I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
p
ute
r
E
ng
in
ee
ring
(
I
J
E
CE
)
Vo
l.
11
,
No
.
3
,
J
u
n
e
2021
,
p
p
.
2
4
5
7
~
2
4
6
6
I
SS
N:
2
0
8
8
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v
1
1
i
3
.
p
p
2
4
5
7
-
2
4
6
6
2457
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Ca
ss
ini
-
H
uy
g
ens
m
iss
io
n
i
m
a
g
es
cl
a
ss
ificatio
n f
ra
m
e
w
o
rk
by
d
eep
l
ea
rning
a
dv
a
nced a
ppro
a
ch
Ash
ra
f
Al
Da
bb
a
s
,
Z
o
lt
a
n
G
a
l
De
p
a
rtme
n
t
o
f
In
f
o
rm
a
ti
o
n
T
e
c
h
n
o
lo
g
y
S
y
ste
m
s a
n
d
Ne
tw
o
rk
s
,
F
a
c
u
lt
y
o
f
In
f
o
r
m
a
ti
c
s,
Un
iv
e
rsit
y
o
f
De
b
re
c
e
n
,
Hu
n
g
a
ry
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ju
l
1
9
,
2
0
20
R
ev
i
s
ed
D
ec
12
,
20
20
A
cc
ep
ted
D
ec
2
8
,
2
0
20
De
v
e
lo
p
in
g
a
d
e
e
p
lea
rn
in
g
(D
L)
m
o
d
e
l
f
o
r
i
m
a
g
e
c
las
si
f
ica
ti
o
n
c
o
m
m
o
n
l
y
d
e
m
a
n
d
s
a
c
ru
c
ial
a
r
c
h
it
e
c
tu
re
o
rg
a
n
iza
ti
o
n
.
P
lan
e
tary
e
x
p
e
d
it
io
n
s
p
ro
d
u
c
e
a
m
a
ss
iv
e
q
u
a
n
ti
ty
o
f
d
a
ta
a
n
d
i
m
a
g
e
s.
Ho
w
e
v
e
r,
m
a
n
u
a
ll
y
a
n
a
l
y
z
in
g
a
n
d
c
las
si
fy
in
g
f
li
g
h
t
m
issio
n
s
im
a
g
e
d
a
tab
a
se
s
w
it
h
h
u
n
d
re
d
s
o
f
th
o
u
sa
n
d
s
o
f
im
a
g
e
s
is
u
n
g
a
in
l
y
a
n
d
y
ield
we
a
k
a
c
c
u
ra
c
y
.
In
th
is
p
a
p
e
r,
w
e
sp
e
c
u
late
a
n
e
ss
e
n
ti
a
l
to
p
ic
re
late
d
t
o
t
h
e
c
l
a
ss
if
i
c
a
ti
o
n
o
f
re
m
o
tel
y
se
n
se
d
im
a
g
e
s,
in
w
h
ich
th
e
p
ro
c
e
ss
o
f
f
e
a
tu
re
c
o
d
in
g
a
n
d
e
x
trac
ti
o
n
a
re
d
e
c
isi
v
e
p
ro
c
e
d
u
re
s.
Div
e
rse
fe
a
tu
re
e
x
tr
a
c
ti
o
n
tec
h
n
iq
u
e
s
a
re
in
ten
d
e
d
t
o
sti
m
u
late
a
d
isc
rim
in
a
ti
v
e
i
m
a
g
e
c
las
si
f
ie
r.
F
e
a
tu
re
s
e
x
trac
ti
o
n
is
th
e
p
rim
a
r
y
e
n
g
a
g
e
m
e
n
t
in
ra
w
d
a
ta
p
ro
c
e
ss
i
n
g
w
it
h
th
e
p
u
r
p
o
se
o
f
d
a
ta
c
las
sif
ica
ti
o
n
;
w
h
e
n
it
c
o
m
e
s
a
c
ro
ss
th
e
tas
k
o
f
a
n
a
ly
sis
o
f
v
a
st
a
n
d
v
a
ried
d
a
ta,
th
e
se
k
in
d
s
o
f
tas
k
s
a
re
c
o
n
sid
e
re
d
a
s
ti
m
e
-
c
o
n
su
m
in
g
a
n
d
h
a
rd
to
b
e
trea
ted
w
it
h
.
M
o
st
o
f
th
e
se
c
las
si
f
iers
a
r
e
e
it
h
e
r,
in
p
rin
c
i
p
le,
q
u
it
e
i
n
tri
c
a
te
o
r
v
irt
u
a
ll
y
u
n
a
tt
a
in
a
b
le
t
o
c
a
lcu
late
f
o
r
m
a
ss
iv
e
d
a
tas
e
ts.
S
ti
m
u
late
d
b
y
th
is
p
e
rc
e
p
ti
o
n
,
w
e
p
u
t
f
o
r
w
a
r
d
a
stra
ig
h
t
f
o
rwa
rd
,
e
ff
icie
n
t
c
las
si
f
ier
b
a
se
d
o
n
f
e
a
tu
re
e
x
trac
ti
o
n
b
y
a
n
a
l
y
z
in
g
th
e
c
e
ll
o
f
ten
so
rs
v
i
a
la
y
e
re
d
M
a
p
Re
d
u
c
e
f
ra
m
e
w
o
rk
b
e
sid
e
m
e
ta
-
le
a
rn
in
g
L
S
T
M
f
o
ll
o
w
e
d
b
y
a
S
o
f
tM
a
x
c
las
si
f
ier.
Ex
p
e
rime
n
t
re
su
lt
s
sh
o
w
th
a
t
t
h
e
p
ro
v
id
e
d
m
o
d
e
l
a
tt
a
in
s
a
c
l
a
ss
if
i
c
a
ti
o
n
a
c
c
u
ra
c
y
o
f
9
6
.
7
%
,
w
h
ich
m
a
k
e
s
th
e
p
ro
v
id
e
d
m
o
d
e
l
q
u
it
e
v
a
li
d
fo
r
d
iv
e
rse
im
a
g
e
d
a
tab
a
se
s
w
it
h
v
a
r
y
in
g
siz
e
s.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
M
ac
h
i
n
e
lear
n
i
n
g
Me
ta
-
l
ea
r
n
in
g
R
e
m
o
te
s
e
n
s
in
g
d
ataset
s
S
atu
r
n
i
m
a
g
es
cla
s
s
i
f
ica
tio
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
:
Ash
r
a
f
A
l
Dab
b
as
Dep
ar
t
m
en
t o
f
I
n
f
o
r
m
atio
n
T
e
ch
n
o
lo
g
y
S
y
s
te
m
s
an
d
Net
w
o
r
k
s
Facu
lt
y
o
f
I
n
f
o
r
m
a
tics
,
U
n
i
v
er
s
it
y
o
f
Deb
r
ec
en
H
-
4
0
2
8
Deb
r
ec
en
,
Hu
n
g
ar
y
E
m
ail:
Ash
r
a
f
.
Dab
b
as@
i
n
f
.
u
n
id
eb
.
h
u
1.
I
NT
RO
D
UCT
I
O
N
Dee
p
lear
n
in
g
(
D
L
)
[
1
]
is
p
r
o
tr
u
d
ed
as
a
n
o
v
el
s
co
p
e
o
f
m
ac
h
in
e
lear
n
i
n
g
,
ap
p
lied
,
an
d
af
f
ec
ti
n
g
s
ev
er
al
d
o
m
a
in
s
i
n
o
u
r
d
ail
y
li
f
e
s
u
c
h
a
s
m
ed
ical
i
m
a
g
e
p
r
o
ce
s
s
in
g
[
2
]
,
p
r
ed
ictio
n
[
3
]
,
m
o
b
ile
tr
a
f
f
ic
class
i
f
icatio
n
[
4
]
,
co
m
p
u
ter
v
is
io
n
[
5
]
,
co
m
p
u
ter
n
e
t
wo
r
k
s
[
6
]
.
DL
'
s
co
n
ce
p
t
ca
n
b
e
d
escr
ib
ed
as
s
tr
aig
h
t
f
o
r
w
ar
d
a
s
t
h
e
p
r
o
ce
s
s
o
f
f
ea
t
u
r
e
lear
n
i
n
g
b
y
m
ac
h
in
e
s
a
s
t
h
e
y
o
r
d
in
ar
il
y
v
er
y
w
ell
i
n
t
h
e
f
ield
o
f
class
i
f
icatio
n
.
R
e
m
o
te
s
e
n
s
in
g
(
RS
)
is
a
s
cie
n
ti
f
ic
d
e
f
y
w
h
e
r
e
s
ce
n
es
ar
e
in
s
p
ec
ted
an
d
c
o
n
s
tr
u
ed
b
y
r
e
m
o
te
m
ea
n
s
[
7
]
.
T
h
is
t
er
m
co
m
p
r
is
es
t
h
e
co
n
v
e
n
tio
n
a
l
R
S
s
co
p
es
,
s
u
c
h
as
i
m
a
g
es
o
f
t
h
e
s
a
tellit
e.
R
e
m
o
tel
y
s
e
n
s
ed
d
ata
an
al
y
s
is
an
d
cla
s
s
i
f
icati
o
n
o
v
er
w
h
e
l
m
in
g
l
y
i
m
p
le
m
e
n
ted
b
y
s
u
p
er
v
i
s
ed
f
ilter
s
o
r
class
i
f
ier
s
.
T
h
i
s
n
ec
es
s
itates
a
s
u
f
f
icie
n
t
v
o
l
u
m
e
o
f
lab
eled
s
a
m
p
le
s
to
p
r
o
v
id
e
th
e
n
ee
d
ed
tr
ain
i
n
g
f
o
r
th
e
class
if
ier
.
T
h
e
s
p
ac
ec
r
af
t
o
f
C
as
s
i
n
i
-
Hu
y
g
en
s
is
co
n
s
id
er
ed
th
e
g
r
ea
test
a
n
d
th
e
m
o
s
t
i
n
tr
icate
i
n
ter
p
lan
etar
y
m
is
s
io
n
ev
e
r
co
n
s
tr
u
cted
,
an
d
it
w
as
ab
le
to
ac
q
u
ir
e
a
d
etailed
im
ag
e
w
it
h
i
n
d
if
f
er
e
n
t
p
lan
etar
y
c
o
n
d
itio
n
s
;
s
c
ien
c
e
in
v
e
s
ti
g
atio
n
s
i
n
t
h
is
r
esp
ec
t
h
av
e
p
ar
tic
u
lar
p
a
y
atte
n
tio
n
to
th
e
th
eo
r
etica
l
p
ar
t.
Ou
r
w
o
r
k
w
ill
f
o
cu
s
o
n
t
h
e
A
I
p
ar
t
to
s
u
p
p
o
r
t
h
i
g
h
-
r
is
e
d
e
m
an
d
o
n
co
m
p
u
ter
ized
s
cie
n
tif
ic
a
s
p
ec
ts
,
r
ed
u
ci
n
g
e
f
f
o
r
t,
t
i
m
e,
a
n
d
h
ar
d
to
b
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
3
,
J
u
n
e
2
0
2
1
:
2
4
5
7
-
2466
2458
ac
co
m
p
li
s
h
ed
b
y
t
h
e
h
u
m
a
n
,
w
ith
th
e
p
o
in
t
o
f
v
ie
w
o
f
a
n
al
y
s
i
s
an
d
cla
s
s
i
f
icat
io
n
a
v
er
y
m
a
s
s
i
v
e
i
m
a
g
e
d
atasets
t
h
at
h
a
v
e
b
ee
n
g
e
n
er
ated
b
y
th
e
C
a
s
s
i
n
i
-
H
u
y
g
e
n
s
m
is
s
io
n
.
Me
ta
-
lear
n
i
n
g
o
f
f
er
s
an
id
ea
l
p
r
e
f
er
en
ce
in
w
h
ic
h
th
e
m
o
d
el
o
f
m
ac
h
in
e
lear
n
i
n
g
ea
r
n
s
k
n
o
w
led
g
e
o
v
er
n
u
m
er
o
u
s
lear
n
i
n
g
e
p
o
ch
s
,
f
r
eq
u
e
n
tl
y
en
co
m
p
as
s
a
n
allo
ca
tio
n
o
f
ass
o
ciate
d
tas
k
s
.
I
t
also
u
tili
ze
s
th
is
k
n
o
w
led
g
e
to
e
n
h
an
ce
its
f
o
r
th
co
m
in
g
lear
n
in
g
e
f
f
icien
c
y
.
W
e
p
r
o
v
id
e
an
ap
p
r
o
ac
h
th
at
p
in
p
o
in
ts
th
e
f
r
ailt
y
o
f
n
eu
r
al
n
et
w
o
r
k
s
tr
ain
ed
v
ia
t
h
e
o
p
ti
m
izatio
n
o
f
a
g
r
ad
ien
t.
W
e
p
u
t
f
o
r
w
ar
d
an
o
p
ti
m
izer
m
eta
-
lear
n
er
th
at
is
e
m
a
n
ated
w
it
h
L
ST
M.
W
ith
in
th
e
tr
ain
i
n
g
p
r
o
ce
s
s
,
th
e
class
i
f
ier
'
s
p
er
f
o
r
m
an
ce
i
s
en
h
an
ce
d
b
y
t
h
e
s
u
p
p
o
r
t
p
r
o
v
id
ed
b
y
th
e
o
p
ti
m
i
ze
r
.
T
h
e
p
r
o
p
o
s
e
d
m
eta
-
lear
n
er
at
tain
s
to
g
eth
er
s
h
o
r
t
-
ter
m
lear
n
ed
d
ata
a
m
o
n
g
t
h
e
p
er
f
o
r
m
ed
m
i
s
s
io
n
,
a
n
d
lo
n
g
-
ter
m
ac
q
u
ir
ed
ex
p
er
ien
ce
-
r
ep
o
r
ted
k
n
o
w
led
g
e
o
cc
u
r
r
in
g
w
i
th
i
n
t
h
e
w
h
o
le
p
r
ev
io
u
s
ta
s
k
s
.
