I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
37
,
No
.
1
,
J
an
u
ar
y
20
25
,
p
p
.
2
6
9
~
2
7
8
I
SS
N:
2
502
-
4
7
52
,
DOI
: 1
0
.
1
1
5
9
1
/ijee
cs
.v
37.
i
1
.
p
p
269
-
2
7
8
269
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs
.
ia
esco
r
e.
co
m
La
nd
scene clas
sif
ica
tion usin
g
dive
rsity
prom
o
ting
metric
lea
rning
-
co
nv
o
lut
io
na
l neural net
w
o
rk
K
a
m
pa
Ra
t
na
B
a
bu
1
,
K
a
m
p
a
K
a
nthi K
um
a
r
2
,
Ak
ula
Su
neet
ha
3
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
En
g
i
n
e
e
r
i
n
g
,
M
.
B
.
T.
S
.
G
o
v
e
r
n
me
n
t
P
o
l
y
t
e
c
h
n
i
c
,
G
u
n
t
u
r
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s
a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
T
i
r
u
mal
a
E
n
g
i
n
e
e
r
i
n
g
C
o
l
l
e
g
e
,
N
a
r
a
s
a
r
a
o
p
e
t
,
I
n
d
i
a
3
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
K
K
R
&
K
S
R
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
a
n
d
S
c
i
e
n
c
e
s
,
G
u
n
t
u
r
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
23
,
2
0
2
4
R
ev
is
ed
Sep
3
,
202
4
Acc
ep
ted
Sep
7
,
2
0
2
4
Th
e
lan
d
sc
e
n
e
c
las
sifica
ti
o
n
b
y
re
m
o
te
se
n
sin
g
ima
g
e
s
p
re
d
icts
se
m
a
n
ti
c
c
las
s
o
f
ima
g
e
b
lo
c
k
s
b
y
re
m
o
v
in
g
v
isu
a
l
p
rimiti
v
e
s
in
re
m
o
t
e
se
n
sin
g
ima
g
e
s.
Ho
we
v
e
r,
th
e
re
is
a
p
ro
b
lem
o
f
with
i
n
-
c
las
s
d
i
v
e
rsity
a
n
d
b
e
twe
e
n
-
c
las
s
sim
il
a
rit
y
th
a
t
d
e
g
ra
d
e
s
a
p
e
rfo
rm
a
n
c
e
o
f
sc
e
n
e
c
las
sifica
ti
o
n
.
I
n
t
h
is
re
se
a
rc
h
,
th
e
d
iv
e
rsity
p
ro
m
o
ti
n
g
m
e
tri
c
lea
rn
in
g
–
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
(DPM
L
-
CNN
)
m
e
th
o
d
is
p
ro
p
o
se
d
f
o
r
c
las
sify
i
n
g
la
n
d
sc
e
n
e
ima
g
e
s.
Th
e
m
e
tri
c
lea
rn
i
n
g
with
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
(
C
NN
)
m
a
p
s
th
e
sa
m
e
sc
e
n
e
ima
g
e
c
las
s
c
lo
se
r
a
n
d
th
e
d
iffere
n
t
c
las
s
sc
e
n
e
s
a
s
fa
r
a
s
p
o
ss
ib
le
wh
ich
m
a
k
e
s
th
e
m
e
t
h
o
d
m
u
c
h
d
isc
rimin
a
ti
o
n
.
T
h
e
d
i
v
e
rsity
p
ro
m
o
ti
n
g
i
n
m
e
tri
c
lea
rn
in
g
is
u
se
d
to
re
d
u
c
e
th
e
o
v
e
rlap
p
in
g
o
f
th
e
sa
m
e
sc
e
n
e
c
las
s
b
y
u
n
c
o
rre
latio
n
o
f
e
v
e
ry
p
a
ra
m
e
ter
a
n
d
p
r
o
v
i
d
e
s
u
n
i
q
u
e
in
fo
rm
a
ti
o
n
f
o
r
th
o
se
p
a
ra
m
e
ters
.
Th
e
UC
M
e
rc
e
d
,
AID
,
a
n
d
NWP
U
RES
IS
C4
5
d
a
tas
e
ts
a
re
u
ti
li
z
e
d
i
n
th
is
re
se
a
rc
h
fo
r
e
v
a
l
u
a
ti
n
g
th
e
p
ro
p
o
se
d
DPM
L
-
CNN
m
e
th
o
d
with
e
v
a
l
u
a
ti
o
n
m
e
tri
c
s
li
k
e
a
c
c
u
ra
c
y
a
n
d
k
a
p
p
a
c
o
e
fficie
n
t.
Th
e
D
PML
-
CNN
m
e
th
o
d
re
a
c
h
e
d
h
ig
h
e
st
a
c
c
u
ra
c
y
o
f
9
9
.
2
7
%
a
n
d
9
9
.
8
4
%
f
o
r
5
0
%
a
n
d
8
0
%
train
in
g
ra
ti
o
s
o
n
th
e
UC
M
e
rc
e
d
d
a
tas
e
t
wh
e
n
c
o
m
p
a
re
d
t
o
o
t
h
e
r
e
x
isti
n
g
m
e
th
o
d
s
li
k
e
m
u
lt
i
-
lev
e
l
se
m
a
n
ti
c
fe
a
tu
re
c
lu
ste
rin
g
a
tt
e
n
ti
o
n
(M
L
F
C
-
Ne
t)
a
n
d
g
lo
b
a
l
c
o
n
tex
t
sp
a
ti
a
l
a
tt
e
n
ti
o
n
(G
CS
A
-
Ne
t).
K
ey
w
o
r
d
s
:
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Div
er
s
ity
p
r
o
m
o
tin
g
m
etr
ic
lear
n
in
g
L
an
d
s
ce
n
e
class
if
icatio
n
R
em
o
te
s
en
s
in
g
s
ce
n
e
im
ag
es
Un
co
r
r
elatio
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
:
Kam
p
a
R
atn
a
B
ab
u
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
E
n
g
i
n
ee
r
in
g
,
M.
B
.
T
.
S.
Go
v
er
n
m
en
t Po
ly
tech
n
ic
Gu
n
tu
r
,
I
n
d
ia
E
m
ail:
k
r
atn
ab
ab
u
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
r
em
o
te
s
en
s
in
g
im
a
g
e
s
ce
n
e
class
if
icatio
n
m
ap
s
s
ce
n
e
i
m
ag
es
to
p
a
r
ticu
lar
h
i
g
h
-
lev
el
s
em
an
tic
class
es
an
d
en
ab
les
th
e
ac
q
u
is
itio
n
o
f
h
ig
h
-
le
v
el
s
em
an
tic
d
a
ta
[
1
]
.
R
ec
en
tly
,
it
h
as
b
ee
n
a
p
r
o
m
is
in
g
r
esear
ch
f
ield
in
h
ig
h
-
r
eso
lu
tio
n
im
ag
e
class
if
icatio
n
o
f
r
em
o
te
s
en
s
i
n
g
[
2
]
.
I
t
is
m
ajo
r
ly
u
tili
ze
d
i
n
an
aly
zin
g
n
atu
r
al
r
eso
u
r
ce
s
,
lan
d
u
s
e
a
n
d
class
if
icatio
n
o
f
lan
d
co
v
e
r
ag
e,
d
ete
ctio
n
o
f
d
is
aster
s
,
an
d
p
lan
n
in
g
o
f
u
r
b
an
[
3
]
.
T
h
e
class
if
icatio
n
o
f
s
ce
n
es
s
u
p
p
o
r
ts
p
eo
p
le
in
u
n
d
er
s
tan
d
in
g
i
m
ag
e
co
n
ten
t,
w
h
ich
is
h
ig
h
l
y
co
n
v
e
n
ien
t
to
th
e
liv
es
o
f
p
eo
p
le
in
v
a
r
io
u
s
ap
p
licatio
n
s
lik
e
s
m
ar
t
cities,
d
etec
tio
n
o
f
r
em
o
te
s
en
s
in
g
[4
]
-
[
6
]
.
Ho
wev
er
,
in
tr
a
-
class
v
ar
iatio
n
s
ar
e
d
if
f
icu
lt,
an
d
in
ter
-
class
s
im
ilar
ity
o
f
o
b
jects
in
o
r
ig
in
al
s
en
s
o
r
s
m
ax
im
izes
th
e
d
ata
co
m
b
in
atio
n
an
d
lo
g
ical
r
ea
s
o
n
in
ev
e
r
y
s
ce
n
e
[
7
]
.
Fo
r
th
e
class
if
icatio
n
o
f
s
ce
n
e
im
ag
es,
th
e
im
ag
e
is
in
itially
ch
ar
ac
ter
ized
th
r
o
u
g
h
a
f
ea
tu
r
e
en
co
d
er
a
n
d
n
e
x
t
cl
ass
if
ied
b
y
a
class
if
ier
[
8
]
.
C
o
m
m
o
n
ly
,
th
er
e
a
r
e
v
ar
iatio
n
s
an
d
i
n
co
n
s
is
ten
cies
am
o
n
g
d
ata
ex
tr
ac
ted
f
r
o
m
v
is
u
al
in
f
o
r
m
atio
n
an
d
co
m
p
r
eh
e
n
s
io
n
o
f
p
eo
p
le
in
s
i
m
il
a
r
d
a
t
a
t
h
a
t
c
a
u
s
e
s
e
m
a
n
tic
g
a
p
s
a
m
o
n
g
f
e
a
t
u
r
e
r
e
p
r
e
s
e
n
t
a
t
i
o
n
a
n
d
u
n
d
e
r
s
t
a
n
d
i
n
g
o
f
h
i
g
h
-
l
e
v
e
l
f
e
a
t
u
r
e
s
[
9
]
.
I
n
r
e
c
e
n
t
t
i
m
e
s
,
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
N
N
s
)
h
a
v
e
d
e
v
e
l
o
p
e
d
i
n
a
r
e
a
o
f
i
m
a
g
e
s
c
e
n
e
c
l
a
s
s
i
f
i
c
a
t
i
o
n
[
1
0
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
37
,
No
.
1
,
J
an
u
ar
y
20
25
:
269
-
2
7
8
270
Var
ian
t
C
NN
-
d
ep
en
d
e
n
t
tech
n
iq
u
es
h
a
v
e
b
ee
n
r
u
lin
g
in
th
e
ar
ea
o
f
r
em
o
te
s
en
s
in
g
s
ce
n
e
class
if
icatio
n
[
1
1
]
.
T
h
e
g
r
ea
t
ac
h
iev
e
m
en
t
o
f
d
if
f
er
en
t
C
NN
-
b
ased
im
ag
e
s
ce
n
e
class
if
icatio
n
tech
n
iq
u
es
is
h
ig
h
ly
attr
ib
u
ted
to
u
tili
za
tio
n
o
f
d
ee
p
C
NN
tech
n
iq
u
es
lik
e
VGGN
et,
Go
o
g
L
eNe
t,
an
d
Alex
Net.
T
h
ese
tec
h
n
iq
u
es
e
x
tr
ac
t
th
e
m
ea
n
in
g
f
u
l
im
ag
e
f
ea
t
u
r
e
r
e
p
r
esen
tatio
n
s
th
at
ar
e
m
u
c
h
d
i
s
cr
im
in
ativ
e
th
an
h
a
n
d
cr
af
te
d
lo
w
-
lev
el
f
ea
tu
r
es
lik
e
tex
tu
r
e,
s
p
ec
tr
al,
an
d
co
lo
r
f
ea
tu
r
es
[
1
2
]
.
T
h
o
u
g
h
p
e
r
f
o
r
m
an
ce
o
f
s
ce
n
e
im
ag
e
class
if
icatio
n
is
en
h
an
ce
d
th
r
o
u
g
h
m
ea
n
in
g
f
u
l
f
ea
tu
r
es
lear
n
ed
th
r
o
u
g
h
d
ee
p
C
NN
tech
n
iq
u
es
[
1
3
]
,
[
1
4
]
,
th
er
e
a
r
e
two
m
ain
is
s
u
es
:
with
in
-
class
d
iv
er
s
ity
an
d
b
etwe
en
-
class
s
im
ilar
ity
.
T
h
ese
i
s
s
u
es
d
eg
r
ad
e
th
e
p
er
f
o
r
m
an
c
e
o
f
r
em
o
te
s
en
s
in
g
im
ag
e
s
ce
n
e
class
if
icatio
n
wh
ich
n
ee
d
s
to
b
e
tack
led
.
W
an
g
et
a
l.
[
1
5
]
s
u
g
g
ested
a
m
u
lti
-
lev
el
s
em
an
tic
f
ea
tu
r
e
c
lu
s
ter
in
g
atten
tio
n
(
ML
FC
-
Net)
m
eth
o
d
d
ep
en
d
i
n
g
o
n
d
ee
p
co
n
v
o
lu
tio
n
n
e
u
r
al
n
etwo
r
k
s
(
DC
NNs)
wh
ich
ex
tr
ac
te
d
m
u
ch
m
o
r
e
a
cc
u
r
ate
f
ea
tu
r
e
d
ata.
T
h
e
s
u
g
g
ested
m
eth
o
d
u
tili
ze
d
h
ig
h
s
p
atial
d
ata
i
n
r
e
m
o
te
s
en
s
in
g
im
ag
es
co
m
b
in
in
g
co
m
m
o
n
s
em
an
tic
f
ea
tu
r
e
d
ata
with
clu
s
ter
ed
d
ata.
T
h
en
,
R
ea
r
r
an
g
e
th
e
r
esp
ec
tiv
e
d
ata
weig
h
t
lik
e
f
ea
tu
r
e
m
ap
s
an
d
ten
s
o
r
b
lo
ck
s
b
y
atten
tio
n
m
ec
h
an
i
s
m
.
T
h
e
s
u
g
g
ested
m
eth
o
d
i
m
p
r
o
v
e
d
t
h
e
r
e
p
r
esen
tatio
n
o
f
v
ar
io
u
s
d
if
f
icu
lt
asp
ec
ts
with
m
in
im
u
m
co
m
p
u
tatio
n
al
co
s
t
an
d
b
etter
p
o
r
t
ab
ilit
y
.
Ho
wev
er
,
th
e
s
u
g
g
est
ed
m
eth
o
d
was
n
o
t
g
o
o
d
at
d
is
cr
im
in
atio
n
ca
p
ab
ilit
y
am
o
n
g
v
ar
ian
t
s
ce
n
e
ca
teg
o
r
ies.
T
h
ir
u
m
alad
ev
i
et
a
l
.
[
1
6
]
p
r
esen
ted
a
m
eth
o
d
th
r
o
u
g
h
in
teg
r
ates
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
an
d
t
r
an
s
f
er
lear
n
in
g
to
lan
d
s
ce
n
e
class
if
icatio
n
.
T
h
e
ca
teg
o
r
izatio
n
o
f
im
ag
es
was
e
n
h
an
ce
d
f
o
r
s
ce
n
e
class
if
icatio
n
ac
c
u
r
a
cy
th
r
o
u
g
h
tr
an
s
f
e
r
lear
n
in
g
with
n
etwo
r
k
s
lik
e
Alex
Net
an
d
v
is
u
al
g
eo
m
etr
y
g
r
o
u
p
(
VGG)
an
d
co
m
p
ar
ed
to
co
n
v
en
tio
n
al
f
ea
tu
r
e
e
x
tr
ac
tio
n
tech
n
iq
u
es.
I
n
itially
,
f
ea
tu
r
es
wer
e
r
ec
o
v
e
r
ed
f
r
o
m
th
e
n
etwo
r
k
’
s
n
ex
t
f
u
lly
co
n
n
ec
ted
lay
er
an
d
ass
ig
n
ed
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
class
if
icatio
n
.
Nex
t,
th
e
f
in
al
lay
e
r
s
o
f
th
e
n
etwo
r
k
s
wer
e
s
u
b
s
titu
ted
d
u
r
i
n
g
tr
an
s
f
er
lea
r
n
in
g
to
class
if
y
n
ew
d
atasets
.
T
h
e
p
r
esen
ted
m
eth
o
d
ca
p
tu
r
ed
th
e
b
o
u
n
d
a
r
y
o
f
s
ce
n
e
im
ag
es
an
d
r
ec
o
g
n
ize
d
it.
