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.
4
3
,
No
.
1
,
Ju
ly
2
0
2
6
,
p
p
.
207
~
21
8
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
4
3
.i
1
.
pp
207
-
21
8
207
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Co
ra
l clas
sifica
tion in un
derwa
te
r i
ma
g
es using
a dua
l
-
bra
nch
deep learning
fra
mewo
rk
P
ra
cha
ra
t
Sa
-
ng
a
ds
up
,
Cha
wa
n K
o
o
pip
a
t
D
e
p
a
r
t
me
n
t
o
f
I
mag
i
n
g
a
n
d
P
r
i
n
t
i
n
g
Te
c
h
n
o
l
o
g
y
,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
,
C
h
u
l
a
l
o
n
g
k
o
r
n
U
n
i
v
e
r
si
t
y
,
B
a
n
g
k
o
k
,
T
h
a
i
l
a
n
d
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
9
,
2
0
2
6
R
ev
is
ed
J
u
n
3
,
2
0
2
6
Acc
ep
ted
J
u
n
2
7
,
2
0
2
6
Au
to
m
a
ted
c
o
ra
l
c
las
sifica
ti
o
n
f
ro
m
u
n
d
e
rwa
ter
ima
g
e
ry
is
e
ss
e
n
ti
a
l
fo
r
larg
e
-
sc
a
le
re
e
f
m
o
n
it
o
ri
n
g
b
u
t
re
m
a
in
s
c
h
a
ll
e
n
g
i
n
g
d
u
e
to
c
o
l
o
r
a
t
ten
u
a
ti
o
n
,
il
lu
m
in
a
ti
o
n
v
a
riab
il
it
y
,
a
n
d
d
i
ffe
re
n
c
e
s
b
e
twe
e
n
tex
tu
re
-
fo
c
u
se
d
c
lo
se
-
ra
n
g
e
ima
g
e
s
a
n
d
m
o
r
p
h
o
lo
g
y
-
fo
c
u
se
d
c
o
lo
n
y
-
le
v
e
l
o
b
se
r
v
a
ti
o
n
s.
To
a
d
d
re
ss
th
e
se
c
h
a
ll
e
n
g
e
s,
t
h
is
st
u
d
y
p
r
o
p
o
se
s
a
d
u
a
l
-
b
ra
n
c
h
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
th
a
t
i
n
teg
ra
tes
in
fo
rm
a
ti
o
n
fr
o
m
th
e
CIEL
AB
(L
AB)
c
o
lo
r
sp
a
c
e
with
str
u
c
tu
ra
l
d
e
sc
rip
t
o
rs
d
e
riv
e
d
fro
m
th
e
d
isc
re
te
wa
v
e
let
tran
sfo
rm
(DWT
).
RG
B
ima
g
e
s
a
re
first
c
o
n
v
e
rted
t
o
th
e
LAB
c
o
l
o
r
sp
a
c
e
to
se
p
a
ra
te
lu
m
in
a
n
c
e
a
n
d
c
h
ro
m
a
ti
c
c
o
m
p
o
n
e
n
ts
i
n
se
p
a
ra
te
c
h
a
n
n
e
ls
,
e
n
a
b
li
n
g
t
h
e
m
o
d
e
l
to
e
x
p
l
o
it
c
o
lo
r
a
n
d
li
g
h
tn
e
ss
i
n
fo
rm
a
ti
o
n
m
o
re
e
x
p
li
c
it
l
y
.
S
tr
u
c
tu
ra
l
i
n
fo
rm
a
ti
o
n
is
e
x
trac
ted
fro
m
th
e
lu
m
i
n
a
n
c
e
c
h
a
n
n
e
l
u
sin
g
wa
v
e
let
d
e
c
o
m
p
o
siti
o
n
t
o
c
a
p
tu
re
h
i
g
h
-
fre
q
u
e
n
c
y
m
o
r
p
h
o
lo
g
ica
l
p
a
tt
e
rn
s.
T
h
e
two
re
p
re
se
n
ta
ti
o
n
s
a
re
p
r
o
c
e
ss
e
d
th
ro
u
g
h
p
a
ra
ll
e
l
c
o
n
v
o
lu
ti
o
n
a
l
b
ra
n
c
h
e
s
a
n
d
f
u
se
d
a
t
t
h
e
fe
a
tu
re
le
v
e
l
fo
r
c
las
sifica
ti
o
n
.
Ex
p
e
rime
n
ts
c
o
n
d
u
c
ted
o
n
a
u
n
ifi
e
d
c
o
ra
l
d
a
tas
e
t
c
o
n
tai
n
in
g
tex
t
u
re
-
d
o
m
in
a
n
t
a
n
d
m
o
rp
h
o
l
o
g
y
-
fo
c
u
se
d
ima
g
e
ry
a
c
ro
ss
1
4
c
las
se
s
sh
o
w
th
a
t
th
e
p
ro
p
o
se
d
m
e
th
o
d
a
c
h
iev
e
s
9
6
.
5
2
%
a
c
c
u
ra
c
y
o
n
th
e
tex
t
u
r
e
-
fo
c
u
se
d
RS
M
AS
d
a
tas
e
t
a
n
d
8
8
.
3
3
%
a
c
c
u
ra
c
y
o
n
t
h
e
m
o
rp
h
o
l
o
g
y
-
fo
c
u
se
d
s
tru
c
tu
re
RS
M
AS
d
a
tas
e
t,
re
d
u
c
i
n
g
th
e
c
ro
ss
-
d
o
m
a
in
p
e
rfo
rm
a
n
c
e
g
a
p
fr
o
m
2
2
.
4
6
%
to
8
.
1
9
%
c
o
m
p
a
re
d
wit
h
RG
B
b
a
se
li
n
e
s,
d
e
m
o
n
stra
ti
n
g
imp
ro
v
e
d
c
ro
ss
-
d
o
m
a
in
r
o
b
u
stn
e
ss
f
o
r
c
o
ra
l
c
las
sifica
ti
o
n
u
n
d
e
r
h
e
ter
o
g
e
n
e
o
u
s
u
n
d
e
rwa
ter
ima
g
in
g
c
o
n
d
it
i
o
n
s.
K
ey
w
o
r
d
s
:
C
I
E
L
AB
co
lo
r
s
p
ac
e
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
C
o
r
al
im
ag
e
class
if
icatio
n
C
o
r
al
r
ee
f
m
o
n
ito
r
in
g
Dis
cr
ete
wav
elet
tr
an
s
f
o
r
m
Featu
r
e
f
u
s
io
n
Un
d
er
wate
r
im
ag
e
a
n
aly
s
is
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Pra
ch
ar
at
Sa
-
n
g
ad
s
u
p
Dep
ar
tm
en
t o
f
I
m
ag
in
g
an
d
Pr
in
tin
g
T
ec
h
n
o
lo
g
y
,
Facu
lty
o
f
Scien
ce
,
C
h
u
lalo
n
g
k
o
r
n
U
n
iv
er
s
ity
B
an
g
k
o
k
1
0
3
3
0
,
T
h
ailan
d
E
m
ail: p
r
ac
h
ar
at.
s
a@
ch
u
la.
ac
.
th
1.
I
NT
RO
D
UCT
I
O
N
C
o
r
al
r
ee
f
s
ar
e
am
o
n
g
th
e
m
o
s
t
b
io
lo
g
ically
d
iv
er
s
e
m
ar
in
e
ec
o
s
y
s
tem
s
an
d
p
r
o
v
id
e
ess
en
tial
ec
o
lo
g
ical
an
d
ec
o
n
o
m
ic
s
er
v
ices,
in
clu
d
in
g
co
astal
p
r
o
tectio
n
,
f
is
h
er
ies
s
u
p
p
o
r
t
,
an
d
b
io
d
iv
er
s
ity
p
r
eser
v
atio
n
[
1
]
.
Ho
we
v
er
,
c
o
r
al
r
ee
f
s
wo
r
ld
wid
e
ar
e
in
c
r
ea
s
in
g
l
y
th
r
ea
ten
ed
b
y
clim
ate
ch
an
g
e
,
o
ce
an
war
m
in
g
,
an
d
an
t
h
r
o
p
o
g
e
n
ic
d
is
tu
r
b
an
ce
s
,
wh
ich
ac
ce
ler
ate
co
r
al
b
leac
h
in
g
an
d
ec
o
s
y
s
tem
d
eg
r
ad
atio
n
[
2
]
-
[
4
]
.
T
r
ad
itio
n
al
f
ield
s
u
r
v
ey
s
ar
e
lab
o
r
-
i
n
ten
s
iv
e
an
d
s
p
atially
lim
ited
,
m
ak
in
g
a
u
to
m
ated
im
ag
e
-
b
ased
an
aly
s
is
a
n
in
cr
ea
s
in
g
ly
im
p
o
r
tan
t to
o
l f
o
r
r
ee
f
m
o
n
ito
r
i
n
g
at
s
ca
le
[
5
]
.
E
ar
ly
s
tu
d
ies
o
n
c
o
r
al
im
a
g
e
class
if
icatio
n
p
r
im
ar
ily
r
el
ied
o
n
h
a
n
d
cr
a
f
ted
f
ea
tu
r
e
ex
tr
ac
tio
n
co
m
b
in
ed
with
class
ical
m
ac
h
in
e
lear
n
in
g
m
eth
o
d
s
s
u
ch
as
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
es
(
SVM
s
)
an
d
tex
t
u
r
e
-
b
ased
d
escr
ip
to
r
s
[
6
]
-
[
9
]
.
W
h
ile
th
ese
ap
p
r
o
ac
h
es
d
em
o
n
s
tr
ated
p
r
o
m
is
in
g
r
esu
lts
,
th
eir
p
er
f
o
r
m
an
ce
is
o
f
ten
co
n
s
tr
ai
n
ed
b
y
th
e
lim
ited
r
e
p
r
esen
t
atio
n
al
ca
p
ac
ity
o
f
m
an
u
ally
d
esig
n
ed
f
ea
tu
r
es.
C
las
s
ical
m
ac
h
in
e
lear
n
in
g
m
eth
o
d
s
h
a
v
e
also
b
ee
n
ex
p
lo
r
ed
in
m
o
r
e
r
ec
e
n
t
wo
r
k
[
1
0
]
.
R
ec
en
t
ad
v
an
ce
s
i
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
207
-
21
8
208
d
ee
p
lear
n
in
g
,
p
ar
ticu
lar
ly
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
et
wo
r
k
s
(
C
NNs),
h
av
e
s
ig
n
if
ican
tly
im
p
r
o
v
e
d
p
er
f
o
r
m
an
ce
in
u
n
d
e
r
wate
r
im
ag
e
class
if
icatio
n
task
s
b
y
en
ab
lin
g
au
t
o
m
atic
h
ier
ar
c
h
ic
al
f
ea
tu
r
e
lear
n
in
g
[
1
1
]
-
[
1
7
]
.
Desp
ite
th
ese
ad
v
an
ce
s
,
co
r
al
im
ag
e
class
if
icatio
n
in
u
n
d
er
wate
r
en
v
ir
o
n
m
en
ts
r
em
ain
s
c
h
allen
g
in
g
.
Un
d
er
wate
r
im
a
g
es
o
f
te
n
s
u
f
f
er
f
r
o
m
co
l
o
r
d
is
to
r
tio
n
,
r
ed
u
ce
d
co
n
tr
ast,
an
d
s
ca
tter
in
g
ca
u
s
ed
b
y
lig
h
t
ab
s
o
r
p
tio
n
a
n
d
s
u
s
p
e
n
d
ed
p
a
r
ticles
[
1
8
]
.
Sev
e
r
al
m
et
h
o
d
s
h
av
e
b
ee
n
p
r
o
p
o
s
ed
to
a
d
d
r
ess
th
ese
is
s
u
es,
in
clu
d
in
g
p
h
y
s
ics
-
b
ased
r
est
o
r
atio
n
tech
n
iq
u
es
an
d
lear
n
in
g
-
b
ased
co
lo
r
c
o
r
r
ec
tio
n
m
o
d
els
s
u
c
h
as
W
ater
GAN
[
1
9
]
.
I
n
ad
d
itio
n
,
u
n
d
er
wate
r
im
a
g
e
en
h
a
n
ce
m
en
t
tech
n
iq
u
es
h
a
v
e
b
ee
n
d
ev
elo
p
ed
to
r
esto
r
e
co
lo
r
an
d
co
n
t
r
ast in
d
eg
r
a
d
ed
m
ar
in
e
im
ag
e
r
y
[
2
0
]
.
An
o
th
er
c
h
allen
g
e
ar
is
es
f
r
o
m
v
ar
iatio
n
s
in
s
p
atial
s
ca
le
ac
r
o
s
s
co
r
al
im
ag
e
d
atasets
.
C
lo
s
e
-
r
an
g
e
im
ag
er
y
ty
p
ically
em
p
h
asizes
f
in
e
s
u
r
f
ac
e
tex
tu
r
es,
wh
er
ea
s
co
lo
n
y
-
lev
el
im
ag
e
r
y
h
ig
h
lig
h
ts
lar
g
e
r
m
o
r
p
h
o
lo
g
ical
s
t
r
u
ctu
r
es.
W
h
en
d
is
cr
im
in
ativ
e
cu
es
s
h
if
t
f
r
o
m
tex
tu
r
e
-
d
o
m
in
ated
p
atter
n
s
to
b
r
o
ad
er
s
tr
u
ctu
r
al
ch
ar
ac
ter
is
tics
,
C
N
N
m
o
d
els
tr
ain
ed
p
r
im
ar
ily
o
n
tex
tu
r
e
-
r
ic
h
im
ag
er
y
m
ay
ex
h
ib
it
r
ed
u
ce
d
g
en
er
aliza
tio
n
p
er
f
o
r
m
an
ce
.
