1
0
.
3
3
9
0
/
b
e
v
e
r
a
g
e
s6
0
4
0
0
7
0
.
[
9
]
V
.
S
o
ma
u
d
o
n
a
n
d
T
.
K
e
r
d
c
h
a
r
o
e
n
,
“
U
n
l
o
c
k
i
n
g
t
h
e
a
r
o
ma
p
r
o
f
i
l
e
s
o
f
c
o
f
f
e
e
r
o
a
st
i
n
g
l
e
v
e
l
s
w
i
t
h
a
n
e
l
e
c
t
r
o
n
i
c
n
o
se
c
o
u
p
l
e
d
w
i
t
h
mac
h
i
n
e
l
e
a
r
n
i
n
g
,
”
i
n
2
0
2
4
2
1
s
t
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
i
n
t
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
S
o
f
t
w
a
re
En
g
i
n
e
e
ri
n
g
(
J
C
S
S
E)
,
I
EEE,
Ju
n
.
2
0
2
4
,
p
p
.
6
7
8
–
6
8
1
,
d
o
i
:
1
0
.
1
1
0
9
/
JC
S
S
E6
1
2
7
8
.
2
0
2
4
.
1
0
6
1
3
7
1
5
.
[
1
0
]
R
.
G
.
P
e
r
e
i
r
a
N
e
t
o
,
P
.
M
.
d
e
S
o
u
s
a
,
L
.
F
.
R
.
M
o
r
e
i
r
a
,
P
.
I
.
V
.
G
.
G
o
d
,
a
n
d
J
.
F
.
M
a
r
i
,
“
En
h
a
n
c
i
n
g
g
r
e
e
n
c
o
f
f
e
e
q
u
a
l
i
t
y
a
sse
ss
me
n
t
t
h
r
o
u
g
h
d
e
e
p
l
e
a
r
n
i
n
g
,
”
i
n
A
n
a
i
s
d
o
X
VI
I
I
Wo
r
k
sh
o
p
d
e
Vi
sã
o
C
o
m
p
u
t
a
c
i
o
n
a
l
(
WV
C
2
0
2
3
)
,
S
o
c
i
e
d
a
d
e
B
r
a
si
l
e
i
r
a
d
e
C
o
m
p
u
t
a
ç
ã
o
-
S
B
C
,
N
o
v
.
2
0
2
3
,
p
p
.
8
4
–
89
,
d
o
i
:
1
0
.
5
7
5
3
/
w
v
c
.
2
0
2
3
.
2
7
5
3
7
.
[
1
1
]
Q.
-
H
.
W
u
a
n
d
D
.
-
F
.
S
h
e
n
,
“
C
o
l
o
r
v
i
d
e
o
a
n
d
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
d
e
e
p
l
e
a
r
n
i
n
g
b
a
se
d
r
e
a
l
-
t
i
me
a
g
t
r
o
n
b
a
k
i
n
g
l
e
v
e
l
e
st
i
mat
i
o
n
me
t
h
o
d
,
”
i
n
N
e
w
T
re
n
d
s
i
n
C
o
m
p
u
t
e
r
T
e
c
h
n
o
l
o
g
i
e
s
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
2
0
1
9
,
p
p
.
1
9
7
–
206
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
9
8
1
-
13
-
9
1
9
0
-
3
_
2
1
.
[
1
2
]
P
.
W
a
n
g
,
H
.
-
W
.
T
s
e
n
g
,
T
.
-
C
.
C
h
e
n
,
a
n
d
C
.
-
H
.
H
si
a
,
“
D
e
e
p
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
f
o
r
c
o
f
f
e
e
b
e
a
n
i
n
s
p
e
c
t
i
o
n
,
”
S
e
n
so
rs a
n
d
Ma
t
e
r
i
a
l
s
,
v
o
l
.
3
3
,
n
o
.
7
,
J
u
l
.