Ho
wev
er
,
th
e
m
eth
o
d
w
as
d
if
f
icu
lt
in
d
escr
ib
in
g
h
i
g
h
s
em
an
tic
d
ata
in
r
em
o
te
s
en
s
in
g
im
ag
es.
C
h
en
et
a
l.
[
1
7
]
d
e
v
elo
p
e
d
a
m
eth
o
d
th
at
d
e
p
en
d
e
d
o
n
g
lo
b
al
co
n
t
ex
t
s
p
atial
atten
tio
n
(
GC
SA)
an
d
d
en
s
ely
co
n
n
ec
t
ed
co
n
v
o
lu
tio
n
al
n
etwo
r
k
s
f
o
r
ex
tr
ac
tin
g
m
u
ltip
le
s
ca
le
s
ce
n
e
f
ea
tu
r
es
n
am
ed
GC
SANe
t.
T
h
e
m
ix
tu
r
e
p
r
o
c
ess
was
u
til
ized
f
o
r
im
p
r
o
v
in
g
s
p
atial
m
ix
tu
r
e
in
f
o
r
m
atio
n
o
f
r
em
o
te
s
en
s
in
g
im
ag
es
an
d
d
is
cr
ete
s
am
p
le
s
p
ac
e
was
r
ed
u
ce
d
to
en
h
an
c
e
s
m
o
o
th
n
ess
in
d
ata
s
p
ac
e
n
eig
h
b
o
r
h
o
o
d
.
T
h
e
GC
SA
was
im
p
lem
en
ted
in
a
d
en
s
ely
co
n
n
ec
ted
n
etwo
r
k
f
o
r
e
n
co
d
i
n
g
c
o
n
tex
tu
al
d
ata
o
f
r
e
m
o
te
s
en
s
in
g
s
ce
n
e
im
ag
es
in
to
lo
ca
l
f
ea
tu
r
es.
T
h
e
d
ev
elo
p
ed
m
et
h
o
d
h
a
s
h
ig
h
r
o
b
u
s
tn
ess
an
d
s
tab
ilit
y
,
b
y
u
s
in
g
m
ix
-
u
p
o
p
er
atio
n
s
th
at
en
h
a
n
ce
th
e
c
lass
if
icatio
n
ac
cu
r
ac
y
an
d
s
m
o
o
th
n
ess
o
f
th
e
m
eth
o
d
.
Ho
wev
er
,
th
e
is
s
u
e
o
f
in
ter
class
s
im
ilar
i
ty
lies
in
o
v
er
lap
p
in
g
o
f
s
im
ilar
s
u
r
f
ac
es b
e
twee
n
v
ar
io
u
s
s
ce
n
es.
L
v
et
a
l
.
[
1
8
]
in
c
o
r
p
o
r
ated
th
e
b
en
e
f
its
o
f
m
u
ltip
le
s
ca
l
e
an
d
m
u
ltip
le
le
v
el
f
ea
tu
r
es
an
d
d
ev
el
o
p
ed
a
m
eth
o
d
th
at
co
m
b
in
e
d
g
lo
b
al
f
ea
tu
r
es
f
o
r
id
e
n
tify
in
g
g
lo
b
al
atten
tio
n
f
ea
tu
r
es
an
d
lear
n
in
g
th
e
m
u
lti
-
d
ee
p
d
ep
e
n
d
en
cie
s
am
o
n
g
v
ar
iatio
n
s
in
s
p
atial
s
ca
le.
T
wo
v
ar
i
o
u
s
f
ea
tu
r
e
ad
ap
tiv
e
f
u
s
io
n
m
eth
o
d
s
wer
e
im
p
lem
e
n
ted
f
o
r
ex
p
lo
r
in
g
co
m
p
lem
en
tar
y
ass
o
ciatio
n
s
o
f
lo
ca
l
an
d
g
lo
b
al
ag
g
r
eg
ate
f
ea
tu
r
es
th
at
ac
q
u
ir
e
v
ar
io
u
s
im
ag
e
s
ce
n
es.
T
h
e
m
eth
o
d
e
x
p
lo
r
e
d
th
e
n
atu
r
e
o
f
g
lo
b
al
an
d
lo
ca
l
f
ea
tu
r
es
f
o
r
co
m
p
r
eh
en
s
iv
ely
d
escr
ib
in
g
th
e
im
ag
e
s
ce
n
es.
Ho
wev
er
,
d
u
e
to
lim
ited
d
ata,
th
e
m
eth
o
d
ten
d
s
to
is
s
u
e
o
v
e
r
-
f
itti
n
g
an
d
less
f
ea
tu
r
e
g
en
er
a
lizatio
n
ca
p
ab
ilit
y
.
Ma
et
a
l.
[
1
9
]
p
r
esen
ted
th
e
h
o
m
o
-
h
ete
r
o
g
e
n
o
u
s
tr
an
s
f
o
r
m
e
r
lear
n
in
g
(
HHT
L
)
m
eth
o
d
to
class
if
y
R
S
s
ce
n
es.
I
n
itially
,
a
p
atch
g
en
er
atio
n
m
eth
o
d
was
d
ev
elo
p
ed
f
o
r
g
en
er
atin
g
p
atch
es
o
f
h
o
m
o
g
e
n
eo
u
s
an
d
h
eter
o
g
en
e
o
u
s
d
ata
with
in
R
S
s
ce
n
es.
Nex
t,
a
d
o
u
b
le
-
p
h
ase
f
ea
tu
r
e
lear
n
in
g
m
o
d
u
le
(
FL
M)
was
in
tr
o
d
u
ce
d
f
o
r
h
o
m
o
g
en
eo
u
s
an
d
h
eter
o
g
en
eo
u
s
d
ata
with
in
R
S
s
ce
n
es.
I
n
t
h
e
FLM
-
d
ep
e
n
d
en
t
v
is
io
n
tr
an
s
f
o
r
m
er
,
b
o
th
g
lo
b
al
d
ata
an
d
lo
ca
l
r
eg
i
o
n
s
an
d
th
eir
co
n
tex
t
d
ata
wer
e
ca
p
tu
r
ed
.
At
last
,
d
ev
elo
p
ed
a
cl
ass
if
icatio
n
m
eth
o
d
th
at
h
as
a
f
u
s
io
n
s
u
b
m
o
d
u
le
an
d
m
etr
ic
lear
n
i
n
g
.
Ho
wev
e
r
,
it
h
as
d
if
f
icu
lt
in
tr
a
-
class
v
ar
iatio
n
s
an
d
in
ter
-
class
s
im
ilar
it
ies
in
o
r
ig
in
al
s
c
en
es
th
at
m
ax
im
ize
t
h
e
co
m
p
l
ex
ity
o
f
d
ata
in
te
g
r
atio
n
in
ev
er
y
s
ce
n
e.
X
u
et
a
l.
[
2
0
]
in
tr
o
d
u
ce
d
a
L
ie
Gr
o
u
p
d
ee
p
lear
n
in
g
m
et
h
o
d
to
class
if
y
r
em
o
te
s
en
s
in
g
s
ce
n
es.
I
n
itially
,
ex
tr
ac
te
d
s
h
allo
wer
an
d
h
ig
h
-
lev
el
f
ea
tu
r
es
f
r
o
m
im
a
g
es
d
ep
e
n
d
ed
o
n
lie
g
r
o
u
p
m
ac
h
in
e
lear
n
in
g
(
L
GM
L
)
an
d
d
ee
p
lear
n
in
g
f
o
r
m
ax
im
izin
g
th
e
ca
p
ac
ity
o
f
f
ea
t
u
r
e
r
ep
r
esen
t
atio
n
o
f
th
e
m
eth
o
d
.
T
h
en
,
t
h
e
s
p
atial
atten
tio
n
m
ec
h
an
is
m
en
h
a
n
ce
d
lo
ca
l
s
em
an
tic
f
ea
tu
r
es
a
n
d
c
o
m
p
r
ess
ed
ir
r
elev
a
n
ce
f
ea
tu
r
e
d
ata.
Fin
ally
,
f
ea
tu
r
e
-
lev
e
l
f
u
s
io
n
was
em
p
lo
y
ed
to
m
in
im
ize
r
ed
u
n
d
an
t
f
ea
tu
r
es
an
d
en
h
an
ce
ex
ec
u
tio
n
p
er
f
o
r
m
a
n
ce
.
Ad
d
itio
n
ally
,
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
f
u
n
ctio
n
with
lab
el
s
m
o
o
th
in
g
was
u
tili
ze
d
to
im
p
r
o
v
e
class
if
icat
io
n
ac
cu
r
ac
y
.
T
h
e
in
tr
o
d
u
ce
d
m
eth
o
d
m
in
im
ize
d
th
e
in
f
lu
e
n
ce
o
f
h
u
g
e
s
im
ilar
ity
cl
ass
es
o
n
s
ce
n
e
class
if
ica
tio
n
.
Ho
wev
er
,
th
e
m
eth
o
d
ca
n
’
t
b
e
co
n
ce
n
tr
ate
d
o
n
th
e
s
p
atial
r
elatio
n
s
h
ip
o
f
lo
ca
l
f
ea
tu
r
es.
Fro
m
th
e
o
v
er
all
an
aly
s
is
,
th
e
ex
is
tin
g
m
eth
o
d
s
h
av
e
lim
itatio
n
s
lik
e
d
if
f
icu
lty
in
d
escr
ib
in
g
th
e
h
ig
h
s
em
an
tic
d
ata
d
u
e
to
lim
ited
r
ep
r
esen
tatio
n
,
is
s
u
e
o
f
in
ter
cl
ass
s
im
ilar
ity
,
d
if
f
icu
lty
in
in
tr
a
-
class
v
ar
iatio
n
s
an
d
in
ter
-
class
s
im
ilar
ities
an
d
ca
n
’
t
co
n
ce
n
tr
ate
o
n
th
e
s
p
atial
r
elatio
n
s
h
ip
o
f
lo
ca
l
f
ea
tu
r
es.
T
h
e
p
r
o
b
lem
o
f
with
in
-
c
lass
d
iv
er
s
ity
an
d
b
etwe
en
-
class
s
im
ilar
ity
d
eg
r
a
d
es
th
e
p
er
f
o
r
m
a
n
ce
o
f
r
e
m
o
t
e
s
en
s
in
g
im
ag
e
s
ce
n
e
class
if
icatio
n
wh
ich
n
ee
d
s
to
b
e
tack
led
.
I
n
th
is
r
esear
ch
,
th
e
p
r
o
p
o
s
ed
DPML
-
C
NN
m
eth
o
d
m
a
x
im
izes
in
ter
class
v
ar
iatio
n
an
d
m
in
im
izes
th
e
in
te
r
class
v
ar
ia
tio
n
.
T
h
e
m
et
r
ic
lear
n
i
n
g
in
C
NN
m
ak
es
th
e
m
eth
o
d
m
u
ch
m
o
r
e
d
is
cr
im
in
ativ
e
an
d
ca
p
tu
r
es
b
o
th
g
lo
b
al
an
d
lo
ca
l
f
ea
tu
r
es.
T
h
e
d
iv
er
s
i
ty
p
r
o
m
o
tio
n
in
m
etr
ic
lear
n
in
g
m
in
im
izes
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
La
n
d
s
ce
n
e
cla
s
s
ifica
tio
n
u
s
in
g
d
ivers
ity
p
r
o
mo
tin
g
metric le
a
r
n
in
g
-
co
n
vo
lu
tio
n
a
l
… (
K
a
mp
a
R
a
tn
a
B
a
b
u
)
271
o
v
er
lap
p
i
n
g
o
f
in
ter
class
s
im
ilar
ity
b
y
u
n
c
o
r
r
elatio
n
o
f
ev
e
r
y
s
ce
n
e
p
ar
am
eter
.
T
h
e
s
ig
n
if
ican
t
co
n
tr
ib
u
tio
n
s
o
f
th
e
r
esear
c
h
ar
e
m
e
n
tio
n
ed
as f
o
llo
ws:
−
T
h
e
d
iv
er
s
ity
p
r
o
m
o
tin
g
m
et
r
ic
lear
n
in
g
–
C
NN
(
DPML
-
C
NN)
m
eth
o
d
is
p
r
o
p
o
s
ed
t
o
class
if
y
lan
d
s
ce
n
e
r
em
o
te
s
en
s
in
g
im
ag
es.
T
h
e
C
NN
with
d
iv
er
s
ity
-
p
r
o
m
o
tin
g
m
etr
ic
lear
n
in
g
m
in
im
iz
ed
th
e
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
an
d
m
ad
e
th
e
m
et
h
o
d
m
u
ch
m
o
r
e
d
is
cr
im
in
ativ
e
.
−
T
h
e
m
etr
ic
lear
n
in
g
r
e
g
u
lar
izatio
n
u
s
ed
th
e
d
iv
er
s
ity
p
r
o
m
o
tin
g
ter
m
wh
ich
u
n
c
o
r
r
elate
d
ev
er
y
p
ar
am
eter
a
n
d
g
iv
es u
n
iq
u
e
in
f
o
r
m
atio
n
f
o
r
th
o
s
e
p
ar
a
m
eter
s
.
T
h
is
m
in
im
izes
th
e
o
v
e
r
lap
p
in
g
o
f
s
im
ilar
s
ce
n
es a
n
d
en
h
a
n
ce
s
th
e
m
eth
o
d
'
s
p
er
f
o
r
m
an
ce
.
−
T
h
e
s
ca
le
in
v
ar
ian
t
f
ea
tu
r
e
tr
an
s
f
o
r
m
(
SIFT
)
an
d
lo
ca
l
b
in
ar
y
p
atter
n
s
(
L
B
P)
tech
n
iq
u
e
s
ar
e
u
s
ed
f
o
r
ex
tr
ac
tin
g
m
ea
n
in
g
f
u
l
f
ea
tu
r
es
f
r
o
m
im
a
g
es
wh
ich
e
n
h
a
n
ce
d
th
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
o
f
th
e
DPML
-
C
NN
m
eth
o
d
.
T
h
is
m
an
u
s
cr
ip
t
is
o
r
g
an
ized
as
f
o
llo
ws:
Sectio
n
2
ex
p
lain
s
th
e
p
r
o
ce
s
s
o
f
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
.
Sectio
n
3
g
iv
es
th
e
r
esu
lts
an
d
co
m
p
ar
is
o
n
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
with
e
x
is
tin
g
m
eth
o
d
s
.
T
h
e
co
n
clu
s
io
n
o
f
th
is
r
esear
ch
is
g
iv
en
in
Sectio
n
4
.
2.
P
RO
P
O
SE
D
M
E
T
H
O
D
T
h
is
r
esear
ch
,
p
r
o
p
o
s
ed
a
DPML
-
C
NN
m
eth
o
d
to
class
if
y
th
e
lan
d
s
ce
n
e
im
ag
es
f
r
o
m
r
em
o
te
s
en
s
in
g
s
ce
n
e
im
ag
es.
T
h
e
U
C
Me
r
ce
d
,
AI
D,
an
d
NW
PU
R
E
SIS
C
4
5
d
atasets
ar
e
u
tili
z
ed
in
th
is
r
esear
ch
an
d
th
e
s
ig
n
if
ican
t
f
ea
tu
r
es
f
r
o
m
th
e
im
ag
es
ar
e
ex
tr
ac
ted
b
y
S
I
F
T
a
n
d
L
B
P
m
e
t
h
o
d
s
.
T
h
e
n
,
t
h
e
e
x
t
r
a
c
t
e
d
f
e
a
t
u
r
e
s
a
r
e
c
l
ass
i
f
i
e
d
b
y
u
s
in
g
t
h
e
D
P
M
L
-
C
N
N
m
e
t
h
o
d
.
F
i
g
u
r
e
1
d
escr
ib
es
th
e
p
r
o
ce
s
s
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
f
r
am
ewo
r
k
.
Fig
u
r
e
1
.
Pro
ce
s
s
o
f
p
r
o
p
o
s
ed
m
eth
o
d
f
r
am
ewo
r
k
2
.
1
.
Da
t
a
s
et
T
h
e
d
atasets
u
s
ed
in
th
e
r
esea
r
ch
f
o
r
lan
d
s
ce
n
e
im
ag
e
class
if
icatio
n
ar
e
UC
M
[
2
1
]
,
AI
D
[
2
2
]
,
a
n
d
NW
PU
R
E
SIS
C
4
5
[
2
3
]
d
atasets
.