T
o
ad
d
r
ess
th
is
lim
itatio
n
,
alter
n
ativ
e
r
ep
r
esen
tatio
n
s
tr
ateg
ies
h
av
e
b
ee
n
ex
p
lo
r
ed
.
T
r
an
s
f
o
r
m
in
g
im
ag
es
in
to
th
e
C
I
E
L
AB
(
L
AB
)
co
lo
r
s
p
ac
e
r
e
p
r
esen
ts
lu
m
in
an
ce
an
d
c
h
r
o
m
atic
co
m
p
o
n
en
ts
in
s
ep
ar
ate
ch
an
n
els,
wh
ich
ca
n
r
ed
u
ce
s
en
s
itiv
ity
to
illu
m
in
atio
n
v
ar
iab
i
lity
[
1
8
]
,
[
2
1
]
.
I
n
p
ar
allel,
f
r
e
q
u
en
c
y
-
d
o
m
ain
a
n
aly
s
is
u
s
in
g
th
e
d
is
cr
ete
wa
v
elet
tr
an
s
f
o
r
m
(
DW
T
)
e
n
ab
l
es
m
u
lti
-
r
eso
lu
tio
n
ch
ar
ac
ter
izatio
n
o
f
s
tr
u
ctu
r
al
p
atter
n
s
[
2
2
]
-
[
2
4
]
.
Alth
o
u
g
h
m
u
lti
-
b
r
an
c
h
an
d
f
e
atu
r
e
-
f
u
s
io
n
a
r
ch
itectu
r
es
h
av
e
d
em
o
n
s
tr
ated
im
p
r
o
v
e
d
p
er
f
o
r
m
an
ce
in
g
en
er
al
im
ag
e
class
if
ic
atio
n
b
y
p
r
o
ce
s
s
in
g
co
m
p
lem
e
n
tar
y
r
ep
r
esen
tatio
n
s
in
p
ar
allel
[
2
5
]
,
[
2
6
]
,
th
ese
d
esig
n
s
ar
e
ty
p
ically
ev
alu
ated
in
n
at
u
r
al
-
im
ag
e
r
ec
o
g
n
itio
n
s
ettin
g
s
an
d
d
o
n
o
t
e
x
p
licitly
a
d
d
r
es
s
th
e
p
h
y
s
ical
an
d
d
o
m
ain
-
s
p
ec
if
ic
c
h
ar
ac
ter
is
tics
o
f
u
n
d
e
r
wate
r
co
r
al
im
ag
er
y
,
in
cl
u
d
in
g
wa
v
elen
g
t
h
-
d
ep
e
n
d
en
t
c
o
lo
r
atten
u
atio
n
,
illu
m
in
atio
n
v
ar
i
ab
ilit
y
,
an
d
ch
a
n
g
es
in
v
is
u
al
s
ca
le.
E
x
is
tin
g
co
r
al
class
if
icatio
n
m
eth
o
d
s
p
r
ed
o
m
i
n
an
tly
r
ely
o
n
s
in
g
le
-
in
p
u
t
R
GB
r
ep
r
esen
tatio
n
s
[
1
1
]
-
[
1
7
]
.
I
n
R
GB
im
ag
es,
ch
an
g
es
in
illu
m
in
atio
n
,
s
h
ad
in
g
,
wate
r
atte
n
u
atio
n
,
an
d
ca
m
er
a
e
x
p
o
s
u
r
e
d
ir
ec
tly
af
f
ec
t
th
e
r
e
d
,
g
r
ee
n
,
a
n
d
b
lu
e
i
n
ten
s
ity
ch
an
n
els,
m
ak
in
g
b
r
ig
h
tn
ess
an
d
co
lo
r
v
ar
iatio
n
d
if
f
icu
lt
t
o
s
ep
ar
ate.
B
y
co
n
t
r
ast,
C
I
E
L
AB
r
ep
r
es
en
ts
lig
h
tn
ess
in
th
e
L
*
ch
an
n
el
an
d
c
h
r
o
m
aticity
alo
n
g
th
e
o
p
p
o
n
en
t
-
c
o
lo
r
a
x
e
s
a
*
an
d
b
*
.
I
n
p
r
in
cip
le,
r
e
g
io
n
s
with
s
im
ilar
ch
r
o
m
aticity
ca
n
r
e
m
ain
m
o
r
e
co
m
p
ar
a
b
le
u
n
d
er
m
o
d
er
ate
lig
h
tin
g
d
if
f
er
en
ce
s
w
h
en
lu
m
in
an
ce
is
r
ep
r
esen
ted
s
ep
ar
ately
.
T
h
is
p
r
o
p
er
ty
is
u
s
ef
u
l
f
o
r
c
o
r
al
class
if
icatio
n
b
ec
au
s
e
im
ag
es
ac
q
u
ir
ed
at
d
if
f
e
r
en
t
s
ca
les
an
d
u
n
d
er
d
if
f
er
e
n
t
u
n
d
er
wate
r
co
n
d
itio
n
s
m
ay
v
ar
y
s
tr
o
n
g
ly
in
b
r
ig
h
tn
ess
wh
il
e
r
etain
in
g
s
p
ec
ies
-
r
elate
d
co
lo
r
c
h
ar
ac
ter
is
tics
.
Ho
wev
er
,
co
lo
r
r
ep
r
esen
tatio
n
alo
n
e
d
o
es
n
o
t
ca
p
tu
r
e
c
o
l
o
n
y
-
lev
el
m
o
r
p
h
o
lo
g
y
.
C
o
lo
r
-
co
r
r
ec
tio
n
m
eth
o
d
s
s
u
ch
as
W
ater
GA
N
[
1
9
]
ca
n
im
p
r
o
v
e
v
is
u
al
ap
p
ea
r
an
ce
,
b
u
t
th
ey
d
o
n
o
t
d
ir
ec
tly
en
co
d
e
m
o
r
p
h
o
lo
g
y
-
r
elate
d
s
tr
u
ctu
r
al
d
escr
ip
to
r
s
f
o
r
ta
x
o
n
o
m
ic
d
is
c
r
im
in
atio
n
.
Alth
o
u
g
h
Gó
m
ez
-
R
ío
s
et
a
l.
[
1
5
]
r
ep
o
r
ted
r
esu
lts
o
n
b
o
th
tex
tu
r
e
-
f
o
cu
s
ed
an
d
m
o
r
p
h
o
lo
g
y
-
f
o
c
u
s
ed
co
r
a
l
d
atasets
,
th
eir
ev
alu
ati
o
n
u
s
ed
a
two
-
lev
el
class
if
ier
u
n
d
e
r
d
if
f
er
en
t
e
x
p
er
im
en
t
al
co
n
d
itio
n
s
.
T
h
is
leav
es
o
p
en
th
e
q
u
esti
o
n
o
f
wh
eth
er
a
s
in
g
le
r
e
p
r
esen
tatio
n
f
r
am
ewo
r
k
ca
n
m
ai
n
tain
p
e
r
f
o
r
m
an
ce
wh
en
th
e
v
is
u
al
cu
es
s
h
if
t
f
r
o
m
clo
s
e
-
r
an
g
e
tex
t
u
r
e
t
o
c
o
l
ony
-
lev
el
m
o
r
p
h
o
lo
g
y
.
Prio
r
wo
r
k
h
as
n
o
t
s
y
s
tem
atica
lly
is
o
lated
th
e
in
d
iv
id
u
al
an
d
co
m
b
in
ed
e
f
f
ec
ts
o
f
L
AB
co
lo
r
r
ep
r
esen
tatio
n
,
wav
elet
-
d
er
iv
ed
s
tr
u
ctu
r
al
d
escr
ip
to
r
s
,
an
d
d
u
al
-
b
r
an
c
h
f
ea
tu
r
e
f
u
s
io
n
o
n
p
er
f
o
r
m
an
ce
ac
r
o
s
s
co
r
al
im
ag
e
d
o
m
ain
s
.
I
n
th
is
s
tu
d
y
,
th
e
ter
m
d
o
m
ai
n
g
a
p
r
e
f
er
s
to
th
e
p
er
f
o
r
m
a
n
ce
d
if
f
e
r
en
ce
o
b
s
er
v
e
d
wh
e
n
a
m
o
d
el
is
ev
alu
ated
ac
r
o
s
s
co
r
al
im
ag
e
d
o
m
ain
s
with
d
if
f
er
en
t
v
is
u
al
ch
ar
ac
ter
is
tics
,
p
ar
ticu
lar
ly
te
x
tu
r
e
-
f
o
cu
s
ed
clo
s
e
-
r
an
g
e
im
ag
e
r
y
an
d
m
o
r
p
h
o
lo
g
y
-
f
o
cu
s
ed
c
o
lo
n
y
-
lev
el
im
ag
e
r
y
.
T
h
is
g
ap
is
in
ter
p
r
ete
d
as
an
in
d
icato
r
o
f
h
o
w
s
tr
o
n
g
ly
a
m
o
d
el
d
e
p
en
d
s
o
n
cu
es
th
at
m
ay
n
o
t
tr
an
s
f
er
ac
r
o
s
s
im
ag
in
g
s
ca
le,
v
iewp
o
in
t,
illu
m
in
atio
n
,
an
d
s
tr
u
ctu
r
al
co
n
te
x
t.
A
s
m
aller
d
o
m
ain
g
ap
th
er
ef
o
r
e
s
u
g
g
ests
b
etter
cr
o
s
s
-
d
o
m
ain
g
en
er
al
izatio
n
,
esp
ec
ial
ly
wh
en
p
er
f
o
r
m
a
n
ce
o
n
t
h
e
m
o
r
p
h
o
lo
g
y
-
f
o
cu
s
ed
d
ataset
im
p
r
o
v
es
with
o
u
t
s
u
b
s
tan
tially
r
e
d
u
cin
g
ac
cu
r
ac
y
o
n
th
e
tex
tu
r
e
-
f
o
cu
s
ed
d
ataset.
T
o
ad
d
r
ess
th
is
g
ap
,
t
h
e
p
r
ese
n
t
s
tu
d
y
p
r
o
p
o
s
es
a
d
u
al
-
b
r
an
ch
d
ee
p
lear
n
in
g
f
r
am
ewo
r
k
i
n
wh
ich
th
e
L
AB
b
r
an
ch
r
e
p
r
esen
ts
lu
m
in
an
ce
an
d
ch
r
o
m
atic
co
m
p
o
n
e
n
ts
in
s
ep
ar
ate
ch
an
n
els,
wh
ile
th
e
DW
T
b
r
an
ch
ex
tr
ac
ts
m
u
lti
-
r
eso
lu
tio
n
s
tr
u
c
tu
r
al
d
etail
f
r
o
m
th
e
lu
m
i
n
an
c
e
ch
an
n
el.
T
h
e
two
r
e
p
r
esen
ta
tio
n
s
ar
e
p
r
o
c
ess
ed
th
r
o
u
g
h
p
ar
allel
co
n
v
o
lu
tio
n
a
l
b
r
an
ch
es
an
d
f
u
s
ed
b
ef
o
r
e
class
if
icat
io
n
.
T
h
e
f
r
am
ewo
r
k
is
ev
alu
ated
b
y
d
ir
ec
tly
co
m
p
ar
in
g
R
GB
,
L
A
B
,
D
W
T
-
o
n
ly
,
R
GB
+D
W
T
,
an
d
L
AB
+D
W
T
co
n
f
ig
u
r
atio
n
s
u
n
d
er
a
u
n
if
ie
d
ev
alu
atio
n
p
r
o
to
co
l.
T
h
e
m
ain
co
n
tr
ib
u
tio
n
s
o
f
th
is
s
tu
d
y
ar
e
s
u
m
m
a
r
ized
as f
o
llo
ws:
−
A
d
u
al
-
b
r
an
c
h
r
ep
r
esen
tatio
n
lear
n
in
g
f
r
am
ewo
r
k
th
at
co
m
b
in
es
L
AB
co
lo
r
r
ep
r
esen
tatio
n
with
DW
T
-
d
er
iv
ed
s
tr
u
ctu
r
al
d
escr
ip
to
r
s
;
−
A
s
y
s
tem
atic
ev
alu
atio
n
o
f
R
GB
,
L
AB
,
D
W
T
-
o
n
ly
,
R
GB
+
DW
T
,
an
d
L
AB
+D
W
T
co
n
f
ig
u
r
atio
n
s
ac
r
o
s
s
tex
tu
r
e
-
d
o
m
in
an
t a
n
d
m
o
r
p
h
o
l
o
g
y
-
f
o
cu
s
ed
c
o
r
al
im
ag
e
d
atasets
; a
n
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:
2502
-
4
7
5
2
C
o
r
a
l c
la
s
s
ifica
tio
n
in
u
n
d
erw
a
ter ima
g
es u
s
in
g
a
d
u
a
l
-
b
r
a
n
ch
d
ee
p
le
a
r
n
in
g
…
(
P
r
a
ch
a
r
a
t S
a
-
n
g
a
d
s
u
p
)
209
−
A
d
o
m
ain
-
g
ap
a
n
aly
s
is
th
at
q
u
an
tifie
s
cr
o
s
s
-
d
o
m
ain
g
e
n
er
aliza
tio
n
as
th
e
a
b
s
o
lu
te
d
if
f
er
e
n
ce
in
class
if
icatio
n
ac
cu
r
ac
y
b
etwe
en
th
e
tex
tu
r
e
-
f
o
c
u
s
ed
R
S
MA
S
d
ataset
an
d
th
e
m
o
r
p
h
o
lo
g
y
-
f
o
c
u
s
ed
s
tr
u
ctu
r
e
R
SMAS d
ataset.