2
0
2
1
,
d
o
i
:
1
0
.
1
8
4
9
4
/
S
A
M
.
2
0
2
1
.
3
2
7
7
.
[
1
3
]
N
.
K
.
N
a
i
k
a
n
d
P
.
K
.
S
e
t
h
y
,
“
R
o
a
st
e
d
c
o
f
f
e
e
b
e
a
n
s
c
l
a
ss
i
f
i
c
a
t
i
o
n
b
a
se
d
o
n
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
,
”
i
n
2
0
2
2
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
F
u
t
u
ri
s
t
i
c
T
e
c
h
n
o
l
o
g
i
e
s (I
N
C
O
FT)
,
I
EEE,
N
o
v
.
2
0
2
2
,
p
p
.
1
–
3
,
d
o
i
:
1
0
.
1
1
0
9
/
I
N
C
O
F
T
5
5
6
5
1
.
2
0
2
2
.
1
0
0
9
4
3
7
8
.
[
1
4
]
E.
I
.
M
u
h
l
i
si
n
,
R
.
R
.
N
u
r
m
a
l
a
sari
,
L
.
K
a
mel
i
a
,
a
n
d
R
.
W
.
S
u
r
u
r
i
e
,
“
I
mp
l
e
m
e
n
t
a
t
i
o
n
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
(
C
N
N
)
i
n
T
h
e
A
n
d
r
o
i
d
-
b
a
se
d
a
p
p
l
i
c
a
t
i
o
n
f
o
r
d
e
t
e
c
t
i
n
g
c
o
f
f
e
e
b
e
a
n
mat
u
r
i
t
y
,
”
i
n
2
0
2
4
1
0
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
Wi
r
e
l
e
ss a
n
d
T
e
l
e
m
a
t
i
c
s
(
I
C
WT)
,
I
EEE,
Ju
l
.
2
0
2
4
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
W
T
6
2
0
8
0
.
2
0
2
4
.
1
0
6
7
4
6
7
6
.
[
1
5
]
K
.
S
a
d
d
a
mi
,
N
.
A
u
l
i
a
,
V
.
M
a
u
l
i
d
i
a
,
a
n
d
A
n
d
r
i
a
n
sy
a
h
,
“
C
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
si
s
o
f
l
i
g
h
t
w
e
i
g
h
t
p
r
e
-
t
r
a
i
n
e
d
C
N
N
mo
d
e
l
s
f
o
r
c
o
f
f
e
e
b
e
a
n
r
o
a
st
i
n
g
l
e
v
e
l
i
d
e
n
t
i
f
i
c
a
t
i
o
n
,
”
i
n
2
0
2
3
2
n
d
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
S
y
s
t
e
m
,
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
a
n
d
E
l
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
(
C
O
S
I
T
E)
,
I
EEE,
A
u
g
.
2
0
2
3
,
p
p
.
1
3
–
18
,
d
o
i
:
1
0
.
1
1
0
9
/
C
O
S
I
T
E6
0
2
3
3
.
2
0
2
3
.
1
0
2
4
9
5
4
6
.
[
1
6
]
E.
H
a
ssa
n
,
“
En
h
a
n
c
i
n
g
c
o
f
f
e
e
b
e
a
n
c
l
a
ssi
f
i
c
a
t
i
o
n
:
a
c
o
mp
a
r
a
t
i
v
e
a
n
a
l
y
si
s o
f
p
r
e
-
t
r
a
i
n
e
d
d
e
e
p
l
e
a
r
n
i
n
g
mo
d
e
l
s,”
N
e
u
r
a
l
C
o
m
p
u
t
i
n
g
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
6
,
n
o
.
1
6
,
p
p
.
9
0
2
3
–
9
0
5
2
,
Ju
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
5
2
1
-
0
2
4
-
0
9
6
2
3
-
z.
[
1
7
]
I
.
A
.
F
.
A
n
t
o
,
J.