T
h
ese
th
r
ee
d
atasets
co
n
tai
n
r
em
o
te
s
en
s
in
g
s
ce
n
e
im
a
g
e
s
wh
ich
ar
e
u
tili
ze
d
in
th
is
r
esear
ch
to
class
if
y
lan
d
s
ce
n
es.
T
h
e
d
etailed
e
x
p
lan
a
tio
n
o
f
th
r
ee
d
atasets
is
d
escr
ib
ed
b
elo
w:
2
.
1
.
1
.
UCM da
t
a
s
et
T
h
e
UC
M
d
ataset
i
s
av
ailab
l
e
th
r
o
u
g
h
Un
iv
er
s
ity
o
f
C
alif
o
r
n
ia
Me
r
ce
d
a
n
d
it
h
as
2
1
0
0
r
em
o
te
s
en
s
in
g
s
ce
n
e
im
ag
es.
T
h
e
r
em
o
te
s
en
s
in
g
im
ag
es
ar
e
s
e
p
ar
ated
in
to
2
1
s
ce
n
es
an
d
e
v
er
y
class
h
as
1
0
0
im
ag
es
o
f
s
ize
2
5
6
x
2
5
6
p
ix
e
ls
.
T
h
e
s
p
atial
r
eso
lu
tio
n
o
f
i
m
ag
e
is
0
.
3
m
p
er
p
ix
el
i
n
co
l
o
r
s
p
ac
e
o
f
r
ed
g
r
ee
n
bl
u
e
(
R
GB
)
.
2
.
1
.
2
.
AID da
t
a
s
et
T
h
e
AI
D
d
ataset
is
d
ev
elo
p
e
d
th
r
o
u
g
h
W
u
h
an
Un
i
v
er
s
ity
an
d
it
h
as
1
0
,
0
0
0
r
em
o
te
s
en
s
in
g
s
ce
n
e
im
ag
es.
T
h
e
r
em
o
te
s
en
s
in
g
s
ce
n
e
im
ag
es
ar
e
s
ep
a
r
ated
i
n
to
3
0
s
ce
n
e
class
es
with
th
e
s
ize
o
f
600
×
600
p
ix
els
an
d
t
h
e
n
u
m
b
er
o
f
im
ag
es
in
ev
er
y
class
ar
e
v
a
r
ied
f
r
o
m
2
0
0
to
4
0
0
.
T
h
e
s
p
atial
r
eso
lu
tio
n
v
a
r
ied
f
r
o
m
8
m
to
0
.
5
m
p
er
p
ix
el.
2
.
1
.
3
.
NWP
URES
I
SC4
5
da
t
a
s
et
T
h
e
NW
PU
d
atase
t
is
d
ev
el
o
p
ed
th
r
o
u
g
h
No
r
th
wester
n
Po
ly
tech
n
ical
Un
iv
er
s
ity
an
d
co
n
tain
s
3
1
,
5
0
0
r
em
o
te
-
s
en
s
in
g
s
ce
n
e
i
m
ag
es
in
co
lo
r
s
p
ac
e
o
f
R
GB
.
T
h
e
r
em
o
te
s
en
s
in
g
s
ce
n
e
im
ag
es
ar
e
s
ep
ar
ated
in
to
4
5
s
ce
n
e
class
es
an
d
ev
er
y
class
co
n
tain
s
7
0
0
s
ce
n
e
im
ag
es
with
th
e
s
ize
o
f
2
5
6
x
2
5
6
p
ix
els.
T
h
e
s
p
atial
r
eso
lu
tio
n
v
ar
ied
f
r
o
m
0
.
2
m
to
3
0
m
.
T
ab
le
1
r
ep
r
esen
ts
th
e
d
escr
ip
tio
n
o
f
all
th
r
ee
d
a
tasets
.
T
h
e
im
ag
es
f
r
o
m
th
e
t
h
r
ee
d
atasets
ar
e
g
iv
en
as
r
aw
in
p
u
t
i
m
ag
es
to
th
e
f
ea
tu
r
e
ex
tr
ac
tio
n
s
tag
e
to
e
x
tr
ac
t
th
e
lo
ca
l
a
n
d
p
ix
el
-
b
ased
f
ea
tu
r
es f
o
r
class
if
icatio
n
.
T
h
e
d
etailed
ex
p
lan
ati
o
n
o
f
f
ea
tu
r
e
e
x
tr
ac
tio
n
is
d
es
cr
ib
ed
as f
o
llo
ws.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
37
,
No
.
1
,
J
an
u
ar
y
20
25
:
269
-
2
7
8
272
T
ab
le
1
.
Data
s
et
d
escr
ip
tio
n
D
a
t
a
s
e
t
To
t
a
l
i
m
a
g
e
s
N
o
.
o
f
sc
e
n
e
s
S
i
z
e
o
f
i
ma
g
e
U
C
M
[
2
0
]
2
1
0
0
21
2
5
6
x
2
5
6
A
I
D
[
2
1
]
1
0
,
0
0
0
30
6
0
0
x
6
0
0
N
W
P
U
R
ESI
S
C
4
5
[
2
2
]
3
1
,
5
0
0
45
2
5
6
x
2
5
6
2
.
2
.
F
e
a
t
ure
ex
t
r
a
ct
io
n
Featu
r
e
ex
tr
ac
tio
n
is
a
p
r
o
ce
s
s
o
f
co
n
v
er
tin
g
r
aw
im
ag
e
d
a
ta
to
ce
r
tain
m
ea
n
in
g
f
u
l
r
ep
r
e
s
en
tatio
n
s
wh
ich
m
in
im
izes
th
e
d
im
en
s
io
n
ality
r
ed
u
ctio
n
.
I
n
Gen
er
al,
th
e
h
ig
h
d
im
en
s
io
n
al
n
atu
r
e
o
f
im
ag
es
ca
n
lead
to
a
r
ed
u
ctio
n
in
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
.
T
h
e
f
ea
tu
r
e
e
x
tr
a
ctio
n
co
n
v
er
ts
h
ig
h
-
d
im
en
s
io
n
al
d
ata
to
lo
w
-
lev
el
d
ata
wh
en
ex
tr
ac
tin
g
s
ig
n
if
ica
n
t
d
ata
f
r
o
m
s
ce
n
e
im
ag
es.
I
n
th
is
r
esear
ch
,
SIFT
an
d
L
B
P
f
ea
tu
r
e
ex
tr
ac
tio
n
tech
n
iq
u
es a
r
e
u
s
ed
to
ex
tr
ac
t
lo
ca
l a
n
d
p
i
x
el
-
b
ased
f
ea
t
u
r
es.
2
.
2
.
1
.
L
o
ca
l
bin
a
ry
pa
t
t
er
ns
(
L
B
P
)
T
h
e
L
B
P
id
en
tifie
d
u
n
if
o
r
m
L
B
P
as
a
cr
itical
f
ea
tu
r
e
th
at
r
ep
r
esen
ts
im
a
g
e
te
x
tu
r
e.
T
h
e
u
n
if
o
r
m
L
B
P
is
em
p
lo
y
ed
f
o
r
g
en
er
at
in
g
o
cc
u
r
r
en
ce
s
o
f
h
is
to
g
r
am
s
f
o
r
th
e
r
ep
r
esen
tatio
n
o
f
te
x
tu
r
e
f
ea
t
u
r
es.
T
h
e
L
B
P
ch
ar
ac
ter
ized
th
e
im
ag
e
b
y
s
p
atial
d
ata
o
f
im
ag
e
tex
tu
r
e
s
tr
u
ctu
r
e
[
2
4
]
,
[
2
5
]
.
T
h
e
L
B
P
i
s
m
ea
s
u
r
ed
b
y
th
r
esh
o
ld
in
g
t
h
e
n
eig
h
b
o
r
{
}
=
0
−
1
p
ix
els
alo
n
g
th
e
m
id
d
le
p
ix
el
f
o
r
ex
ec
u
tin
g
a
n
n
-
b
it
b
in
ar
y
n
u
m
b
er
th
at
is
ch
an
g
ed
to
d
ec
i
m
al
an
d
n
u
m
er
ical
ex
p
r
ess
io
n
is
g
iv
en
as
(
1
)
.
T
h
e
=
(
−
)
r
ep
r
esen
ts
v
ar
i
an
ce
b
etwe
en
n
eig
h
b
o
r
a
n
d
ce
n
ter
p
ix
els
r
ep
r
esen
tin
g
th
e
s
p
atial
ar
ch
itectu
r
e
o
f
ce
n
ter
p
o
s
itio
n
with
lo
ca
l
v
ar
ian
ce
v
ec
to
r
[
0
,
1
,
…
,
−
1
]
.
T
h
e
L
B
P g
en
er
ate
d
a
h
is
to
g
r
am
an
d
wh
ich
is
g
iv
e
n
in
(
2
)
.
,
(
)
=
∑
(
−
)
2
=
∑
(
)
2
,
(
)
=
{
1
,
>
0
0
,
<
0
−
1
=
0
−
1
=
0
(
1
)
(
(
)
)
=
∑
∑
(
,
,
∈
[
0
,
]
)
,
(
,
)
=
{
1
,
0
,
=
0
=
0
(
2
)
T
h
e
r
ep
r
esen
ts
th
e
h
i
g
h
est
p
a
tter
n
n
u
m
b
er
o
f
L
B
P.
T
h
e
L
B
P
p
ar
titi
o
n
s
th
e
im
ag
e
to
t
h
e
f
i
x
ed
s
ize
o
f
g
r
id
ce
lls
f
o
r
ac
co
m
p
lis
h
in
g
th
e
p
o
o
lin
g
o
f
lo
ca
l
tex
tu
r
e
d
escr
ip
to
r
s
.
T
h
e
L
B
P
ch
ar
ac
ter
ized
th
e
im
ag
e
b
y
s
p
atial
d
ata
o
f
im
ag
e
tex
t
u
r
e
s
tr
u
ctu
r
e.
2
.
2
.
2
.
Sca
le
inv
a
ria
nt
f
e
a
t
ure
t
ra
ns
f
o
rm
(
SI
F
T
)
T
h
e
SIFT
m
eth
o
d
is
a
m
ajo
r
l
y
u
tili
ze
d
s
h
ap
e
f
ea
tu
r
e
ex
tr
a
ctio
n
tech
n
iq
u
e.
T
h
e
tech
n
iq
u
e
is
a
k
ey
p
o
in
t
d
etec
to
r
a
n
d
d
escr
ip
to
r
tech
n
iq
u
e
f
o
r
ex
tr
ac
tin
g
m
ea
n
i
n
g
f
u
l
f
ea
tu
r
es
f
r
o
m
im
a
g
es.
T
h
at
is
h
u
g
ely
r
o
b
u
s
t
to
s
ca
lin
g
an
d
o
r
ien
tatio
n
o
f
th
e
im
ag
e
an
d
is
in
v
ar
ian
t
f
o
r
illu
m
in
atio
n
ch
an
g
es.
T
h
e
SIFT
ex
tr
ac
ted
th
e
h
ig
h
est
f
ea
tu
r
es
f
r
o
m
lo
w
-
lev
el
r
eso
lu
tio
n
im
ag
es
[
2
6
]
,
[
2
7
]
.
T
h
e
SIFT
ex
tr
ac
te
d
th
e
1
2
8
f
ea
tu
r
es
f
r
o
m
r
em
o
te
s
en
s
in
g
s
ce
n
e
im
ag
es
b
y
f
ilter
in
g
m
eth
o
d
th
at
p
r
o
ce
s
s
ed
in
f
o
u
r
p
h
ases
.
T
h
e
in
itial
p
h
ase
d
etec
ted
s
ig
n
if
ican
t
p
o
s
itio
n
s
f
r
o
m
an
im
ag
e
b
y
d
if
f
er
en
ce
o
f
th
e
Gau
s
s
ian
(
Do
G)
m
eth
o
d
.
Nex
t,
lo
ca
lizatio
n
is
p
r
o
ce
s
s
ed
to
d
ete
r
m
in
e
s
ig
n
i
f
ican
t
f
ea
tu
r
es.
I
n
f
in
al
p
h
as
e,
ex
ec
u
ted
k
ey
p
o
in
ts
ar
e
c
h
an
g
ed
to
f
ea
tu
r
e
v
ec
to
r
s
.
T
h
e
e
x
tr
ac
ted
f
ea
t
u
r
es f
r
o
m
t
h
e
L
B
P a
n
d
SIFT
m
eth
o
d
s
ar
e
f
u
s
ed
f
o
r
t
h
e
class
if
icatio
n
p
r
o
ce
s
s
.
2
.
3
.
Cla
s
s
if
ica
t
io
n by
DP
M
L
-
CNN
m
et
ho
d
T
h
e
class
if
icatio
n
o
f
lan
d
s
ce
n
es
is
p
er
f
o
r
m
ed
b
y
C
NN
with
d
iv
er
s
ity
-
p
r
o
m
o
tin
g
m
etr
ic
lear
n
in
g
.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
s
ce
n
e
c
lass
if
icatio
n
is
en
h
an
ce
d
b
ec
au
s
e
o
f
m
ea
n
in
g
f
u
l
f
ea
tu
r
es
lear
n
ed
b
y
C
NN
m
eth
o
d
s
.
T
h
e
p
r
o
p
o
s
ed
C
NN
with
d
iv
er
s
ity
-
p
r
o
m
o
tin
g
m
etr
ic
lear
n
in
g
m
i
n
im
ized
th
e
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
an
d
lear
n
ed
t
h
e
m
o
d
el
to
b
e
d
is
cr
im
in
ativ
e.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
m
ap
s
th
e
im
ag
es
f
r
o
m
s
im
ilar
s
ce
n
e
class
es
clo
s
est
an
d
im
ag
es o
f
v
ar
io
u
s
ca
teg
o
r
ies
as
f
ar
as
p
o
s
s
ib
le
[
2
8
]
,
[
2
9
]
.
A
d
etailed
e
x
p
lan
atio
n
o
f
d
is
cr
im
in
ativ
e
C
NN
an
d
d
iv
er
s
ity
-
p
r
o
m
o
tin
g
m
etr
ic
lear
n
in
g
is
d
escr
ib
ed
b
elo
w.
Fig
u
r
e
2
d
escr
ib
es
th
e
p
r
o
ce
s
s
o
f
th
e
p
r
o
p
o
s
ed
DPML
-
C
NN
m
eth
o
d
.
2
.
3
.
1
.
Div
er
s
it
y
pro
mo
t
ing
met
ric
lea
rning
–
CNN
(
DP
M
L
-
CNN)
T
h
e
p
r
o
p
o
s
ed
DPML
-
C
NN
i
s
u
s
ed
f
o
r
lan
d
s
ce
n
e
class
if
icatio
n
to
tack
le
is
s
u
es
o
f
w
ith
in
-
class
d
iv
er
s
ity
an
d
b
etwe
en
-
class
d
iv
er
s
ity
.
Her
e
ex
tr
ac
ted
th
e
m
ea
n
in
g
f
u
l
C
NN
f
ea
tu
r
es
f
o
r
en
h
a
n
cin
g
th
e
p
er
f
o
r
m
an
ce
o
f
class
if
icatio
n
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
m
in
i
m
ized
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
wh
ich
is
an
er
r
o
r
o
f
s
o
f
tm
ax
class
if
icatio
n
f
r
o
m
th
e
last
f
u
lly
co
n
n
ec
ted
(
FC
)
lay
er
u
tili
ze
d
in
class
ica
l
C
NN
m
eth
o
d
s
.
T
h
e
d
iv
er
s
ity
-
p
r
o
m
o
tin
g
m
etr
ic
lear
n
in
g
o
n
C
NN
f
ea
tu
r
es
m
a
k
es
th
e
m
et
h
o
d
m
u
c
h
m
o
r
e
d
is
cr
im
in
ativ
e
an
d
m
a
p
s
th
e
s
im
ilar
s
ce
n
e
class
es c
lo
s
e
r
to
g
eth
er
an
d
d
is
tan
t scen
e
class
es a
s
f
ar
as p
o
s
s
ib
le.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
La
n
d
s
ce
n
e
cla
s
s
ifica
tio
n
u
s
in
g
d
ivers
ity
p
r
o
mo
tin
g
metric le
a
r
n
in
g
-
co
n
vo
lu
tio
n
a
l
… (
K
a
mp
a
R
a
tn
a
B
a
b
u
)
273
Fig
u
r
e
2
.