T
h
e
m
eta
-
lear
n
e
r
alg
o
r
ith
m
w
it
h
i
n
o
u
r
co
n
ce
p
tio
n
is
tr
ain
ed
to
i
m
p
e
l
t
h
e
clas
s
if
ier
,
s
o
it
te
n
d
s
to
m
ee
t
a
t
a
p
o
in
t
o
f
co
n
v
er
g
en
ce
to
w
ar
d
a
n
o
p
tim
a
l so
lu
tio
n
s
w
i
f
tl
y
f
o
r
ev
er
y
ta
s
k
,
t
h
i
s
is
t
h
e
ai
m
o
f
ch
o
o
s
in
g
th
i
s
alg
o
r
it
h
m
.
T
h
is
r
esear
ch
'
s
r
esid
u
al
is
ar
r
an
g
ed
i
n
th
e
f
o
llo
w
i
n
g
o
r
d
er
:
Sectio
n
2
in
cl
u
d
es
a
d
escr
ip
t
io
n
o
f
th
e
u
s
ed
d
ataset.
Sectio
n
3
co
n
tai
n
s
m
ater
ials
a
n
d
m
et
h
o
d
s
alo
n
g
w
it
h
t
h
e
ad
o
p
ted
alg
o
r
ith
m
p
o
r
tio
n
s
d
escr
ip
tio
n
.
Sectio
n
4
o
f
t
h
e
r
e
s
ea
r
ch
co
m
p
r
is
es
t
h
e
r
es
u
lt
s
a
n
d
an
a
l
y
s
is
d
is
cu
s
s
io
n
.
I
n
t
h
e
la
s
t
Sec
tio
n
5
co
n
c
lu
s
io
n
s
a
n
d
f
u
tu
r
e
w
o
r
k
ar
e
p
r
o
v
id
ed
.
2.
DATAS
E
T
UT
I
L
I
Z
E
D
I
N
T
H
E
ANAL
YSI
S
C
ass
in
i
-
H
u
y
g
e
n
s
s
p
ac
ec
r
a
f
t
w
as
s
o
f
ar
th
e
m
o
s
t
asp
ir
an
t
ex
p
ed
itio
n
u
p
till
n
o
w
s
e
n
t
to
o
u
ter
s
p
ac
e,
s
tu
f
f
ed
w
i
th
a
g
r
o
u
p
o
f
r
o
b
u
s
t
d
ev
ices
an
d
ca
m
er
a
s
.
C
as
s
in
i
-
H
u
y
g
e
n
s
w
er
e
elig
ib
le
f
o
r
g
ath
er
i
n
g
d
elica
te
m
ea
s
u
r
e
m
e
n
t
s
ite
m
ized
i
m
a
g
e
s
w
it
h
i
n
s
e
v
er
al
at
m
o
s
p
h
er
ic
cir
cu
m
s
ta
n
ce
s
.
T
h
e
s
p
ac
ec
r
af
t
h
ad
t
w
o
p
ar
ts
:
t
h
e
Hu
y
g
e
n
s
p
r
o
b
e
an
d
C
ass
in
i o
r
b
iter
,
C
ass
i
n
i
-
Hu
y
g
en
s
ar
r
iv
e
at
Satu
r
n
in
2
0
0
4
,
tr
an
s
m
itti
n
g
p
r
ec
io
u
s
d
ata
b
ac
k
to
u
s
,
w
h
ic
h
i
m
p
r
o
v
ed
o
u
r
co
m
p
r
e
h
e
n
s
io
n
o
f
Sat
u
r
n
a
n
d
its
m
o
o
n
s
.
H
u
y
g
e
n
s
s
tep
i
n
s
id
e
T
itan
'
s
at
m
o
s
p
h
er
e,
Satu
r
n
'
s
m
ass
iv
e
m
o
o
n
,
f
all
d
o
w
n
w
ar
d
th
r
o
u
g
h
a
p
ar
ac
h
u
te
to
th
e
f
u
r
t
h
est
p
o
in
t
s
o
f
ar
,
lan
d
o
n
its
s
u
r
f
ac
e,
tak
e
s
a
m
p
le
s
a
n
d
an
al
y
ze
th
e
m
,
a
n
d
s
e
n
d
t
h
e
r
es
u
lt
s
to
C
ass
i
n
i,
w
h
ic
h
w
i
ll
s
e
n
d
t
h
e
m
later
to
t
h
e
E
ar
t
h
.
C
ass
in
i
i
n
s
tr
u
m
en
t
s
o
f
r
e
m
o
te
s
en
s
i
n
g
co
llected
d
ata
r
em
o
tel
y
f
r
o
m
e
n
o
r
m
o
u
s
d
is
tan
ce
s
.
Af
ter
t
w
e
n
t
y
y
ea
r
s
s
p
en
t
i
n
o
u
ter
s
p
ac
e
an
d
t
h
ir
t
ee
n
y
ea
r
s
to
u
r
in
g
Sat
u
r
n
,
th
e
o
r
b
iter
"
C
ass
in
i"
d
r
ain
ed
o
u
t
o
f
en
er
g
y
.
C
a
s
s
i
n
i
w
a
s
i
m
m
er
s
ed
i
n
Sat
u
r
n
's
at
m
o
s
p
h
er
e
o
n
1
5
o
f
Sep
te
m
b
e
r
2
0
1
7
,
an
d
th
is
is
h
o
w
th
e
m
is
s
io
n
e
n
d
ed
.
T
h
e
ac
q
u
ir
ed
i
m
ag
e
s
d
ata
h
as
b
ee
n
g
e
n
er
ated
b
y
t
h
e
i
m
a
g
i
n
g
s
cien
ce
s
u
b
s
y
s
te
m
(
I
S
S),
w
h
ich
h
as
t
h
e
b
est
r
eso
lu
tio
n
f
o
r
th
e
ac
q
u
ir
ed
i
m
ag
e
s
.
T
h
e
I
SS
is
co
m
p
o
s
ed
o
f
2
d
etac
h
ed
ca
m
er
as
w
id
e
-
an
g
le
ca
m
er
a
an
d
a
n
ar
r
o
w
-
an
g
le
ca
m
er
a.
I
SS
i
m
ag
e
v
o
lu
m
e
s
d
ataset
is
co
m
p
o
s
ed
o
f
a
m
a
s
s
i
v
e
n
u
m
b
er
o
f
i
m
a
g
es
an
d
th
ei
r
r
elate
d
lab
els th
at
h
o
ld
th
e
i
m
ag
es
'
m
etad
ata.
T
h
e
d
ata
s
et
is
p
u
b
licl
y
av
a
ilab
le
at
th
e
r
e
f
er
en
ce
[
8
]
.
3.
M
AT
E
RIAL
S AN
D
M
E
T
H
O
DS
T
h
e
m
eta
-
lear
n
in
g
ap
p
r
o
ac
h
p
r
o
p
o
s
ed
in
[
9
,
1
0
]
w
o
r
k
s
v
i
a
ex
ec
u
ti
n
g
a
f
e
w
-
s
h
o
t
d
atase
t
s
a
m
p
li
n
g
f
r
o
m
a
n
i
n
te
n
d
ed
tas
k
a
n
d
ac
c
li
m
ati
n
g
t
h
e
ap
p
r
o
ac
h
i
n
n
er
p
o
r
tr
ay
al
s
a
m
o
n
g
g
r
ad
ien
t
d
esc
en
t
ce
r
tai
n
s
tep
s
.
I
t
m
u
s
t
b
e
s
tated
t
h
at
th
is
r
e
s
ea
r
ch
i
s
a
s
u
p
p
le
m
en
t
o
f
p
ast
in
v
esti
g
at
io
n
e
x
er
tio
n
ass
o
ciate
d
w
i
th
th
e
C
a
s
s
i
n
i
-
Hu
y
g
e
n
s
p
r
o
j
ec
t
d
ataset
[
11
-
1
3
]
.
R
ec
u
r
r
en
t
m
o
d
el
m
eta
-
le
ar
n
in
g
is
a
p
ar
t
o
f
t
h
e
ad
o
p
te
d
m
o
d
el
tailo
r
ed
to
lo
n
g
s
h
o
r
t
ter
m
m
e
m
o
r
y
(
L
S
T
M
(
.
W
ith
th
is
s
u
b
-
f
r
am
e
w
o
r
k
,
th
e
alg
o
r
ith
m
o
f
m
eta
-
lea
r
n
in
g
s
h
all
t
r
a
in
th
e
L
ST
M
m
o
d
el
,
w
h
ich
in
its
p
a
r
t
m
u
s
t
p
er
f
o
r
m
th
e
n
ee
d
e
d
d
atas
et
p
r
o
c
ess
in
g
co
n
s
e
cu
tiv
e
l
y
,
an
d
s
u
b
s
e
q
u
en
tly
p
r
o
ce
s
s
th
e
in
co
m
in
g
d
ata
as
n
e
w
in
p
u
ts
to
th
e
S
o
f
tMa
x
class
if
i
er
,
w
h
ich
r
eq
u
i
r
es
p
as
s
in
g
th
e
ex
tr
a
ct
ed
f
ea
tu
r
es
w
ith
th
e
(
im
ag
e,
la
b
e
l)
p
a
ir
s
s
et
f
o
r
e
ac
h
b
atch
o
f
th
e
d
at
aset
.
Fig
u
r
e
1
illu
s
tr
at
es
a
m
e
ta
-
lear
n
i
n
g
la
y
o
u
t
wh
ile
t
h
e
ad
o
p
te
d
f
r
am
ew
o
r
k
i
s
s
h
o
w
n
in
Fig
u
r
e
2
,
w
h
ich
in
clu
d
es
th
r
e
e
m
o
d
u
les,
a
f
ea
tu
r
es
ex
t
r
a
ct
o
r
(
G
)
,
a
m
eta
-
le
ar
n
er
L
S
T
M
(
M
)
,
an
d
a
m
ap
-
r
e
d
u
ce
r
d
is
cr
im
in
ato
r
(
D
)
,
al
l o
f
th
em
ar
e
a
cq
u
i
r
in
g
th
e
k
n
o
w
led
g
e
alt
o
g
eth
e
r
.
Fr
o
m
o
n
e
s
id
e,
w
e
an
tici
p
ate
th
e
f
ea
tu
r
es
ex
tr
ac
t
o
r
(
G
)
to
ex
t
r
ac
t
a
ll
th
e
r
elat
e
d
d
at
a
d
u
r
in
g
its
task
b
y
ca
p
tu
r
in
g
h
ig
h
v
a
lu
e
d
f
ea
tu
r
es
,
w
h
ich
w
ill
clu
e
th
e
m
eta
-
l
ea
r
n
er
L
S
T
M
(
M)
t
o
c
a
r
r
y
o
u
t
.
On
th
e
o
th
e
r
s
id
e,
it
is
l
o
g
ica
l
th
at
th
e
f
e
atu
r
es
ex
t
r
ac
to
r
(
G
)
w
ill
b
e
r
ein
f
o
r
ce
d
v
ia
th
e
m
ap
-
r
e
d
u
c
er
d
is
c
r
im
in
ato
r
(
D
)
o
v
e
r
co
n
s
e
q
u
en
t
t
ask
s
o
n
a
b
ig
d
a
taset
(
B
d
)
.
A
f
ter
d
ea
lin
g
w
ith
a
m
ass
iv
e
n
u
m
b
er
o
f
d
at
a
a
n
d
its
f
ea
tu
r
es,
th
e
f
ea
tu
r
es
ex
t
r
a
ct
o
r
(
G
)
p
r
o
g
r
ess
iv
ely
lear
n
s
an
d
a
cq
u
i
r
es
th
e
k
n
o
w
led
g
e
to
ex
tr
ac
t
f
ea
tu
r
es
f
r
o
m
r
aw
i
m
a
g
e
d
a
t
a
;
t
h
i
s
m
a
p
p
i
n
g
p
r
o
c
e
s
s
w
i
l
l
c
o
n
s
i
d
e
r
a
b
l
y
f
a
c
i
l
i
t
a
t
e
t
h
e
m
e
t
a
-
l
e
a
r
n
i
n
g
i
m
p
l
em
e
n
t
a
t
i
o
n
.
B
y
c
o
n
s
i
d
e
r
i
n
g
t
h
e
d
o
m
a
i
n
b
i
g
d
a
t
a
(
B
d
)
o
f
p
o
s
s
i
b
l
e
e
x
p
e
r
i
e
n
c
e
s
E
T
b
y
a
r
i
t
h
m
e
t
i
c
a
p
p
r
o
a
c
h
,
w
e
d
e
v
i
s
e
t
h
e
s
u
b
s
e
q
u
e
n
t
o
p
t
i
m
i
z
a
t
i
o
n
i
s
s
u
e
i
n
(
1
)
m
eth
o
d
ic
ally
:
ⅈ
,
,
~
(
)
,
(
,
)
~
[
(
(
,
)
,
(
,
)
(
,
)
]
(
1
)
w
h
er
e
,
,
r
ep
r
esen
t
th
e
co
n
g
r
u
e
n
t
m
o
d
u
le
s
p
ar
a
m
eter
s
.
T
h
e
task
o
f
m
eta
-
lear
n
i
n
g
(
T
)
h
as
a
d
is
tr
ib
u
tio
n
p
(
T
)
,
w
h
ile
(
x
,
y
)
s
y
m
b
o
lizes a
s
a
m
p
led
lab
eled
in
s
ta
n
ce
f
r
o
m
t
h
e
B
d
.
T
h
e
p
u
r
p
o
s
e
is
to
d
ec
r
ea
s
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
C
a
s
s
in
i
–
Hu
yg
en
s
mis
s
io
n
ima
g
es c
la
s
s
ifica
tio
n
fr
a
mewo
r
k
b
y
d
ee
p
l
ea
r
n
in
g
...
(
A
s
h
r
a
f A
l
Da
b
b
a
s
)
2459
th
e
j
o
in
t
ex
p
ec
tatio
n
i
n
d
icate
d
b
y
J
.