W
.
W
i
b
o
w
o
,
T
.
I
.
S
a
l
i
m,
a
n
d
A
.
M
u
n
a
n
d
a
r
,
“
I
mp
l
e
me
n
t
a
t
i
o
n
o
f
i
m
a
g
e
p
r
o
c
e
ssi
n
g
a
n
d
C
N
N
f
o
r
r
o
a
st
e
d
-
c
o
f
f
e
e
l
e
v
e
l
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
I
n
d
o
n
e
si
a
n
J
o
u
r
n
a
l
o
f
El
e
c
t
ri
c
a
l
E
n
g
i
n
e
e
r
i
n
g
a
n
d
I
n
f
o
rm
a
t
i
c
s
(
I
J
EEI)
,
v
o
l
.
1
2
,
n
o
.
4
,
p
p
.
1
0
0
5
–
1
0
1
8
,
D
e
c
.
2
0
2
4
,
d
o
i
:
1
0
.
5
2
5
4
9
/
i
j
e
e
i
.
v
1
2
i
4
.
5
5
3
1
.
[
1
8
]
M
.
B
o
l
k
a
a
n
d
S
.
Em
i
r
e
,
“
Ef
f
e
c
t
s
o
f
c
o
f
f
e
e
r
o
a
st
i
n
g
t
e
c
h
n
o
l
o
g
i
e
s
o
n
c
u
p
q
u
a
l
i
t
y
a
n
d
b
i
o
a
c
t
i
v
e
c
o
m
p
o
u
n
d
s
o
f
s
p
e
c
i
a
l
t
y
c
o
f
f
e
e
b
e
a
n
s,”
Fo
o
d
S
c
i
N
u
tr
,
v
o
l
.
8
,
n
o
.
1
1
,
p
p
.
6
1
2
0
–
6
1
3
0
,
N
o
v
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
2
/
f
s
n
3
.
1
9
0
4
.
[
1
9
]
F
.
d
e
C
a
r
v
a
l
h
o
P
i
r
e
s
e
t
a
l
.
,
“
F
e
a
si
b
i
l
i
t
y
o
f
u
si
n
g
c
o
l
o
r
i
me
t
r
i
c
d
e
v
i
c
e
s
f
o
r
w
h
o
l
e
a
n
d
g
r
o
u
n
d
c
o
f
f
e
e
r
o
a
st
i
n
g
d
e
g
r
e
e
s
p
r
e
d
i
c
t
i
o
n
,
”
J
o
u
rn
a
l
o
f
t
h
e
S
c
i
e
n
c
e
o
f
Fo
o
d
a
n
d
A
g
ri
c
u
l
t
u
re
,
v
o
l
.
1
0
4
,
n
o
.
9
,
p
p
.
5
4
3
5
–
5
4
4
1
,
Ju
l
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
2
/
j
sf
a
.
1
3
3
7
6
.
[
2
0
]
M
.
M
ü
n
c
h
o
w
,
J.
A
l
st
r
u
p
,
I
.
S
t
e
e
n
,
a
n
d
D
.
G
i
a
c
a
l
o
n
e
,
“
R
o
a
st
i
n
g
c
o
n
d
i
t
i
o
n
s
a
n
d
c
o
f
f
e
e
f
l
a
v
o
r
:
A
mu
l
t
i
-
st
u
d
y
e
mp
i
r
i
c
a
l
i
n
v
e
st
i
g
a
t
i
o
n
,
”
Be
v
e
ra
g
e
s
,
v
o
l
.
6
,
n
o
.
2
,
M
a
y
2
0
2
0
,
d
o
i
:
1
0
.
3
3
9
0
/
b
e
v
e
r
a
g
e
s6
0
2
0
0
2
9
.
[
2
1
]
A
.
K
r
i
z
h
e
v
sk
y
,
I
.
S
u
t
sk
e
v
e
r
,
a
n
d
G
.