Pro
ce
s
s
o
f
p
r
o
p
o
s
ed
DPML
-
C
NN
m
eth
o
d
2
.
3
.
2
.
L
ea
rning
dis
cr
im
ina
t
i
v
e
CNNs
C
o
n
s
id
er
=
{
|
=
1
,
2
,
…
,
}
as
a
s
et
o
f
tr
ain
i
n
g
s
am
p
les
an
d
=
{
|
=
1
,
2
,
…
,
}
as
a
s
et
o
f
lab
els
in
X,
wh
er
e
N
r
e
p
r
ese
n
ts
wh
o
le
co
u
n
t
o
f
tr
ain
i
n
g
s
a
m
p
les,
∈
r
ep
r
esen
ts
lab
el
v
ec
t
o
r
o
f
g
r
o
u
n
d
tr
u
th
in
s
am
p
le
alo
n
g
o
n
e
co
m
p
o
n
en
t
an
d
C
r
ep
r
esen
ts
th
e
wh
o
le
co
u
n
t
o
f
im
ag
e
s
ce
n
e
class
es.
T
h
e
DPML
-
C
NN
m
eth
o
d
co
n
tain
s
+
1
lay
er
s
.
T
h
e
n
etwo
r
k
p
ar
am
eter
s
ar
e
=
{
1
,
…
,
,
+
1
}
an
d
=
{
1
,
…
,
,
+
1
}
,
h
er
e
r
ep
r
esen
ts
f
ilter
weig
h
ts
o
f
ℎ
lay
er
an
d
r
ep
r
esen
ts
r
esp
ec
tiv
e
b
iases
,
=
1
,
…
,
+
1
.
T
h
e
(
+
1
)
lay
er
r
ep
r
esen
ts
s
o
f
tm
ax
class
if
icatio
n
an
d
ℎ
lay
er
r
e
p
r
esen
ts
r
esu
lt
o
f
DPML
-
C
NN
m
eth
o
d
.
T
h
e
in
p
u
t
an
d
r
esu
lt
o
f
s
o
f
tm
ax
lay
er
+
1
(
)
,
(
)
r
ep
r
esen
ts
DPML
-
C
NN
f
ea
tu
r
e
lay
er
an
d
w
h
o
le
in
ter
m
e
d
iate
lay
er
s
(
)
is
g
iv
en
as (
3
)
–
(
5
)
.
+
1
(
)
=
+
1
(
+
1
(
)
+
+
1
)
(
3
)
(
)
=
(
−
1
(
)
+
)
‖
(
−
1
(
)
+
)
‖
2
(
4
)
(
)
=
(
(
)
+
)
(
5
)
T
h
e
r
ep
r
esen
ts
elem
en
t
-
wis
e
n
o
n
lin
ea
r
ac
tiv
atio
n
f
u
n
ctio
n
s
lik
e
s
o
f
tm
ax
.
T
h
e
+
1
r
ep
r
esen
ts
a
s
o
f
tm
ax
f
u
n
cti
o
n
,
n
e
x
t
f
o
r
e
x
ec
u
tin
g
th
e
d
is
tan
ce
in
DPML
-
C
NN
is
L
2
n
o
r
m
alize
d
f
o
r
elim
in
atin
g
s
ca
le
v
ar
ian
ce
.
T
h
is
r
esear
ch
p
r
o
p
o
s
ed
a
ter
m
th
at
co
n
s
is
t
s
o
f
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
,
d
iv
er
s
ity
p
r
o
m
o
tin
g
m
etr
ic
lear
n
in
g
,
an
d
weig
h
t
d
ec
ay
ter
m
.
Nu
m
er
ical
ex
p
r
ess
io
n
is
g
iv
en
as
(
6
)
.
T
h
e
1
an
d
2
r
ep
r
ese
n
ts
p
ar
am
eter
s
o
f
tr
ad
eo
f
f
wh
ich
co
n
tr
o
ls
th
e
r
elativ
e
s
ig
n
if
ican
ce
o
f
th
r
ee
ter
m
s
.
T
h
e
d
etailed
ex
p
lan
atio
n
o
f
th
r
ee
ter
m
s
is
d
escr
ib
ed
in
th
e
b
elo
w
s
u
b
s
ec
t
io
n
s
.
=
min
(
1
(
,
,
)
+
1
2
2
(
,
,
)
+
2
2
3
(
,
)
(
6
)
A.
C
r
o
s
s
-
en
tr
o
p
y
lo
s
s
T
h
e
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
is
d
ete
r
m
in
ed
as
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
f
u
n
ctio
n
wh
ich
m
in
im
ized
clas
s
if
icatio
n
er
r
o
r
i
n
tr
ain
in
g
s
am
p
les.
T
h
e
(
,
l
og
+
1
(
)
)
r
ep
r
esen
ts
th
e
in
s
id
e
p
r
o
d
u
c
t
o
f
an
d
l
og
+
1
(
)
an
d
N
r
ep
r
esen
ts
th
e
co
u
n
t o
f
t
r
ain
in
g
s
am
p
les in
X.
an
d
n
u
m
e
r
ical
ex
p
r
ess
io
n
is
g
iv
en
as (
7
)
.
1
(
,
,
)
=
−
1
∑
(
,
l
og
+
1
(
)
)
=
1
(
7
)
B.
Div
er
s
ity
p
r
o
m
o
tin
g
m
etr
ic
le
ar
n
in
g
T
h
is
lear
n
in
g
en
f
o
r
ce
d
a
C
NN
m
eth
o
d
to
b
e
m
u
c
h
m
o
r
e
d
is
cr
im
in
ativ
e,
wh
ich
en
h
an
c
es
a
f
ea
tu
r
e
r
ep
r
esen
tatio
n
th
at
h
as littl
e
s
c
atter
o
f
in
tr
ac
lass
an
d
h
u
g
e
s
e
p
ar
atio
n
o
f
in
ter
class
.
Giv
en
e
v
er
y
p
air
e
d
tr
ain
in
g
s
am
p
le
(
,
)
,
its
d
is
tan
ce
o
f
p
air
-
wis
e
f
ea
tu
r
e
is
ca
lcu
lated
th
r
o
u
g
h
e
x
ec
u
tin
g
E
u
clid
ea
n
d
is
tan
ce
am
o
n
g
C
NN
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
.
T
h
e
n
u
m
e
r
ical
ex
p
r
ess
io
n
f
o
r
co
m
p
u
tin
g
E
u
clid
ea
n
d
is
tan
ce
is
g
iv
en
as (
8
)
.
(
,
)
=
‖
(
)
−
(
)
‖
2
(
8
)
T
o
ex
p
lo
it
d
is
cr
im
in
ativ
e
f
ea
tu
r
e
r
ep
r
esen
tatio
n
in
o
u
tp
u
t
lay
er
o
f
th
e
DPML
-
C
NN
m
eth
o
d
,
th
e
d
is
tan
ce
s
am
o
n
g
th
e
s
am
e
p
ai
r
s
ar
e
less
er
th
an
d
is
s
im
ilar
p
air
s
an
d
th
er
e
is
a
h
u
g
e
m
ar
g
i
n
am
o
n
g
s
im
ilar
an
d
d
is
s
im
ilar
p
air
s
d
ev
elo
p
e
d
b
y
im
ag
e
s
ce
n
e
ca
teg
o
r
ies
Par
ticu
lar
ly
,
wh
et
h
er
two
im
ag
es
s
h
ar
ed
s
im
ilar
s
ce
n
e
lab
el
it
is
co
n
s
id
er
in
g
as
s
im
ilar
p
air
o
r
else
co
n
s
id
er
as
d
is
s
im
ilar
p
air
.
Fo
r
p
r
o
tectin
g
th
e
is
s
u
e
o
f
d
is
s
im
ilar
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
37
,
No
.
1
,
J
an
u
ar
y
20
25
:
269
-
2
7
8
274
p
air
s
b
ein
g
ch
o
s
en
with
a
s
im
ilar
n
u
m
b
er
o
f
th
e
s
am
e
p
air
s
.
T
h
e
an
d
r
ep
r
esen
ts
s
im
ilar
s
ce
n
e
class
,
its
f
ea
tu
r
e
d
is
tan
ce
is
r
ep
r
esen
ted
as
(
,
)
is
s
m
al
ler
th
an
u
p
-
m
ar
g
in
1
wh
eth
er
an
d
r
ep
r
esen
ts
v
ar
i
o
u
s
s
ce
n
e
ca
teg
o
r
ies,
its
f
ea
tu
r
e
d
i
s
tan
ce
is
h
ig
h
er
th
an
t
h
e
d
o
w
n
-
m
ar
g
in
2
.
Nu
m
er
ical
ex
p
r
ess
io
n
is
g
iv
en
as (
9
)
.
{
2
(
,
)
<
1
,
=
2
(
,
)
>
2
,
≠
(
9
)
T
h
e
1
is
u
tili
ze
d
f
o
r
p
en
alizin
g
s
am
e
-
p
air
d
is
tan
ce
s
,
2
is
u
tili
ze
d
f
o
r
co
n
s
tr
ain
in
g
d
is
s
im
ilar
d
is
tan
ce
p
air
s
in
tr
ain
in
g
an
d
2
is
h
ig
h
er
th
an
1
.
Fo
r
m
in
im
izi
n
g
th
e
co
u
n
t
o
f
p
ar
am
ete
r
s
in
th
e
p
r
o
p
o
s
ed
m
eth
o
d
,
im
p
lem
en
ted
in
te
r
m
e
d
iate
p
ar
am
eter
f
o
r
m
e
r
g
in
g
1
an
d
2
.
Par
ticu
lar
ly
,
s
et
1
=
−
0
.
05
an
d
2
=
+
0
.
05
an
d
n
u
m
e
r
ical
ex
p
r
ess
io
n
is
g
iv
en
as
(
1
0
)
.
T
h
e
r
ep
r
esen
ts
lab
el
in
d
icato
r
f
o
r
p
air
e
d
d
a
ta
(
,
)
an
d
n
u
m
er
ical
ex
p
r
ess
io
n
is
g
iv
en
as (
1
1
)
.
0
.
05
−
(
−
2
(
,
)
)
<
0
(
1
0
)
=
{
+
1
,
=
−
1
,
≠
(
1
1
)
T
h
er
e
is
a
co
n
n
ec
tio
n
am
o
n
g
ev
er
y
s
im
ilar
an
d
d
is
s
im
il
ar
p
air
lear
n
ed
in
C
NN
f
ea
t
u
r
e
s
p
ac
e.
T
h
r
o
u
g
h
em
p
lo
y
in
g
th
e
c
o
n
s
tr
ain
t
to
e
v
er
y
s
im
ilar
an
d
d
is
s
im
ilar
p
air
i
n
tr
ain
i
n
g
p
h
ase
o
b
tain
ed
h
in
g
e
lo
s
s
f
u
n
ctio
n
o
f
d
iv
er
s
ity
p
r
o
m
o
tin
g
m
etr
ic
lear
n
in
g
an
d
n
u
m
er
ic
al
ex
p
r
ess
io
n
is
g
iv
en
as (
1
2
)
.
2
(
,
,
)
=
∑
ℎ
(
0
.
05
−
(
−
‖
(
)
−
(
)
‖
2
2
)
)
,
(
1
2
)
C.
Div
er
s
ity
p
r
o
m
o
tin
g
an
d
weig
h
t d
ea
cy
T
h
e
g
en
er
al
r
em
o
te
s
en
s
in
g
s
ce
n
es
ca
n
’
t
g
iv
e
en
o
u
g
h
s
am
p
l
es
f
o
r
tr
ain
in
g
,
th
is
r
esear
ch
en
co
u
r
a
g
es
th
e
m
etr
ic
p
ar
am
eter
f
ac
to
r
s
t
o
b
e
d
iv
er
s
if
ied
f
o
r
en
h
an
ci
n
g
ca
p
ab
ilit
y
o
f
r
ep
r
esen
tatio
n
an
d
ca
p
tu
r
es
m
u
ch
d
is
cr
im
in
ativ
e
d
ata
f
r
o
m
t
h
e
r
estricte
d
co
u
n
t
o
f
s
am
p
les
f
o
r
ac
q
u
ir
i
n
g
g
o
o
d
p
er
f
o
r
m
a
n
ce
.
T
o
m
ee
t
s
p
ec
if
ic
r
eq
u
ir
em
e
n
ts
,
th
e
d
iv
er
s
ity
-
p
r
o
m
o
tin
g
p
r
io
r
s
ar
e
im
p
lem
en
ted
as
r
eg
u
lar
izatio
n
f
ac
to
r
s
in
p
ar
am
eter
s
f
o
r
b
etter
r
ep
r
esen
tatio
n
.
I
t
f
o
cu
s
ed
o
n
d
iv
er
s
if
icatio
n
am
o
n
g
lear
n
ed
p
ar
am
eter
s
.
B
y
im
p
lem
en
tin
g
d
iv
er
s
ity
-
p
r
o
m
o
tin
g
p
r
io
r
s
in
ML
,
e
v
er
y
p
ar
am
eter
f
ac
to
r
r
ep
r
esen
ts
u
n
iq
u
e
i
n
f
o
r
m
atio
n
f
r
o
m
s
ce
n
es
an
d
all
f
ac
to
r
s
ar
e
co
m
b
in
ed
in
to
a
lar
g
e
p
r
o
p
o
r
ti
o
n
o
f
tr
ain
in
g
d
ata.
He
n
ce
,
m
u
ch
d
ata
is
ca
p
t
u
r
ed
t
h
r
o
u
g
h
d
iv
er
s
if
ied
m
eth
o
d
s
,
an
d
p
er
f
o
r
m
an
ce
is
en
h
an
ce
d
.
I
t
is
d
ev
elo
p
ed
f
o
r
m
in
im
izin
g
weig
h
t
m
ag
n
itu
d
es
o
f
W
an
d
B
an
d
it
i
s
u
s
ed
to
p
r
ev
en
t
o
v
er
-
f
itti
n
g
,
th
e
n
u
m
e
r
ical
ex
p
r
ess
io
n
is
g
iv
en
as (
1
3
)
.
3
(
,
)
=
∑
(
‖
‖
2
+
‖
‖
2
2
)
+
1
=
1
(
1
3
)
B
y
u
t
i
li
z
i
n
g
d
i
v
e
r
s
i
t
y
-
p
r
o
m
o
t
in
g
m
e
t
r
i
c
l
e
a
r
n
i
n
g
i
n
C
N
N
,
i
t
t
a
c
k
l
es
t
h
e
is
s
u
e
o
f
wi
t
h
i
n
-
c
la
s
s
d
i
v
e
r
s
i
t
y
a
n
d
b
e
t
w
e
e
n
-
c
l
a
s
s
s
i
m
i
l
a
r
i
t
y
.
T
h
e
o
v
e
r
l
a
p
p
i
n
g
o
f
s
i
m
i
l
a
r
i
t
y
s
c
e
n
es
i
s
t
a
c
k
l
e
d
b
y
u
s
i
n
g
d
i
v
e
r
s
i
t
y
p
r
o
m
o
t
i
n
g
i
n
m
e
t
r
i
c
le
a
r
n
i
n
g
w
h
i
c
h
u
n
c
o
r
r
e
l
a
t
e
d
e
v
e
r
y
f
a
c
t
o
r
a
n
d
r
e
p
r
es
e
n
ts
u
n
i
q
u
e
i
n
f
o
r
m
a
ti
o
n
f
o
r
e
v
e
r
y
s
c
e
n
e
.
T
h
is
m
i
n
i
m
i
z
es
t
h
e
p
r
o
b
a
b
i
l
i
t
y
o
f
o
v
e
r
l
a
p
p
i
n
g
a
n
d
i
n
c
r
e
as
e
s
t
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
t
h
e
m
e
t
h
o
d
.