2.
P
RO
P
O
SE
D
M
E
T
H
OD
T
h
e
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
in
t
eg
r
ates
L
AB
co
lo
r
r
ep
r
esen
tatio
n
with
DW
T
-
d
er
iv
ed
d
escr
ip
to
r
s
f
o
r
co
r
al
tax
o
n
o
m
y
.
T
h
e
o
v
e
r
all
wo
r
k
f
lo
w
co
n
s
is
ts
o
f
th
r
ee
m
ain
s
tag
es:
r
ep
r
esen
tatio
n
tr
an
s
f
o
r
m
atio
n
,
d
u
al
-
b
r
an
ch
f
ea
tu
r
e
e
x
tr
ac
tio
n
,
a
n
d
f
ea
tu
r
e
f
u
s
io
n
f
o
r
class
if
icatio
n
,
as illu
s
tr
ated
in
Fig
u
r
e
1
.
Fig
u
r
e
1
.
Ov
e
r
v
iew
o
f
th
e
d
u
a
l
-
b
r
an
ch
ar
ch
itectu
r
e
2
.
1
.
Repre
s
ent
a
t
io
n t
ra
ns
f
o
r
m
a
t
io
n
Un
d
er
wate
r
im
ag
es
o
f
ten
ex
h
ib
it
s
ig
n
if
ican
t
co
lo
r
d
is
to
r
t
io
n
d
u
e
to
wav
elen
g
th
-
d
ep
en
d
en
t
lig
h
t
ab
s
o
r
p
tio
n
a
n
d
s
ca
tter
in
g
.
T
o
r
ep
r
esen
t
lu
m
in
a
n
ce
an
d
ch
r
o
m
atic
co
m
p
o
n
en
ts
m
o
r
e
ex
p
licitly
,
in
p
u
t
R
GB
im
ag
es
ar
e
tr
an
s
f
o
r
m
e
d
in
to
t
h
e
C
I
E
L
AB
(
L
AB
)
co
lo
r
s
p
ac
e
,
w
h
ich
r
ep
r
esen
ts
lu
m
in
a
n
ce
in
f
o
r
m
atio
n
(
L*
)
an
d
ch
r
o
m
atic
co
m
p
o
n
e
n
ts
(
a*
an
d
b*
)
.
I
n
ad
d
itio
n
to
co
lo
r
-
s
p
ac
e
t
r
an
s
f
o
r
m
atio
n
,
s
tr
u
ctu
r
al
in
f
o
r
m
atio
n
is
ex
t
r
ac
ted
u
s
in
g
th
e
DW
T
.
W
av
elet
d
ec
o
m
p
o
s
itio
n
p
e
r
f
o
r
m
s
m
u
lti
-
r
eso
lu
tio
n
a
n
aly
s
is
b
y
s
ep
ar
atin
g
im
ag
e
co
n
ten
t
i
n
to
f
r
e
q
u
en
c
y
s
u
b
-
b
an
d
s
,
wh
e
r
e
l
o
w
-
f
r
eq
u
en
cy
c
o
m
p
o
n
en
ts
r
ep
r
esen
t
co
a
r
s
e
i
m
ag
e
s
tr
u
ctu
r
e
wh
ile
h
ig
h
-
f
r
e
q
u
en
cy
co
m
p
o
n
e
n
ts
ca
p
tu
r
e
e
d
g
es
a
n
d
f
in
e
s
p
atia
l
d
etails
[
2
2
]
-
[
2
4
]
,
[
2
7
]
-
[
2
9
]
.
B
ec
au
s
e
co
r
al
co
lo
n
ies
ex
h
i
b
it
co
m
p
lex
s
u
r
f
ac
e
g
eo
m
etr
y
an
d
b
r
an
c
h
in
g
m
o
r
p
h
o
lo
g
y
,
th
ese
h
ig
h
-
f
r
eq
u
en
c
y
co
m
p
o
n
en
ts
h
ig
h
lig
h
t
p
atte
r
n
s
ass
o
ciate
d
with
co
r
al
g
r
o
wth
f
o
r
m
s
.
C
o
n
s
e
q
u
en
tly
,
th
e
wav
elet
r
e
p
r
es
en
tatio
n
p
r
o
v
id
es
a
co
m
p
a
ct
d
escr
ip
to
r
th
at
co
m
p
lem
en
ts
th
e
L
AB
co
lo
r
in
p
u
t.
Sp
ec
if
ically
,
th
e
Dau
b
ec
h
ies
-
2
(
d
b
2
)
m
o
t
h
er
wav
elet
[
3
0
]
i
s
ap
p
lied
at
a
s
in
g
le
d
ec
o
m
p
o
s
itio
n
lev
el
to
th
e
lu
m
in
an
ce
ch
an
n
el.
T
h
e
d
b
2
wav
elet
was
s
elec
ted
f
o
r
its
b
alan
ce
b
etwe
en
co
m
p
ac
t
s
u
p
p
o
r
t
(
4
c
o
ef
f
icien
ts
)
a
n
d
s
m
o
o
th
n
ess
,
wh
ich
is
s
u
itab
le
f
o
r
ca
p
tu
r
in
g
ed
g
e
an
d
tex
tu
r
e
p
atter
n
s
in
c
o
r
al
s
u
r
f
ac
e
m
o
r
p
h
o
lo
g
y
.
Sin
g
le
-
lev
el
d
e
co
m
p
o
s
itio
n
p
r
eser
v
es
s
p
atial
r
eso
lu
tio
n
wh
ile
ex
tr
ac
tin
g
d
ir
ec
tio
n
al
h
ig
h
-
f
r
eq
u
e
n
cy
in
f
o
r
m
atio
n
.
L
et
I
l
d
e
n
o
te
th
e
n
o
r
m
alize
d
lu
m
in
a
n
ce
ch
a
n
n
el
e
x
tr
ac
t
ed
f
r
o
m
th
e
C
I
E
L
AB
r
ep
r
e
s
en
tatio
n
.
A
s
in
g
le
-
lev
el
two
-
d
im
en
s
io
n
al
DW
T
y
ield
s
f
o
u
r
s
u
b
-
b
an
d
co
ef
f
icien
t m
atr
ices
.
{
,
,
,
}
=
2
(
;
2
)
(
1
)
W
h
er
e
L
L
is
th
e
ap
p
r
o
x
im
ati
o
n
s
u
b
-
b
a
n
d
,
a
n
d
L
H,
HL
,
a
n
d
HH
ar
e
th
e
h
o
r
izo
n
tal,
v
e
r
tical,
an
d
d
iag
o
n
al
d
etail
s
u
b
-
b
an
d
s
,
r
e
s
p
ec
tiv
ely
.
T
h
e
ap
p
r
o
x
im
atio
n
s
u
b
-
b
an
d
L
L
is
d
is
ca
r
d
ed
b
ec
au
s
e
it
p
r
im
ar
ily
co
n
tain
s
s
m
o
o
th
e
d
lu
m
in
a
n
ce
in
f
o
r
m
atio
n
alr
ea
d
y
r
e
p
r
esen
t
ed
in
th
e
L
AB
co
lo
r
b
r
a
n
ch
.
E
ac
h
d
etail
s
u
b
-
b
an
d
is
co
n
v
er
ted
to
m
ag
n
itu
d
e
f
o
r
m
to
r
em
o
v
e
d
ir
ec
tio
n
al
s
ig
n
d
ep
en
d
en
c
y
.
=
|
|
,
=
|
|
,
=
|
|
(
2
)
=
(
+
+
)
(
3
)
W
h
er
e
α
=
0
.
5
is
a
s
ca
lin
g
c
o
n
s
tan
t
th
at
b
ala
n
ce
s
th
e
c
o
m
b
in
ed
m
a
g
n
itu
d
e
o
f
th
e
th
r
ee
d
etail
s
u
b
-
b
an
d
s
.
T
h
e
DW
T
-
d
er
iv
e
d
s
tr
u
ctu
r
al
m
ap
,
,
is
k
ep
t
in
its
co
m
p
u
ted
m
ag
n
itu
d
e
s
ca
le
an
d
s
tan
d
ar
d
ized
im
p
licitly
b
y
th
e
b
atch
-
n
o
r
m
aliza
tio
n
lay
er
s
in
th
e
DW
T
b
r
an
ch
.
B
ec
au
s
e
DW
T
d
ec
o
m
p
o
s
itio
n
r
e
d
u
ce
s
s
p
atial
d
im
en
s
io
n
s
,
th
e
f
u
s
ed
s
tr
u
ctu
r
al
m
ap
is
r
esized
to
2
2
4
×2
2
4
p
i
x
els
u
s
in
g
b
ilin
ea
r
i
n
ter
p
o
latio
n
to
p
r
o
d
u
ce
,
2
24
.
T
h
e
p
r
ep
r
o
ce
s
s
in
g
p
ip
elin
e
th
er
ef
o
r
e
p
r
o
d
u
c
es
two
s
y
n
ch
r
o
n
ized
in
p
u
ts
:
a
th
r
ee
-
ch
an
n
el
L
AB
im
ag
e
∈
ℝ
224
×
224
×
3
f
o
r
t
h
e
co
lo
r
b
r
an
ch
,
an
d
a
s
in
g
le
-
c
h
an
n
el
r
esized
DW
T
-
d
er
iv
e
d
s
tr
u
ctu
r
al
m
ap
,
224
∈
ℝ
224
×
224
×
1
f
o
r
th
e
DW
T
b
r
an
ch
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
207
-
21
8
210
2
.
2.
Dua
l
-
bra
nch f
ea
t
ure
ex
t
ra
ct
io
n
T
h
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
co
n
s
is
ts
o
f
two
p
ar
allel
b
r
an
ch
es
th
at
p
r
o
ce
s
s
co
m
p
lem
en
tar
y
r
ep
r
esen
tatio
n
s
o
f
th
e
in
p
u
t
i
m
ag
e.
T
h
e
f
ir
s
t
b
r
an
ch
r
ec
ei
v
es
th
e
L
AB
co
lo
r
r
ep
r
esen
tatio
n
an
d
ex
tr
ac
ts
co
lo
r
-
r
elate
d
f
ea
tu
r
es
u
s
in
g
a
p
r
etr
ain
ed
R
esNet
-
1
8
b
ac
k
b
o
n
e
[
31
]
,
p
r
o
d
u
cin
g
a
5
1
2
-
d
im
en
s
io
n
al
f
ea
tu
r
e
v
ec
to
r
f
r
o
m
th
e
g
lo
b
al
av
e
r
ag
e
p
o
o
lin
g
lay
er
.
T
h
is
b
r
an
ch
l
ea
r
n
s
h
ier
ar
ch
ical
v
is
u
al
f
ea
tu
r
es
f
r
o
m
th
e
th
r
ee
-
ch
an
n
el
L
AB
r
ep
r
esen
tatio
n
,
wh
er
e
lu
m
in
a
n
ce
an
d
c
h
r
o
m
atic
co
m
p
o
n
en
ts
a
r
e
p
r
o
v
id
ed
as
s
ep
ar
ate
in
p
u
t
ch
an
n
els.
T
h
e
s
ec
o
n
d
b
r
an
ch
p
r
o
ce
s
s
es
th
e
DW
T
-
d
er
iv
ed
s
tr
u
ctu
r
al
m
ap
ex
tr
ac
ted
f
r
o
m
th
e
lu
m
in
an
ce
ch
an
n
el.
B
ec
au
s
e
th
e
wav
elet
r
ep
r
esen
tatio
n
em
p
h
asizes
ed
g
es
an
d
lo
ca
l
s
tr
u
ctu
r
al
p
atter
n
s
r
elate
d
to
co
r
al
m
o
r
p
h
o
lo
g
y
,
it
p
r
o
v
id
es
an
in
p
u
t
f
o
r
e
x
tr
ac
tin
g
s
h
a
p
e
-
r
elate
d
f
ea
tu
r
es
.
A
lig
h
tweig
h
t
cu
s
t
o
m
C
NN
is
u
s
ed
f
o
r
th
is
b
r
an
ch
,
co
n
s
is
tin
g
o
f
th
r
ee
co
n
v
o
lu
tio
n
al
b
lo
c
k
s
ea
ch
co
m
p
r
is
in
g
a
3
×
3
co
n
v
o
lu
tio
n
,
b
atc
h
n
o
r
m
aliza
tio
n
,
an
d
R
eL
U
ac
tiv
atio
n
,
with
f
ilter
s
izes
o
f
1
6
,
3
2
,
a
n
d
6
4
,
r
esp
e
ctiv
ely
.
A
g
lo
b
al
av
er
a
g
e
p
o
o
lin
g
lay
er
th
e
n
p
r
o
d
u
ce
s
a
6
4
-
d
im
e
n
s
io
n
al
f
ea
tu
r
e
v
ec
to
r
.
T
h
e
s
elec
tio
n
o
f
R
esNet
-
1
8
[
3
1
]
as
th
e
b
ac
k
b
o
n
e
f
o
r
th
e
co
lo
r
b
r
an
ch
is
m
o
tiv
ated
b
y
s
ev
er
al
co
n
s
id
er
atio
n
s
.
First,
th
e
r
elativ
ely
s
m
all
co
r
al
d
ataset
(
1
,
1
7
5
to
tal
im
a
g
e
s
)
f
a
v
o
r
s
a
lig
h
tweig
h
t
ar
ch
itectu
r
e
to
r
ed
u
ce
th
e
r
is
k
o
f
o
v
e
r
f
itti
n
g
.