E.
H
i
n
t
o
n
,
“
I
mag
e
N
e
t
c
l
a
ssi
f
i
c
a
t
i
o
n
w
i
t
h
d
e
e
p
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,”
C
o
m
m
u
n
i
c
a
t
i
o
n
s
o
f
t
h
e
AC
M
,
v
o
l
.
6
0
,
n
o
.
6
,
p
p
.
8
4
–
9
0
,
M
a
y
2
0
1
7
,
d
o
i
:
1
0
.
1
1
4
5
/
3
0
6
5
3
8
6
.
[
2
2
]
W
.
H
u
ssai
n
e
t
a
l
.
,
“
En
se
m
b
l
e
g
e
n
e
t
i
c
a
n
d
C
N
N
mo
d
e
l
-
b
a
se
d
i
mag
e
c
l
a
ssi
f
i
c
a
t
i
o
n
b
y
e
n
h
a
n
c
i
n
g
h
y
p
e
r
p
a
r
a
me
t
e
r
t
u
n
i
n
g
,
”
S
c
i
e
n
t
i
f
i
c
Re
p
o
r
t
s
,
v
o
l
.
1
5
,
n
o
.
1
,
J
a
n
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
8
-
0
2
4
-
7
6
1
7
8
-
3.
[
2
3
]
K
.
S
i
mo
n
y
a
n
a
n
d
A
.
Z
i
sse
r
man
,
“
V
e
r
y
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
s
f
o
r
l
a
r
g
e
-
sca
l
e
i
mag
e
r
e
c
o
g
n
i
t
i
o
n
,
”
Pre
p
r.
a
rXi
v
.
1
4
0
9
.
1
5
5
6
,
A
p
r
.
2
0
1
5
.
[
2
4
]
S
.
G
h
o
sh
,
A
.
C
h
a
k
i
,
a
n
d
K
.
S
a
n
t
o
sh
,
“
I
mp
r
o
v
e
d
U
-
N
e
t
a
r
c
h
i
t
e
c
t
u
r
e
w
i
t
h
V
G
G
-
1
6
f
o
r
b
r
a
i
n
t
u
mo
r
se
g
me
n
t
a
t
i
o
n
,
”
P
h
y
s
i
c
a
l
a
n
d
En
g
i
n
e
e
ri
n
g
S
c
i
e
n
c
e
s
i
n
Me
d
i
c
i
n
e
,
v
o
l
.
4
4
,
n
o
.
3
,
p
p
.
7
0
3
–
7
1
2
,
S
e
p
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
3
2
4
6
-
0
2
1
-
0
1
0
1
9
-
w.
[
2
5
]
P
.
U
d
a
y
a
r
a
j
u
,
P
.
Je
y
a
n
t
h
i
,
a
n
d
B
.
V
.
D
.
S
.
S
e
k
h
a
r
,
“
A
h
y
b
r
i
d
mu
l
t
i
l
a
y
e
r
e
d
c
l
a
ssi
f
i
c
a
t
i
o
n
mo
d
e
l
w
i
t
h
V
G
G
-
1
9
n
e
t
f
o
r
r
e
t
i
n
a
l
d
i
se
a
se
s
u
si
n
g
o
p
t
i
c
a
l
c
o
h
e
r
e
n
c
e
t
o
mo
g
r
a
p
h
y
i
mag
e
s,”
S
o
f
t
C
o
m
p
u
t
i
n
g
,
v
o
l
.
2
7
,
n
o
.
1
7
,
p
p
.
1
2
5
5
9
–
1
2
5
7
0
,
S
e
p
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s0
0
5
0
0
-
0
2
3
-
0
8
9
2
8
-
w.
[
2
6
]
K
.
H
e
,
X
.
Z
h
a
n
g
,
S
.
R
e
n
,
a
n
d
J.