B
y
u
s
i
n
g
t
h
e
D
PM
L
-
C
N
N
m
e
t
h
o
d
,
i
t
cl
a
s
s
i
f
ie
d
t
h
e
la
n
d
s
c
e
n
e
s
f
r
o
m
r
e
m
o
t
e
s
e
n
s
i
n
g
i
m
a
g
e
s
wi
t
h
h
i
g
h
c
l
as
s
i
f
i
c
at
io
n
p
e
r
f
o
r
m
a
n
c
e
.
3.
E
XP
E
R
I
M
E
N
T
A
L
ANA
L
Y
SI
S
T
h
e
p
r
o
p
o
s
ed
DPML
-
C
NN
m
eth
o
d
is
s
im
u
lated
with
th
e
en
v
ir
o
n
m
e
n
t
o
f
MA
T
L
AB
2
0
2
0
a
an
d
with
s
y
s
tem
r
eq
u
ir
em
en
ts
ar
e
a
W
in
d
o
ws
1
0
o
p
er
atin
g
s
y
s
tem
,
1
6
GB
o
f
R
AM
,
an
d
an
i5
p
r
o
ce
s
s
o
r
.
T
h
e
ev
alu
atio
n
m
etr
ics
u
s
ed
f
o
r
a
n
aly
zin
g
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
ar
e
Acc
u
r
ac
y
(
%)
a
n
d
Kap
p
a
C
o
ef
f
icien
t
(
%)
with
th
r
ee
d
atasets
s
u
ch
as
UC
Me
r
ce
d
,
AI
D,
an
d
NW
PU
R
E
SIS
C
4
5
.
T
h
e
ac
c
u
r
ac
y
is
r
ep
r
esen
ted
as
a
co
u
n
t
o
f
ac
cu
r
ately
class
if
ied
im
ag
es
s
ep
ar
ated
th
r
o
u
g
h
a
wh
o
le
n
u
m
b
er
o
f
test
in
g
im
ag
es;
th
is
r
ef
lecte
d
th
e
o
v
er
all
class
if
icatio
n
p
er
f
o
r
m
an
ce
o
f
th
e
m
eth
o
d
.
T
h
e
Kap
p
a
co
e
f
f
icien
t
d
ef
in
es
th
e
r
atio
o
f
er
r
o
r
m
in
im
izatio
n
am
o
n
g
class
if
icatio
n
an
d
f
u
lly
r
an
d
o
m
class
if
icatio
n
.
3
.
1
.
Q
ua
ntit
a
t
iv
e
a
nd
qu
a
lita
t
iv
e
a
na
ly
s
is
T
h
e
p
r
o
p
o
s
ed
DPML
-
C
NN
m
eth
o
d
is
ev
alu
ated
with
ev
alu
atio
n
m
etr
ics
o
f
ac
cu
r
a
cy
,
k
ap
p
a
co
ef
f
icien
t,
p
r
ec
is
io
n
,
an
d
f
1
-
s
co
r
e
with
th
r
ee
d
atasets
.
T
h
e
ev
alu
atio
n
is
p
e
r
f
o
r
m
ed
b
ased
o
n
d
if
f
er
en
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
La
n
d
s
ce
n
e
cla
s
s
ifica
tio
n
u
s
in
g
d
ivers
ity
p
r
o
mo
tin
g
metric le
a
r
n
in
g
-
co
n
vo
lu
tio
n
a
l
… (
K
a
mp
a
R
a
tn
a
B
a
b
u
)
275
tr
ain
in
g
r
atio
s
f
o
r
t
h
r
ee
d
if
f
er
en
t
d
atasets
.
T
h
e
s
ep
ar
ated
tr
ain
in
g
r
atio
s
f
o
r
UC
M
d
ataset
ar
e
5
0
%
an
d
8
0
%,
f
o
r
A
I
D
d
ataset
ar
e
2
0
%
an
d
5
0
%,
f
o
r
NW
PU
R
E
SISC
4
5
d
ataset
ar
e
1
0
%
an
d
2
0
%.
T
h
e
ex
is
tin
g
n
eu
r
al
n
etwo
r
k
s
u
s
ed
f
o
r
ev
al
u
atin
g
th
e
p
r
o
p
o
s
ed
DPML
-
C
NN
m
e
th
o
d
ar
e
d
ee
p
n
eu
r
al
n
etwo
r
k
(
DNN)
,
C
NN,
an
d
r
eg
io
n
-
b
ased
C
NN
(
R
C
NN)
.
T
h
r
ee
d
if
f
er
en
t
ta
b
les
ar
e
d
es
cr
ib
ed
b
el
o
w
f
o
r
th
r
ee
d
atasets
.
I
n
T
ab
le
2
,
th
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
o
n
th
e
UC
M
d
ataset
i
s
d
escr
ib
ed
with
ev
alu
atio
n
m
etr
ics
o
f
ac
cu
r
ac
y
,
k
ap
p
a
co
ef
f
icien
t,
p
r
ec
is
io
n
an
d
f
1
-
s
co
r
e.
T
h
e
p
r
o
p
o
s
ed
DPML
-
C
NN
m
eth
o
d
r
ea
ch
ed
th
e
h
i
g
h
est
ac
cu
r
ac
y
o
f
9
9
.
2
7
%
a
n
d
9
9
.
8
4
%
f
o
r
5
0
%
an
d
8
0
%
tr
ain
in
g
r
atio
s
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
attain
ed
a
k
ap
p
a
co
ef
f
icien
t
o
f
9
8
.
6
3
%,
9
9
.
9
4
%
f
o
r
5
0
%
an
d
8
0
%
tr
ain
in
g
r
atio
s
o
n
UC
M
d
ataset.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
s
h
o
ws
ef
f
ec
tiv
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
o
n
lan
d
s
ce
n
es wh
en
co
m
p
ar
ed
to
o
th
er
n
e
u
r
al
n
etwo
r
k
s
.
I
n
T
ab
le
3
,
th
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
o
n
AI
D
d
ataset
is
d
escr
ib
ed
with
ev
alu
atio
n
m
etr
ics
o
f
ac
cu
r
ac
y
,
k
a
p
p
a
co
ef
f
icien
t,
p
r
ec
is
io
n
an
d
f
1
-
s
co
r
e.
T
h
e
p
r
o
p
o
s
ed
DPML
-
C
NN
m
eth
o
d
r
ea
ch
ed
th
e
h
ig
h
est
ac
cu
r
ac
y
o
f
9
6
.
2
3
%
an
d
9
7
.
9
1
%
f
o
r
tr
ai
n
in
g
r
atio
s
o
f
2
0
%
an
d
5
0
%.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
attain
ed
a
k
ap
p
a
co
ef
f
icien
t
o
f
9
4
.
9
3
%,
9
8
.
0
3
%
f
o
r
tr
ain
in
g
r
atio
s
o
f
2
0
%
an
d
5
0
%
o
n
AI
D
d
ataset.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
s
h
o
ws ef
f
ec
tiv
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
o
n
lan
d
s
ce
n
es wh
en
co
m
p
ar
ed
t
o
o
th
e
r
n
eu
r
al
n
etwo
r
k
s
.
I
n
T
ab
le
4
,
t
h
e
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
o
n
NW
PU
R
E
SIS
C
4
5
d
ataset
is
d
escr
ib
ed
with
ev
alu
atio
n
m
etr
ics
o
f
ac
cu
r
ac
y
,
k
ap
p
a
co
ef
f
icien
t,
p
r
ec
is
io
n
,
an
d
f
1
-
s
c
o
r
e.
T
h
e
p
r
o
p
o
s
ed
DPML
-
C
NN
m
eth
o
d
r
ea
c
h
ed
th
e
h
ig
h
est
ac
cu
r
ac
y
o
f
9
3
.
5
2
%
an
d
9
5
.
2
1
%
f
o
r
tr
ain
i
n
g
r
atio
s
o
f
1
0
%
an
d
2
0
%.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
attain
ed
a
k
ap
p
a
co
e
f
f
icien
t
o
f
9
2
.
7
8
%,
an
d
9
3
.
9
2
%
f
o
r
tr
ain
in
g
r
atio
s
o
f
2
0
%
an
d
5
0
%
o
n
NW
PU
R
E
SIS
C
4
5
d
ataset.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
s
h
o
ws
ef
f
ec
tiv
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
o
n
lan
d
s
ce
n
es
wh
en
co
m
p
ar
ed
to
o
th
e
r
n
e
u
r
al
n
etw
o
r
k
s
.
T
ab
le
2
.
C
lass
if
icatio
n
p
er
f
o
r
m
an
ce
o
n
UC
M
d
ataset
M
e
t
h
o
d
s
A
c
c
u
r
a
c
y
(
%)
K
a
p
p
a
C
o
e
f
f
i
c
i
e
n
t
(
%)
P
r
e
c
i
s
i
o
n
(
%)
F1
-
sc
o
r
e
(
%)
5
0
%
8
0
%
5
0
%
8
0
%
5
0
%
8
0
%
5
0
%
8
0
%
DNN
9
6
.
4
5
9
7
.
0
2
9
6
.
1
7
9
6
.
9
3
9
3
.
8
9
9
4
.
5
6
9
4
.
5
7
9
5
.
2
6
C
N
N
9
7
.
0
3
9
7
.
8
4
9
6
.
8
9
9
7
.
2
6
9
4
.
6
7
9
5
.
7
8
9
5
.
5
3
9
6
.
3
9
R
C
N
N
9
7
.
7
8
9
8
.
6
3
9
7
.
2
6
9
8
.
7
7
9
5
.
7
1
9
6
.
2
5
9
6
.
6
2
9
7
.
7
4
D
P
M
L
-
C
N
N
9
9
.
2
7
9
9
.
8
4
9
8
.
6
3
9
9
.
9
4
9
7
.
4
3
9
8
.
2
7
9
7
.
7
1
9
8
.
5
6
T
ab
le
3
.
C
lass
if
icatio
n
p
er
f
o
r
m
an
ce
o
n
AI
D
d
ataset
M
e
t
h
o
d
s
A
c
c
u
r
a
c
y
(
%)
K
a
p
p
a
C
o
e
f
f
i
c
i
e
n
t
(
%)
P
r
e
c
i
s
i
o
n
(
%)
F1
-
sc
o
r
e
(
%)
2
0
%
5
0
%
2
0
%
5
0
%
2
0
%
5
0
%
2
0
%
5
0
%
DNN
9
3
.
8
2
9
4
.
4
7
9
2
.
8
4
9
4
.
7
3
9
3
.
7
3
9
5
.
4
9
9
2
.
7
1
9
4
.
5
2
C
N
N
9
4
.
7
8
9
5
.
5
8
9
3
.
4
5
9
6
.
0
4
9
4
.
6
8
9
6
.
8
2
9
3
.
5
9
9
5
.
7
2
R
C
N
N
9
5
.
3
7
9
6
.
8
8
9
4
.
0
2
9
7
.
1
6
9
5
.
5
6
9
7
.
6
1
9
4
.
8
3
9
6
.
6
9
D
P
M
L
-
C
N
N
9
6
.
2
3
9
7
.
9
1
9
4
.
9
3
9
8
.
0
3
9
6
.
6
2
9
8
.
4
5
9
5
.
8
2
9
7
.
2
8
T
ab
le
4
.
C
lass
if
icatio
n
p
er
f
o
r
m
an
ce
o
n
NW
PU RE
SISC
4
5
d
ataset
M
e
t
h
o
d
s
A
c
c
u
r
a
c
y
(
%)
K
a
p
p
a
C
o
e
f
f
i
c
i
e
n
t
(
%)
P
r
e
c
i
s
i
o
n
(
%)
F1
-
sc
o
r
e
(
%)
1
0
%
2
0
%
1
0
%
2
0
%
1
0
%
2
0
%
1
0
%
2
0
%
DNN
9
1
.
4
2
9
2
.
7
9
8
9
.
2
8
9
0
.
5
8
8
8
.
3
5
8
9
.
9
2
9
0
.
5
6
9
1
.
2
7
C
N
N
9
2
.
0
3
9
3
.
3
7
9
0
.
4
6
9
1
.
6
5
8
9
.
5
6
9
1
.
4
9
9
1
.
6
6
9
2
.
7
8
R
C
N
N
9
2
.
7
7
9
4
.
6
8
9
1
.
8
6
9
2
.
7
6
9
1
.
4
5
9
2
.
7
1
9
2
.
7
8
9
3
.
6
3
D
P
M
L
-
C
N
N
9
3
.
5
2
9
5
.
2
1
9
2
.
7
8
9
3
.
9
2
9
2
.
1
7
9
3
.
4
7
9
3
.
1
2
9
4
.
0
7
3
.
2
.
Co
m
pa
ra
t
iv
e
a
na
ly
s
is
T
h
e
class
if
icatio
n
p
er
f
o
r
m
a
n
c
e
o
f
th
e
DPML
-
C
NN
m
eth
o
d
is
co
m
p
ar
ed
with
ex
is
tin
g
lan
d
s
ce
n
e
m
eth
o
d
s
lik
e
ML
FC
-
Net
[
1
4
]
,
T
L
[
1
5
]
,
GC
SA
-
Net
[
1
6
]
,
GL
FAF
-
Net
[
1
7
]
,
HHT
L
[
1
8
]
,
an
d
L
GM
L
[
1
9
]
.
T
h
r
ee
d
if
f
er
en
t
co
m
p
ar
is
o
n
tab
les
ar
e
d
escr
ib
ed
in
th
is
s
ec
tio
n
f
o
r
th
r
ee
d
atasets
with
d
if
f
er
en
t
tr
ain
in
g
r
atio
s
.
B
y
u
tili
zin
g
d
iv
e
r
s
ity
-
p
r
o
m
o
tin
g
m
etr
ic
lear
n
in
g
in
C
NN,
it
tack
les
th
e
is
s
u
e
o
f
with
in
-
class
d
iv
er
s
ity
an
d
b
etwe
en
-
class
s
im
ilar
ity
.
T
h
e
o
v
er
lap
p
in
g
o
f
s
im
ilar
ity
s
ce
n
es
is
tack
led
b
y
u
s
in
g
d
iv
er
s
ity
p
r
o
m
o
tin
g
in
m
etr
ic
lear
n
in
g
wh
ich
u
n
co
r
r
elate
d
ev
er
y
f
ac
to
r
an
d
r
ep
r
e
s
en
ts
u
n
iq
u
e
i
n
f
o
r
m
atio
n
f
o
r
ev
er
y
s
ce
n
e.
T
h
is
m
in
im
izes
th
e
p
r
o
b
ab
ilit
y
o
f
o
v
er
lap
p
in
g
a
n
d
in
cr
ea
s
es
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
m
eth
o
d
.
B
y
u
s
in
g
th
e
DPML
-
C
NN
m
eth
o
d
,
it
class
if
ied
th
e
lan
d
s
ce
n
es
f
r
o
m
r
em
o
te
s
en
s
in
g
im
ag
es
with
h
ig
h
class
if
icatio
n
p
er
f
o
r
m
an
ce
.
T
ab
le
5
d
escr
ib
es c
o
m
p
ar
is
o
n
o
f
DPML
-
C
NN
m
eth
o
d
with
o
th
er
class
if
icatio
n
m
eth
o
d
s
lik
e
ML
FC
-
Net
[
1
4
]
,
T
L
[
1
5
]
,
GC
SA
-
Net
[
1
6
]
,
GL
FAF
-
Net
[
1
7
]
,
HHT
L
[
1
8
]
a
n
d
L
GM
L
[
1
9
]
o
n
th
e
UC
M
d
a
taset.
T
h
e
p
r
o
p
o
s
ed
m
e
t
h
o
d
i
s
c
o
m
p
a
r
e
d
w
i
t
h
e
v
a
lu
a
t
i
o
n
m
e
t
r
i
cs
o
f
a
c
c
u
r
a
c
y
a
n
d
k
a
p
p
a
c
o
e
f
f
i
c
i
e
n
t
o
n
5
0
%
a
n
d
8
0
%
t
r
a
i
n
i
n
g
r
a
t
i
o
s
.
F
r
o
m
T
a
b
l
e
5
,
i
t
i
s
cl
e
a
r
t
h
at
t
h
e
p
r
o
p
o
s
e
d
m
e
t
h
o
d
p
e
r
f
o
r
m
e
d
m
o
r
e
e
f
f
e
c
t
i
v
e
l
y
t
h
a
n
e
x
is
ti
n
g
a
l
g
o
r
i
t
h
m
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
37
,
No
.