T
h
is
j
o
in
t
w
ill
b
e
ap
p
lied
o
n
b
o
th
lo
s
s
es:
th
e
lo
s
s
L
T
(
θ
M
,
θ
G
)
,
w
h
ich
i
s
r
elate
d
to
th
e
task
o
f
m
eta
-
lear
n
in
g
,
an
d
o
n
th
e
lo
s
s
L
(x,
y)
(θ
D
,
θ
G
)
w
h
ich
i
s
r
elate
d
to
m
ap
-
r
ed
u
ce
r
d
is
cr
i
m
i
n
ato
r
(
D)
.
A
m
eta
-
le
ar
n
er
ca
n
also
co
m
m
e
n
ce
a
lear
n
er
to
b
r
in
g
u
p
to
d
ate
l
ea
r
n
er
v
ia
g
r
ad
ien
t
d
esc
en
t,
w
h
ic
h
h
as
a
s
tead
y
lear
n
i
n
g
r
ate
(
R
)
o
v
er
ev
er
y
ta
s
k
,
g
r
ad
ien
t
d
esce
n
t
b
eg
in
s
f
r
o
m
i
n
cip
ien
t
p
ar
am
eter
s
θ
0,
an
d
af
ter
w
ar
d
,
i
t c
ar
r
ies o
u
t th
e
s
u
b
s
eq
u
e
n
t
u
p
d
ate
in
(
2
)
:
=
−
1
–
−
1
(
2)
T
h
e
p
r
ev
io
u
s
eq
u
atio
n
is
u
t
te
r
l
y
an
alo
g
ica
l
to
th
e
ce
ll
s
tate
u
p
d
ate
o
f
th
e
L
ST
M.
W
ith
ea
ch
m
eta
-
lear
n
in
g
p
r
o
ce
s
s
ep
o
ch
s
,
t
h
e
a
s
s
o
ciate
d
tas
k
i
n
cl
u
d
es
a
tr
ai
n
in
g
s
et
tr
ain
(
T
)
an
d
an
o
t
h
er
o
n
e
f
o
r
tes
tin
g
(
T
)
.
Fo
r
th
e
m
ap
-
r
ed
u
ce
r
d
is
cr
i
m
i
n
ato
r
(
D)
it
s
h
o
u
ld
b
e
ab
le
to
eli
m
i
n
ate
t
h
e
id
en
t
ical
i
m
ag
e
s
o
r
b
atch
es
a
m
o
n
g
th
e
B
d
an
d
p
r
o
v
id
e
lab
eled
b
atch
es
a
n
d
i
m
a
g
es.
T
h
e
m
ap
-
r
ed
u
ce
r
d
is
cr
i
m
in
a
to
r
(
D)
,
an
d
th
e
m
eta
-
lear
n
er
L
ST
M
(
M)
h
av
e
a
j
o
in
t
lo
s
s
w
it
h
t
h
e
p
u
r
p
o
s
e
o
f
r
ed
u
ci
n
g
th
e
ex
p
ec
ted
lo
s
s
w
i
th
t
h
e
ta
s
k
o
f
m
a
p
-
r
ed
u
ce
r
d
is
cr
i
m
i
n
atio
n
:
(
,
)
(
,
)
=
(
◦
(
)
,
)
(
3
)
w
h
er
e
L
s
ca
n
b
e
an
y
ap
p
r
o
p
r
iate
lo
s
s
f
u
n
ctio
n
f
o
r
m
ap
-
r
ed
u
ce
r
d
is
cr
i
m
in
at
io
n
,
t
h
e
f
ea
t
u
r
es
ex
tr
ac
to
r
(
G)
is
s
u
p
p
lied
w
it
h
th
e
r
eq
u
ir
ed
tr
ain
i
n
g
to
e
x
tr
ac
t
t
h
e
n
ee
d
ed
d
ata
f
r
o
m
t
h
e
i
m
ag
e
b
atc
h
es,
w
h
ile
m
eta
-
lear
n
er
o
b
tain
ex
p
er
ien
ce
a
n
d
lear
n
ed
to
ca
r
r
y
o
u
t
t
h
e
i
m
a
g
e
a
n
d
b
atch
lab
elin
g
.
T
h
e
m
ap
-
r
ed
u
ce
r
d
is
cr
i
m
i
n
ato
r
(
D)
,
w
h
ic
h
h
as
g
iv
e
n
t
h
e
p
ar
am
ete
r
θ
D
,
is
in
ten
d
ed
to
f
o
r
etell
im
ag
e
lab
els
cr
ea
ted
b
y
G,
an
d
it
is
p
er
f
o
r
m
ed
v
i
a
g
ated
n
eu
r
al
n
e
t
w
o
r
k
s
.
T
h
e
f
ea
tu
r
es
ex
tr
ac
to
r
(
G)
h
as
g
iv
en
th
e
p
ar
a
m
eter
θ
G
an
d
is
ca
r
r
ied
o
u
t
v
ia
g
ated
n
eu
r
al
n
et
w
o
r
k
s
.
T
h
e
m
eta
-
le
ar
n
er
L
ST
M
(
M)
,
h
as
g
iv
e
n
t
h
e
p
ar
a
m
eter
θ
M
,
a
n
d
its
m
is
s
io
n
is
to
g
ai
n
t
h
e
k
n
o
w
led
g
e
a
n
d
p
ass
f
r
o
m
a
ta
s
k
(
T
)
to
an
o
th
er
a
m
o
n
g
th
e
t
r
ain
in
g
p
r
o
ce
s
s
.
Ov
er
a
n
y
p
r
o
v
id
ed
tas
k
(
T
)
,
th
e
r
elate
d
co
n
ce
p
ts
i
n
cl
u
d
e
m
et
a
-
lear
n
er
(
M)
th
at
ad
j
u
s
ts
a
l
ea
r
n
er
A
T
f
o
r
a
g
i
v
en
tas
k
w
it
h
a
s
p
ec
i
f
ic
ai
m
,
w
h
ic
h
is
to
r
e
d
u
ce
th
e
a
n
ticip
a
ted
lo
s
s
w
ith
t
h
e
tas
k
o
f
m
eta
-
l
ea
r
n
in
g
,
an
d
p
r
o
v
id
ed
b
y
(
4
)
:
(
,
)
=
1
|
t
es
t
(
T
)
|
∑
Ls
(
A
T
◦
G
(
x
)
,
y
)
(
x
,
y
)
∈
t
es
t
(
T
)
(
4
)
I
m
p
lem
en
tin
g
So
f
tMa
x
o
n
a
h
u
g
e
d
ataset
p
r
o
d
u
ce
a
m
o
r
e
r
e
lia
b
l
e
L
S
T
M,
f
o
r
cl
as
s
if
icat
io
n
,
o
r
d
in
ar
ily
,
S
o
f
tMa
x
f
u
n
ctio
n
(
5
)
is
u
til
ize
d
b
y
:
=
ex
p
(
V
k
)
∑
ex
p
(
Vn
)
=
(
5)
w
h
er
e
P
k
in
d
icat
es
th
e
ch
an
c
e
o
r
th
e
p
r
o
b
a
b
ilit
y
th
a
t
th
e
v
e
cto
r
k
is
a
p
ar
t
o
f
a
g
r
o
u
p
ter
m
ed
as
clas
s
v
.
I
n
m
u
lticla
s
s
cla
s
s
i
f
icatio
n
,
t
h
er
e
is
a
n
ee
d
to
co
m
p
u
te
th
e
l
o
s
s
r
elate
d
to
ev
er
y
clas
s
lab
el
f
o
r
ea
ch
p
r
o
ce
s
s
o
b
s
er
v
atio
n
,
an
d
th
e
o
u
tco
m
e
ca
n
b
e
ag
g
r
e
g
ated
as a
s
u
m
b
y
cr
o
s
s
-
e
n
tr
o
p
y
,
w
h
ic
h
is
e
x
p
r
ess
ed
b
y
(
6
)
.
=
−
∑
=
1
P
k
(
6
)
w
h
er
e
Z
-
r
e
p
r
esen
ts
th
e
n
u
m
b
e
r
o
f
cl
ass
es
th
at
m
a
y
in
clu
d
e
(
Satu
r
n
,
R
i
n
g
s
,
T
itan
,
I
c
y
Satel
lites
,
S
m
a
ll
s
atellite
s
(
r
o
ck
s
)
,
Sk
y
)
,
th
e
lo
g
in
d
ica
tes
th
e
t
r
a
d
i
ti
o
n
al
l
o
g
,
A
t
th
e
s
a
m
e
ti
m
e
B
is
a
b
in
ar
y
r
ef
er
en
ce
(
1
o
r
0
)
th
at
s
h
o
w
s
w
h
eth
er
a
cl
ass
l
ab
el
k
is
th
e
p
r
o
p
e
r
c
lass
if
i
c
ati
o
n
f
o
r
a
g
iv
en
o
b
s
e
r
v
at
io
n
.
T
h
e
in
d
i
ca
t
o
r
P
r
e
p
r
esen
ts
th
e
p
r
e
d
i
cte
d
p
r
o
b
a
b
il
ity
o
f
a
g
iv
en
o
b
s
er
v
a
ti
o
n
t
h
at
b
el
o
n
g
s
t
o
a
c
lass
k
.
T
o
r
e
d
u
ce
c
r
o
s
s
-
en
tr
o
p
y
lo
s
s
,
th
e
p
r
o
p
o
s
e
d
n
etw
o
r
k
is
in
d
o
ct
r
in
a
te
d
to
p
r
esen
t
th
e
r
e
s
u
lt
v
ec
to
r
k
n
ea
r
its
r
el
ate
d
o
n
e
-
h
o
t
v
ec
t
o
r
.
I
t
is
cr
u
c
ial
t
o
p
ay
att
en
ti
o
n
th
a
t
th
e
r
ig
h
t
r
esu
lts
o
f
th
e
t
a
r
g
et
v
ec
to
r
s
w
ith
in
th
e
n
etw
o
r
k
a
r
e
s
t
e
ad
y
th
r
o
u
g
h
o
u
t
th
e
tr
a
in
in
g
p
r
o
c
ess
.
3
.
1
.
T
he
al
g
o
rit
h
m
Af
ter
r
ep
r
esen
t
in
g
t
h
e
f
r
a
m
e
w
o
r
k
,
a
m
o
n
g
d
ee
p
m
eta
-
lear
n
in
g
L
ST
M,
th
e
p
h
a
s
e
o
f
ill
u
s
tr
atin
g
o
u
r
h
ar
m
o
n
ized
al
g
o
r
ith
m
i
s
n
ec
es
s
ar
y
.
T
h
e
al
g
o
r
ith
m
o
f
s
to
ch
a
s
tic
g
r
ad
ie
n
t d
esce
n
t c
o
u
ld
b
e
u
tili
ze
d
to
o
p
ti
m
ize
th
e
p
r
ev
io
u
s
ai
m
s
,
b
u
t
w
it
h
o
u
r
th
e
m
at
ic
m
o
d
el,
w
e
g
e
n
er
a
ted
a
m
o
d
if
ied
v
er
s
io
n
o
f
th
e
s
to
ch
ast
ic
g
r
ad
ien
t
d
escen
t
m
eth
o
d
,
th
e
elab
o
r
ated
ap
p
r
o
ac
h
is
s
u
m
m
ar
ized
in
Fig
u
r
e
1
.
T
h
e
u
p
p
er
s
ec
tio
n
in
d
icate
s
th
e
s
et
o
f
m
eta
tr
ain
i
n
g
D
(
m
eta
-
trai
n)
w
h
er
e
ea
ch
n
u
m
b
er
ed
b
o
x
r
ep
r
esen
ts
a
d
if
f
er
e
n
t
b
atch
o
f
th
e
Bd
th
at
is
co
m
p
o
s
ed
o
f
th
e
tr
ain
i
n
g
s
et
d
en
o
ted
as
D
train
a
n
d
D
test
.
T
h
e
m
eta
-
te
s
t
s
et
w
h
ic
h
is
in
d
icate
d
in
t
h
e
illu
s
t
r
atio
n
w
it
h
D
(
m
eta
-
test
)
is
also
d
em
o
n
s
tr
ated
in
th
e
s
a
m
e
m
et
h
o
d
,
b
u
t
v
ia
a
v
ar
io
u
s
d
ataset
t
h
at
i
n
cl
u
d
e
s
b
atch
es
t
h
at
ar
e
n
o
t
av
ailab
le
i
n
a
n
y
o
f
th
e
o
t
h
er
b
atc
h
es
in
D
(
m
eta
-
trai
n).
F
u
r
th
er
m
o
r
e,
th
er
e
is
a
s
et
o
f
m
eta
-
v
a
lid
atio
n
w
h
ic
h
is
e
x
p
lo
ited
to
s
p
ec
if
y
ad
d
itio
n
al
lab
els
a
n
d
f
ea
t
u
r
es.
T
h
e
a
d
o
p
te
d
f
r
am
e
w
o
r
k
is
p
r
es
en
te
d
in
F
ig
u
r
e
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
3
,
J
u
n
e
2
0
2
1
:
2
4
5
7
-
2466
2460
Fig
u
r
e
1
.