S
u
n
,
“
D
e
e
p
r
e
si
d
u
a
l
l
e
a
r
n
i
n
g
f
o
r
i
mag
e
r
e
c
o
g
n
i
t
i
o
n
,
”
i
n
2
0
1
6
I
EE
E
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
(
C
V
PR)
,
I
EEE,
J
u
n
.
2
0
1
6
,
p
p
.
7
7
0
–
7
7
8
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
6
.
9
0
.
[
2
7
]
J.
G
u
o
,
H
.
W
a
n
g
,
X
.
X
u
e
,
M
.
L
i
,
a
n
d
Z
.
M
a
,
“
R
e
a
l
‐
t
i
me
c
l
a
ss
i
f
i
c
a
t
i
o
n
o
n
o
r
a
l
u
l
c
e
r
i
mag
e
s
w
i
t
h
r
e
si
d
u
a
l
n
e
t
w
o
r
k
a
n
d
i
mag
e
e
n
h
a
n
c
e
me
n
t
,
”
I
ET
I
m
a
g
e
Pro
c
e
ss
i
n
g
,
v
o
l
.
1
6
,
n
o
.
3
,
p
p
.
6
4
1
–
6
4
6
,
F
e
b
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
4
9
/
i
p
r
2
.
1
2
1
4
4
.
[
2
8
]
A
.
G
.
H
o
w
a
r
d
et
a
l
.
,
“
M
o
b
i
l
e
N
e
t
s:
Ef
f
i
c
i
e
n
t
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
f
o
r
mo
b
i
l
e
v
i
si
o
n
a
p
p
l
i
c
a
t
i
o
n
s,
”
Pre
p
r
.
a
rXi
v
.
1
7
0
4
.
0
4
8
6
1
,
A
p
r
.
2
0
1
7
.
[
2
9
]
E.
El
f
a
t
i
mi
,
R
.
Er
y
i
g
i
t
,
a
n
d
L
.
El
f
a
t
i
mi
,
“
B
e
a
n
s
l
e
a
f
d
i
se
a
se
s
c
l
a
ssi
f
i
c
a
t
i
o
n
u
s
i
n
g
M
o
b
i
l
e
N
e
t
mo
d
e
l
s,”
I
EEE
A
c
c
e
ss
,
v
o
l
.
1
0
,
p
p
.
9
4
7
1
–
9
4
8
2
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
2
.
3
1
4
2
8
1
7
.
[
3
0
]
M
.
S
a
n
d
l
e
r
,
A
.
H
o
w
a
r
d
,
M
.
Z
h
u
,
A
.
Zh
mo
g
i
n
o
v
,
a
n
d
L
.
-
C
.
C
h
e
n
,
“
M
o
b
i
l
e
N
e
t
V
2
:
I
n
v
e
r
t
e
d
r
e
si
d
u
a
l
s
a
n
d
l
i
n
e
a
r
b
o
t
t
l
e
n
e
c
k
s,”
P
re
p
r.
a
rXi
v
.
1
8
0
1
.
0
4
3
8
1
,
M
a
r
.
2
0
1
9
.
[
3
1
]
M
.
T
a
n
a
n
d
Q
.
V
.
L
e
,
“
Ef
f
i
c
i
e
n
t
N
e
t
:
R
e
t
h
i
n
k
i
n
g
mo
d
e
l
sca
l
i
n
g
f
o
r
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,”
Pre
p
r.
a
rX
i
v
.
1
9
0
5
.
1
1
9
4
6
,
S
e
p
.
2
0
2
0
.
[
3
2
]
K
.
S
u
n
,
X
.
W
a
n
g
,
X
.
M
i
a
o
,
a
n
d
Q
.
Z
h
a
o
,
“
A
r
e
v
i
e
w
o
f
A
I
e
d
g
e
d
e
v
i
c
e
s
a
n
d
l
i
g
h
t
w
e
i
g
h
t
C
N
N
a
n
d
L
L
M
d
e
p
l
o
y
m
e
n
t
,
”
N
e
u
ro
c
o
m
p
u
t
i
n
g
,
v
o
l
.