1
,
J
an
u
ar
y
20
25
:
269
-
2
7
8
276
T
ab
le
5
.
C
o
m
p
a
r
is
o
n
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
n
UC
M
d
ataset
M
e
t
h
o
d
s
P
e
r
f
o
r
ma
n
c
e
M
e
t
r
i
c
s
Tr
a
i
n
i
n
g
r
a
t
i
o
5
0
%
8
0
%
M
LFC
-
N
e
t
[
1
4
]
A
c
c
u
r
a
c
y
(
%)
9
8
.
5
3
9
9
.
6
6
TL
[
1
5
]
A
c
c
u
r
a
c
y
(
%)
N
/
A
95
G
C
S
A
-
N
e
t
[
1
6
]
A
c
c
u
r
a
c
y
(
%)
9
8
.
3
2
9
9
.
3
1
G
LFA
F
-
N
e
t
[
1
7
]
A
c
c
u
r
a
c
y
(
%)
9
7
.
5
2
N
/
A
H
H
TL
[
1
8
]
A
c
c
u
r
a
c
y
(
%)
9
8
.
8
7
9
9
.
4
8
LG
M
L
[
1
9
]
A
c
c
u
r
a
c
y
(
%)
9
8
.
6
7
9
9
.
7
8
K
a
p
p
a
c
o
e
f
f
i
c
i
e
n
t
(
%)
9
8
.
3
1
9
9
.
7
6
P
r
o
p
o
se
d
D
P
M
L
-
C
N
N
A
c
c
u
r
a
c
y
(
%)
9
9
.
2
7
9
9
.
8
4
K
a
p
p
a
C
o
e
f
f
i
c
i
e
n
t
(
%)
9
8
.
6
3
9
9
.
9
4
T
ab
le
6
d
escr
ib
es
co
m
p
ar
is
o
n
o
f
th
e
DPML
-
C
NN
m
eth
o
d
with
o
th
er
class
if
icatio
n
m
eth
o
d
s
lik
e
GC
SA
-
Net
[
1
6
]
,
HHT
L
[
1
8
]
an
d
L
GM
L
[
1
9
]
o
n
th
e
AI
D
d
ataset.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
is
co
m
p
ar
ed
with
ev
alu
atio
n
m
etr
ics
o
f
ac
cu
r
ac
y
an
d
k
a
p
p
a
c
o
ef
f
icien
t
o
n
2
0
%
an
d
5
0
%
tr
ain
i
n
g
r
atio
s
.
Fro
m
T
ab
le
6
,
it
is
clea
r
th
at
th
e
p
r
o
p
o
s
ed
m
eth
o
d
p
er
f
o
r
m
ed
m
o
r
e
ef
f
ec
tiv
ely
t
h
an
ex
is
tin
g
alg
o
r
ith
m
s
.
T
ab
le
6
.
C
o
m
p
a
r
is
o
n
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
n
th
e
AI
D
d
a
taset
M
e
t
h
o
d
s
P
e
r
f
o
r
ma
n
c
e
M
e
t
r
i
c
s
Tr
a
i
n
i
n
g
r
a
t
i
o
2
0
%
5
0
%
G
C
S
A
-
N
e
t
[
1
6
]
A
c
c
u
r
a
c
y
(
%)
9
5
.
9
6
9
7
.
5
3
H
H
TL
[
1
8
]
A
c
c
u
r
a
c
y
(
%)
9
5
.
6
2
9
6
.
8
8
LG
M
L
[
1
9
]
A
c
c
u
r
a
c
y
(
%)
9
4
.
7
9
9
7
.
7
2
K
a
p
p
a
c
o
e
f
f
i
c
i
e
n
t
(
%)
9
4
.
5
7
9
7
.
6
1
P
r
o
p
o
se
d
D
P
M
L
-
C
N
N
A
c
c
u
r
a
c
y
(
%)
9
6
.
2
3
9
7
.
9
1
K
a
p
p
a
c
o
e
f
f
i
c
i
e
n
t
(
%)
9
4
.
9
3
9
8
.
0
5
T
ab
le
7
d
escr
ib
es
co
m
p
ar
is
o
n
o
f
th
e
DPML
-
C
NN
m
eth
o
d
with
o
th
er
class
if
icatio
n
m
eth
o
d
s
lik
e
GC
SA
-
Net
[
1
6
]
,
HHT
L
[
1
8
]
an
d
L
GM
L
[
1
9
]
o
n
th
e
NW
PU
R
E
SISC
4
5
d
ataset.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
is
co
m
p
ar
ed
b
y
ev
al
u
atio
n
m
etr
i
cs
o
f
ac
cu
r
ac
y
an
d
k
ap
p
a
c
o
ef
f
icien
t
o
n
tr
ain
in
g
r
atio
s
o
f
1
0
%
an
d
2
0
%.
Fro
m
T
ab
le
7
,
it is
clea
r
th
at
th
e
p
r
o
p
o
s
ed
m
eth
o
d
p
e
r
f
o
r
m
ed
m
o
r
e
ef
f
ec
tiv
ely
th
a
n
ex
is
tin
g
alg
o
r
ith
m
s
.
T
ab
le
7
.
C
o
m
p
a
r
is
o
n
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
n
NW
PU RE
SISC
4
5
d
ataset
M
e
t
h
o
d
s
P
e
r
f
o
r
ma
n
c
e
M
e
t
r
i
c
s
Tr
a
i
n
i
n
g
r
a
t
i
o
1
0
%
2
0
%
G
C
S
A
-
N
e
t
[
1
6
]
A
c
c
u
r
a
c
y
(
%)
9
3
.
3
9
9
4
.
9
5
H
H
TL
[
1
8
]
A
c
c
u
r
a
c
y
(
%)
9
2
.
0
7
9
4
.
2
1
LG
M
L
[
1
9
]
A
c
c
u
r
a
c
y
(
%)
9
2
.
6
2
9
4
.
4
9
K
a
p
p
a
C
o
e
f
f
i
c
i
e
n
t
(
%)
9
2
.
2
5
9
4
.
3
1
P
r
o
p
o
se
d
D
P
M
L
-
C
N
N
A
c
c
u
r
a
c
y
(
%)
9
3
.
5
2
9
5
.
2
1
K
a
p
p
a
C
o
e
f
f
i
c
i
e
n
t
(
%)
9
2
.
7
8
9
3
.
9
2
3
.
3
.
Dis
cus
s
io
n
T
h
e
r
esu
lts
o
f
DPML
-
C
NN
m
eth
o
d
ar
e
ev
al
u
ated
with
ex
is
tin
g
m
eth
o
d
s
lik
e
DNN,
C
NN
an
d
R
C
N
N.
T
h
e
p
r
o
p
o
s
ed
DPML
-
C
NN
m
eth
o
d
ac
h
ie
v
ed
th
e
h
ig
h
est
ac
cu
r
ac
ies
o
n
th
ese
d
at
asets
,
with
9
9
.
2
7
%
an
d
9
9
.
8
4
%
f
o
r
5
0
%
an
d
8
0
%
tr
ain
in
g
r
atio
s
o
n
th
e
UC
Me
r
ce
d
d
ataset,
9
6
.
2
3
%
an
d
9
7
.
9
1
%
f
o
r
2
0
%
an
d
5
0
%
tr
ain
in
g
r
atio
s
o
n
t
h
e
A
I
D
d
ataset,
an
d
9
3
.
5
2
%
an
d
9
5
.
2
1
%
f
o
r
1
0
%
an
d
2
0
%
tr
ain
in
g
r
atio
s
o
n
th
e
NW
PU
R
E
SISC
4
5
d
ataset.
A
d
d
itio
n
ally
,
th
e
p
er
f
o
r
m
a
n
ce
o
f
DPML
-
C
NN
m
eth
o
d
is
co
m
p
ar
ed
with
ML
FC
-
Net
[
1
4
]
,
T
L
[
1
5
]
,
GC
SA
-
N
et
[
1
6
]
,
GL
FAF
-
Net
[
1
7
]
,
HHT
L
[
1
8
]
,
an
d
L
GM
L
[
1
9
]
.
T
h
e
DPML
-
C
NN
m
eth
o
d
is
p
r
o
p
o
s
ed
f
o
r
th
e
c
lass
if
icatio
n
o
f
lan
d
s
ce
n
es
f
r
o
m
r
em
o
te
s
en
s
in
g
s
ce
n
e
im
ag
es.
T
h
e
e
x
is
tin
g
m
eth
o
d
s
h
av
e
d
r
awb
ac
k
s
o
f
n
o
t
b
ein
g
g
o
o
d
at
d
is
cr
im
in
atio
n
ca
p
a
b
ilit
y
am
o
n
g
v
ar
ia
n
t
s
c
en
e
ca
teg
o
r
ies
[
1
4
]
.
Dif
f
icu
lt
in
d
escr
ib
in
g
h
ig
h
s
em
an
tic
d
ata
in
r
em
o
te
s
en
s
in
g
im
ag
es
[
1
5
]
.
T
h
e
is
s
u
e
o
f
in
ter
class
s
im
ilar
ity
lies
in
th
e
o
v
er
lap
p
in
g
o
f
s
im
ilar
s
u
r
f
ac
es
am
o
n
g
v
ar
io
u
s
s
ce
n
es
[
1
6
]
.
Du
e
to
lim
ited
d
a
ta,
th
e
m
eth
o
d
h
as
is
s
u
es
o
f
o
v
er
-
f
itti
n
g
an
d
less
f
ea
tu
r
e
g
en
er
aliza
tio
n
ca
p
ab
i
lity
[
1
7
]
.
Dif
f
icu
lty
in
in
tr
a
-
c
lass
v
ar
iatio
n
s
an
d
i
n
t
e
r
-
c
l
as
s
s
i
m
i
la
r
i
t
ie
s
i
n
o
r
i
g
i
n
a
l
s
c
e
n
e
s
m
a
x
i
m
i
z
es
t
h
e
c
o
m
p
l
e
x
i
t
y
o
f
d
a
t
a
i
n
t
e
g
r
a
t
i
o
n
i
n
e
v
e
r
y
s
c
e
n
e
[
1
8
]
,
c
a
n
’
t
b
e
c
o
n
c
e
n
t
r
a
t
e
d
o
n
t
h
e
s
p
a
ti
a
l
r
e
l
a
t
i
o
n
s
h
i
p
o
f
l
o
c
a
l
f
e
a
t
u
r
es
[
1
9
]
.
T
h
e
p
r
o
b
l
e
m
o
f
w
i
t
h
i
n
-
c
l
a
s
s
d
i
v
e
r
s
i
t
y
a
n
d
b
e
t
w
e
e
n
-
c
l
ass
s
i
m
i
l
a
r
i
t
y
d
e
g
r
ad
e
s
t
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
i
m
a
g
e
s
c
e
n
e
c
l
as
s
i
f
i
c
a
ti
o
n
w
h
i
c
h
n
e
e
d
s
t
o
b
e
t
a
c
k
l
e
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
La
n
d
s
ce
n
e
cla
s
s
ifica
tio
n
u
s
in
g
d
ivers
ity
p
r
o
mo
tin
g
metric le
a
r
n
in
g
-
co
n
vo
lu
tio
n
a
l
… (
K
a
mp
a
R
a
tn
a
B
a
b
u
)
277
T
h
is
r
esear
ch
p
r
o
p
o
s
ed
a
DPML
-
C
NN
m
eth
o
d
th
at
m
ax
im
izes
in
ter
class
v
ar
iatio
n
an
d
m
in
im
izes
th
e
in
ter
class
v
ar
iatio
n
.
T
h
e
m
etr
ic
lear
n
in
g
in
C
NN
m
ak
es
th
e
m
eth
o
d
m
u
ch
m
o
r
e
d
is
cr
im
in
ativ
e
an
d
ca
p
tu
r
es b
o
th
g
l
o
b
al
an
d
l
o
ca
l f
ea
tu
r
es.
T
h
e
d
iv
e
r
s
ity
p
r
o
m
o
t
io
n
in
m
etr
ic
lear
n
in
g
m
in
im
iz
es th
e
o
v
er
lap
p
in
g
o
f
in
ter
class
s
im
ilar
ity
b
y
u
n
co
r
r
elatio
n
o
f
ev
er
y
s
ce
n
e
p
a
r
am
eter
.
B
y
u
tili
zin
g
d
i
v
er
s
ity
-
p
r
o
m
o
tin
g
m
etr
ic
lear
n
in
g
in
C
NN,
it
tack
les
th
e
is
s
u
e
o
f
with
in
-
class
d
iv
er
s
ity
an
d
b
etwe
en
-
class
s
im
ilar
it
y
.
T
h
e
o
v
er
lap
p
in
g
o
f
s
im
ilar
ity
s
ce
n
es
is
tack
led
b
y
u
s
in
g
d
i
v
er
s
ity
p
r
o
m
o
ti
n
g
in
m
etr
ic
lear
n
in
g
wh
ich
u
n
co
r
r
elate
d
ev
er
y
f
ac
to
r
an
d
r
ep
r
esen
ts
u
n
iq
u
e
in
f
o
r
m
atio
n
f
o
r
e
v
er
y
s
ce
n
e.
T
h
is
m
in
im
izes
th
e
p
r
o
b
ab
ilit
y
o
f
o
v
er
la
p
p
in
g
a
n
d
in
cr
ea
s
es
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
m
eth
o
d
.
B
y
u
s
in
g
th
e
D
PML
-
C
NN
m
eth
o
d
,
it
clas
s
if
ied
th
e
lan
d
s
ce
n
es
f
r
o
m
r
e
m
o
te
s
en
s
in
g
im
ag
es w
ith
h
ig
h
class
if
icatio
n
p
er
f
o
r
m
an
ce
.
4.
CO
NCLU
SI
O
N
T
h
e
DPML
-
C
NN
m
eth
o
d
is
p
r
o
p
o
s
ed
f
o
r
t
h
e
class
if
icatio
n
o
f
lan
d
s
ce
n
es
f
r
o
m
r
em
o
te
s
en
s
in
g
s
ce
n
e
im
ag
es.
T
h
e
m
etr
ic
le
ar
n
in
g
with
th
e
C
NN
m
eth
o
d
en
h
an
ce
d
th
e
d
is
cr
im
in
atio
n
ca
p
ab
ilit
y
o
f
th
e
m
o
d
el,
ef
f
ec
tiv
ely
ca
p
t
u
r
in
g
b
o
th
g
lo
b
al
a
n
d
lo
ca
l
f
ea
t
u
r
es
f
o
r
im
p
r
o
v
ed
class
if
icatio
n
.
T
h
e
d
iv
er
s
ity
p
r
o
m
o
te
d
in
m
etr
ic
lear
n
in
g
r
ed
u
ce
d
th
e
o
v
er
lap
p
in
g
o
f
in
ter
class
s
im
ilar
ity
th
r
o
u
g
h
t
h
e
u
n
c
o
r
r
elatio
n
o
f
ev
er
y
s
ce
n
e
p
ar
am
eter
.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
ad
d
r
ess
es
th
e
is
s
u
e
o
f
class
d
iv
er
s
ity
an
d
b
etwe
en
-
class
s
im
ilar
ity
b
y
m
ap
p
in
g
th
e
s
a
m
e
s
ce
n
e
class
clo
s
er
t
o
g
eth
er
an
d
d
if
f
e
r
en
t
class
es
as
f
ar
a
p
ar
t
as
p
o
s
s
ib
le.
T
h
e
p
r
o
p
o
s
ed
DPML
-
C
NN
m
eth
o
d
ac
h
iev
ed
th
e
h
ig
h
est
ac
cu
r
a
cies
o
n
th
ese
d
atasets
,
with
9
9
.
2
7
%
an
d
9
9
.
8
4
%
f
o
r
5
0
%
an
d
8
0
%
tr
ain
in
g
r
ati
o
s
o
n
UC
Me
r
ce
d
d
ataset,
9
6
.
2
3
%
an
d
9
7
.
9
1
% f
o
r
2
0
% a
n
d
5
0
%
tr
ai
n
in
g
r
atio
s
o
n
AI
D
d
ataset,
an
d
9
3
.
5
2
%
an
d
9
5
.