I
llu
s
tr
atio
n
o
f
m
e
ta
-
l
ea
r
n
in
g
la
y
o
u
t
T
r
a
i
n
i
n
g
D
o
m
a
i
n
T
es
t
i
n
g
D
o
m
a
i
n
L
a
be
l
ed
T
a
r
get
F
ea
t
u
r
es
O
p
er
a
t
i
n
g
Cl
a
s
s
i
f
i
c
a
t
i
o
n
T
a
s
k
O
u
t
-
Co
m
e
s
O
b
t
a
i
n
L
e
a
r
n
er
L
o
s
s
M
et
a
-
L
ea
r
n
er
L
S
T
M
Cl
a
s
s
i
f
i
e
r
B
i
g
Da
t
as
e
t
M
a
p
P
r
oce
s
s
M
a
p
P
r
oce
s
s
M
a
p
P
r
oce
s
s
M
a
p
P
r
oce
s
s
R
e
d
u
c
e
P
r
o
c
e
s
s
R
e
d
u
c
e
P
r
o
c
e
s
s
F
ea
t
u
r
e M
a
ps
Co
n
c
a
t
en
a
t
i
o
n
M
e
ta
K
n
o
w
l
e
d
g
e
L
a
b
e
le
d
I
ma
g
e
D
a
t
a
a
n
d
B
a
t
c
h
e
s
S
o
f
t
M
a
x
Cl
a
ss
i
f
i
e
r
Cr
o
ss
-
E
n
t
r
o
p
y
O
n
e
-
H
o
t
L
a
b
el
F
ea
t
u
r
es
E
x
t
r
a
c
t
o
r
(
G
)
E
x
t
r
a
c
t
F
e
a
t
u
r
es
wi
t
h
D
ee
p
N
et
s
T
r
a
i
n
ed o
n
V
a
r
i
o
u
s
T
a
s
k
s
R
el
a
t
i
o
n
S
c
o
r
e
O
n
e
-
h
o
t
V
ec
t
o
r
C
la
s
s
if
ie
d
I
ma
g
e
s
a
n
d
B
a
t
c
h
e
s
A
n
a
l
y
s
i
s
o
f
B
l
o
c
k
o
f
T
en
so
r
s
v
i
a
l
a
y
er
e
d
M
a
p
-
R
edu
c
e
U
p
d
at
e
M
e
t
a
-
L
e
ar
ne
r
P
ar
am
e
t
e
r
M
a
p
-
R
edu
c
er
D
i
sc
r
i
m
i
n
a
t
o
r
(
D
)
T
r
a
i
n
i
n
g D
o
m
a
i
n
T
es
t
i
n
g
D
o
m
a
i
n
L
a
be
l
ed
S
o
u
r
c
e
F
ea
t
u
r
es
I
n
p
u
t
s
O
u
t
p
u
t
s
F
e
a
t
u
r
e
E
x
t
r
a
c
t
i
o
n
+
Le
a
r
n
i
n
g
Fig
u
r
e
2
.
Ov
er
v
i
ew
o
f
th
e
a
d
o
p
te
d
f
r
am
e
w
o
r
k
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
C
a
s
s
in
i
–
Hu
yg
en
s
mis
s
io
n
ima
g
es c
la
s
s
ifica
tio
n
fr
a
mewo
r
k
b
y
d
ee
p
l
ea
r
n
in
g
...
(
A
s
h
r
a
f A
l
Da
b
b
a
s
)
2461
W
e
ac
q
u
ir
ed
th
e
r
ev
elatio
n
al
g
o
r
ith
m
b
y
g
r
ad
ien
t
d
escen
t
t
h
at
r
ep
r
esen
ted
in
(
1
)
,
if
w
e
as
s
u
m
ed
th
at
th
er
e
is
a
lear
n
er
cla
s
s
i
f
icatio
n
p
r
o
ce
s
s
t
h
at
h
as
a
p
ar
a
m
ete
r
θ
w
h
ic
h
it
n
ee
d
s
to
b
e
tr
ai
n
ed
o
n
th
e
s
et
D
train
.
T
h
e
d
ec
is
iv
e
al
g
o
r
ith
m
w
h
ic
h
m
a
k
es
t
h
e
b
est
o
r
m
o
s
t
e
f
f
ec
tiv
e
u
s
e
o
f
(
a
s
it
u
atio
n
,
o
p
p
o
r
tu
n
it
y
,
o
r
r
eso
u
r
ce
)
an
d
also
e
x
p
lo
ited
to
tr
ain
th
e
n
e
u
r
al
clas
s
i
f
ier
is
ap
p
r
o
x
i
m
atel
y
d
iv
er
g
e
n
t
o
f
g
r
ad
ien
t
d
escen
t,
t
h
at
u
tili
ze
s
u
p
d
ates,
w
h
er
e
th
e
ex
p
r
ess
io
n
o
f
th
e
(
θ
t−
1
)
r
ep
r
esen
ts
th
e
lear
n
er
p
ar
am
eter
s
i
n
ac
co
r
d
an
ce
w
it
h
(
t
-
1
)
u
p
d
ates
,
an
d
(
R
)
w
h
ich
is
o
b
tai
n
ed
d
u
r
in
g
t
h
e
ti
m
e.
p
er
io
d
t,
L
t
d
e
n
o
tes
t
h
e
lo
s
s
,
w
h
ich
is
b
r
o
u
g
h
t
to
o
p
tim
izatio
n
v
ia
th
e
lear
n
er
t
h
r
o
u
g
h
n
th
n
u
m
b
er
o
f
u
p
d
ate,
∇
θt
−
1
L
t
s
p
ec
if
ie
s
lo
s
s
g
r
ad
ien
t
co
r
r
esp
o
n
d
in
g
to
t
h
e
as
s
o
ciate
d
p
ar
am
eter
s
θ
t
−
1
,
a
n
d
θ
t
d
e
m
o
n
s
tr
ates
t
h
e
lear
n
er
u
p
d
ated
p
ar
am
e
ter
s
,
t
h
e
p
i
v
o
tal
r
u
m
in
at
io
n
th
at
w
e
p
ar
am
eter
ized
at
th
i
s
s
p
o
t
th
i
s
u
p
d
ate
w
h
ich
h
as
co
m
m
o
n
f
ea
t
u
r
es
th
at
lo
o
k
o
r
s
ee
m
lik
e
L
ST
M
ce
ll
s
tate
.
T
h
e
ch
ar
ac
ter
is
tic
s
o
f
th
e
ti
m
e
d
if
f
er
en
ce
∆
t
b
et
w
ee
n
co
n
s
ec
u
ti
v
e
s
a
m
p
li
n
g
s
a
s
s
h
o
w
n
i
n
F
ig
u
r
e
3
.
T
h
e
m
ea
n
an
d
s
tan
d
ar
d
d
ev
iatio
n
o
f
t
h
e
∆t
is
1
,
0
5
3
s
an
d
1
2
,
2
8
2
s
,
r
es
p
ec
tiv
el
y
.
T
h
is
e
x
a
m
p
le
f
i
g
u
r
e
d
em
o
n
s
tr
a
te
s
h
o
w
th
e
ex
tr
ac
ted
i
m
ag
e
f
ea
t
u
r
es (
l
ea
r
n
ed
f
ea
tu
r
e
s
)
ca
n
b
e
u
tili
ze
d
to
tr
ain
th
e
class
if
ier
.
Fig
u
r
e
3
.
F
ea
tu
r
es
o
f
th
e
c
o
n
s
e
cu
tiv
e
s
am
p
lin
g
in
t
er
v
a
ls
at
C
a
s
s
in
i a
n
d
its
h
is
to
g
r
am
T
h
er
e
a
r
e
s
ev
er
al
p
a
r
am
eter
s
o
f
s
im
ilar
ity
w
h
ich
co
u
ld
b
e
ex
p
l
o
it
e
d
w
ith
in
o
u
r
ap
p
r
o
ac
h
,
an
d
we
co
n
s
i
d
e
r
th
e
p
r
o
c
ess
o
f
im
ag
e
cl
ass
if
ic
ati
o
n
b
y
an
aly
zin
g
s
ev
er
al
p
a
r
am
ete
r
s
as
a
p
r
e
p
r
o
c
ess
in
g
p
h
as
e
f
o
r
im
ag
e
class
if
ic
ati
o
n
an
d
an
al
y
s
e
s
o
f
a
b
lo
ck
o
f
te
n
s
o
r
s
v
i
a
lay
er
ed
Ma
p
R
e
d
u
c
e
(
L
MR)
f
o
r
im
ag
e
p
atch
es
class
if
i
ca
t
io
n
w
h
er
e
th
ese
in
v
e
s
tig
ati
o
n
p
r
o
ce
s
s
es p
ass
e
d
in
t
o
(
L
ST
M
)
b
l
o
ck
s
,
w
e
ar
e
u
s
in
g
L
ST
Ms
b
l
o
ck
u
n
it
as
it
ex
p
lo
its
th
e
m
em
o
r
y
d
ep
en
d
en
cies
c
o
m
p
o
n
en
t
t
o
r
eg
u
late
th
e
in
f
o
r
m
atio
n
f
lo
w
,
as
w
e
a
r
e
d
ea
l
in
g
w
ith
a
B
d
m
eta
-
lear
n
i
n
g
L
ST
M
is
a
p
e
r
f
ec
t
ch
o
i
ce
.
T
h
e
p
r
im
e
f
a
cto
r
to
DL
'
s
p
r
o
s
p
er
it
y
is
w
i
t
h
in
i
ts
p
o
ten
tiali
t
y
to
w
ar
d
s
ac
q
u
ir
i
n
g
k
n
o
w
led
g
e
f
r
o
m
h
ier
ar
ch
ical
f
ea
t
u
r
es
v
i
a
en
o
r
m
o
u
s
a
m
o
u
n
ts
o
f
r
a
w
d
ata
s
u
c
h
as
te
x
t
o
r
i
m
a
g
es
.
E
x
tr
ac
ted
f
ea
t
u
r
es
t
h
r
o
ugh
DL
m
eth
o
d
s
h
a
v
e
b
ee
n
co
n
f
ir
m
ed
to
b
e
o
u
ts
ta
n
d
in
g
,
an
d
b
etter
t
h
a
n
tr
ad
itio
n
al
ap
p
r
o
ac
h
es in
m
a
n
y
d
if
f
er
en
t a
s
p
ec
ts
s
u
ch
a
s
p
r
ec
is
en
e
s
s
a
n
d
r
ed
u
ce
d
r
eso
u
r
ce
ch
alle
n
g
i
n
g
[
1
4
]
.
3
.
1
.
1
.
P
ri
m
a
ry
a
lg
o
rit
h
m
o
f
m
et
a
-
lea
rning
L
ST
M
T
h
e
lear
n
in
g
al
g
o
r
ith
m
w
i
ll
in
clu
d
e
an
i
n
p
u
t,
w
h
ich
h
as
a
n
a
i
m
to
p
er
f
o
r
m
tr
ain
in
g
v
ia
a
tr
ain
i
n
g
s
et
r
ef
er
r
ed
to
as
D
train
=
[
(
X
t
,
Y
t
)]
.
A
ls
o
,
it
h
as
a
n
o
u
tp
u
t
w
i
th
a
p
ar
am
eter
θ,
w
h
ic
h
w
il
l
m
o
d
el
th
e
lear
n
er
(
A
T
)
,
th
e
ai
m
o
f
t
h
e
lear
n
in
g
al
g
o
r
it
h
m
to
g
ai
n
a
r
o
b
u
s
t p
er
f
o
r
m
an
ce
p
er
f
o
r
m
ed
o
n
th
e
te
s
t se
t D
t
est
=
(
X,
Y)
.
T
a
b
le
1
d
is
p
la
y
s
t
h
e
p
r
i
m
ar
y
al
g
o
r
it
h
m
o
f
t
h
e
m
e
ta
-
lear
n
i
n
g
L
ST
M.
3
.
1
.
2
.
T
ra
in o
f
m
et
a
-
lea
rning
L
ST
M
a
lg
o
rit
h
m
T
h
e
m
e
ta
-
lear
n
i
n
g
L
ST
M
alg
o
r
ith
m
w
ill
i
n
cl
u
d
e
an
i
n
p
u
t,
w
h
ic
h
h
a
s
a
n
ai
m
to
p
er
f
o
r
m
t
r
ain
in
g
a
n
d
test
i
n
g
v
ia
a
m
eta
-
v
al
id
atio
n
tr
ain
i
n
g
s
et
r
ef
er
r
ed
to
as
D
m
e
t
a
-
v
a
lid
atio
n
tr
ai
n
-
L
T
SM
tr
ain
=[
(
t
ra
in
,
t
e
st
)
]=
1
;
also
it
h
as
an
o
u
tp
u
t
w
it
h
a
p
ar
a
m
eter
φ,
w
h
ic
h
w
i
ll
m
o
d
el
th
e
m
eta
-
lear
n
i
n
g
L
ST
M
alg
o
r
ith
m
,
th
i
s
p
r
o
ce
s
s
ai
m
s
to
g
ai
n
a
r
o
b
u
s
t
p
er
f
o
r
m
an
ce
v
ia
t
h
e
s
et
o
f
(
m
eta
-
test
)
D
Me
ta
L
S
T
M
t
e
st
=
(
D
t
r
a
i
n
,
D
t
e
st
)
.
T
ab
le
2
p
r
esen
ts
th
e
tr
ain
in
g
al
g
o
r
ith
m
o
f
t
h
e
m
eta
-
lear
n
i
n
g
L
ST
M.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
3
,
J
u
n
e
2
0
2
1
:
2
4
5
7
-
2466
2462
T
ab
le
1
.
P
r
im
ar
y
al
g
o
r
ith
m
o
f
m
eta
-
lear
n
i
n
g
L
ST
M
In
p
u
t
:
t
a
sk
d
i
st
r
i
b
u
t
i
o
n
P
(
T
)
,
l
a
b
e
l
e
d
i
mag
e
b
i
g
d
a
t
a
se
t
B
d
,
l
e
a
r
n
i
n
g
r
a
t
e
R
Ou
t
p
u
t
:
,
,
w
h
i
l
e
t
r
u
e
:
I
n
i
t
i
a
l
i
z
e
,
,
f
o
r
t
a
sk
T
=
1
,
2
.
.
.
d
o
c
a
l
c
u
l
a
t
e
c
o
r
r
e
sp
o
n
d
i
n
g
l
o
ss o
f
me
t
a
-
l
e
a
r
n
i
n
g
:
L
T
(
θ
M
, θ
G
);
c
a
l
c
u
l
a
t
e
c
o
r
r
e
sp
o
n
d
i
n
g
l
o
ss o
f
map
-
r
e
d
u
c
e
r
d
i
scri
mi
n
a
t
o
r
(
D
)
:
L
(
x
,
y
)
(
θ
D
, θ
G
);
,
,
(θ
G
, θ
D
, θ
M
)
-
R
∇
[
(
(
,
)
,
(
,
)
(
,
)
]
;
e
n
d
w
h
i
l
e
e
n
d
f
o
r
T
ab
le
2
.