C
o
m
p
a
r
ed
with
d
ee
p
er
v
ar
ian
ts
s
u
ch
as
R
esNet
-
5
0
,
R
esNet
-
1
8
p
r
o
v
i
d
es
s
u
f
f
icien
t r
ep
r
esen
tatio
n
al
ca
p
ac
ity
wh
ile
r
eq
u
ir
in
g
f
ewe
r
p
a
r
am
eter
s
an
d
lo
wer
co
m
p
u
tatio
n
al
co
s
t,
m
ak
in
g
it
s
u
itab
le
f
o
r
th
e
p
r
o
p
o
s
ed
d
u
al
-
b
r
an
ch
f
r
am
ewo
r
k
.
T
h
e
DW
T
b
r
an
ch
em
p
lo
y
s
a
s
h
allo
w
th
r
ee
-
lay
er
C
NN
r
ath
er
th
an
a
d
ee
p
e
r
p
r
et
r
ain
ed
b
ac
k
b
o
n
e
b
ec
au
s
e
th
e
DW
T
-
d
er
iv
ed
s
tr
u
ctu
r
al
m
ap
is
a
s
in
g
le
-
c
h
an
n
el
r
ep
r
esen
tatio
n
th
at
d
if
f
er
s
f
u
n
d
am
e
n
tally
f
r
o
m
n
atu
r
al
R
GB
im
ag
e
s
,
m
a
k
in
g
I
m
ag
eNe
t
-
p
r
et
r
ain
ed
weig
h
ts
u
n
s
u
itab
le
f
o
r
th
is
in
p
u
t
m
o
d
ality
.
T
h
e
p
r
o
g
r
ess
iv
e
f
ilter
ex
p
an
s
io
n
f
r
o
m
1
6
t
o
6
4
ca
p
tu
r
es
i
n
cr
ea
s
in
g
ly
c
o
m
p
lex
lo
ca
l
p
atter
n
s
,
wh
ile
g
lo
b
al
av
er
a
g
e
p
o
o
lin
g
p
r
o
d
u
ce
s
a
co
m
p
ac
t
6
4
-
d
im
en
s
io
n
al
f
ea
tu
r
e
v
ec
t
o
r
.
T
h
is
asy
m
m
etr
ic
d
esig
n
,
with
a
p
r
etr
ain
e
d
b
ac
k
b
o
n
e
f
o
r
th
e
L
AB
b
r
a
n
c
h
an
d
a
s
h
allo
w
c
u
s
to
m
n
etwo
r
k
f
o
r
th
e
DW
T
b
r
a
n
ch
,
r
ef
lects
th
e
d
if
f
er
en
t
co
m
p
lex
ity
r
e
q
u
ir
em
e
n
ts
o
f
ea
ch
r
ep
r
esen
tatio
n
.
2
.
3
.
F
e
a
t
ure
f
us
io
n a
nd
cla
s
s
if
ica
t
io
n
Featu
r
e
f
u
s
io
n
co
m
b
in
es
co
lo
r
-
r
elate
d
a
n
d
DW
T
-
d
er
iv
ed
f
ea
tu
r
es
b
y
co
n
ca
te
n
atin
g
th
e
512
-
d
im
en
s
io
n
al
L
AB
b
r
a
n
ch
o
u
tp
u
t
with
th
e
6
4
-
d
i
m
en
s
io
n
al
DW
T
b
r
an
c
h
o
u
tp
u
t,
y
iel
d
in
g
a
576
-
d
im
en
s
io
n
al
f
u
s
ed
v
ec
to
r
.
T
h
is
v
ec
t
o
r
is
p
ass
ed
th
r
o
u
g
h
a
b
atch
n
o
r
m
aliza
tio
n
lay
er
,
a
f
u
lly
co
n
n
ec
ted
lay
er
with
2
5
6
u
n
its
f
o
llo
we
d
b
y
R
eL
U
ac
tiv
atio
n
a
n
d
d
r
o
p
o
u
t
(
p
=
0
.
3
)
,
an
d
a
f
in
a
l
f
u
lly
co
n
n
ec
ted
lay
er
with
1
4
o
u
tp
u
t
u
n
its
co
r
r
esp
o
n
d
in
g
t
o
th
e
c
o
r
al
c
lass
es.
A
S
o
f
t
M
ax
f
u
n
ctio
n
p
r
o
d
u
ce
s
th
e
f
in
al
class
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
ti
on.
Featu
r
e
-
lev
el
co
n
ca
ten
atio
n
was
ch
o
s
en
o
v
er
alter
n
ativ
e
f
u
s
io
n
s
tr
ateg
ies,
s
u
ch
as
ele
m
en
t
-
wis
e
ad
d
itio
n
o
r
atten
tio
n
-
b
ased
f
u
s
io
n
,
b
ec
au
s
e
it
p
r
eser
v
es
th
e
f
u
ll
o
u
tp
u
t
o
f
b
o
th
b
r
a
n
ch
es
with
o
u
t
ass
u
m
in
g
d
ir
ec
t
f
ea
tu
r
e
alig
n
m
en
t.
T
h
e
b
atch
n
o
r
m
aliza
tio
n
lay
er
ap
p
lied
af
ter
co
n
ca
ten
atio
n
n
o
r
m
alize
s
th
e
f
ea
tu
r
e
s
ca
les
f
r
o
m
th
e
two
b
r
an
ch
es,
an
d
d
r
o
p
o
u
t
(
p
=
0
.
3
)
r
ed
u
c
es
co
-
ad
ap
tatio
n
b
etwe
en
b
r
a
n
ch
f
ea
tu
r
es
d
u
r
in
g
tr
ain
in
g
.
3.
M
E
T
H
O
D
3
.
1
.
Da
t
a
s
et
des
cr
iptio
n
E
x
p
er
im
en
tal
ev
alu
atio
n
w
as
co
n
d
u
cted
u
s
in
g
a
u
n
if
ied
d
ataset
c
o
n
s
tr
u
cted
f
r
o
m
two
co
m
p
lem
en
tar
y
co
r
al
im
ag
e
c
o
llectio
n
s
:
th
e
R
SMA
S
d
atase
t
an
d
th
e
s
tr
u
ctu
r
e
R
SMAS
d
ataset.
T
h
e
R
SMA
S
d
ataset
co
n
tain
s
clo
s
e
-
r
an
g
e
co
r
al
im
ag
er
y
em
p
h
asizin
g
f
i
n
e
s
u
r
f
ac
e
te
x
tu
r
es,
wh
ile
th
e
s
tr
u
ctu
r
e
R
SMAS
d
ataset
f
o
cu
s
es
o
n
c
o
lo
n
y
-
lev
el
m
o
r
p
h
o
lo
g
y
ca
p
tu
r
ed
f
r
o
m
g
r
ea
ter
v
iewin
g
d
is
tan
ce
s
an
d
m
o
r
e
co
m
p
lex
s
ce
n
e
co
n
tex
ts
[
1
5
]
.
A
s
u
m
m
a
r
y
o
f
t
h
e
d
ataset
co
m
p
o
s
itio
n
i
s
p
r
esen
ted
in
T
ab
le
1
[
4
]
,
[
1
5
]
.
T
ab
le
1
.
Data
s
et
s
u
m
m
ar
y
D
a
t
a
s
et
C
l
a
s
ses
I
mag
e
s
I
mag
e
t
y
p
e
R
S
M
A
S
[
4
]
14
7
6
6
T
e
x
t
u
r
e
-
f
o
c
u
se
d
S
t
r
u
c
t
u
r
e
R
S
M
A
S
[
1
5
]
14
4
0
9
M
o
r
p
h
o
l
o
g
y
-
f
o
c
u
se
d
All
im
ag
es
wer
e
co
llected
u
n
d
er
u
n
co
n
t
r
o
lled
u
n
d
er
wa
ter
co
n
d
itio
n
s
an
d
e
x
h
ib
it
s
u
b
s
tan
tial
v
ar
iatio
n
i
n
illu
m
in
atio
n
,
c
o
lo
r
atten
u
atio
n
,
s
ca
le,
an
d
b
ac
k
g
r
o
u
n
d
clu
tter
.
E
x
am
p
le
im
ag
es
f
r
o
m
b
o
th
d
atasets
ar
e
p
r
esen
ted
in
Fig
u
r
e
2
.
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:
2502
-
4
7
5
2
C
o
r
a
l c
la
s
s
ifica
tio
n
in
u
n
d
erw
a
ter ima
g
es u
s
in
g
a
d
u
a
l
-
b
r
a
n
ch
d
ee
p
le
a
r
n
in
g
…
(
P
r
a
ch
a
r
a
t S
a
-
n
g
a
d
s
u
p
)
211
R
S
M
A
S
S
t
r
u
c
t
u
r
e
R
S
M
A
S
Fig
u
r
e
2
.
E
x
am
p
le
c
o
r
al
im
ag
es f
r
o
m
th
e
R
SMAS d
ataset
(
tex
tu
r
e
-
f
o
cu
s
ed
im
ag
e
r
y
)
a
n
d
t
h
e
s
tr
u
ctu
r
e
R
SMA
S d
ataset
(
m
o
r
p
h
o
l
o
g
y
-
f
o
cu
s
ed
im
ag
e
r
y
)
3
.
2
.
T
ra
ini
ng
co
nfig
ura
t
io
n
All
m
o
d
el
tr
ain
in
g
an
d
class
if
icatio
n
ex
p
er
im
en
ts
wer
e
im
p
lem
en
ted
u
s
in
g
MA
T
L
AB
R
2
0
2
5
a
with
th
e
d
ee
p
lear
n
in
g
to
o
lb
o
x
an
d
im
ag
e
p
r
o
ce
s
s
in
g
to
o
l
b
o
x
o
n
a
wo
r
k
s
tatio
n
e
q
u
ip
p
ed
with
an
NVI
DI
A
GPU.
I
n
p
u
t
im
a
g
es
wer
e
r
esized
to
2
2
4
×2
2
4
p
ix
els,
an
d
th
e
p
r
ep
r
o
ce
s
s
in
g
p
ip
elin
e
g
en
er
ated
t
wo
p
ar
allel
in
p
u
ts
:
a
th
r
ee
-
ch
an
n
el
L
AB
co
lo
r
im
ag
e
f
o
r
th
e
R
esNet
-
1
8
b
r
an
ch
a
n
d
a
s
in
g
le
-
ch
an
n
el
DW
T
-
d
er
iv
ed
s
tr
u
ctu
r
al
m
ap
f
o
r
th
e
DW
T
b
r
an
ch
.
T
h
e
tr
ai
n
in
g
co
n
f
ig
u
r
atio
n
is
s
u
m
m
ar
ized
in
T
ab
le
2
.
I
n
f
er
e
n
ce
-
tim
e
b
en
ch
m
ar
k
in
g
was
co
n
d
u
cte
d
s
ep
ar
ately
u
n
d
er
C
PU
-
o
n
ly
ex
ec
u
tio
n
,
as d
escr
ib
ed
in
Sectio
n
4
.
3
.
Mo
d
el
tr
ain
in
g
u
s
ed
th
e
Ad
am
o
p
tim
izer
with
an
in
itial
lear
n
in
g
r
ate
o
f
1
×1
0
⁻⁴
an
d
a
m
in
i
-
b
atch
s
ize
o
f
3
2
.
T
h
e
n
etwo
r
k
was
tr
ain
ed
f
o
r
3
0
e
p
o
ch
s
,
an
d
th
e
m
o
d
el
ch
ec
k
p
o
in
t
with
th
e
l
o
west
v
alid
atio
n
lo
s
s
was
s
elec
ted
f
o
r
ev
al
u
atio
n
(
b
est
-
v
alid
atio
n
-
lo
s
s
r
ec
o
v
er
y
)
.
Data
au
g
m
en
tatio
n
in
clu
d
e
d
r
a
n
d
o
m
r
o
tatio
n
(
±
1
8
0
°),
r
an
d
o
m
h
o
r
iz
o
n
tal
r
ef
lectio
n
,
an
d
r
an
d
o
m
v
er
tical
r
ef
lectio
n
,
ap
p
lied
o
n
ly
d
u
r
in
g
tr
ain
in
g
.
No
co
lo
r
jitt
er
in
g
o
r
b
r
ig
h
tn
ess
au
g
m
en
tatio
n
was
ap
p
lied
,
as
th
e
L
AB
tr
a
n
s
f
o
r
m
atio
n
r
e
p
r
esen
ts
lu
m
in
an
ce
an
d
ch
r
o
m
atic
co
m
p
o
n
en
ts
in
s
ep
ar
ate
ch
a
n
n
els.
T
ab
le
2
.
T
r
ai
n
in
g
p
ar
am
eter
s
P
a
r
a
me
t
e
r
V
a
l
u
e
B
a
c
k
b
o
n
e
R
e
sN
e
t
-
18
O
p
t
i
mi
z
e
r
A
d
a
m
Le
a
r
n
i
n
g
r
a
t
e
1
×
10
−
4
Ep
o
c
h
30
B
a
t
c
h
si
z
e
32
I
n
p
u
t
si
z
e
2
2
4
×
2
2
4
3
.
3
.
Cro
s
s
-
v
a
lid
a
t
io
n pro
t
o
c
o
l
T
o
ev
alu
ate
m
o
d
el
s
tab
ilit
y
an
d
r
ed
u
ce
d
ep
e
n
d
en
ce
o
n
a
s
in
g
le
d
ata
p
ar
titi
o
n
,
f
iv
e
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
was
p
e
r
f
o
r
m
ed
.