2
1
%
f
o
r
1
0
%
an
d
2
0
%
tr
ain
in
g
r
atio
s
o
n
NW
PU
R
E
SIS
C
4
5
d
ataset.
I
n
th
e
f
u
tu
r
e,
v
ar
i
o
u
s
d
ee
p
lear
n
i
n
g
m
eth
o
d
s
ca
n
b
e
u
s
ed
to
cla
s
s
if
y
th
e
lan
d
s
ce
n
e
im
ag
es
f
o
r
f
u
r
th
er
im
p
r
o
v
in
g
th
e
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
.
RE
F
E
R
E
NC
E
S
[
1
]
T.
B
h
o
s
a
l
e
a
n
d
S
.
P
u
s
h
k
a
r
,
“
A
Ti
T
H
i
:
d
e
e
p
l
e
a
r
n
i
n
g
a
n
d
h
y
b
r
i
d
o
p
t
i
m
i
z
a
t
i
o
n
f
o
r
a
c
c
u
r
a
t
e
t
o
u
r
i
st
d
e
s
t
i
n
a
t
i
o
n
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
I
n
t
e
l
l
i
g
e
n
t
S
y
st
e
m
s
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
i
n
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
1
2
,
n
o
.
1
0
,
2
0
2
4
.
[
2
]
A
.
H
a
mza
e
t
a
l
.
,
“
A
n
i
n
t
e
g
r
a
t
e
d
p
a
r
a
l
l
e
l
i
n
n
e
r
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
s i
n
f
o
r
m
a
t
i
o
n
f
u
s
i
o
n
w
i
t
h
B
a
y
e
si
a
n
o
p
t
i
m
i
z
a
t
i
o
n
f
o
r
l
a
n
d
sc
e
n
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
i
n
s
a
t
e
l
l
i
t
e
i
m
a
g
e
s,
”
I
E
EE
J
o
u
r
n
a
l
o
f
S
e
l
e
c
t
e
d
T
o
p
i
c
s
i
n
A
p
p
l
i
e
d
E
a
rt
h
O
b
s
e
rv
a
t
i
o
n
s
a
n
d
Re
m
o
t
e
S
e
n
s
i
n
g
,
v
o
l
.
1
6
,
p
p
.
9
8
8
8
–
9
9
0
3
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
JS
TA
R
S
.
2
0
2
3
.
3
3
2
4
4
9
4
.
[
3
]
H
.
W
a
n
g
,
X
.
Li
,
G
.
Zh
o
u
,
W
.
C
h
e
n
,
a
n
d
L.
W
a
n
g
,
“
E
d
g
e
e
n
h
a
n
c
e
d
c
h
a
n
n
e
l
a
t
t
e
n
t
i
o
n
-
b
a
s
e
d
g
r
a
p
h
c
o
n
v
o
l
u
t
i
o
n
n
e
t
w
o
r
k
f
o
r
s
c
e
n
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
c
o
m
p
l
e
x
l
a
n
d
sc
a
p
e
s,
”
I
EEE
J
o
u
r
n
a
l
o
f
S
e
l
e
c
t
e
d
T
o
p
i
c
s
i
n
Ap
p
l
i
e
d
E
a
rt
h
O
b
ser
v
a
t
i
o
n
s
a
n
d
R
e
m
o
t
e
S
e
n
si
n
g
,
v
o
l
.
1
6
,
p
p
.
3
8
3
1
–
3
8
4
9
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
JS
TA
R
S
.
2
0
2
3
.
3
2
6
5
6
7
7
.
[
4
]
C
.
C
.
Y
u
,
T.
Y
.
C
h
e
n
,
C
.
W
.
H
su
,
a
n
d
H
.
Y
.
C
h
e
n
g
,
“
I
n
c
r
e
me
n
t
a
l
sc
e
n
e
c
l
a
ssi
f
i
c
a
t
i
o
n
u
s
i
n
g
d
u
a
l
k
n
o
w
l
e
d
g
e
d
i
st
i
l
l
a
t
i
o
n
a
n
d
c
l
a
ss
i
f
i
e
r
d
i
s
c
r
e
p
a
n
c
y
o
n
n
a
t
u
r
a
l
a
n
d
r
e
mo
t
e
se
n
si
n
g
i
m
a
g
e
s,
”
El
e
c
t
ro
n
i
c
s
(
S
w
i
t
zer
l
a
n
d
)
,
v
o
l
.
1
3
,
n
o
.
3
,
p
.
5
8
3
,
Jan
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
e
l
e
c
t
r
o
n
i
c
s
1
3
0
3
0
5
8
3
.
[
5
]
J.
G
a
w
l
i
k
o
w
sk
i
,
P
.
Eb
e
l
,
M
.
S
c
h
mi
t
t
,
a
n
d
X
.
X
.
Z
h
u
,
“
Ex
p
l
a
i
n
i
n
g
t
h
e
e
f
f
e
c
t
s
o
f
c
l
o
u
d
s
o
n
r
e
m
o
t
e
se
n
si
n
g
sce
n
e
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
I
EEE
J
o
u
r
n
a
l
o
f
S
e
l
e
c
t
e
d
T
o
p
i
c
s
i
n
A
p
p
l
i
e
d
E
a
rt
h
O
b
se
rv
a
t
i
o
n
s
a
n
d
R
e
m
o
t
e
S
e
n
s
i
n
g
,
v
o
l
.
1
5
,
p
p
.
9
9
7
6
–
9
9
8
6
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
JS
TA
R
S
.
2
0
2
2
.
3
2
2
1
7
8
8
.
[
6
]
J.
N
i
,
K
.
S
h
e
n
,
Y
.
C
h
e
n
,
W
.
C
a
o
,
a
n
d
S
.
X
.
Y
a
n
g
,
“
A
n
i
mp
r
o
v
e
d
d
e
e
p
n
e
t
w
o
r
k
-
b
a
se
d
sce
n
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
me
t
h
o
d
f
o
r
se
l
f
-
d
r
i
v
i
n
g
c
a
r
s,”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
I
n
st
r
u
m
e
n
t
a
t
i
o
n
a
n
d
M
e
a
s
u
rem
e
n
t
,
v
o
l
.
7
1
,
p
p
.
1
–
1
4
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
TI
M
.
2
0
2
2
.
3
1
4
6
9
2
3
.
[
7
]
M
.
N
.
R
a
z
a
l
i
,
E.
O
.
N
.
T
o
n
y
,
A
.
A
.
A
.
I
b
r
a
h
i
m,
R
.
H
a
n
a
p
i
,
a
n
d
Z
.
I
sw
a
n
d
o
n
o
,
“
La
n
d
mar
k
r
e
c
o
g
n
i
t
i
o
n
m
o
d
e
l
f
o
r
smar
t
t
o
u
r
i
sm
u
si
n
g
l
i
g
h
t
w
e
i
g
h
t
d
e
e
p
l
e
a
r
n
i
n
g
a
n
d
l
i
n
e
a
r
d
i
scri
m
i
n
a
n
t
a
n
a
l
y
si
s
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
A
d
v
a
n
c
e
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
4
,
n
o
.
2
,
p
p
.
1
9
8
–
2
1
3
,
2
0
2
3
,
d
o
i
:
1
0
.
1
4
5
6
9
/
I
JA
C
S
A
.
2
0
2
3
.
0
1
4
0
2
2
5
.
[
8
]
K
.
Li
u
,
J.
Y
a
n
g
,
a
n
d
S
.
Li
,
“
R
e
mo
t
e
-
sen
s
i
n
g
c
r
o
s
s
-
d
o
ma
i
n
sc
e
n
e
c
l
a
ssi
f
i
c
a
t
i
o
n
:
a
d
a
t
a
s
e
t
a
n
d
b
e
n
c
h
mar
k
,
”
R
e
m
o
t
e
S
e
n
si
n
g
,
v
o
l
.
1
4
,
n
o
.
1
8
,
p
.
4
6
3
5
,
S
e
p
.
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
r
s
1
4
1
8
4
6
3
5
.
[
9
]
C
.
X
u
,
J.
S
h
u
,
a
n
d
G
.
Z
h
u
,
“
S
c
e
n
e
c
l
a
ssi
f
i
c
a
t
i
o
n
b
a
s
e
d
o
n
h
e
t
e
r
o
g
e
n
e
o
u
s
f
e
a
t
u
r
e
s
o
f
mu
l
t
i
-
so
u
r
c
e
d
a
t
a
,
”
R
e
m
o
t
e
S
e
n
si
n
g
,
v
o
l
.
1
5
,
n
o
.
2
,
p
.
3
2
5
,
J
a
n
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
r
s1
5
0
2
0
3
2
5
.
[
1
0
]
S
.
D
.
K
h
a
n
a
n
d
S
.
B
a
sal
a
ma
h
,
“
M
u
l
t
i
-
b
r
a
n
c
h
d
e
e
p
l
e
a
r
n
i
n
g
f
r
a
m
e
w
o
r
k
f
o
r
l
a
n
d
sc
e
n
e
c
l
a
ssi
f
i
c
a
t
i
o
n
i
n
s
a
t
e
l
l
i
t
e
i
mag
e
r
y
,
”
Re
m
o
t
e
S
e
n
si
n
g
,
v
o
l
.
1
5
,
n
o
.
1
3
,
p
.
3
4
0
8
,
J
u
l
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
r
s1
5
1
3
3
4
0
8
.
[
1
1
]
N
.
G
u
o
e
t
a
l
.
,
“
H
F
C
C
-
N
e
t
:
a
d
u
a
l
-
b
r
a
n
c
h
h
y
b
r
i
d
f
r
a
mew
o
r
k
o
f
C
N
N
a
n
d
C
a
p
sN
e
t
f
o
r
l
a
n
d
-
u
se
sc
e
n
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
R
e
m
o
t
e
S
e
n
si
n
g
,
v
o
l
.
1
5
,
n
o
.
2
0
,
p
.
5
0
4
4
,
O
c
t
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
r
s1
5
2
0
5
0
4
4
.
[
1
2
]
S
.
W
a
n
g
e
t
a
l
.
,
“
H
S
C
N
e
t
+
+
:
h
i
e
r
a
r
c
h
i
c
a
l
sc
e
n
e
c
o
o
r
d
i
n
a
t
e
c
l
a
ssi
f
i
c
a
t
i
o
n
a
n
d
r
e
g
r
e
ssi
o
n
f
o
r
v
i
s
u
a
l
l
o
c
a
l
i
z
a
t
i
o
n
w
i
t
h
t
r
a
n
sf
o
r
mer
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
e
r
Vi
si
o
n
,
v
o
l
.
1
3
2
,
n
o
.
7
,
p
p
.
2
5
3
0
–
2
5
5
0
,
F
e
b
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
2
6
3
-
023
-
0
1
9
8
2
-
9.
[
1
3
]
Y
.
Y
u
e
t
a
l
.
,
“
C
2
-
C
a
p
sV
i
T:
c
r
o
ss
-
c
o
n
t
e
x
t
a
n
d
c
r
o
s
s
-
sca
l
e
c
a
p
su
l
e
v
i
si
o
n
t
r
a
n
sf
o
r
mers
f
o
r
r
e
m
o
t
e
se
n
si
n
g
i
ma
g
e
sc
e
n
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
I
E
EE
G
e
o
s
c
i
e
n
c
e
a
n
d
Re
m
o
t
e
S
e
n
s
i
n
g
L
e
t
t
e
rs
,
v
o
l
.
1
9
,
p
p
.
1
–
5
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
LG
R
S
.
2
0
2
2
.
3
1
8
5
4
5
4
.
[
1
4
]
K
.
C
.
G
.
G
a
n
a
sh
r
e
e
,
R
.
H
e
ma
v
a
t
h
y
,
a
n
d
M
.
R
.
A
n
a
l
a
,
“
La
n
d
s
c
e
n
e
c
l
a
ssi
f
i
c
a
t
i
o
n
f
r
o
m
r
e
m
o
t
e
se
n
si
n
g
i
m
a
g
e
s
u
si
n
g
i
m
p
r
o
v
e
d
a
r
t
i
f
i
c
i
a
l
b
e
e
c
o
l
o
n
y
o
p
t
i
m
i
z
a
t
i
o
n
a
l
g
o
r
i
t
h
m
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
El
e
c
t
ri
c
a
l
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
ri
n
g
,
v
o
l
.
1
4
,
n
o
.
1
,
p
p
.
3
4
7
–
3
5
7
,
F
e
b
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
c
e
.
v
1
4
i
1
.
p
p
3
4
7
-
3
5
7
.
[
1
5
]
D
.
W
a
n
g
,
C
.
Zh
a
n
g
,
a
n
d
M
.
H
a
n
,
“
M
LFC
-
n
e
t
:
A
m
u
l
t
i
-
l
e
v
e
l
f
e
a
t
u
r
e
c
o
mb
i
n
a
t
i
o
n
a
t
t
e
n
t
i
o
n
mo
d
e
l
f
o
r
r
e
m
o
t
e
se
n
si
n
g
sce
n
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
C
o
m
p
u
t
e
rs
a
n
d
G
e
o
s
c
i
e
n
c
e
s
,
v
o
l
.
1
6
0
,
p
.
1
0
5
0
4
2
,
M
a
r
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
a
g
e
o
.
2
0
2
2
.
1
0
5
0
4
2
.
[
1
6
]
S
.
Th
i
r
u
m
a
l
a
d
e
v
i
,
K
.
V
e
e
r
a
S
w
a
m
y
,
a
n
d
M
.
S
a
i
l
a
j
a
,
“
R
e
m
o
t
e
se
n
si
n
g
i
m
a
g
e
sc
e
n
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
b
y
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
t
o
a
u
g
m
e
n
t
t
h
e
a
c
c
u
r
a
c
y
,
”
Me
a
su
r
e
m
e
n
t
:
S
e
n
s
o
rs
,
v
o
l
.
2
5
,
p
.
1
0
0
6
4
5
,
F
e
b
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
me
a
se
n
.
2
0
2
2
.
1
0
0
6
4
5
.
[
1
7
]
W
.
C
h
e
n
,
S
.
O
u
y
a
n
g
,
W
.
T
o
n
g
,
X
.
Li
,
X
.
Z
h
e
n
g
,
a
n
d
L
.
W
a
n
g
,
“
G
C
S
A
N
e
t
:
a
g
l
o
b
a
l
c
o
n
t
e
x
t
sp
a
t
i
a
l
a
t
t
e
n
t
i
o
n
d
e
e
p
l
e
a
r
n
i
n
g
n
e
t
w
o
r
k
f
o
r
r
e
mo
t
e
s
e
n
s
i
n
g
sc
e
n
e
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
I
EE
E
J
o
u
r
n
a
l
o
f
S
e
l
e
c
t
e
d
T
o
p
i
c
s
i
n
A
p
p
l
i
e
d
Ea
r
t
h
O
b
ser
v
a
t
i
o
n
s
a
n
d
Re
m
o
t
e
S
e
n
si
n
g
,
v
o
l
.
1
5
,
p
p
.
1
1
5
0
–
1
1
6
2
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
JS
TA
R
S
.
2
0
2
2
.
3
1
4
1
8
2
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
37
,
No
.
1
,
J
an
u
ar
y
20
25
:
269
-
2
7
8
278
[
1
8
]
G
.
L
v
,
L
.
D
o
n
g
,
W
.
Zh
a
n
g
,
a
n
d
W
.
X
u
,
“
A
g
l
o
b
a
l
-
l
o
c
a
l
f
e
a
t
u
r
e
a
d
a
p
t
i
v
e
f
u
si
o
n
n
e
t
w
o
r
k
f
o
r
i
ma
g
e
sc
e
n
e
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
Mu
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
8
3
,
n
o
.
3
,
p
p
.
6
5
2
1
–
6
5
5
4
,
Ju
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
1
0
4
2
-
0
2
3
-
1
5
5
1
9
-
2.
[
1
9
]
J.
M
a
,
M
.
L
i
,
X
.
T
a
n
g
,
X
.
Z
h
a
n
g
,
F
.
Li
u
,
a
n
d
L.