Me
ta
-
lear
n
in
g
L
ST
M
tr
ain
in
g
al
g
o
r
ith
m
In
p
u
t
:
t
a
sk
d
i
st
r
i
b
u
t
i
o
n
P
(
T
)
,
l
a
b
e
l
e
d
i
mag
e
B
i
g
d
a
t
a
se
t
(
B
d
)
,
i
mag
e
s
b
a
t
c
h
si
z
e
w
i
t
h
n
t
a
sk
s
,
i
mag
e
b
a
t
c
h
si
z
e
w
i
t
h
m i
n
s
t
a
n
c
e
s,
me
t
a
-
l
e
a
r
n
e
r
L
S
T
M
w
i
t
h
I
n
i
t
i
a
l
i
z
a
t
i
o
n
Φ
a
s
a
p
r
i
m
a
r
y
r
a
n
d
o
m
p
a
r
a
me
t
e
r
v
e
c
t
o
r
l
e
a
r
n
e
r
(
A
T
)
a
c
c
o
mp
a
n
i
e
d
b
y
t
h
e
p
a
r
a
me
t
e
r
θ,
me
t
a
-
t
r
a
i
n
i
n
g
se
t
(
−
)
,
me
t
a
-
l
e
a
r
n
e
r
M
a
c
c
o
m
p
a
n
i
e
d
b
y
t
h
e
p
a
r
a
me
t
e
r
φ
.
Ou
t
p
u
t
:
,
,
=
{
φ
,
r
}
d
a
t
a
T
r
a
i
n
=
L
O
A
D
d
a
t
a
T
e
st
=
L
O
A
D
f
o
r
1
,
…
n
i
t
e
r
a
t
i
o
n
s
d
o
t
ra
i
n
,
t
e
s
t
h
a
p
h
a
z
a
r
d
d
a
t
a
se
t
f
r
o
m D
me
t
a
-
t
r
a
i
n
i
n
i
t
i
a
l
i
z
e
φ
,
r
,
,
sp
l
i
t
t
i
n
g
d
a
t
a
se
t
t
o
t
r
a
i
n
i
n
g
a
n
d
t
e
st
i
n
g
w
i
t
h
a
se
q
u
e
n
t
i
a
l
mo
d
e
l
t
r
a
i
n
a
n
d
v
a
l
i
d
a
t
e
C
e
l
l
0
i
n
a
u
g
u
r
a
t
e
l
e
a
r
n
e
r
p
a
r
a
me
t
e
r
A
c
t
i
v
a
t
i
o
n
=
si
g
mo
i
d
f
o
r
e
v
e
r
y
t
a
sk
T
d
o
X
t
, Y
t
r
a
n
d
o
m
b
a
t
c
h
o
b
t
a
i
n
e
d
f
r
o
m t
h
e
t
r
a
i
n
i
n
g
d
a
t
a
se
t
(
D
t
r
a
i
n
)
L
t
L
(
A
T
(
X
t
;
θ
t
-
1
)
,
Y
t
)
→
o
b
t
a
i
n
l
e
a
r
n
e
r
l
o
ss o
n
t
r
a
i
n
b
a
t
c
h
c
t
M
(
(
∇
θ
t
-
1
L
t
,
L
t
)
;
φ
d
-
1
)
→
o
b
t
a
i
n
me
t
a
-
l
e
a
r
n
e
r
o
u
t
p
u
t
θ
t
c
t
b
r
i
n
g
u
p
t
o
d
a
t
e
l
e
a
r
n
e
r
p
a
r
a
me
t
e
r
s
X
,
Y
D
t
e
s
t
L
t
e
s
t
L
(
A
T
(
X
;
θ
T
)
,
Y
)
→
o
b
t
a
i
n
l
e
a
r
n
e
r
l
o
ss o
n
t
e
st
b
a
t
c
h
u
p
d
a
t
e
φ
by
-
u
t
i
l
i
z
i
n
g
(
−
1
)
→b
r
i
n
g
u
p
t
o
d
a
t
e
me
t
a
-
l
e
a
r
n
e
r
p
a
r
a
me
t
e
r
s
e
n
d
f
o
r
e
n
d
f
o
r
3
.
2
.
F
ea
t
ure
ex
t
ra
ct
io
n
a
nd
la
belin
g
Featu
r
e
e
x
tr
ac
tio
n
d
ep
icts
t
h
e
af
f
in
ed
s
h
ap
e
d
etails
th
at
co
m
p
r
i
s
ed
in
r
ec
o
g
n
it
io
n
o
f
a
p
atter
n
,
th
e
co
n
ce
p
t
o
f
f
ea
tu
r
e
ex
tr
ac
tio
n
co
n
s
id
er
ed
as
d
is
cr
im
i
n
ato
r
y
k
i
n
d
o
f
d
i
m
en
s
io
n
ali
t
y
r
e
d
u
ctio
n
,
w
h
er
e
th
e
p
u
r
p
o
s
e
o
f
th
e
ex
tr
ac
tio
n
p
r
o
ce
s
s
is
to
attain
t
h
e
s
i
g
n
if
ica
n
t
p
er
tin
e
n
t
d
ata
f
r
o
m
th
e
s
o
u
r
ce
o
f
th
at
d
ata
to
ap
p
o
in
t
th
e
m
in
a
m
i
n
i
m
ized
d
i
m
en
s
io
n
al
it
y
.
R
ef
er
r
i
n
g
to
t
i
m
e
o
r
cir
cu
m
s
ta
n
ce
,
th
e
i
n
p
u
t
d
ata
is
s
o
h
u
g
e
t
o
b
e
h
an
d
led
b
ased
o
n
th
e
f
o
llo
w
ed
s
et
o
f
r
u
les
i
n
an
al
g
o
r
ith
m
,
s
o
in
p
u
t
d
ata
s
h
o
u
ld
b
e
m
e
ta
m
o
r
p
h
o
s
ed
in
to
a
m
i
n
i
m
ized
f
o
r
m
o
f
f
ea
t
u
r
es.
T
h
e
p
r
o
ce
s
s
o
f
m
eta
m
o
r
p
h
o
s
in
g
i
n
p
u
t
d
ata
i
n
to
a
g
r
o
u
p
o
f
f
ea
tu
r
e
s
ca
n
b
e
d
ef
in
ed
as
f
ea
t
u
r
e
ex
tr
ac
tio
n
.
T
h
e
ex
tr
ac
tio
n
p
r
o
ce
s
s
in
v
o
lv
es
ca
p
tu
r
i
n
g
es
s
en
tial
d
etails
an
d
d
is
ti
n
cti
v
e
ch
ar
ac
ter
is
tic
s
f
r
o
m
r
a
w
d
ata,
w
h
er
e
ea
ch
f
ea
tu
r
e
i
s
ex
e
m
p
lif
ied
t
h
r
o
u
g
h
a
f
ea
tu
r
e
v
ec
to
r
th
at
tu
r
n
s
i
n
to
its
id
en
tit
y
[
1
5
]
.
T
ab
le
3
s
h
o
w
s
t
h
e
e
x
tr
ac
tio
n
a
n
d
l
ab
elin
g
al
g
o
r
ith
m
o
f
t
h
e
m
e
ta
-
lear
n
i
n
g
L
ST
M.
3
.
3
.
M
a
pR
educe
f
ra
m
ew
o
rk
Ma
p
R
ed
u
ce
m
o
d
el
co
n
ta
in
s
t
w
o
tas
k
s
;
Ma
p
-
ta
s
k
a
n
d
r
ed
u
ce
tas
k
.
Ma
p
R
ed
u
ce
m
o
d
el
i
s
d
ed
u
ce
d
f
r
o
m
th
e
f
u
n
ctio
n
a
m
al
g
a
m
at
io
n
o
f
a
m
ap
an
d
r
ed
u
ce
s
;
th
is
m
o
d
el
is
e
x
ce
ed
in
g
l
y
u
til
ized
to
p
r
o
ce
s
s
an
en
o
r
m
o
u
s
d
ataset.
Ma
p
R
ed
u
c
e
u
s
e
s
th
e
s
et
o
f
(
k
e
y
/v
al
u
e)
as
a
d
ata
ca
teg
o
r
y
[
1
6
]
.
A
r
e
p
r
esen
tatio
n
o
f
t
h
e
Ma
p
R
ed
u
ce
f
r
a
m
e
w
o
r
k
i
s
d
is
p
lay
ed
i
n
Fi
g
u
r
e
2
.
T
h
e
m
aj
o
r
s
tag
es
o
f
th
e
Ma
p
R
ed
u
ce
p
l
atf
o
r
m
ar
e
Ma
p
p
er
,
R
ed
u
ce
r
,
an
d
a
m
id
d
le
p
h
ase
k
n
o
w
n
as
s
h
u
f
f
le,
all
o
f
t
h
e
m
ar
e
in
tr
o
d
u
ce
d
in
t
h
e
s
ec
tio
n
:
Ma
p
p
er
f
u
n
ctio
n
:
h
an
d
les
in
p
u
t
d
a
ta
an
d
c
ar
r
y
o
u
t
a
f
ew
ca
lcu
l
ati
o
n
s
o
n
th
at
in
p
u
t
th
en
g
en
er
ate
s
in
ter
m
ed
i
at
e
o
u
tc
o
m
es
w
ith
th
e
ar
r
an
g
em
en
t
o
f
(
k
ey
/v
alu
e)
[
1
7
]
.
R
ed
u
ce
r
f
u
n
ctio
n
:
a
g
r
o
u
p
o
f
k
ey
v
alu
es
an
d
av
er
ag
e
d
k
ey
,
it
m
in
g
les
all
v
alu
es
w
ith
ea
ch
o
t
h
e
r
to
c
r
e
at
e
v
alu
es
o
f
a
lo
w
er
s
et
[
1
8
]
.
Sh
u
f
f
le
p
h
ase:
w
ith
in
th
e
Ma
p
R
ed
u
ce
p
latf
o
r
m
,
af
te
r
th
e
task
o
f
th
e
Ma
p
,
o
r
d
in
ar
il
y
co
n
s
id
er
ab
le
v
o
lu
m
es
o
f
m
i
d
d
le
d
at
a
n
e
e
d
to
b
e
s
h
if
ted
t
o
R
e
d
u
ce
o
p
e
r
at
io
n
f
r
o
m
th
e
Ma
p
o
p
er
ati
o
n
,
th
e
s
h
u
f
f
le
m
o
v
es
th
e
av
ai
l
ab
le
d
at
a
f
r
o
m
th
e
Ma
p
p
e
r
t
o
th
e
R
ed
u
c
e
r
p
h
as
e,
af
t
e
r
b
e
in
g
ar
r
an
g
ed
v
i
a
th
e
k
ey
s
.
T
h
u
s
th
e
w
h
o
le
p
a
ir
s
w
ith
an
id
e
n
tical
k
ey
s
h
all
b
e
o
r
g
an
ize
d
an
d
g
r
o
u
p
e
d
w
ith
e
ac
h
o
th
er
an
d
th
en
t
r
an
s
f
e
r
r
e
d
[
1
9
]
.
Ma
p
R
e
d
u
ce
'
s
f
r
am
ew
o
r
k
ca
r
r
ies
o
u
t
th
o
s
e
f
u
n
ctio
n
s
s
id
e
b
y
s
id
e
ef
f
icie
n
tly
,
ev
en
in
m
an
y
d
ev
ices
[
2
0
]
.
T
h
e
m
eta
-
L
STM
co
u
l
d
b
e
co
n
s
i
d
e
r
e
d
as
p
ast
lea
r
n
in
g
o
f
th
e
s
em
an
tic
s
tr
u
ctu
r
e,
an
d
th
e
p
iv
o
t
al
L
S
T
M
is
th
e
s
u
b
s
eq
u
en
t
k
n
o
w
led
g
e
.
C
o
n
s
e
q
u
en
tly
,
th
e
lea
r
n
e
d
m
eta
-
L
S
T
M
o
f
f
e
r
s
a
c
o
m
p
eten
t
m
eth
o
d
o
f
im
p
lem
en
tin
g
t
r
an
s
f
er
l
ea
r
n
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
C
a
s
s
in
i
–
Hu
yg
en
s
mis
s
io
n
ima
g
es c
la
s
s
ifica
tio
n
fr
a
mewo
r
k
b
y
d
ee
p
l
ea
r
n
in
g
...
(
A
s
h
r
a
f A
l
Da
b
b
a
s
)
2463
T
ab
le
3
.