I
n
ea
ch
f
o
ld
,
th
e
d
ataset
was
d
iv
id
ed
in
t
o
tr
ain
in
g
a
n
d
v
alid
atio
n
s
u
b
s
ets
wh
ile
m
ain
tain
in
g
class
b
alan
ce
ac
r
o
s
s
th
e
p
ar
titi
o
n
s
.
Fo
r
ea
ch
f
o
ld
,
a
m
o
d
el
was
t
r
ain
ed
in
d
ep
e
n
d
en
tly
a
n
d
ev
a
lu
ated
o
n
th
e
co
r
r
esp
o
n
d
i
n
g
v
alid
atio
n
s
u
b
s
et.
Fin
al
p
er
f
o
r
m
an
ce
m
etr
ics
wer
e
o
b
tain
ed
b
y
av
e
r
ag
in
g
t
h
e
r
esu
lts
ac
r
o
s
s
th
e
f
iv
e
f
o
ld
s
.
T
h
is
p
r
o
ce
d
u
r
e
p
r
o
v
id
es a
m
o
r
e
r
eliab
le
esti
m
ate
o
f
m
o
d
el
p
er
f
o
r
m
an
ce
u
n
d
er
v
a
r
y
in
g
d
ata
s
p
li
ts
.
3
.
4
.
E
v
a
lua
t
i
o
n m
et
rics
C
las
s
if
icatio
n
p
er
f
o
r
m
an
ce
w
as
ev
alu
ated
u
s
in
g
ac
cu
r
ac
y
a
n
d
m
ac
r
o
F1
-
s
co
r
e.
Acc
u
r
ac
y
m
ea
s
u
r
es
th
e
p
r
o
p
o
r
tio
n
o
f
co
r
r
ec
tly
class
if
ied
s
am
p
les ac
r
o
s
s
all
clas
s
es,
p
r
o
v
id
in
g
an
o
v
er
all
m
ea
s
u
r
e
o
f
class
if
icatio
n
p
er
f
o
r
m
an
ce
.
Ma
cr
o
F1
-
s
co
r
e
p
r
o
v
id
es
a
b
alan
ce
d
ev
alu
atio
n
ac
r
o
s
s
class
e
s
b
y
co
m
p
u
tin
g
th
e
F1
-
s
co
r
e
in
d
ep
en
d
en
tly
f
o
r
ea
c
h
class
a
n
d
th
en
av
er
a
g
in
g
th
e
r
esu
lts
.
T
h
is
m
etr
ic
is
p
ar
ticu
lar
ly
u
s
e
f
u
l
f
o
r
m
u
lti
-
class
class
if
icatio
n
task
s
b
ec
au
s
e
it t
r
ea
ts
all
class
e
s
eq
u
ally
r
eg
ar
d
less
o
f
class
f
r
eq
u
en
cy
.
Acc
u
r
ac
y
is
d
ef
in
ed
as:
=
(
+
)
(
+
+
+
)
(
4
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
207
-
21
8
212
wh
er
e
,
,
,
an
d
r
ep
r
esen
t
th
e
n
u
m
b
e
r
o
f
tr
u
e
p
o
s
itiv
es,
tr
u
e
n
eg
ativ
es,
f
alse
p
o
s
itiv
es,
an
d
f
alse
n
eg
ativ
es,
r
esp
ec
tiv
ely
.
Fo
r
m
u
lti
-
class
clas
s
if
icatio
n
,
ac
cu
r
ac
y
is
co
m
p
u
ted
in
a
o
n
e
-
vs
-
r
est
m
an
n
e
r
,
wh
er
e
th
ese
v
alu
es a
r
e
a
g
g
r
eg
ated
ac
r
o
s
s
all
class
es.
T
h
e
F1
-
s
co
r
e
c
o
m
b
in
es
p
r
ec
is
io
n
an
d
r
ec
all
in
t
o
a
s
in
g
le
h
ar
m
o
n
ic
m
ea
n
m
ea
s
u
r
e
f
o
r
ea
ch
class
.
I
n
th
is
s
tu
d
y
,
th
e
m
ac
r
o
F1
-
s
co
r
e
is
co
m
p
u
ted
b
y
av
er
a
g
in
g
th
e
F1
-
s
co
r
es a
cr
o
s
s
all
class
e
s
:
1
=
(
1
)
∑
1
=
1
(
5
)
wh
er
e
K
r
ep
r
esen
ts
th
e
n
u
m
b
e
r
o
f
class
es a
n
d
F1
k
d
en
o
tes th
e
F1
-
s
co
r
e
f
o
r
class
k
.
Fo
r
ea
c
h
class
,
1
=
2
+
(
6
)
wh
er
e
P
k
an
d
R
k
d
e
n
o
te
p
r
ec
is
i
o
n
an
d
r
ec
all
f
o
r
class
k
,
r
esp
e
ctiv
ely
.
3
.5
.
Co
nfusi
o
n m
a
t
rix
ev
a
lua
t
io
n
T
o
an
aly
ze
class
-
lev
el
class
if
i
ca
tio
n
b
eh
av
io
r
,
co
n
f
u
s
io
n
m
a
tr
ices
wer
e
g
en
er
ated
s
ep
ar
ately
f
o
r
ea
ch
d
ataset
d
o
m
ain
.
R
o
w
-
wis
e
n
o
r
m
aliza
tio
n
was
ap
p
lie
d
s
o
th
at
ea
ch
r
o
w
r
ep
r
esen
ts
th
e
p
r
o
p
o
r
tio
n
o
f
p
r
ed
ictio
n
s
r
elativ
e
to
th
e
tr
u
e
class
.
T
h
is
ev
alu
atio
n
en
ab
les
id
en
tific
atio
n
o
f
c
o
r
al
ca
teg
o
r
ies
th
at
ar
e
d
if
f
icu
lt
to
d
i
s
tin
g
u
is
h
an
d
p
r
o
v
id
es
in
s
ig
h
t
in
to
m
is
class
if
icatio
n
p
atter
n
s
o
b
s
er
v
ed
in
b
o
th
tex
tu
r
e
-
d
o
m
in
an
t
an
d
m
o
r
p
h
o
lo
g
y
-
f
o
c
u
s
ed
im
ag
er
y
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
ab
le
3
s
u
m
m
ar
izes
th
e
clas
s
if
icatio
n
p
er
f
o
r
m
a
n
ce
o
f
b
as
elin
e
an
d
p
r
o
p
o
s
ed
m
o
d
els
a
cr
o
s
s
b
o
th
co
r
al
im
ag
e
d
atasets
.
C
o
n
v
e
n
tio
n
al
R
GB
m
o
d
els
ac
h
iev
ed
h
ig
h
ac
c
u
r
ac
y
u
n
d
er
te
x
tu
r
e
-
r
ich
R
SMAS
co
n
d
itio
n
s
b
u
t
s
h
o
wed
lo
wer
ac
cu
r
ac
y
o
n
m
o
r
p
h
o
lo
g
y
-
f
o
c
u
s
ed
s
tr
u
ctu
r
e
R
SMAS
im
ag
es.
I
n
co
n
tr
ast,
th
e
p
r
o
p
o
s
ed
L
AB
+D
W
T
m
o
d
el
i
m
p
r
o
v
e
d
s
tr
u
ctu
r
e
R
SMAS
ac
cu
r
ac
y
an
d
r
ed
u
ce
d
t
h
e
d
ataset
-
lev
el
d
o
m
ain
g
a
p
b
y
co
m
b
in
in
g
L
AB
-
b
ased
c
o
lo
r
r
ep
r
esen
tatio
n
with
wav
ele
t
-
d
er
iv
ed
s
tr
u
ctu
r
al
d
escr
ip
to
r
s
.
Fo
r
q
u
an
titativ
e
co
m
p
ar
is
o
n
,
th
e
d
o
m
ain
g
ap
was
co
m
p
u
ted
as
th
e
ab
s
o
lu
te
d
if
f
e
r
en
ce
in
class
if
icatio
n
ac
cu
r
ac
y
b
etwe
en
th
e
two
d
atasets
.
T
h
is
co
m
p
a
r
is
o
n
s
h
o
ws
th
at
m
ix
e
d
-
d
o
m
ain
ac
cu
r
ac
y
a
n
d
d
o
m
ai
n
-
g
ap
r
e
d
u
c
tio
n
ar
e
r
elate
d
b
u
t
d
is
tin
ct
ev
alu
atio
n
o
b
jectiv
es.
T
ab
le
3
.
Ma
in
c
o
m
p
ar
ativ
e
an
d
ab
latio
n
p
er
f
o
r
m
an
ce
o
f
b
as
elin
e
an
d
p
r
o
p
o
s
ed
m
o
d
els u
n
d
e
r
th
e
s
in
g
le
b
est
-
f
o
ld
ev
alu
atio
n
p
r
o
to
co
l
M
o
d
e
l
a
r
c
h
i
t
e
c
t
u
r
e
C
o
l
o
r
S
t
r
u
c
t
u
r
e
R
S
M
A
S
a
c
c
u
r
a
c
y
(
%)
R
S
M
A
S
F1
S
t
r
u
c
t
u
r
e
R
S
M
A
S
a
c
c
u
r
a
c
y
(
%)
S
t
r
u
c
t
u
r
e
F1
D
o
ma
i
n
g
ap
(
%)
C
V
a
cc
u
r
a
c
y
(
%)
C
V
F
1
D
W
T
-
o
n
l
y
—
D
W
T(
L)
2
0
.
0
0
0
.
1
3
8
1
5
.
0
0
0
.
1
0
9
5
.
0
0
1
7
.
0
3
±
1
.
9
2
0
.
1
3
0
R
G
B
(
R
e
sN
e
t
-
1
8
)
R
G
B
—
9
8
.
2
6
0
.
9
8
2
8
3
.
3
3
0
.
8
2
3
1
4
.
9
3
9
2
.
8
5
±
1
.
5
2
0
.
9
2
7
R
G
B
(
R
e
sN
e
t
-
5
0
)
R
G
B
—
9
9
.
1
3
0
.
9
8
4
7
6
.
6
7
0
.
7
6
3
2
2
.
4
6
9
1
.
6
6
±
2
.
8
6
0
.
9
1
6
R
G
B
+
D
W
T
R
G
B
D
W
T(
L)
9
4
.
7
8
0
.
9
3
7
7
6
.
6
7
0
.
7
5
2
1
8
.
1
1
9
3
.
0
5
±
2
.
2
1
0
.
9
2
7
LA
B
(
R
e
sN
e
t
-
1
8
)
L*
a
*
b
*
—
9
8
.
2
6
0
.
9
8
2
8
6
.
6
7
0
.
8
8
2
1
1
.
5
9
9
0
.
2
9
±
2
.
2
3
0
.
9
0
2
P
r
o
p
o
se
d
LA
B
+
D
W
T
L*
a
*
b
*
D
W
T(
L)
9
6
.
5
2
0
.
9
7
4
8
8
.
3
3
0
.
8
8
8
8
.
1
9
8
9
.
4
0
±
2
.
0
8
0
.
8
8
9
N
o
t
e
:
T
e
st
-
se
t
a
c
c
u
ra
c
y
,
F
1
-
sc
o
re,
a
n
d
d
o
m
a
i
n
g
a
p
a
r
e
r
e
p
o
rt
e
d
u
n
d
e
r
t
h
e
si
n
g
l
e
b
e
s
t
-
f
o
l
d
e
v
a
l
u
a
t
i
o
n
p
ro
t
o
c
o
l
;
C
V
Ac
c
u
r
a
c
y
a
n
d
C
V
F1
s
u
m
m
a
r
i
ze
5
-
f
o
l
d
c
r
o
ss
-
v
a
l
i
d
a
t
i
o
n
p
e
r
f
o
rm
a
n
c
e
.
4
.
1
.
P
er
f
o
r
m
a
nce,
a
bla
t
i
o
n study
,
a
nd
co
m
plem
ent
a
rit
y
o
f
L
AB
a
nd
DW
T
T
h
e
ab
latio
n
co
n
f
ig
u
r
atio
n
s
i
n
T
ab
le
3
is
o
late
th
e
c
o
n
tr
ib
u
tio
n
o
f
ea
ch
co
m
p
o
n
e
n
t.
T
h
e
DW
T
-
o
n
ly
co
n
f
ig
u
r
atio
n
ac
h
iev
es
o
n
ly
1
7
.
0
3
%
cr
o
s
s
-
v
alid
atio
n
ac
c
u
r
ac
y
,
in
d
icatin
g
th
at
th
e
s
tr
u
ctu
r
al
d
escr
ip
to
r
s
p
r
o
v
id
e
wea
k
s
tan
d
alo
n
e
d
is
cr
im
in
ativ
e
p
o
wer
an
d
t
h
at
t
h
e
DW
T
b
r
an
ch
f
u
n
ctio
n
s
as
a
co
m
p
lem
en
tar
y
f
ea
tu
r
e
s
o
u
r
ce
r
ath
er
t
h
an
a
s
el
f
-
s
u
f
f
icien
t c
lass
if
ier
.
T
h
e
R
GB
+D
W
T
co
n
f
ig
u
r
atio
n
ac
h
ie
v
es
th
e
h
ig
h
est
m
ix
ed
-
d
o
m
ain
cr
o
s
s
-
v
alid
atio
n
ac
cu
r
ac
y
am
o
n
g
th
e
ev
alu
ate
d
c
o
n
f
ig
u
r
atio
n
s
,
b
u
t
it
d
o
es
n
o
t
r
ed
u
c
e
th
e
d
o
m
ain
g
a
p
.
I
ts
d
o
m
ain
g
ap
o
f
1
8
.
1
1
%
is
la
r
g
er
th
an
th
at
o
f
th
e
R
GB
R
esNet
-
1
8
b
a
s
elin
e
(
1
4
.