6
1
4
,
Ja
n
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m.
2
0
2
4
.
1
2
8
7
9
1
.
[
3
3
]
G
.
H
u
a
n
g
,
Z
.
L
i
u
,
L
.
V
a
n
D
e
r
M
a
a
t
e
n
,
a
n
d
K
.
Q
.
W
e
i
n
b
e
r
g
e
r
,
“
D
e
n
se
l
y
c
o
n
n
e
c
t
e
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
s,
”
i
n
2
0
1
7
I
EE
E
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
V
i
si
o
n
a
n
d
Pa
t
t
e
r
n
R
e
c
o
g
n
i
t
i
o
n
(
C
VPR)
,
I
EEE,
J
u
l
.
2
0
1
7
,
p
p
.
2
2
6
1
–
2
2
6
9
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
7
.
2
4
3
.
[
3
4
]
T
.
Z
h
o
u
,
X
.
Y
e
,
H
.
L
u
,
X
.
Z
h
e
n
g
,
S
.
Q
i
u
,
a
n
d
Y
.
L
i
u
,
“
D
e
n
se
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
a
n
d
i
t
s
a
p
p
l
i
c
a
t
i
o
n
i
n
me
d
i
c
a
l
i
m
a
g
e
a
n
a
l
y
s
i
s,”
Bi
o
Me
d
Re
se
a
r
c
h
I
n
t
e
r
n
a
t
i
o
n
a
l
,
v
o
l
.
2
0
2
2
,
n
o
.
1
,
Ja
n
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
2
/
2
3
8
4
8
3
0
.
[
3
5
]
P
.
K
.
M
a
l
l
et
a
l
.
,
“
A
c
o
mp
r
e
h
e
n
si
v
e
r
e
v
i
e
w
o
f
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
me
d
i
c
a
l
i
mag
e
p
r
o
c
e
ssi
n
g
:
R
e
c
e
n
t
d
e
v
e
l
o
p
me
n
t
s
a
n
d
f
u
t
u
r
e
o
p
p
o
r
t
u
n
i
t
i
e
s,”
H
e
a
l
t
h
c
a
re
A
n
a
l
y
t
i
c
s
,
v
o
l
.
4
,
D
e
c
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
h
e
a
l
t
h
.
2
0
2
3
.
1
0
0
2
1
6
.
[
3
6
]
X
.
W
a
n
g
a
n
d
W
.
Ji
a
,
“
O
p
t
i
m
i
z
i
n
g
e
d
g
e
A
I
:
A
c
o
mp
r
e
h
e
n
si
v
e
su
r
v
e
y
o
n
d
a
t
a
,
mo
d
e
l
,
a
n
d
sy
st
e
m
st
r
a
t
e
g
i
e
s,”
Pr
e
p
r
.
a
rXi
v
.
2
5
0
1
.
0
3
2
6
5
,
J
a
n
.
2
0
2
5
.
[
3
7
]
K
.
M
.
T
i
n
g
,
“
C
o
n
f
u
si
o
n
mat
r
i
x
,
”
i
n
E
n
c
y
c
l
o
p
e
d
i
a
o
f
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
,
B
o
s
t
o
n
,
M
A
:
S
p
r
i
n
g
e
r
U
S
,
2
0
1
1
,
p
p
.
2
0
9
–
2
0
9
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
0
-
3
8
7
-
3
0
1
6
4
-
8
_
1
5
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
C
o
mp
a
r
a
tive
p
erfo
r
ma
n
ce
a
n
a
lysi
s
o
f c
o
n
vo
lu
tio
n
a
l n
eu
r
a
l
…
(
I
r
fa
n
A
s
fy
F
a
kh
r
y
A
n
to
)
1599
[
3
8
]
K
.