E
x
tr
ac
tio
n
an
d
lab
elin
g
al
g
o
r
ith
m
e
x
t
r
a
c
t
_
f
e
a
t
u
r
e
s
i
mg
e
=
l
o
a
d
I
mag
e
(
"
*
.
j
p
e
g
"
)
l
o
a
d
P
i
x
e
l
s(
)
;
/
/
p
i
x
e
l
l
o
o
p
i
n
g
a
n
d
t
h
e
n
p
e
r
f
o
r
m
l
o
o
p
i
n
g
t
h
r
o
u
g
h
a
l
l
c
o
l
u
m
n
s a
n
d
r
o
w
s
f
o
r
(
i
n
t
C
o
l
u
mn
=
0
;
C
o
l
u
mn
<
w
i
d
t
h
;
C
o
l
u
mn
+
+
)
/
/
L
o
o
p
o
v
e
r
p
i
x
e
l
c
o
l
u
mn
f
o
r
(
i
n
t
R
o
w
=
0
;
R
o
w
<
h
e
i
g
h
t
;
R
o
w
+
+
)
/
/
L
o
o
p
o
v
e
r
p
i
x
e
l
r
o
w
f
o
r
(
i
n
t
p
i
x
=
0
;
p
i
x
<
p
i
x
e
l
s.
l
e
n
g
t
h
;
p
i
x
+
+
)
/
/
a
c
q
u
i
r
i
n
g
p
i
x
e
l
v
a
l
u
e
w
i
t
h
r
a
n
g
e
0
t
o
2
5
5
c
o
l
o
r
=
g
r
a
y
sca
l
e
p
i
x
e
l
s [p
i
x
]
=
c
o
l
o
r
b
a
se
d
o
n
r
a
n
g
e
[
0
-
2
5
5
]
/
/
u
p
d
a
t
e
p
i
x
e
l
s
i
m
a
g
e
_
si
z
e
=
(
i
mag
e
-
w
i
d
t
h
,
i
mag
e
-
h
e
i
g
h
t
)
,
samp
l
e
-
si
z
e
=
b
a
t
c
h
-
si
z
e
,
D
a
t
a
=
D
a
t
a
se
t
b
a
t
c
h
e
s
d
a
t
a
-
b
a
t
c
h
=
L
O
A
D
map
-
d
a
t
a
d
a
t
a
b
a
t
c
h
-
si
z
e
=
1
2
8
i
f
d
a
t
a
si
z
e
>
s
p
e
c
i
f
i
e
d
d
a
t
a
s
a
m
p
l
e
t
h
e
n
d
a
t
a
-
b
a
t
c
h
=
s
p
l
i
t
-
b
a
t
c
h
(
r
e
d
u
c
e
)
k
=
0
f
o
r
(
i
n
p
u
t
s
-
s
a
mp
l
e
,
l
a
b
e
l
s
-
samp
l
e
)
:
f
e
a
t
u
r
e
s_
b
a
t
c
h
=
L
S
T
M
.
p
r
e
d
i
c
t
(
i
n
p
u
t
s
-
sa
mp
l
e
)
f
e
a
t
u
r
e
s
[
k
*
sam
p
l
e
_
si
z
e
:
(
k
+
1
)
*
sam
p
l
e
-
si
z
e
]
=
f
e
a
t
u
r
e
s
-
sam
p
l
e
l
a
b
e
l
s [
k
*
sam
p
l
e
_
si
z
e
:
(
k
+
1
)
*
sam
p
l
e
-
si
z
e
]
=
l
a
b
e
l
s
sam
p
l
e
k
+
=
1
i
f
k
*
s
a
m
p
l
e
-
si
z
e
>
=
sam
p
l
e
-
n
u
me
r
a
t
e
:
b
r
e
a
k
r
e
t
u
r
n
l
a
b
e
l
s,
f
e
a
t
u
r
e
s
t
r
a
i
n
_
l
a
b
e
l
s &
f
e
a
t
u
r
e
s=
e
x
t
r
a
c
t
_
f
e
a
t
u
r
e
s
(
t
r
a
i
n
-
d
i
r
e
c
t
o
r
y
,
t
r
a
i
n
-
s
i
z
e
)
/
/
i
n
i
t
i
a
t
e
t
r
a
i
n
i
n
g
d
i
r
e
c
t
o
r
y
v
a
l
i
d
a
t
i
o
n
_
f
e
a
t
u
r
e
s,
v
a
l
i
d
a
t
i
o
n
_
l
a
b
e
l
s
=
e
x
t
r
a
c
t
_
f
e
a
t
u
r
e
s (v
a
l
i
d
a
t
i
o
n
-
d
i
r
e
c
t
o
r
y
,
v
a
l
i
d
a
t
i
o
n
-
si
z
e
)
t
e
st
_
f
e
a
t
u
r
e
s,
t
e
st
_
l
a
b
e
l
s =
e
x
t
r
a
c
t
_
f
e
a
t
u
r
e
s (t
e
st
-
d
i
r
e
c
t
o
r
y
,
t
e
st
-
si
z
e
)
v
a
l
i
d
a
t
i
o
n
_
d
a
t
a
=
(
v
a
l
i
d
a
t
i
o
n
-
f
e
a
t
u
r
e
s,
v
a
l
i
d
a
t
i
o
n
-
l
a
b
e
l
s)
#
C
o
mp
i
l
e
mo
d
e
l
mo
d
e
l
.
c
o
mp
i
l
e
(
o
p
t
i
mi
z
e
r
=
A
d
a
m ()
,
/
/
U
sag
e
o
f
S
t
o
c
h
a
st
i
c
g
r
a
d
i
e
n
t
d
e
sce
n
t
o
p
t
i
mi
z
e
r
.
l
o
ss='b
i
n
a
r
y
-
c
r
o
sse
n
t
r
o
p
y
',
me
t
r
i
c
s
=
[
't
a
r
g
e
t
_
t
e
n
so
r
s
-
a
c
c
u
r
a
c
y
']
)
mo
d
e
l
.
a
d
d
(
A
c
t
i
v
a
t
i
o
n
(
'so
f
t
m
a
x
')
)
r
e
c
o
r
d
=
mo
d
e
l
.
f
i
t
(
t
r
a
i
n
_
f
e
a
t
u
r
e
s,
t
r
a
i
n
-
l
a
b
e
l
s,
e
p
o
c
h
s=e
p
o
c
h
s,
sam
p
l
e
-
si
z
e
=
b
a
t
c
h
-
si
z
e
,
v
a
l
i
d
a
t
i
o
n
-
d
a
t
a
=
(
v
a
l
i
d
a
t
i
o
n
_
l
a
b
e
l
s,v
a
l
i
d
a
t
i
o
n
_
f
e
a
t
u
r
e
s)
/
/
M
o
d
e
l
T
r
a
i
n
e
n
d
f
o
r
e
n
d
f
o
r
e
n
d
f
o
r
e
n
d
f
o
r
4.
RE
SU
L
T
S AN
D
AN
AL
Y
SI
S DIS
C
USS
I
O
N
T
h
e
ess
en
tial
co
n
s
id
er
ed
f
ac
to
r
s
o
f
q
u
an
ti
tati
v
e
i
m
a
g
e
an
a
l
y
s
is
ar
e
p
r
o
ce
s
s
in
g
an
d
an
al
y
s
i
s
.
Am
o
n
g
th
e
ch
al
len
g
es
t
h
at
w
ill
f
ac
e
an
y
r
esear
ch
er
,
ar
e
s
o
f
t
w
ar
e
an
d
h
ar
d
w
ar
e
li
m
itatio
n
s
.
Du
r
in
g
o
u
r
d
ataset
p
r
o
ce
s
s
in
g
a
n
d
in
s
p
ec
tio
n
,
w
e
en
co
u
n
ter
ed
th
ese
k
in
d
s
o
f
r
estrictio
n
s
.
An
d
w
e
w
er
e
ab
le
to
o
v
er
co
m
e
th
e
m
,
as
w
e
ar
e
d
ea
li
n
g
w
i
th
t
h
e
B
d
,
w
e
ca
n
n
o
t
d
ep
en
d
o
n
th
e
r
eg
u
lar
co
m
p
u
ter
h
ar
d
w
ar
e,
s
o
w
e
u
s
ed
g
r
ap
h
ica
l
p
r
o
ce
s
s
in
g
u
n
i
ts
(
GP
Us);
th
e
y
ar
e
a
h
ar
d
w
ar
e
ap
p
lian
c
e
th
at
is
m
o
s
t
e
f
f
ec
ti
v
e
f
o
r
p
ar
allel
an
d
r
ap
i
d
p
r
o
ce
s
s
in
g
.
GP
Us
p
r
o
v
id
e
D
L
w
i
th
t
h
e
ab
ilit
y
to
p
er
f
o
r
m
s
e
p
ar
ated
co
m
p
u
tat
io
n
s
f
r
o
m
t
h
e
ce
n
tr
al
p
r
o
ce
s
s
o
r
(
th
at
i
s
s
er
ial
tas
k
s
d
ed
icate
d
)
an
d
ad
eq
u
atel
y
f
u
l
f
i
lli
n
g
co
m
p
lex
co
m
p
u
tatio
n
s
.
I
n
o
u
r
r
esear
ch
,
w
e
w
i
ll
u
s
e
th
e
GP
U
to
b
e
elig
ib
le
to
p
r
o
c
ess
a
b
i
g
d
ata
s
et
o
f
Sat
u
r
n
i
m
ag
es
t
h
at
co
n
tai
n
m
o
r
e
t
h
an
4
0
0
,
0
0
0
im
a
g
es.
T
o
in
itiate
t
h
e
p
r
o
ce
s
s
o
f
tr
ain
i
n
g
,
w
e
i
n
d
is
cr
i
m
i
n
atel
y
s
a
m
p
led
5
0
K
i
m
a
g
e
s
f
r
o
m
th
e
ad
o
p
ted
d
ataset
v
o
lu
m
es.
E
v
er
y
i
m
ag
e
i
s
co
r
r
elate
d
to
o
n
e
o
r
m
o
r
e
ca
teg
o
r
ies,
ar
r
an
g
ed
in
6
o
b
s
er
v
ab
le
d
e
n
o
m
in
atio
n
s
co
n
tai
n
i
n
g
as
r
ep
r
esen
ted
in
Fi
g
u
r
e
4
.
Fro
m
lef
t
to
r
ig
h
t:
Sat
u
r
n
,
R
i
n
g
s
,
T
itan
,
I
c
y
Sate
llit
es,
S
m
all
Sate
llit
es
(
r
o
ck
s
)
,
S
k
y
.
T
h
e
lab
el
s
elec
tio
n
m
e
th
o
d
is
b
ased
o
n
ea
ch
i
m
ag
e
co
n
te
n
t,
as
it
is
d
elin
ea
ted
to
an
in
ter
p
r
etativ
e
w
o
r
d
j
u
s
t
as
C
as
s
in
i
tea
m
s
ad
o
p
ted
.
Fig
u
r
e
4
p
r
o
v
id
ed
a
s
cr
ee
n
s
h
o
t
of
th
e
class
i
f
icat
io
n
p
r
o
ce
s
s
ad
o
p
te
d
,
w
h
ich
s
h
o
w
s
t
h
e
i
m
a
g
e
w
it
h
its
tar
g
et
w
it
h
co
r
r
ec
t c
lass
i
f
icatio
n
ac
cu
r
ac
y
,
e
v
en
w
it
h
n
o
t e
x
is
ti
n
g
b
ef
o
r
e
i
m
ag
es.
T
h
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
is
c
ap
ab
le
o
f
p
r
o
ce
s
s
i
n
g
at
h
i
g
h
s
p
ee
d
.
A
ls
o
,
t
h
e
e
f
f
ec
ti
v
e
n
e
s
s
o
f
t
h
e
ad
o
p
ted
ap
p
r
o
ac
h
to
tr
an
s
f
er
l
ea
r
n
in
g
i
s
n
o
ticed
.
T
ab
le
4
s
h
o
w
s
th
e
ad
o
p
ted
class
e
s
a
n
d
th
e
n
u
m
b
er
o
f
i
m
a
g
e
s
u
tili
ze
d
i
n
th
e
p
r
i
m
ar
y
p
r
o
ce
s
s
es o
f
tr
ain
i
n
g
an
d
test
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
3
,
J
u
n
e
2
0
2
1
:
2
4
5
7
-
2466
2464
Fig
u
r
e
4
.
C
ass
i
n
i
m
is
s
io
n
i
m
a
g
es sa
m
p
li
n
g
p
r
ese
n
ti
n
g
t
h
e
a
d
o
p
ted
c
o
n
ten
t la
b
eli
n
g
cla
s
s
e
s
T
ab
le
4
.
Nu
m
b
er
o
f
u
s
ed
i
m
ag
es f
o
r
th
e
i
n
itiated
p
r
o
ce
s
s
o
f
t
r
ain
in
g
a
n
d
test
i
n
g
C
l
a
ss
S
a
t
u
r
n
R
i
n
g
s
T
i
t
a
n
I
c
y
S
a
t
e
l
l
i
t
e
s
S
mal
l
S
a
t
e
l
l
i
t
e
s (r
o
c
k
s)
S
k
y
T
r
a
i
n
i
n
g
I
mag
e
s
9
8
6
5
8
3
6
9
9
5
6
8
7
4
6
0
7
8
7
9
6
8
5
9
T
e
st
i
n
g
i
mag
e
s
1
9
8
3
1
8
7
0
2
1
2
0
1
9
8
7
2
0
6
0
1
5
8
0
T
h
e
f
r
a
m
e
w
o
r
k
o
f
m
eta
-
lear
n
i
n
g
co
u
ld
b
e
ex
ec
u
ted
to
an
y
te
ch
n
iq
u
e
w
h
ich
i
s
tr
ain
ed
v
ia
t
h
e
s
ch
e
m
e
o
f
m
eta
-
lear
n
i
n
g
.
T
h
e
p
u
r
p
o
s
e
is
i
m
p
ar
tin
g
an
in
cl
u
s
iv
e
ap
p
r
o
ac
h
th
at
ca
n
r
ea
d
ily
b
e
o
r
d
er
ed
to
ac
h
iev
e
th
e
b
est
o
r
th
e
d
esire
d
p
e
r
f
o
r
m
a
n
c
e
w
it
h
an
y
r
eq
u
ir
ed
task
.
T
h
e
an
al
y
s
es
o
f
t
h
e
ce
ll
o
f
te
n
s
o
r
s
ar
e
s
tr
aig
h
t
f
o
r
w
ar
d
e
m
b
ed
d
in
g
v
ia
th
e
la
y
er
o
f
m
a
p
-
r
ed
u
ce
r
d
is
cr
i
m
in
ato
r
(
D)
,
w
h
ic
h
i
s
ac
ti
n
g
as
a
p
o
o
lin
g
la
y
er
th
a
t
r
ed
u
ce
s
th
e
m
ap
p
ed
f
ea
tu
r
es.
A
t
th
e
s
a
m
e
ti
m
e,
th
e
m
eta
-
lear
n
i
n
g
L
STM
w
ill
d
ec
o
n
s
tr
u
ct
ea
ch
g
en
er
ated
ten
s
o
r
in
to
f
o
u
r
ce
ll
ten
s
o
r
s
:
(
in
p
u
t
ten
s
o
r
ce
ll),
(
f
o
r
g
et
ten
s
o
r
ce
ll),
(
ce
ll
s
tate
ten
s
o
r
ce
ll),
an
d
(
o
u
tp
u
t
ten
s
o
r
ce
ll).