9
3
%),
s
u
g
g
esti
n
g
t
h
at
s
tr
u
ctu
r
al
d
escr
ip
to
r
s
p
a
ir
e
d
with
R
GB
d
o
n
o
t
s
u
f
f
icien
tly
ad
d
r
ess
cr
o
s
s
-
d
o
m
ain
v
ar
iatio
n
.
I
n
co
n
tr
ast,
th
e
p
r
o
p
o
s
ed
L
AB
+D
W
T
ac
h
iev
es
th
e
s
m
allest
d
o
m
ain
g
ap
am
o
n
g
th
e
co
lo
r
-
b
ased
class
if
ier
s
(
8
.
1
9
%)
an
d
th
e
h
ig
h
est
s
tr
u
ctu
r
e
R
SMAS
ac
cu
r
ac
y
(
8
8
.
3
3
%),
alth
o
u
g
h
it d
o
es n
o
t
m
ax
im
ize
m
ix
ed
-
d
o
m
ain
c
r
o
s
s
-
v
alid
atio
n
ac
cu
r
ac
y
.
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:
2502
-
4
7
5
2
C
o
r
a
l c
la
s
s
ifica
tio
n
in
u
n
d
erw
a
ter ima
g
es u
s
in
g
a
d
u
a
l
-
b
r
a
n
ch
d
ee
p
le
a
r
n
in
g
…
(
P
r
a
ch
a
r
a
t S
a
-
n
g
a
d
s
u
p
)
213
T
h
ese
r
esu
lts
in
d
icate
th
at
DW
T
is
n
o
t
u
n
iv
er
s
ally
b
en
ef
i
cial
b
y
its
elf
.
L
AB
r
ep
r
esen
ts
lu
m
in
an
ce
an
d
c
h
r
o
m
aticity
i
n
s
ep
ar
at
e
ch
an
n
els,
wh
ile
DW
T
ca
p
tu
r
es
h
ig
h
-
f
r
e
q
u
en
c
y
s
tr
u
ct
u
r
al
cu
es
f
r
o
m
th
e
lu
m
in
an
ce
ch
a
n
n
el.
T
h
e
a
b
latio
n
s
h
o
ws
th
at
DW
T
co
n
tr
ib
u
tes
m
o
s
t
ef
f
ec
tiv
ely
wh
en
p
air
ed
with
th
e
L
AB
r
ep
r
esen
tatio
n
,
s
u
g
g
esti
n
g
t
h
at
s
tr
u
ctu
r
al
d
escr
ip
to
r
s
an
d
lu
m
in
an
ce
/ch
r
o
m
atic
s
ep
ar
ati
o
n
ac
t
t
o
g
eth
er
to
r
ed
u
ce
th
e
d
o
m
ain
g
ap
ac
r
o
s
s
im
ag
in
g
co
n
d
itio
n
s
.
4
.
2
.
E
ns
em
ble
ev
a
l
ua
t
io
n a
n
d do
m
a
in
-
g
a
p a
na
ly
s
is
T
ab
le
4
r
e
p
o
r
ts
th
e
f
iv
e
-
f
o
ld
e
n
s
em
b
le
ac
cu
r
ac
y
,
en
s
em
b
le
g
ain
,
an
d
en
s
em
b
le
d
o
m
ai
n
g
ap
f
o
r
ea
ch
co
n
f
ig
u
r
atio
n
.
E
n
s
em
b
le
av
er
ag
in
g
im
p
r
o
v
ed
s
tr
u
ctu
r
e
R
SMAS
ac
cu
r
ac
y
ac
r
o
s
s
m
o
s
t
co
n
f
ig
u
r
atio
n
s
,
m
o
s
t
s
u
b
s
tan
tially
f
o
r
R
GB
R
e
s
Net
-
5
0
(
+1
0
.
0
0
%),
f
o
llo
we
d
b
y
R
GB
+D
W
T
(
+5
.
0
0
%).
Fo
r
th
e
p
r
o
p
o
s
ed
L
AB
+D
W
T
,
en
s
em
b
le
g
ain
o
n
s
tr
u
ctu
r
e
R
SMAS
was
m
o
d
est
(
+1
.
6
7
%),
r
ef
lectin
g
th
at
t
h
e
s
in
g
le
b
est
-
f
o
ld
m
o
d
el
alr
ea
d
y
ac
h
iev
es h
i
g
h
s
tr
u
ctu
r
e
R
SMAS a
cc
u
r
ac
y
(
8
8
.
3
3
%).
Am
o
n
g
th
e
c
o
lo
r
-
b
ased
cla
s
s
if
ier
s
,
th
e
p
r
o
p
o
s
ed
L
AB
+D
W
T
ac
h
iev
es
th
e
s
m
allest
en
s
em
b
le
d
o
m
ain
g
ap
(
9
.
1
3
%)
an
d
th
e
h
i
g
h
e
s
t
s
tr
u
ctu
r
e
R
SMA
S
en
s
em
b
le
ac
cu
r
ac
y
(
9
0
.
0
0
%).
T
h
e
co
m
p
ar
is
o
n
with
R
GB
+
DW
T
is
p
ar
ticu
lar
ly
in
f
o
r
m
ativ
e:
R
GB
+D
W
T
ac
h
iev
es
h
ig
h
e
r
m
i
x
ed
-
d
o
m
ain
cr
o
s
s
-
v
alid
atio
n
accu
r
ac
y
(
9
3
.
0
5
%
v
s
.
8
9
.
4
0
%)
b
u
t
a
lar
g
er
en
s
em
b
le
d
o
m
ai
n
g
ap
(
1
6
.
5
9
%
v
s
.
9
.
1
3
%),
in
d
icatin
g
th
at
m
ix
e
d
-
d
o
m
ain
ac
c
u
r
ac
y
a
n
d
cr
o
s
s
-
d
o
m
ain
r
o
b
u
s
tn
ess
r
ef
lect
d
is
tin
ct
asp
ec
ts
o
f
m
o
d
el
b
e
h
av
io
r
.
T
ab
le
4
.
E
n
s
em
b
le
ev
alu
atio
n
an
d
d
o
m
ain
-
g
a
p
an
aly
s
is
ac
r
o
s
s
ev
alu
ated
co
n
f
ig
u
r
atio
n
s
M
o
d
e
l
R
S
M
A
S
e
n
s
e
mb
l
e
(
%)
R
S
M
A
S
s
i
n
g
l
e
g
a
i
n
(
%)
S
t
r
u
c
t
u
r
e
e
n
s
e
mb
l
e
(
%)
S
t
r
u
c
t
u
r
e
g
a
i
n
(
%)
En
se
mb
l
e
g
a
p
(
%)
D
W
T
-
o
n
l
y
2
1
.
7
4
+
1
.
7
4
1
8
.
3
3
+
3
.
3
3
3
.
4
1
R
G
B
(
R
e
sN
e
t
-
1
8
)
9
9
.
1
3
+
0
.
8
7
8
3
.
3
3
0
.
0
0
1
5
.
8
0
R
G
B
(
R
e
sN
e
t
-
5
0
)
9
9
.
1
3
0
.
0
0
8
6
.
6
7
+
1
0
.
0
0
1
2
.
4
6
R
G
B
+
D
W
T
9
8
.
2
6
+
3
.
4
8
8
1
.
6
7
+
5
.
0
0
1
6
.
5
9
LA
B
(
R
e
sN
e
t
-
1
8
)
9
8
.
2
6
0
.
0
0
8
6
.
6
7
0
.
0
0
1
1
.
5
9
P
r
o
p
o
se
d
LA
B
+
D
W
T
9
9
.
1
3
+
2
.
6
1
9
0
.
0
0
+
1
.
6
7
9
.
1
3
N
o
t
e
:
En
s
e
m
b
l
e
G
a
p
=
|
RS
MA
S
E
n
s
e
m
b
l
e
(
%)
−
S
t
r
u
c
t
u
re
En
s
e
m
b
l
e
(
%
)
|
.
G
a
i
n
=
e
n
sem
b
l
e
a
c
c
u
r
a
c
y
−
si
n
g
l
e
b
e
s
t
-
f
o
l
d
a
c
c
u
r
a
c
y
(
f
ro
m
T
a
b
l
e
3
)
.
D
WT
-
o
n
l
y
i
s
i
n
c
l
u
d
e
d
f
o
r
a
b
l
a
t
i
o
n
c
o
m
p
l
e
t
e
n
e
s
s
b
u
t
i
s
n
o
t
c
o
n
s
i
d
e
re
d
a
p
ra
c
t
i
c
a
l
c
l
a
ss
i
f
i
e
r
.
4
.
3
.
Co
m
pa
riso
n wit
h
prio
r
m
et
ho
ds
T
h
e
p
r
o
p
o
s
ed
L
AB
+D
W
T
f
r
am
ewo
r
k
ad
d
r
ess
es
th
e
cr
o
s
s
-
d
o
m
ain
g
a
p
d
ir
ec
tly
.
C
o
m
p
ar
ed
with
L
u
m
i
n
i
et
a
l.
[
8
]
,
wh
ich
r
ep
o
r
ted
9
9
.
2
0
%
ac
cu
r
ac
y
o
n
R
SMAS,
th
e
p
r
o
p
o
s
ed
L
AB
+D
W
T
m
o
d
el
ac
h
iev
es
a
s
lig
h
tly
lo
wer
R
SMA
S a
cc
u
r
ac
y
o
f
9
6
.
5
2
%.
Ho
wev
er
,
th
e
p
r
o
p
o
s
ed
m
o
d
el
also
ac
h
iev
es 8
8
.
3
3
% a
cc
u
r
ac
y
o
n
s
tr
u
ctu
r
e
R
SMAS
(
T
ab
le
5
)
,
wh
ich
was
n
o
t
ev
al
u
ated
b
y
L
u
m
in
i
et
a
l.
[
8
]
,
an
d
r
e
d
u
ce
s
th
e
d
o
m
ain
g
ap
to
8
.
1
9
%.
T
h
is
tr
ad
e
-
o
f
f
in
d
icate
s
th
at
th
e
p
r
o
p
o
s
ed
d
u
al
-
b
r
an
c
h
r
ep
r
esen
tatio
n
ac
ce
p
ts
a
s
m
all
r
ed
u
ctio
n
in
to
p
-
lin
e
R
SMAS
ac
cu
r
ac
y
wh
ile
im
p
r
o
v
in
g
g
en
er
aliza
tio
n
to
m
o
r
p
h
o
lo
g
y
-
f
o
cu
s
ed
im
a
g
er
y
u
n
d
er
a
u
n
if
ied
ev
alu
atio
n
p
r
o
to
co
l.
T
ab
le
5
.
C
o
m
p
a
r
is
o
n
with
p
r
i
o
r
m
eth
o
d
s
o
n
R
SMAS a
n
d
s
tr
u
ctu
r
e
R
SMAS d
atasets
M
e
t
h
o
d
I
n
p
u
t
R
S
M
A
S
a
c
c
u
r
a
c
y
(
%)
S
t
r
u
c
t
u
r
e
R
S
M
A
S
a
c
c
u
r
a
c
y
(
%)
D
o
ma
i
n
g
a
p
(
%)
S
h
i
h
a
v
u
d
d
i
n
e
t
a
l
.
(
2
0
1
3
)
[
9
]
R
G
B
9
2
.
7
4
%
—
—
G
ó
mez
-
R
í
o
s
e
t
a
l
.
(
2
0
1
9
)
[
1
5
]
R
G
B
9
8
.
3
6
%
8
5
.
2
6
%
1
3
.
1
0
%
Lu
m
i
n
i
e
t
a
l
.
(
2
0
2
3
)
[
8
]
R
G
B
(
C
N
N
e
n
sem
b
l
e
)
9
9
.
2
0
%
—
—
P
r
o
p
o
se
d
LA
B
+
D
W
T
L*
a
*
b
*
+
D
W
T
9
6
.
5
2
%
8
8
.
3
3
%
8
.
1
9
%
P
r
o
p
o
se
d
LA
B
+
D
W
T
+
En
s
e
m
b
l
e
L*
a
*
b
*
+
D
W
T
9
9
.
1
3
%
9
0
.
0
0
%
9
.
1
3
%
N
o
t
e
:
Me
t
h
o
d
s
n
o
t
e
v
a
l
u
a
t
e
d
o
n
a
d
a
t
a
se
t
a
re
m
a
rk
e
d
a
s
“
—
”
.
D
o
m
a
i
n
g
a
p
i
s
c
o
m
p
u
t
e
d
a
s
t
h
e
a
b
so
l
u
t
e
d
i
f
f
e
r
e
n
c
e
i
n
a
c
c
u
r
a
c
y
b
e
t
w
e
e
n
t
h
e
t
w
o
d
a
t
a
s
e
t
s
.
T
h
e
e
n
s
e
m
b
l
e
r
o
w
re
p
o
rt
s
t
h
e
f
i
v
e
-
f
o
l
d
e
n
sem
b
l
e
r
e
su
l
t
f
o
r
b
o
t
h
R
S
MA
S
a
n
d
S
t
r
u
c
t
u
r
e
R
S
MA
S
.
4
.
4
.
St
a
t
is
t
ica
l
s
ig
nifica
nce
T
ab
le
6
r
e
p
o
r
ts
p
air
e
d
t
-
test
s
ac
r
o
s
s
th
e
5
-
f
o
l
d
cr
o
s
s
-
v
al
id
atio
n
ac
cu
r
ac
y
to
ass
ess
wh
eth
er
th
e
p
r
o
p
o
s
ed
L
AB
+D
W
T
m
o
d
el
s
ig
n
if
ican
tly
im
p
r
o
v
es
m
ix
ed
-
d
o
m
ain
v
alid
atio
n
p
e
r
f
o
r
m
a
n
ce
co
m
p
ar
ed
with
th
e
b
aselin
e
an
d
ab
latio
n
co
n
f
ig
u
r
atio
n
s
.