W
a
n
,
B
.
F
e
n
g
,
L
.
X
i
e
,
a
n
d
Y
.
D
i
n
g
,
“
R
e
c
o
n
c
i
l
i
n
g
f
e
a
t
u
r
e
-
r
e
u
se
a
n
d
o
v
e
r
f
i
t
t
i
n
g
i
n
D
e
n
se
N
e
t
w
i
t
h
s
p
e
c
i
a
l
i
z
e
d
d
r
o
p
o
u
t
,
”
Pr
e
p
r
.
a
rXi
v
.
1
8
1
0
.
0
0
0
9
1
,
S
e
p
.
2
0
1
8
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Ir
fa
n
As
fy
Fa
k
h
r
y
An
to
w
a
s
re
c
e
iv
e
d
h
is
Ba
c
h
e
lo
r
o
f
C
o
m
p
u
ter
S
c
ien
c
e
f
ro
m
th
e
Un
iv
e
rsity
o
f
Bra
w
ij
a
y
a
in
2
0
0
9
,
a
n
d
h
is
M
a
ste
r’s
d
e
g
re
e
in
S
c
h
o
o
l
o
f
El
e
c
tri
c
a
l
En
g
in
e
e
rin
g
a
n
d
I
n
f
o
rm
a
ti
c
s
f
r
o
m
th
e
Ba
n
d
u
n
g
In
st
it
u
te
o
f
T
e
c
h
n
o
lo
g
y
(S
T
EI
-
I
T
B),
Ba
n
d
u
n
g
,
In
d
o
n
e
sia
,
in
2
0
1
9
.
He
is
c
u
rre
n
tl
y
a
Re
se
a
rc
h
e
r
a
t
th
e
Re
se
a
rc
h
Ce
n
ter
f
o
r
Da
ta
a
n
d
In
f
o
rm
a
ti
o
n
S
c
ien
c
e
,
Na
ti
o
n
a
l
Re
se
a
rc
h
a
n
d
In
n
o
v
a
ti
o
n
,
In
d
o
n
e
sia
.
His c
u
rre
n
t
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
d
a
ta
sc
ien
c
e
,
a
rti
f
icia
l
in
telli
g
e
n
c
e
,
in
f
o
rm
a
ti
o
n
re
tri
e
v
a
l,
a
n
d
in
tern
e
t
o
f
th
in
g
s.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
irf
a
0
1
3
@b
ri
n
.
g
o
.
id
.
J
o
n
y
W
i
n
a
r
y
o
W
ib
o
w
o
w
a
s
c
o
m
p
lete
d
h
is
Ba
c
h
e
lo
r
o
f
S
c
ien
c
e
in
Co
m
p
u
ter
S
c
ien
c
e
f
ro
m
P
a
d
jaja
ra
n
Un
iv
e
rsity
o
f
Ba
n
d
u
n
g
,
I
n
d
o
n
e
sia
,
i
n
2
0
0
7
.
His
M
a
ste
r
o
f
In
f
o
rm
a
ti
c
s
En
g
in
e
e
rin
g
d
e
g
re
e
f
r
o
m
In
stit
u
t
T
e
k
n
o
lo
g
i
Ba
n
d
u
n
g
(I
T
B),
In
d
o
n
e
sia
,
in
2
0
1
0
.
He
is
a
Re
se
a
r
c
h
e
r
a
t
th
e
Re
se
a
r
c
h
Ce
n
ter
f
o
r
S
m
a
rt
M
e
c
h
a
tro
n
ics
,
Na
ti
o
n
a
l
Re
se
a
rc
h
a
n
d
In
n
o
v
a
ti
o
n
A
g
e
n
c
y
o
f
th
e
Re
p
u
b
l
ic
o
f
In
d
o
n
e
sia
.