Ou
r
m
o
d
el
co
m
p
le
x
it
y
is
d
eter
m
in
ed
b
y
t
h
e
r
ev
elatio
n
o
f
L
ev
in
co
m
p
le
x
it
y
d
e
f
i
n
itio
n
[
2
1
]
:
(
)
=
ⅈ
{
(
)
: if
p
r
o
g
r
a
m
p
s
o
lv
e
s
P
an
d
th
en
ce
ases
d
u
r
i
n
g
ti
m
e
}
(7
)
w
h
er
e
(
)
=
1
(
)
+
(
(
)
)
(8
)
T
h
e
p
r
o
b
lem
t
h
at
n
ee
d
s
to
b
e
s
o
lv
ed
is
r
ep
r
esen
ted
b
y
P
,
w
h
ile
l(
p
)
is
th
e
p
r
o
g
r
a
m
p
len
g
th
,
an
d
t(
p
)
r
ep
r
esen
ts
t
h
e
ti
m
e
t
h
at
is
co
n
s
u
m
ed
b
y
p
to
s
o
l
v
e
P
.
T
r
an
s
f
er
r
in
g
k
n
o
w
led
g
e
ac
q
u
ir
ed
f
r
o
m
a
s
in
g
le
tas
k
w
it
h
t
h
e
ab
u
n
d
an
ce
o
f
lab
eled
d
ata
to
s
o
m
e
o
th
er
ta
s
k
s
w
i
th
s
l
ig
h
t
lab
eled
d
ata,
th
e
le
v
el
o
f
p
r
o
g
r
ess
io
n
o
f
p
er
f
o
r
m
in
g
its
m
i
s
s
io
n
r
elies
o
n
h
o
w
p
er
tin
e
n
t
is
t
h
e
f
o
r
m
er
t
ask
o
f
b
i
g
-
s
ca
le
i
m
ag
e
r
ec
o
g
n
i
tio
n
to
th
e
cu
r
r
e
n
t
task
[
2
2
]
.
I
n
th
e
s
it
u
atio
n
o
f
m
eta
-
lear
n
i
n
g
L
ST
Ms,
w
it
h
t
h
e
ep
ilo
g
u
e
o
f
ea
c
h
ta
s
k
,
th
e
ex
p
er
ien
ce
i
s
g
ai
n
ed
an
d
k
ep
t
i
n
t
h
e
m
e
m
o
r
y
o
f
t
h
e
L
ST
M
ce
ll.
T
o
co
n
f
ir
m
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
e
f
f
icien
c
y
,
t
h
is
m
o
d
el
is
s
et
to
w
ei
g
h
w
it
h
o
th
er
co
m
m
o
n
l
y
k
n
o
w
n
m
e
th
o
d
s
o
f
i
m
ag
e
cla
s
s
i
f
icatio
n
.
Ou
r
co
n
d
u
cted
ex
p
er
im
e
n
t
r
es
u
lt
s
ar
e
d
em
o
n
s
tr
ated
i
n
T
ab
le
5
.
I
t
is
clea
r
b
y
th
e
ev
id
en
ce
th
at
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
i
n
e
(
SVM
)
an
d
r
an
d
o
m
f
o
r
est
m
o
d
els
n
o
tab
l
y
ha
v
e
les
s
ac
c
u
r
ac
y
t
h
an
t
h
e
o
th
er
m
o
d
els.
T
h
e
j
u
s
ti
f
icat
io
n
o
f
t
h
is
lie
s
i
n
th
e
e
m
p
h
atic
f
ea
t
u
r
e
e
x
tr
ac
tio
n
p
er
f
o
r
m
ed
b
y
o
u
r
p
r
o
v
id
ed
m
o
d
el.
A
ls
o
,
o
u
r
m
o
d
el
ad
o
p
ts
DL
w
i
th
in
its
o
p
ti
m
al
o
p
ti
m
iz
atio
n
an
d
p
ar
a
m
eter
in
itial
izatio
n
.
Fi
g
u
r
e
5
s
h
o
w
s
t
h
e
d
ia
g
r
a
m
o
f
tr
ain
i
n
g
ac
c
u
r
ac
y
v
er
s
u
s
v
alid
atio
n
ac
c
u
r
ac
y
o
v
er
th
e
n
u
m
b
er
o
f
ep
o
ch
s
.
W
ith
i
n
th
e
lo
s
s
p
lo
t,
it
is
clea
r
th
at
t
h
e
m
o
d
el
h
o
ld
s
a
co
m
p
ar
ab
le
ef
f
icie
n
c
y
o
n
t
h
e
tr
ain
i
n
g
d
ata
an
d
v
alid
atio
n
d
ata.
T
h
e
co
n
f
u
s
io
n
m
atr
ix
f
o
r
ca
lc
u
lati
n
g
th
e
o
v
er
all
clas
s
if
ier
ac
cu
r
ac
y
is
s
h
o
w
n
i
n
Fi
g
u
r
e
6
,
w
h
ic
h
is
e
v
alu
a
ted
u
s
i
n
g
th
e
5
0
th
o
u
s
a
n
d
i
m
a
g
es d
atase
t.
T
h
e
ac
q
u
ir
ed
p
er
ce
n
tag
e
is
9
6
.
7
%.
T
ab
le
5
.
A
co
m
p
ar
is
o
n
a
m
o
n
g
p
r
io
r
r
elev
an
t
w
o
r
k
M
o
d
e
l
N
a
me
S
V
M
[
2
3
]
R
a
n
d
o
m
F
o
r
e
st
[
2
3
]
F
u
z
z
y
C
l
u
st
e
r
i
n
g
[
2
4
]
O
p
t
i
mi
z
e
d
F
u
z
z
y
s
y
st
e
m
[
2
5
]
S
w
e
e
p
I
mag
e
T
r
a
n
sf
o
r
mat
i
o
n
T
e
c
h
n
i
q
u
e
[
2
6
]
G
r
a
y
L
e
v
e
l
Co
-
o
c
c
u
r
r
e
n
c
e
M
a
t
r
i
c
e
s
[
2
7
]
O
u
r
P
r
o
p
o
se
d
M
o
d
e
l
A
c
c
u
r
a
c
y
5
2
.
6
%
7
2
.
3
%
8
8
.
7
8
%
9
3
.
0
7
%
a
n
d
9
5
.
2
5
%
9
3
.
3
4
%
9
0
%
9
6
.
7
%
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
C
a
s
s
in
i
–
Hu
yg
en
s
mis
s
io
n
ima
g
es c
la
s
s
ifica
tio
n
fr
a
mewo
r
k
b
y
d
ee
p
l
ea
r
n
in
g
...
(
A
s
h
r
a
f A
l
Da
b
b
a
s
)
2465
Fig
u
r
e
5
.
P
lo
t r
e
p
r
esen
tatio
n
o
f
th
e
m
o
d
el
ac
cu
r
ac
y
a
n
d
lo
s
s
o
n
tr
ain
,
v
al
id
atio
n
Fig
u
r
e
6
.
P
lo
t r
e
p
r
esen
tatio
n
o
f
th
e
m
o
d
el
co
n
f
u
s
io
n
m
atr
ix
5.
CO
NCLU
SI
O
N
AND
F
U
T
U
RE
WO
RK
T
h
e
i
m
a
g
e
clas
s
i
f
icatio
n
p
r
o
ce
s
s
i
s
a
co
m
p
lex
also
t
i
m
e
-
e
f
f
o
r
t d
r
ain
i
n
g
o
p
er
atio
n
,
an
d
ca
u
s
e
ex
tr
e
m
e
p
h
y
s
ical
o
r
m
en
ta
l
f
ati
g
u
e
;
it
n
ee
d
s
f
u
r
th
er
s
p
ac
e
an
d
t
i
m
e
t
o
co
m
p
lete
a
s
in
g
le
tas
k
,
al
s
o
ad
d
itio
n
al
ef
f
o
r
t
to
ex
p
r
ess
a
n
d
ex
tr
ac
t
f
ea
t
u
r
es
f
r
o
m
i
m
ag
e
s
o
r
to
cr
ea
te
a
n
e
u
r
al
n
et
w
o
r
k
to
clas
s
if
y
i
m
a
g
es,
esp
ec
iall
y
w
h
e
n
t
h
e
s
ize
o
f
th
e
d
ata
s
et
i
s
h
u
g
e,
s
in
g
le
d
e
v
ice
ca
p
ab
ilit
ies
w
il
l
n
o
t
b
e
e
n
o
u
g
h
to
co
m
p
l
y
w
it
h
s
p
ac
e
o
r
ti
m
e
ex
ig
e
n
cie
s
to
p
er
f
o
r
m
i
m
a
g
e
class
i
f
icatio
n
.
T
h
e
p
ar
allel
an
al
y
s
i
s
o
f
th
e
t
h
r
ee
m
o
d
u
le
s
:
t
h
e
f
ea
t
u
r
es
ex
tr
ac
to
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
3
,
J
u
n
e
2
0
2
1
:
2
4
5
7
-
2466
2466
(
G)
,
th
e
m
eta
-
lear
n
er
L
ST
M
(
M)
,
an
d
m
ap
-
r
ed
u
ce
r
d
is
cr
i
m
i
n
ato
r
(
D)
:
all
to
g
et
h
er
d
em
o
n
s
tr
ates
a
r
o
b
u
s
t
,
o
p
tim
ized
n
e
u
r
al
n
et
w
o
r
k
f
r
a
m
e
w
o
r
k
t
h
at
h
as
s
h
o
w
n
a
m
eli
o
r
ated
tr
ain
in
g
s
p
ee
d
,
an
d
h
er
e
w
e
co
n
clu
d
e
t
h
at
u
s
i
n
g
p
o
ten
t
i
m
a
g
e
d
a
ta
an
a
l
y
zi
n
g
f
r
a
m
e
w
o
r
k
,
o
f
f
er
s
th
e
ab
ilit
y
to
p
r
o
ce
s
s
b
ig
d
ata
w
it
h
m
o
r
e
p
r
ec
is
e
class
i
f
icatio
n
r
es
u
lts
.
T
h
e
ad
o
p
ted
f
r
a
m
e
w
o
r
k
ac
q
u
ir
ed
a
cl
ass
i
f
icatio
n
p
r
ec
is
io
n
o
f
9
6
.
7
%
.
T
h
e
s
u
b
s
eq
u
en
t
p
h
ase
f
o
r
t
h
is
w
o
r
k
ca
n
b
e
to
ad
d
an
o
th
er
la
y
er
o
f
clas
s
i
f
ic
atio
n
to
o
u
r
m
o
d
el
,
s
u
c
h
as
c
o
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
et
w
o
r
k
s
(
C
NN
s
)
,
to
ac
q
u
ir
e
a
h
ig
h
er
clas
s
if
icatio
n
ac
c
u
r
ac
y
.
RE
F
E
R
E
NC
E
S
[1
]
C.
A
g
g
a
r
w
a
l,
“
N
e
u
ra
l
Ne
tw
o
rk
s a
n
d
De
e
p
L
e
a
rn
in
g
:
A
T
e
x
tb
o
o
k
,
”
1
st ed
.
S
p
ri
n
g
e
r
,
Be
rli
n
,
G
e
rm
a
n
y
,
2
0
1
8
.
[2
]
X
.
C
h
e
n
g
,
X.
L
in
,
a
n
d
Y.
Z
h
e
n
g
,
“
De
e
p
si
m
il
a
rit
y
lea
rn
in
g
f
o
r
m
u
lt
im
o
d
a
l
m
e
d
ica
l
ima
g
e
s,”
CM
B
BE
:
Ima
g
i
n
g
&
Vi
su
a
li
z
a
ti
o
n
,
v
o
l.
6
,
p
p
.
2
4
8
-
2
5
2
,
2
0
1
8
.
[3
]
A
.
A
l
Da
b
b
a
s,
Z
.
G
a
l,
a
n
d
B
.
Atti
la
,
“
Ne
u
ra
l
Ne
t
w
o
rk
Esti
m
a
ti
o
n
o
f
T
o
u
rism
Cli
m
a
ti
c
In
d
e
x
(TCI)
Ba
se
d
o
n
T
e
m
p
e
r
a
tu
re
-
Hu
m
id
it
y
In
d
e
x
(
THI)
-
Jo
rd
a
n
Re
g
io
n
Us
in
g
S
e
n
se
d
Da
tas
e
ts
,”
Ca
rp
a
th
ia
n
J
o
u
r
n
a
l
o
f
El
e
c
tro
n
ic
a
n
d
Co
mp
u
ter
E
n
g
in
e
e
rin
g
,
v
o
l.
1
1
,
n
o
.
2
,
pp
.
5
0
-
5
5
,
2
0
1
8
.
[4
]
G.
A
c
e
to
,
D.
Ciu
o
n
z
o
,
A
.
M
o
n
ti
e
ri,
a
n
d
A
.
P
e
sc
a
p
é
,
“
T
o
wa
rd
Ef
f
e
c
ti
v
e
M
o
b
il
e
E
n
c
ry
p
ted
T
ra
ff
ic
Clas
s
if
ica
ti
o
n
th
ro
u
g
h
De
e
p
L
e
a
rn
in
g
,”
Ne
u
ro
c
o
mp
u
t
in
g
,
v
o
l
.
4
0
9
,
p
p
.
3
0
6
-
3
1
5
,
2
0
2
0
.
[5
]
P.
P
a
tel
a
n
d
A
.
T
h
a
k
k
a
r
,
“
T
h
e
u
p
su
rg
e
o
f
d
e
e
p
lea
rn
in
g
f
o
r
c
o
m
p
u
ter
v
isio
n
a
p
p
li
c
a
ti
o
n
s
,”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
El
e
c
trica
l
a
n
d
C
o
mp
u
ter
En
g
in
e
e
rin
g
(
IJ
ECE
)
,
v
o
l.