T
h
e
c
o
m
p
a
r
is
o
n
with
DW
T
-
o
n
ly
is
h
ig
h
ly
s
ig
n
i
f
ican
t
(
p
<
0
.
0
0
1
)
,
in
d
icatin
g
th
at
th
e
DW
T
s
tr
u
ctu
r
al
b
r
an
ch
is
n
o
t
s
u
f
f
icien
t
as
a
s
tan
d
alo
n
e
class
if
ier
an
d
r
eq
u
ir
es
th
e
co
lo
r
b
r
an
ch
.
No
s
tatis
tically
s
ig
n
if
ican
t
d
if
f
er
e
n
ce
was
d
etec
ted
a
m
o
n
g
t
h
e
co
m
p
ar
is
o
n
s
b
etwe
e
n
L
AB
+D
W
T
an
d
th
e
co
lo
r
-
b
ased
c
o
n
f
ig
u
r
atio
n
s
at
α
=
0
.
0
5
.
B
ec
au
s
e
th
e
m
a
in
o
b
jectiv
e
o
f
th
is
s
tu
d
y
is
r
e
d
u
cin
g
th
e
d
o
m
ain
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
207
-
21
8
214
g
ap
r
ath
er
th
a
n
m
ax
im
izin
g
m
ix
ed
-
d
o
m
ain
v
alid
atio
n
ac
c
u
r
a
cy
,
th
e
p
air
ed
t
-
test
r
esu
lts
s
h
o
u
ld
b
e
in
ter
p
r
eted
to
g
eth
er
with
th
e
e
n
s
em
b
le
d
o
m
ain
-
g
ap
a
n
aly
s
is
p
r
esen
ted
i
n
T
a
b
le
4
.
T
ab
le
6
.
Statis
tical
s
ig
n
if
ican
ce
o
f
cr
o
s
s
-
v
alid
atio
n
ac
cu
r
ac
y
d
if
f
er
en
ce
s
(
p
air
ed
t
-
test
,
5
-
f
o
l
d
C
V)
C
o
m
p
a
r
i
so
n
M
e
a
n
d
i
f
f
e
r
e
n
c
e
(
%)
t
-
st
a
t
p
-
v
a
l
u
e
9
5
%
C
I
LA
B
+
D
W
T
v
s
.
D
W
T
-
o
n
l
y
+
7
2
.
3
7
6
3
.
3
5
9
<
0
.
0
0
1
[
+
6
9
.
2
0
,
+
7
5
.
5
4
]
LA
B
+
D
W
T
v
s
.
R
G
B
(
R
e
sN
e
t
-
5
0
)
−
2
.
2
6
−
1
.
6
3
1
0
.
1
7
8
[
−
6
.
1
1
,
+
1
.
5
9
]
LA
B
+
D
W
T
v
s
.
R
G
B
(
R
e
sN
e
t
-
1
8
)
−
3
.
4
5
−
2
.
4
1
5
0
.
0
7
3
[
−
7
.
4
2
,
+
0
.
5
2
]
LA
B
+
D
W
T
v
s
.
LA
B
(
R
e
sN
e
t
-
1
8
)
−
0
.
8
9
−
2
.
0
8
7
0
.
1
0
5
[
−
2
.
0
8
,
+
0
.
3
0
]
LA
B
+
D
W
T
v
s
.
R
G
B
+
D
W
T
−
3
.
6
5
−
1
.
9
7
1
0
.
1
2
0
[
−
8
.
7
9
,
+
1
.
4
9
]
4
.
5
.
Co
m
pu
t
a
t
io
na
l
e
f
f
iciency
T
ab
le
7
s
u
m
m
ar
izes
th
e
co
m
p
u
tatio
n
al
co
m
p
le
x
ity
an
d
in
f
er
en
ce
tim
e
o
f
th
e
ev
alu
a
ted
m
o
d
el
co
n
f
ig
u
r
atio
n
s
.
T
im
in
g
was
m
ea
s
u
r
ed
u
s
in
g
th
e
s
elec
ted
s
in
g
le
-
f
o
ld
m
o
d
el
o
n
1
7
5
test
im
ag
es
(
1
1
5
R
SMAS
an
d
6
0
s
tr
u
ctu
r
e
R
SMAS)
an
d
in
clu
d
es
p
r
ep
r
o
ce
s
s
in
g
,
L
AB
co
n
v
er
s
io
n
,
DW
T
co
m
p
u
ta
tio
n
,
an
d
f
o
r
war
d
p
r
o
p
a
g
atio
n
.
T
im
in
g
ex
p
e
r
im
en
ts
wer
e
co
n
d
u
cted
i
n
MA
T
L
AB
R
2
0
2
5
a
o
n
a
W
in
d
o
ws
1
1
Pro
d
esk
to
p
with
an
AM
D
R
y
ze
n
5
5
6
0
0
6
-
co
r
e
/1
2
-
th
r
ea
d
C
PU
at
3
.
5
0
GHz
a
n
d
3
2
GB
R
AM
u
s
in
g
C
P
U
ex
ec
u
tio
n
o
n
ly
.
T
h
e
p
r
o
p
o
s
ed
d
u
al
-
b
r
an
c
h
m
o
d
el
ad
d
s
o
n
ly
0
.
1
8
M
p
ar
am
ete
r
s
(
+1
.
6
%)
o
v
e
r
th
e
s
in
g
le
-
b
r
an
ch
R
esNet
-
1
8
b
aselin
e,
co
n
s
is
tin
g
o
f
th
e
lig
h
tweig
h
t
DW
T
b
r
an
ch
an
d
f
u
s
io
n
h
ea
d
.
T
h
e
esti
m
ated
m
o
d
el
s
ize
o
f
4
3
.
4
MB
is
co
n
s
is
ten
t
with
th
e
s
in
g
le
-
b
r
a
n
ch
R
esNet
-
1
8
b
aselin
e
(
4
2
.
8
MB),
co
n
f
ir
m
in
g
th
at
th
e
d
u
al
-
b
r
an
ch
ex
ten
s
io
n
in
tr
o
d
u
ce
s
n
e
g
lig
ib
le
s
to
r
ag
e
o
v
er
h
ea
d
.
T
r
ain
i
n
g
tim
e
f
o
r
th
e
p
r
o
p
o
s
ed
L
AB
+D
W
T
(
4
4
.
7
m
in
/f
o
ld
)
is
5
6
%
o
f
th
e
R
GB
R
e
s
Net
-
5
0
tr
ain
in
g
tim
e
(
8
0
.
2
m
i
n
/f
o
ld
)
wh
ile
ac
h
i
ev
in
g
a
s
m
aller
en
s
em
b
le
d
o
m
ain
g
ap
.
Per
-
im
a
g
e
in
f
er
en
ce
tim
e
is
co
m
p
ar
ab
le
ac
r
o
s
s
m
o
d
els
u
n
d
er
C
PU
ex
ec
u
tio
n
,
with
th
e
p
r
o
p
o
s
ed
L
AB
+D
W
T
r
eq
u
ir
in
g
0
.
0
3
1
s
ec
o
n
d
s
p
er
im
ag
e.
Fo
r
b
atch
p
r
o
ce
s
s
in
g
o
f
u
n
d
er
wat
er
s
u
r
v
ey
im
ag
er
y
,
wh
er
e
r
ea
l
-
tim
e
p
er
f
o
r
m
a
n
ce
is
n
o
t r
eq
u
ir
e
d
,
th
e
c
o
m
p
u
tatio
n
al
o
v
er
h
ea
d
is
li
m
ited
.
T
ab
le
7
.
Mo
d
el
co
m
p
lex
ity
an
d
in
f
er
e
n
ce
-
tim
e
co
m
p
ar
is
o
n
M
o
d
e
l
I
n
p
u
t
P
a
r
a
me
t
e
r
s
(
M
)
S
i
z
e
(
M
B
)
Tr
a
i
n
i
n
g
t
i
me
(
mi
n
/
f
o
l
d
)
I
n
f
e
r
e
n
c
e
t
i
me
To
t
a
l
(
s)
I
n
f
e
r
e
n
c
e
t
i
me
p
e
r
i
ma
g
e
(
s)
D
W
T
-
o
n
l
y
D
W
T(
L)
0
.
1
8
0
.
7
1
1
.
2
3
.
2
5
0
.
0
1
9
R
G
B
(
R
e
sN
e
t
-
1
8
)
R
G
B
1
1
.
2
1
4
2
.
8
2
8
.
6
5
.
6
7
0
.
0
3
2
R
G
B
(
R
e
sN
e
t
-
5
0
)
R
G
B
2
3
.
5
0
8
9
.
7
8
0
.
2
4
.
5
6
0
.
0
2
6
R
G
B
+
D
W
T
R
G
B
+
D
W
T(
L)
1
1
.
3
8
4
3
.
4
4
7
.
8
4
.
8
2
0
.
0
2
8
LA
B
(
R
e
sN
e
t
-
1
8
)
L*
a
*
b
*
1
1
.
2
1
4
2
.
8
3
6
.
7
4
.
5
6
0
.
0
2
6
P
r
o
p
o
se
d
LA
B
+
D
W
T
L*
a
*
b
*
+
D
W
T(
L)
1
1
.
3
8
4
3
.
4
4
4
.
7
5
.
4
0
0
.
0
3
1
N
o
t
e
:
T
r
a
i
n
i
n
g
t
i
m
e
,
s
i
n
g
l
e
f
o
l
d
,
3
0
e
p
o
c
h
s.
I
n
f
e
re
n
c
e
t
i
m
e
:
1
7
5
t
e
st
i
m
a
g
e
s,
si
n
g
l
e
-
f
o
l
d
m
o
d
e
l
.
Mo
d
e
l
s
i
ze
=
p
a
r
a
m
e
t
e
rs
×
4
b
y
t
e
s.
H
a
rd
w
a
r
e
a
n
d
t
i
m
i
n
g
p
r
o
t
o
c
o
l
a
s
d
e
s
c
ri
b
e
d
a
b
o
v
e
.
S
m
a
l
l
d
i
f
f
e
r
e
n
c
e
s
b
e
t
w
e
e
n
m
o
d
e
l
s ma
y
r
e
f
l
e
c
t
i
m
p
l
e
m
e
n
t
a
t
i
o
n
-
l
e
v
e
l
o
v
e
r
h
e
a
d
.
4
.
6
.
Co
nfusi
o
n m
a
t
rix
a
na
l
y
s
is
An
aly
s
is
o
f
th
e
r
o
w
-
n
o
r
m
aliz
ed
co
n
f
u
s
io
n
m
atr
ices
(
Fig
u
r
e
3
)
r
ev
ea
ls
th
at
clas
s
if
icatio
n
r
eliab
ilit
y
d
ep
en
d
s
s
tr
o
n
g
ly
o
n
m
o
r
p
h
o
lo
g
ical
d
is
tin
ctiv
en
ess
.
Sp
ec
ies
with
u
n
iq
u
e
s
tr
u
ctu
r
al
o
r
tex
tu
r
al
s
ig
n
atu
r
es
ex
h
ib
it
h
ig
h
s
ep
ar
a
b
ilit
y
,
wh
i
le
m
o
r
p
h
o
lo
g
ically
s
im
ilar
tax
a
r
em
ain
p
r
o
n
e
to
m
u
tu
al
c
o
n
f
u
s
io
n
,
r
ef
lectin
g
k
n
o
wn
ch
allen
g
es
in
v
is
u
al
co
r
al
tax
o
n
o
m
y
.
I
n
Fig
u
r
e
3
(
a
)
,
th
e
p
r
o
p
o
s
ed
L
AB
+D
W
T
en
s
em
b
le
ac
h
iev
es
p
er
f
ec
t
class
if
icatio
n
(
1
0
0
%
r
ec
all)
f
o
r
1
3
o
f
1
4
R
SMAS
s
p
ec
ies.
T
h
e
o
n
ly
m
is
class
if
ic
atio
n
in
v
o
lv
es
A
cro
p
o
r
a
p
a
lma
ta
(
APAL
)
,
wh
er
e
o
n
e
s
am
p
le
is
class
if
ie
d
as
A
cro
p
o
r
a
ce
r
vico
r
n
is
(
A
C
E
R
)
.
B
o
th
s
p
ec
ies
b
elo
n
g
to
th
e
g
en
u
s
A
cro
p
o
r
a
an
d
s
h
ar
e
s
im
ilar
b
r
an
ch
in
g
m
o
r
p
h
o
lo
g
y
,
m
ak
in
g
th
is
co
n
f
u
s
io
n
b
io
lo
g
icall
y
p
lau
s
ib
le.
I
n
Fig
u
r
e
3
(
b
)
,
th
e
s
tr
u
c
tu
r
e
R
SMAS
r
esu
lts
s
h
o
w
th
at
class
if
icatio
n
ac
cu
r
ac
y
r
e
m
ain
s
h
ig
h
o
v
er
all
(
8
8
.
3
3
%)
b
u
t
r
ev
ea
ls
in
ter
p
r
etab
le
er
r
o
r
p
atter
n
s
co
n
ce
n
t
r
ated
am
o
n
g
s
p
ec
ies
with
s
im
ilar
co
lo
n
y
-
lev
el
g
r
o
wth
f
o
r
m
s
.