He
is
a
n
a
c
ti
v
e
m
e
m
b
e
r
o
f
th
e
in
telli
g
e
n
t
in
stru
m
e
n
tatio
n
re
se
a
rc
h
g
ro
u
p
,
f
o
c
u
sin
g
h
is
re
se
a
rc
h
in
tere
sts
o
n
tec
h
n
o
lo
g
y
in
stru
m
e
n
tatio
n
a
n
d
c
o
n
tro
l
sy
ste
m
s.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
jo
n
y
0
0
1
@b
ri
n
.
g
o
.
id
.
Ar
is
M
u
n
a
n
d
a
r
e
a
rn
e
d
h
is
b
a
c
h
e
lo
r’s
d
e
g
re
e
in
2
0
0
9
f
ro
m
th
e
F
a
c
u
lt
y
o
f
M
a
th
e
m
a
ti
c
s
a
n
d
Na
tu
ra
l
S
c
ien
c
e
s,
Un
iv
e
rsitas
P
a
d
jad
jara
n
,
m
a
jo
rin
g
in
Co
m
p
u
ter S
c
ien
c
e
.
He
is
c
u
rre
n
tl
y
a
Ju
n
io
r
Re
se
a
rc
h
e
r
a
t
t
h
e
S
m
a
rt
M
e
c
h
a
tro
n
ics
Re
se
a
rc
h
Ce
n
ter,
S
m
a
rt
In
s
tru
m
e
n
tatio
n
Re
se
a
rc
h
G
ro
u
p
,
Na
ti
o
n
a
l
Re
se
a
rc
h
a
n
d
I
n
n
o
v
a
ti
o
n
A
g
e
n
c
y
(BRIN).
His
re
se
a
rc
h
in
tere
sts
in
c
l
u
d
e
in
str
u
m
e
n
tatio
n
,
i
n
tern
e
t
o
f
th
i
n
g
s
(Io
T
),
c
o
m
p
u
ter
v
isio
n
,
a
n
d
m
a
c
h
in
e
lea
rn
in
g
.
He
c
a
n
b
e
c
o
n
t
a
c
ted
a
t
e
m
a
il
:
a
ris0
1
3
@
b
rin
.
g
o
.
i
d
.
Ta
u
fi
k
Ibn
u
S
a
li
m
re
c
ie
v
e
d
h
is
Ba
c
h
e
lo
r’s
d
e
g
re
e
in
El
e
c
tro
n
ics
a
n
d
In
stru
m
e
n
tatio
n
f
ro
m
G
a
d
jah
M
a
d
a
Un
iv
e
rsity
in
2
0
1
0
,
a
n
d
h
is
M
a
ste
r’s
d
e
g
re
e
in
In
stru
m
e
n
tatio
n
a
n
d
Co
n
tr
o
l
f
ro
m
th
e
Ba
n
d
u
n
g
In
stit
u
te
o
f
T
e
c
h
n
o
lo
g
y
,
Ba
n
d
u
n
g
,
In
d
o
n
e
sia
in
2
0
2
1
.
He
is
c
u
rre
n
tl
y
a
re
se
a
rc
h
e
r
a
t
th
e
Re
se
a
rc
h
Ce
n
ter
f
o
r
S
m
a
r
t
M
e
c
h
a
tro
n
ics
,
Na
ti
o
n
a
l
Re
se
a
rc
h
a
n
d
In
n
o
v
a
ti
o
n
A
g
e
n
c
y
,
In
d
o
n
e
sia
.
His
re
se
a
rc
h
in
tere
sts
a
re
re
late
d
to
in
stru
m
e
n
tatio
n
,
I
o
T
,
e
m
b
e
d
d
e
d
m
a
c
h
in
e
lea
rn
in
g
,
a
n
d
a
u
to
n
o
m
o
u
s
sy
ste
m
s.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
tau
f
0
2
1
@b
ri
n
.
g
o
.
id
.
Evaluation Warning : The document was created with Spire.PDF for Python.