1
0
,
n
o
.
1
,
p
p
.
5
3
8
-
5
4
8
,
2
0
2
0
.
[6
]
A.
Bo
u
k
h
a
lf
a
,
e
t
a
l.
,
“
L
S
T
M
d
e
e
p
lea
rn
i
n
g
m
e
th
o
d
f
o
r
n
e
tw
o
rk
in
tru
sio
n
d
e
tec
ti
o
n
sy
ste
m
,”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
El
e
c
trica
l
a
n
d
C
o
mp
u
ter
En
g
in
e
e
rin
g
(
IJ
ECE
)
,
v
o
l.
1
0
,
p
p
.
3
3
1
5
-
3
3
2
2
,
2
0
2
0
.
[7
]
Ba
ll
J
.
E
.
,
A
n
d
e
rso
n
D
.
T
.
,
a
n
d
Ch
a
n
C
.
S.
,
“
Co
m
p
re
h
e
n
siv
e
su
rv
e
y
o
f
d
e
e
p
lea
rn
in
g
in
re
m
o
te
se
n
sin
g
:
th
e
o
ries
,
to
o
ls,
a
n
d
c
h
a
ll
e
n
g
e
s f
o
r
th
e
c
o
m
m
u
n
it
y
,”
J
o
u
rn
a
l
o
f
Ap
p
li
e
d
Rem
o
te S
e
n
si
n
g
,
v
o
l.
11
,
n
o
.
4
,
p
.
0
4
2
6
0
9
,
2
0
1
7
.
[8
]
Na
ti
o
n
a
l
A
e
ro
n
a
u
ti
c
s
a
n
d
S
p
a
c
e
A
d
m
in
istratio
n
o
f
th
e
USA
,
Ca
ss
in
i
IS
S
On
li
n
e
Da
ta
V
o
lu
m
e
s,
I
m
a
g
in
g
S
c
ien
c
e
S
u
b
sy
ste
m
(IS
S
),
S
a
tu
rn
EDR
Da
ta
S
e
ts
(V
o
l
u
m
e
1
-
V
o
l
u
m
e
1
1
6
)
.
[
O
n
li
n
e
]
.
A
v
a
il
a
b
le:
h
tt
p
s:/
/p
d
sim
a
g
in
g
.
jp
l.
n
a
sa
.
g
o
v
/v
o
lu
m
e
s/iss
.
h
tm
l
.
[9
]
C.
F
in
n
,
P
.
A
b
b
e
e
l,
a
n
d
S
.
L
e
v
in
e
,
“
M
o
d
e
l
-
a
g
n
o
stic
m
e
ta
-
lea
rn
i
n
g
f
o
r
f
a
st
a
d
a
p
tatio
n
o
f
d
e
e
p
n
e
tw
o
rk
s
,”
a
rXiv
p
re
p
rin
t
a
rXiv:
1
7
0
3
.
0
3
4
0
0
,
2
0
1
7
.
[1
0
]
A
.
Ni
c
h
o
l,
J
.
A
c
h
ia
m
,
a
n
d
J
.
S
c
h
u
lm
a
n
,
“
On
f
irst
-
o
rd
e
r
m
e
ta
-
lea
rn
in
g
a
lg
o
rit
h
m
s
,”
a
rXiv
p
re
p
rin
t
a
rXiv:
1
8
0
3
.
0
2
9
9
9
,
2
0
1
8
.
[1
1
]
A.
Al
D
a
b
b
a
s
a
n
d
Z
.
G
a
l
,
“
On
th
e
Co
m
p
le
x
Ev
e
n
t
Id
e
n
ti
f
ica
ti
o
n
Ba
se
d
o
n
C
o
g
n
i
t
i
v
e
C
l
a
s
s
i
f
i
c
a
t
i
o
n
P
r
o
c
e
s
s
,
”
2019
1
0
t
h
I
E
E
E
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
C
o
g
n
i
t
i
v
e
I
n
f
o
c
o
m
m
u
n
i
c
a
t
i
o
n
s
(
C
o
g
I
n
f
o
C
o
m
)
,
N
a
p
l
e
s
,
I
t
a
l
y
,
2
0
1
9
,
p
p
.
29
-
3
4
.
[1
2
]
A.
Al
Da
b
b
a
s
a
n
d
Z
.
G
á
l
,
“
G
e
tt
in
g
f
a
c
ts
a
b
o
u
t
in
ter
p
lan
e
tary
m
issio
n
o
f
Ca
ss
in
i
-
Hu
y
g
e
n
s
sp
a
c
e
c
ra
f
t
,
”
in
1
0
th
Hu
n
g
a
ria
n
GIS
Co
n
fer
e
n
c
e
a
n
d
Exh
ib
it
io
n
,
De
b
re
c
e
n
,
H
u
n
g
a
ry
,
2
0
1
9
.
[1
3
]
A.
Al
Da
b
b
a
s
a
n
d
Z
.
G
á
l
,
“
Co
m
p
lex
Ev
e
n
t
P
ro
c
e
ss
in
g
Ba
se
d
A
n
a
l
y
sis
o
f
Ca
ss
in
i
–
Hu
y
g
e
n
s
In
terp
lan
e
tar
y
Da
ta
se
t
,
”
in
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
In
fo
rm
a
t
io
n
,
Co
mm
u
n
ica
ti
o
n
a
n
d
Co
mp
u
t
in
g
T
e
c
h
n
o
lo
g
y
,
S
p
rin
g
e
r,
Ch
a
m
,
2
0
1
9
,
p
p
.
5
1
-
66
.
[1
4
]
Y.
L
e
Cu
n
,
Y
.
Be
n
g
io
,
a
n
d
G
.
Hin
to
n
,
“
De
e
p
lea
rn
in
g
,
”
in
N
a
t
u
re
,
v
o
l.
5
2
1
,
n
o
.
7
5
5
3
,
p
p
.
4
3
6
-
4
4
4
,
2
0
1
5
.
[1
5
]
G.
Ku
m
a
r
a
n
d
P
.
K
.
B
h
a
ti
a
,
“
A
d
e
tailed
re
v
iew
o
f
f
e
a
tu
re
e
x
trac
t
io
n
in
im
a
g
e
p
ro
c
e
ss
in
g
s
y
ste
m
s
,”
2
0
1
4
F
o
u
rt
h
In
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
A
d
v
a
n
c
e
d
C
o
mp
u
ti
n
g
&
Co
mm
u
n
ica
t
io
n
T
e
c
h
n
o
lo
g
ies
,
Ro
h
tak
,
2
0
1
4
,
p
p
.
5
-
12.
[1
6
]
De
a
n
J
.
a
n
d
G
h
e
m
a
w
a
t
S
.
,
“
S
i
m
p
li
f
ied
d
a
ta
p
ro
c
e
ss
in
g
o
n
larg
e
c
lu
ste
rs
,
”
in
S
ixty
c
o
n
fer
e
n
c
e
o
n
S
y
mp
o
siu
m
o
n
Op
e
ra
ti
n
g
S
y
ste
ms
De
sig
n
&
Imp
lem
e
n
ta
ti
o
n
(
OS
DI)
,
Be
rk
e
le
y
,
U
S
A
,
A
CM
,
2
0
0
4
,
p
p
.
1
0
7
-
1
1
3
.
[1
7
]
C.
W
.
Lee
,
K
.
Y
.
Hs
ieh
,
S
.
Y
.
Hs
ieh
,
a
n
d
H
.
C
.
Hs
iao
,
“
A
d
y
n
a
m
i
c
d
a
ta
p
lac
e
m
e
n
t
stra
teg
y
f
o
r
h
a
d
o
o
p
in
h
e
tero
g
e
n
e
o
u
s e
n
v
iro
n
m
e
n
ts
,”
B
ig
Da
t
a
Res
e
a
rc
h
,
v
o
l.
1
,
pp.
14
-
2
2
,
2
0
1
4
.
[1
8
]
S.
A
rid
h
i,
L.
d
’Ora
z
io
,
M.
M
a
d
d
o
u
ri,
E.
M
.
Ng
u
if
o
,
“
De
n
sity
-
b
a
se
d
d
a
ta p
a
rti
ti
o
n
i
n
g
stra
teg
y
to
a
p
p
ro
x
im
a
t
e
larg
e
-
sc
a
le su
b
g
ra
p
h
m
in
in
g
,”
In
f.
S
y
st
,
v
o
l.
4
8
,
p
p
.
2
1
3
-
2
2
3
,
2
0
1
5
.
[1
9
]
L.
Din
g
,
G.
W
a
n
g
,
J.
X
in
,
X.
W
a
n
g
,
S.
Hu
a
n
g
,
R.
Zh
a
n
g
,
“
Co
m
M
a
p
Re
d
u
c
e
:
a
n
im
p
ro
v
e
m
e
n
t
o
f
m
a
p
re
d
u
c
e
w
it
h
li
g
h
tw
e
ig
h
t
c
o
m
m
u
n
ica
ti
o
n
m
e
c
h
a
n
ism
s
,”
Da
ta
Kn
o
wl.
En
g
,
v
o
l.
8
8
,
p
p
.
2
2
4
-
2
4
7
,
2
0
1
3
.
[2
0
]
M
.
C.
S
c
h
a
tz,
“
Clo
u
d
B
u
rst:
h
ig
h
ly
se
n
siti
v
e
re
a
d
m
a
p
p
in
g
w
it
h
M
a
p
Re
d
u
c
e
,”
Bi
o
i
n
fo
rm
a
ti
c
s
,
v
o
l.
25
,
n
o
.
1
1
,
pp.
1
3
6
3
-
1
3
6
9
,
2
0
0
9
.
[2
1
]
M.
L
i
a
n
d
P
.
V
it
á
n
y
i
,
“
A
n
in
tro
d
u
c
ti
o
n
to
Ko
lm
o
g
o
ro
v
c
o
m
p
lex
it
y
a
n
d
it
s
a
p
p
li
c
a
ti
o
n
s
,”
Ne
w
Y
o
rk
:
S
p
ri
n
g
e
r
,
v
o
l.
3
,
2
0
0
8
.
[2
2
]
J.
Yo
sin
sk
i,
J
.
Clu
n
e
,
Y
.
Be
n
g
io
,
a
n
d
H
.
L
ip
so
n
,
“
Ho
w
tran
sf
e
ra
b
le
a
re
f
e
a
tu
re
s
in
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
s?
,
”
in
Ad
v
a
n
c
e
s i
n
n
e
u
r
a
l
i
n
f
o
rm
a
ti
o
n
p
ro
c
e
ss
in
g
sy
ste
ms
,
p
p
.
3
3
2
0
-
3
3
2
8
,
2
0
1
4
.
[2
3
]
L.
Zh
u
a
n
d
P
.
S
p
a
c
h
o
s
,
“
T
o
w
a
rd
s
Im
a
g
e
Clas
sif
i
c
a
ti
o
n
w
it
h
M
a
c
h
in
e
L
e
a
rn
in
g
M
e
t
h
o
d
o
lo
g
ies
f
o
r
S
m
a
rtp
h
o
n
e
s
,”
M
a
c
h
in
e
L
e
a
rn
i
n
g
a
n
d
Kn
o
wled
g
e
Extra
c
ti
o
n
,
v
o
l.
1
,
n
o
.
4
,
p
p
.
1
0
3
9
-
1
0
5
7
,
2
0
1
9
.
[2
4
]
F.
Ya
n
,
W
.
M
e
i,
a
n
d
Z
.
Ch
u
n
q
i
n
,
“
S
A
R
i
m
a
g
e
tar
g
e
t
re
c
o
g
n
it
io
n
b
a
se
d
o
n
Hu
in
v
a
rian
t
m
o
m
e
n
ts
a
n
d
S
V
M
,
”
2
0
0
9
Fi
ft
h
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
In
fo
rm
a
ti
o
n
Assu
r
a
n
c
e
a
n
d
S
e
c
u
rity
,
X
i'
a
n
,
2
0
0
9
,
p
p
.
5
8
5
-
5
8
8
.
[2
5
]
M.
Ra
m
e
z
a
n
if
a
rd
a
n
d
B.
S
.
M
o
u
s
a
v
i
,
“
Dig
it
a
l
i
m
a
g
e
c
las
si
f
ica
ti
o
n
b
y
o
p
ti
m
ise
d
f
u
z
z
y
s
y
st
e
m
,”
In
d
o
n
e
sia
n
J
o
u
rn
a
l
o
f
El
e
c
trica
l
En
g
in
e
e
rin
g
a
n
d
Co
mp
u
ter
S
c
ien
c
e
(
IJ
EE
CS
)
,
v
o
l
.
1
4
,
n
o
.
3
,
p
p
.
1
1
9
6
-
1
2
0
2
,
2
0
1
9
.
[2
6
]
S.
Ib
ra
h
im
,
e
t
a
l.
,
“
R
i
c
e
g
r
a
i
n
c
l
a
s
s
i
f
i
c
a
t
i
o
n
u
s
i
n
g
m
u
l
t
i
-
c
l
a
s
s
s
u
p
p
o
r
t
v
e
c
t
o
r
m
a
c
h
i
n
e
(
S
V
M
)
,”
I
A
E
S
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
I
J
A
I
)
,
v
o
l
.
8
,
n
o
.
3
,
p
p
.
2
1
5
-
220
,
2
0
1
9
.
[2
7
]
Y.
S
a
ri,
P
.
B
.
P
ra
k
o
so
a
n
d
A
.
R
.
B
a
s
k
a
r
a
,
“
A
p
p
l
i
c
a
t
i
o
n
o
f
n
e
u
r
a
l
n
e
t
w
o
r
k
m
e
t
h
o
d
f
o
r
r
o
a
d
c
r
a
c
k
d
e
t
e
c
t
i
o
n
,”
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
.
4
,
p
p
.
1
9
6
2
-
1967
,
2
0
2
0
.
Evaluation Warning : The document was created with Spire.PDF for Python.