J
i
a
o
,
“
H
o
m
o
-
h
e
t
e
r
o
g
e
n
o
u
s
t
r
a
n
sf
o
r
m
e
r
l
e
a
r
n
i
n
g
f
r
a
m
e
w
o
r
k
f
o
r
R
S
s
c
e
n
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
I
EEE
J
o
u
rn
a
l
o
f
S
e
l
e
c
t
e
d
T
o
p
i
c
s
i
n
A
p
p
l
i
e
d
E
a
r
t
h
O
b
s
e
rv
a
t
i
o
n
s
a
n
d
R
e
m
o
t
e
S
e
n
si
n
g
,
v
o
l
.
1
5
,
p
p
.
2
2
2
3
–
2
2
3
9
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
JS
TA
R
S
.
2
0
2
2
.
3
1
5
5
6
6
5
.
[
2
0
]
C
.
X
u
,
G
.
Z
h
u
,
a
n
d
J
.
S
h
u
,
“
A
c
o
mb
i
n
a
t
i
o
n
o
f
l
i
e
g
r
o
u
p
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
f
o
r
r
e
m
o
t
e
se
n
s
i
n
g
s
c
e
n
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
u
si
n
g
mu
l
t
i
-
l
a
y
e
r
h
e
t
e
r
o
g
e
n
e
o
u
s
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
a
n
d
f
u
si
o
n
,
”
Re
m
o
t
e
S
e
n
s
i
n
g
,
v
o
l
.
1
4
,
n
o
.
6
,
p
.
1
4
4
5
,
M
a
r
.
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
r
s
1
4
0
6
1
4
4
5
.
[
2
1
]
“
U
C
m
e
r
c
e
d
l
a
n
d
u
se
d
a
t
a
s
e
t
.
”
h
t
t
p
:
/
/
w
e
e
g
e
e
.
v
i
si
o
n
.
u
c
mer
c
e
d
.
e
d
u
/
d
a
t
a
s
e
t
s
/
l
a
n
d
u
se
.
h
t
ml
.
[
2
2
]
G.
-
S
.
X
i
a
e
t
a
l
.
,
“
A
I
D
:
A
B
e
n
c
h
m
a
r
k
d
a
t
a
se
t
f
o
r
p
e
r
f
o
r
ma
n
c
e
e
v
a
l
u
a
t
i
o
n
o
f
a
e
r
i
a
l
sce
n
e
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
I
EEE
T
r
a
n
s.
o
n
G
e
o
sc
i
e
n
c
e
a
n
d
Re
m
o
t
e
S
e
n
si
n
g
,
v
o
l
.
5
5
,
n
o
.
7
,
p
p
.
3
9
6
5
–
3
9
8
1
,
2
0
1
7
,
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s:
/
/
c
a
p
t
a
i
n
-
w
h
u
.
g
i
t
h
u
b
.
i
o
/
A
I
D
/
.
[
2
3
]
“
N
W
P
U
R
ESI
S
C
4
5
.
”
h
t
t
p
s:
/
/
w
w
w
.
t
e
n
so
r
f
l
o
w
.
o
r
g
/
d
a
t
a
s
e
t
s
/
c
a
t
a
l
o
g
/
r
e
si
s
c
4
5
.
[
2
4
]
G
.
D
h
e
e
p
a
k
,
A
.
C
.
J,
a
n
d
D
.
V
a
i
s
h
a
l
i
,
“
B
r
a
i
n
t
u
mo
r
c
l
a
ss
i
f
i
c
a
t
i
o
n
:
a
n
o
v
e
l
a
p
p
r
o
a
c
h
i
n
t
e
g
r
a
t
i
n
g
G
L
C
M
,
LB
P
a
n
d
c
o
m
p
o
s
i
t
e
f
e
a
t
u
r
e
s,
”
Fr
o
n
t
i
e
rs
i
n
O
n
c
o
l
o
g
y
,
v
o
l
.
1
3
,
Ja
n
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
8
9
/
f
o
n
c
.
2
0
2
3
.
1
2
4
8
4
5
2
.
[
2
5
]
A
.
C
h
a
t
e
r
,
H
.
B
e
n
r
a
d
i
,
a
n
d
A
.
L
a
sf
a
r
,
“
N
e
w
a
p
p
r
o
a
c
h
t
o
s
i
mi
l
a
r
i
t
y
d
e
t
e
c
t
i
o
n
b
y
c
o
m
b
i
n
i
n
g
t
e
c
h
n
i
q
u
e
t
h
r
e
e
-
p
a
t
c
h
l
o
c
a
l
b
i
n
a
r
y
p
a
t
t
e
r
n
s
(
TP
-
LB
P
)
w
i
t
h
s
u
p
p
o
r
t
v
e
c
t
o
r
m
a
c
h
i
n
e
,
”
I
AE
S
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
1
2
,
n
o
.
4
,
p
p
.
1
6
4
4
–
1
6
5
3
,
D
e
c
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
1
2
.
i
4
.
p
p
1
6
4
4
-
1
6
5
3
.
[
2
6
]
J
.
S
.
S
u
j
i
n
a
n
d
S
.
S
o
p
h
i
a
,
“
H
i
g
h
-
p
e
r
f
o
r
m
a
n
c
e
i
m
a
g
e
f
o
r
g
e
r
y
d
e
t
e
c
t
i
o
n
v
i
a
a
d
a
p
t
i
v
e
S
I
F
T
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
f
o
r
l
o
w
-
c
o
n
t
r
a
s
t
o
r
s
m
a
l
l
o
r
s
m
o
o
t
h
c
o
p
y
–
m
o
v
e
r
e
g
i
o
n
i
m
a
g
e
s
,
”
S
o
f
t
C
o
m
p
u
t
i
n
g
,
v
o
l
.
2
8
,
n
o
.
1
,
p
p
.
4
3
7
–
4
4
5
,
A
p
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
5
0
0
-
0
2
3
-
0
8
2
0
9
-
6.
[
2
7
]
X
.
Li
,
G
.
Z
h
u
,
S
.
W
a
n
g
,
Y
.
Zh
o
u
,
a
n
d
X
.
Zh
a
n
g
,
“
D
e
e
p
r
e
v
e
r
se
a
t
t
a
c
k
o
n
S
I
F
T
f
e
a
t
u
r
e
s
w
i
t
h
a
c
o
a
r
se
-
to
-
f
i
n
e
G
A
N
mo
d
e
l
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
C
i
rc
u
i
t
s
a
n
d
S
y
st
e
m
s
f
o
r
Vi
d
e
o
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
3
4
,
n
o
.
7
,
p
p
.
6
3
9
1
–
6
4
0
2
,
J
u
l
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
T
C
S
V
T.
2
0
2
4
.
3
3
6
7
8
0
8
.
[
2
8
]
A
.
S
i
v
a
su
b
r
a
m
a
n
i
a
n
,
V
.
R
.
P
r
a
sh
a
n
t
h
,
T
.
H
a
r
i
,
V
.
S
o
w
my
a
,
E
.
A
.
G
o
p
a
l
a
k
r
i
sh
n
a
n
,
a
n
d
V
.
R
a
v
i
,
“
T
r
a
n
sf
o
r
m
e
r
-
b
a
s
e
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
a
p
p
r
o
a
c
h
f
o
r
r
e
mo
t
e
s
e
n
si
n
g
n
a
t
u
r
a
l
sc
e
n
e
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
R
e
m
o
t
e
S
e
n
si
n
g
A
p
p
l
i
c
a
t
i
o
n
s:
S
o
c
i
e
t
y
a
n
d
En
v
i
r
o
n
m
e
n
t
,
v
o
l
.
3
3
,
p
.
1
0
1
1
2
6
,
Ja
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
r
sase
.
2
0
2
3
.
1
0
1
1
2
6
.
[
2
9
]
T.
L
u
,
L.
W
a
n
,
S
.
Q
i
,
a
n
d
M
.
G
a
o
,
“
La
n
d
c
o
v
e
r
c
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
U
A
V
r
e
mo
t
e
se
n
si
n
g
b
a
se
d
o
n
t
r
a
n
sf
o
r
mer
–
C
N
N
h
y
b
r
i
d
a
r
c
h
i
t
e
c
t
u
r
e
,
”
S
e
n
s
o
rs
,
v
o
l
.
2
3
,
n
o
.
1
1
,
p
.
5
2
8
8
,
J
u
n
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
s2
3
1
1
5
2
8
8
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
K
a
m
p
a
Ra
t
n
a
Ba
b
u
re
c
e
iv
e
d
h
is
P
h
.
D
(C
.
S
.
E)
De
g
re
e
in
t
h
e
y
e
a
r
2
0
1
5
,
M
.
Tec
h
(S
o
ftwa
re
En
g
in
e
e
ri
n
g
)
De
g
re
e
in
th
e
y
e
a
r
2
0
0
3
fro
m
Ja
wa
h
a
rlal
Ne
h
ru
Tec
h
n
o
lo
g
ica
l
Un
iv
e
rsity
Hy
d
e
ra
b
a
d
,
Tela
n
g
a
n
a
,
I
n
d
ia
a
n
d
B.
Tec
h
(C.
S
.
E)
fro
m
Ba
p
a
tl
a
E
n
g
in
e
e
rin
g
Co
ll
e
g
e
,
Na
g
a
rj
u
n
a
Un
iv
e
rsity
,
An
d
h
ra
P
ra
d
e
sh
i
n
t
h
e
y
e
a
r
2
0
0
0
.
He
is
c
u
rre
n
tl
y
wo
r
k
i
n
g
a
s
Lec
tu
re
r
in
C
o
m
p
u
ter
En
g
i
n
e
e
rin
g
De
p
a
rtme
n
t,
M
.
B.
T.
S
.
G
o
v
e
r
n
m
e
n
t
P
o
l
y
tec
h
n
ic,
G
u
n
t
u
r,
An
d
h
ra
P
ra
d
e
sh
.
His
re
se
a
rc
h
in
t
e
re
sts
a
re
ima
g
e
p
ro
c
e
ss
in
g
a
n
d
p
a
tt
e
rn
Re
c
o
g
n
it
io
n
.
He
h
a
s
p
u
b
li
sh
e
d
m
o
re
t
h
a
n
1
5
p
a
p
e
rs
in
In
tern
a
ti
o
n
a
l
a
n
d
Na
ti
o
n
a
l
Jo
u
r
n
a
ls
a
n
d
c
o
n
f
e
re
n
c
e
s.
He
is
h
a
v
in
g
2
2
y
e
a
rs
o
f
Tea
c
h
in
g
Ex
p
e
rien
c
e
.
He
is
a
li
fe
m
e
m
b
e
r
o
f
tec
h
n
ica
l
a
ss
o
c
iatio
n
I
S
TE
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
k
ra
t
n
a
b
a
b
u
@
g
m
a
il
.
c
o
m
.
K
a
m
p
a
K
a
n
t
h
i
K
u
m
a
r
re
c
e
iv
e
d
h
is
P
h
.
D
.
(
El
e
c
tro
n
ics
&
Co
m
m
u
n
ica
ti
o
n
En
g
i
n
e
e
rin
g
)
fr
o
m
Ja
wa
h
a
rlal
N
e
h
ru
Tec
h
n
o
lo
g
ica
l
Un
iv
e
rsit
y
Ka
k
in
a
d
a
,
An
d
h
ra
P
ra
d
e
sh
,
In
d
ia
in
th
e
y
e
a
r
2
0
1
9
,
M
.
Tec
h
.
(Co
m
p
u
ters
a
n
d
Co
m
m
u
n
ica
ti
o
n
s)
De
g
re
e
fro
m
Bh
a
ra
t
h
In
stit
u
te
o
f
Hig
h
e
r
Ed
u
c
a
ti
o
n
a
n
d
Re
se
a
rc
h
,
Ch
e
n
n
a
i,
Tam
il
n
a
d
u
,
In
d
ia
in
th
e
y
e
a
r
2
0
0
5
a
n
d
B.
Tec
h
(El
e
c
tro
n
ics
a
n
d
C
o
m
m
u
n
ica
ti
o
n
E
n
g
i
n
e
e
rin
g
)
fro
m
Ba
p
a
tl
a
En
g
in
e
e
rin
g
Co
ll
e
g
e
,
Na
g
a
rju
n
a
Un
iv
e
rsity
,
An
d
h
ra
P
ra
d
e
sh
i
n
t
h
e
y
e
a
r
2
0
0
2
.
He
is
c
u
rre
n
tl
y
wo
r
k
i
n
g
a
s
a
P
ro
fe
ss
o
r
i
n
ECE
,
Ti
r
u
m
a
la
En
g
i
n
e
e
rin
g
C
o
ll
e
g
e
,
Na
ra
sa
ra
o
p
e
t,
An
d
h
ra
P
ra
d
e
sh
.
His
r
e
se
a
rc
h
in
tere
sts
a
re
wire
les
s
c
o
m
m
u
n
ica
ti
o
n
s
a
n
d
n
e
two
rk
s,
c
o
m
p
u
ter
n
e
two
rk
s,
sig
n
a
l
p
r
o
c
e
ss
in
g
a
n
d
ima
g
e
p
ro
c
e
ss
in
g
.
He
h
a
s
p
u
b
li
s
h
e
d
m
o
re
th
a
n
3
2
p
a
p
e
rs
in
In
tern
a
ti
o
n
a
l
a
n
d
Na
ti
o
n
a
l
Jo
u
rn
a
ls
a
n
d
c
o
n
fe
re
n
c
e
s.
He
h
a
s
1
9
y
e
a
rs
o
f
Tea
c
h
i
n
g
E
x
p
e
rien
c
e
.
He
is
a
li
fe
m
e
m
b
e
r
o
f
tec
h
n
ica
l
a
ss
o
c
iatio
n
IS
TE
,
IE
TE
,
IEAE
,
IARA
IAE.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
k
k
a
n
t
h
i
k
@g
m
a
il
.
c
o
m
.
Aku
la
S
u
n
e
e
th
a
re
c
e
iv
e
d
h
e
r
P
h
.
D
.
(C.
S
.
E)
De
g
re
e
in
th
e
y
e
a
r
2
0
2
2
fr
o
m
Ac
h
a
ry
a
Na
g
a
rju
n
a
U
n
iv
e
rsit
y
,
An
d
h
ra
P
ra
d
e
sh
,
In
d
ia,
M
.
T
e
c
h
(C.
S
.
E)
De
g
re
e
fro
m
Ja
wa
h
a
rlal
Ne
h
ru
Tec
h
n
o
lo
g
ica
l
Un
iv
e
rsity
Hy
d
e
ra
b
a
d
,
Tela
n
g
a
n
a
,
In
d
ia
in
th
e
y
e
a
r
2
0
0
9
a
n
d
B.
Tec
h
(C.
S
.
E)
fr
o
m
S
ir
C
R
Re
d
d
y
Co
ll
e
g
e
o
f
E
n
g
i
n
e
e
rin
g
,
El
u
ru
,
An
d
h
ra
U
n
iv
e
rsit
y
,
A.
P
,
In
d
ia
in
th
e
y
e
a
r
2
0
0
5
.
S
h
e
is
c
u
rre
n
tl
y
wo
rk
in
g
a
s
As
so
c
iate
p
ro
fe
ss
o
r
in
th
e
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
E
n
g
i
n
e
e
rin
g
,
KK
R
a
n
d
K
S
R
I
n
stit
u
te
o
f
Tec
h
n
o
lo
g
y
a
n
d
S
c
ien
c
e
s
,
G
u
n
tu
r,
A
n
d
h
ra
P
ra
d
e
sh
a
ffil
iat
e
d
to
JN
TUK
Un
iv
e
rsity
,
A.
P
,
In
d
ia.
He
r
re
se
a
rc
h
a
re
a
is
F
u
z
z
y
Im
a
g
e
P
ro
c
e
ss
in
g
.
S
h
e
h
a
s
p
u
b
li
sh
e
d
v
a
rio
u
s
p
a
p
e
rs
in
In
tern
a
ti
o
n
a
l
a
n
d
Na
ti
o
n
a
l
Jo
u
rn
a
ls,
p
a
ten
t
p
u
b
li
c
a
ti
o
n
s
a
n
d
c
o
n
fe
re
n
c
e
s.
S
h
e
is
h
a
v
in
g
1
8
y
e
a
rs
o
f
Tea
c
h
in
g
E
x
p
e
rie
n
c
e
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
k
u
las
u
n
e
e
th
a
2
5
@g
m
a
il
.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.