Mil
lep
o
r
a
a
lci
co
r
n
is
(
MA
L
C
)
ex
h
ib
its
th
e
lo
west
r
ec
all,
with
m
is
c
las
s
if
icatio
n
s
d
is
tr
ib
u
ted
ac
r
o
s
s
A
cro
p
o
r
a
ce
r
vico
r
n
is
(
AC
E
R
)
an
d
Sp
o
n
g
e
(
SP
O)
,
all
o
f
wh
ich
ca
n
d
is
p
lay
s
im
ilar
m
ass
iv
e
o
r
b
r
an
ch
in
g
f
o
r
m
s
wh
en
v
iewe
d
at
co
lo
n
y
s
ca
le.
Go
r
g
o
n
ia
n
s
(
GORG)
s
h
o
w
co
n
f
u
s
io
n
with
AC
E
R
,
attr
ib
u
tab
le
to
s
h
ar
ed
b
r
an
ch
in
g
s
tr
u
ct
u
r
es
.
Mea
n
d
r
in
a
mea
n
d
r
ites
(
MM
E
A)
is
o
cc
asio
n
ally
co
n
f
u
s
ed
with
C
o
lp
o
p
h
yllia
n
a
ta
n
s
(
C
NAT
)
,
b
o
th
o
f
wh
ic
h
ex
h
ib
it e
n
cr
u
s
tin
g
o
r
b
r
ain
-
li
k
e
g
r
o
wth
m
o
r
p
h
o
lo
g
y
.
Dif
f
er
en
ce
s
b
etwe
en
th
e
two
d
atasets
h
ig
h
lig
h
t
th
e
r
o
le
o
f
v
is
u
al
s
ca
le.
I
n
th
e
R
SMA
S
d
o
m
ain
,
er
r
o
r
s
a
r
e
p
r
im
a
r
ily
d
r
iv
en
b
y
s
im
ilar
ities
in
f
in
e
-
s
ca
le
te
x
tu
r
e
an
d
p
ig
m
en
tatio
n
p
atte
r
n
s
,
wh
er
ea
s
in
th
e
s
tr
u
ctu
r
e
R
SMAS
d
o
m
ain
,
co
n
f
u
s
io
n
s
h
if
ts
to
war
d
t
ax
a
with
s
im
ilar
co
lo
n
y
-
le
v
el
g
r
o
wth
f
o
r
m
s
.
Nev
er
th
eless
,
ce
r
tain
class
p
air
s
—
p
ar
ticu
lar
ly
MA
L
C
–
AC
E
R
an
d
C
N
AT
–
MMEA
—
r
em
ain
d
if
f
icu
lt
to
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:
2502
-
4
7
5
2
C
o
r
a
l c
la
s
s
ifica
tio
n
in
u
n
d
erw
a
ter ima
g
es u
s
in
g
a
d
u
a
l
-
b
r
a
n
ch
d
ee
p
le
a
r
n
in
g
…
(
P
r
a
ch
a
r
a
t S
a
-
n
g
a
d
s
u
p
)
215
s
ep
ar
ate
ac
r
o
s
s
all
m
o
d
el
co
n
f
ig
u
r
atio
n
s
,
in
d
icatin
g
th
at
th
ese
class
if
icatio
n
am
b
ig
u
ities
ar
is
e
f
r
o
m
in
tr
in
s
ic
b
io
lo
g
ical
s
im
ilar
ity
r
ath
er
t
h
a
n
p
u
r
el
y
alg
o
r
ith
m
ic
lim
itatio
n
s
.
(
a)
(
b
)
Fig
u
r
e
3
.
R
o
w
-
n
o
r
m
alize
d
c
o
n
f
u
s
io
n
m
atr
ices o
f
t
h
e
p
r
o
p
o
s
e
d
L
AB
+D
W
T
en
s
em
b
le:
(
a
)
R
SMAS te
s
t set
an
d
(
b
)
Stru
ctu
r
e
R
SMAS te
s
t
s
et.
Diag
o
n
al
v
alu
es in
d
icate
p
er
-
class
r
ec
all.
Of
f
-
d
iag
o
n
al
en
tr
ies in
d
icate
th
e
p
r
o
p
o
r
tio
n
o
f
m
is
class
if
ied
s
am
p
les r
elativ
e
to
th
e
tr
u
e
c
lass
4
.
7
.
L
im
it
a
t
io
ns
Sev
er
al
lim
itatio
n
s
o
f
th
is
s
tu
d
y
s
h
o
u
ld
b
e
n
o
ted
.
First,
t
h
e
ev
alu
atio
n
is
b
ased
o
n
t
wo
r
elate
d
d
atasets
with
a
lim
ited
to
tal
s
am
p
le
s
ize,
an
d
th
e
s
tr
u
ctu
r
e
R
SMAS
te
s
t
s
et
co
n
tain
s
o
n
ly
6
0
im
a
g
es.
T
h
er
ef
o
r
e,
d
o
m
ain
-
lev
el
ac
c
u
r
ac
y
v
al
u
es
s
h
o
u
ld
b
e
i
n
ter
p
r
eted
with
awa
r
en
ess
o
f
th
e
d
is
cr
ete
1
.
6
7
p
er
ce
n
tag
e
-
p
o
in
t
s
tep
s
ize
p
er
s
am
p
le.
W
ith
s
u
ch
a
s
m
a
ll
test
s
et,
a
s
in
g
le
ad
d
itio
n
al
m
is
class
if
icatio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
207
-
21
8
216
ch
an
g
es
ac
cu
r
a
cy
b
y
1
.
6
7
p
er
ce
n
tag
e
p
o
in
ts
,
wh
ich
lim
its
th
e
p
r
ec
is
io
n
o
f
d
o
m
ai
n
-
g
ap
c
o
m
p
a
r
is
o
n
s
.
Seco
n
d
,
n
o
in
d
ep
en
d
en
t
r
ee
f
-
s
ite
v
alid
atio
n
was
in
clu
d
ed
;
b
o
th
d
atasets
o
r
ig
in
ate
f
r
o
m
C
ar
ib
b
ea
n
r
ee
f
en
v
ir
o
n
m
en
ts
,
an
d
g
en
e
r
aliza
tio
n
to
I
n
d
o
-
Pacif
ic
o
r
o
th
er
b
io
g
eo
g
r
ap
h
ic
r
e
g
io
n
s
r
em
ai
n
s
u
n
test
ed
.
T
h
ir
d
,
th
e
DW
T
p
ar
am
ete
r
s
(
d
b
2
wav
elet,
s
i
n
g
le
-
le
v
el
d
ec
o
m
p
o
s
itio
n
,
α
=
0
.
5
)
wer
e
s
elec
ted
b
ased
o
n
p
r
elim
in
ar
y
ex
p
er
im
en
ts
r
at
h
er
th
a
n
e
x
h
au
s
tiv
e
o
p
tim
iz
atio
n
.
E
a
r
ly
tr
ials
u
s
in
g
f
u
l
l
-
im
ag
e
FF
T
-
b
ased
f
r
eq
u
e
n
cy
r
e
p
r
esen
tatio
n
s
d
id
n
o
t
p
r
o
d
u
ce
m
ea
n
in
g
f
u
l
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
,
wh
ich
m
o
tiv
ated
th
e
u
s
e
o
f
s
p
atially
lo
ca
lized
wav
elet
d
ec
o
m
p
o
s
itio
n
.
Ho
wev
e
r
,
alter
n
ativ
e
wav
elet
f
am
ilies
o
r
m
u
lti
-
lev
el
d
ec
o
m
p
o
s
itio
n
s
wer
e
n
o
t
s
y
s
tem
atica
lly
co
m
p
ar
ed
an
d
m
ay
y
ield
d
if
f
e
r
en
t
r
esu
lts
.
Fo
u
r
th
,
th
e
cu
r
r
en
t
ev
alu
atio
n
d
o
es
n
o
t
in
clu
d
e
co
m
p
ar
is
o
n
with
v
is
io
n
tr
an
s
f
o
r
m
er
s
(
ViT
)
o
r
atten
tio
n
-
b
ased
f
u
s
io
n
m
ec
h
an
is
m
s
,
wh
ich
h
av
e
s
h
o
wn
s
tr
o
n
g
p
er
f
o
r
m
an
ce
in
o
th
er
f
in
e
-
g
r
ai
n
ed
r
ec
o
g
n
itio
n
task
s
.
Fin
ally
,
all
in
f
er
en
ce
tim
in
g
was
m
ea
s
u
r
ed
u
s
in
g
a
C
PU
-
o
n
l
y
MA
T
L
AB
im
p
lem
en
tatio
n
;
GPU
-
ac
ce
ler
ated
d
ep
lo
y
m
e
n
t
wo
u
ld
y
ield
d
if
f
e
r
en
t
r
u
n
tim
e
ch
ar
ac
ter
is
tics
.
Fu
tu
r
e
r
esear
ch
m
ay
also
in
v
es
tig
ate
tr
an
s
f
o
r
m
er
-
b
ased
ar
ch
itectu
r
es
f
o
r
c
o
r
al
class
if
icatio
n
.
ViT
m
o
d
els
h
a
v
e
d
em
o
n
s
tr
ated
s
tr
o
n
g
ca
p
a
b
ilit
y
in
m
o
d
elin
g
lo
n
g
-
r
a
n
g
e
s
p
atial
r
elatio
n
s
h
i
p
s
an
d
m
ay
p
r
o
v
id
e
ad
d
itio
n
a
l
b
e
n
ef
its
f
o
r
c
o
lo
n
y
-
lev
el
im
ag
er
y
,
wh
e
r
e
g
lo
b
al
s
tr
u
ctu
r
al
in
f
o
r
m
atio
n
is
p
ar
tic
u
lar
ly
im
p
o
r
tan
t [
3
2
]
.
5.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
in
v
esti
g
ated
c
o
r
al
im
ag
e
class
if
icatio
n
u
n
d
e
r
t
wo
im
ag
in
g
r
eg
im
es:
tex
tu
r
e
-
d
o
m
in
a
n
t
clo
s
e
-
r
an
g
e
im
a
g
er
y
an
d
m
o
r
p
h
o
lo
g
y
-
f
o
cu
s
ed
co
lo
n
y
-
lev
el
im
ag
er
y
.
R
GB
-
b
ased
n
et
wo
r
k
s
ac
h
iev
e
h
ig
h
ac
cu
r
ac
y
in
tex
tu
r
e
-
r
ich
co
n
d
itio
n
s
b
u
t
ex
h
i
b
it
r
ed
u
ce
d
p
e
r
f
o
r
m
an
ce
wh
en
d
is
cr
im
in
ativ
e
cu
es
s
h
if
t
to
war
d
co
lo
n
y
-
le
v
el
m
o
r
p
h
o
lo
g
y
,
in
d
i
ca
tin
g
a
d
ep
en
d
en
ce
o
n
f
in
e
-
s
ca
le
tex
tu
r
e
f
ea
tu
r
es.
T
o
a
d
d
r
e
s
s
th
is
l
im
itatio
n
,
a
d
u
al
-
b
r
a
n
ch
f
r
am
ewo
r
k
i
n
teg
r
atin
g
C
I
E
L
AB
co
lo
r
r
ep
r
ese
n
tatio
n
with
DW
T
-
d
er
iv
ed
s
tr
u
ctu
r
al
d
escr
ip
to
r
s
was
p
r
o
p
o
s
ed
.
B
y
s
ep
a
r
atin
g
lu
m
in
a
n
ce
a
n
d
ch
r
o
m
atic
co
m
p
o
n
en
ts
a
n
d
in
co
r
p
o
r
at
in
g
h
i
g
h
-
f
r
eq
u
e
n
cy
s
tr
u
ctu
r
al
in
f
o
r
m
atio
n
t
h
r
o
u
g
h
Dau
b
ec
h
ies
-
2
wav
elet
d
ec
o
m
p
o
s
itio
n
,
th
e
ar
ch
i
tectu
r
e
in
tr
o
d
u
ce
s
co
m
p
lem
en
tar
y
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
p
r
io
r
t
o
co
n
v
o
lu
tio
n
al
ab
s
tr
ac
tio
n
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
r
e
q
u
ir
es
o
n
ly
1
.
6
%
ad
d
itio
n
al
p
ar
am
eter
s
o
v
er
th
e
s
in
g
le
-
b
r
an
c
h
R
esNet
-
1
8
b
aselin
e
(
1
1
.
3
8
M
v
s
.
1
1
.
2
1
M)
.
Un
d
er
th
e
s
in
g
le
b
est
-
f
o
ld
ev
alu
atio
n
p
r
o
to
co
l,
t
h
e
p
r
o
p
o
s
ed
L
AB
+D
W
T
m
o
d
el
ac
h
iev
e
d
9
6
.
5
2
%
ac
c
u
r
ac
y
o
n
R
SMAS
an
d
8
8
.
3
3
%
ac
cu
r
ac
y
o
n
s
tr
u
ctu
r
e
R
SMAS,
r
ed
u
cin
g
th
e
d
o
m
ain
g
ap
f
r
o
m
2
2
.
4
6
%
f
o
r
th
e
R
GB
R
esNet
-
5
0
b
aselin
e
to
8
.
1
9
%,
in
d
icatin
g
im
p
r
o
v
e
d
g
en
e
r
aliza
tio
n
wh
ile
m
ain
tain
in
g
a
c
o
m
p
ac
t c
o
m
p
u
tatio
n
al
p
r
o
f
ile.
An
aly
s
is
o
f
co
n
f
u
s
io
n
m
atr
ice
s
s
h
o
ws th
at
r
em
ain
in
g
m
is
class
if
icatio
n
s
ar
e
p
r
im
ar
ily
ass
o
ciate
d
with
m
o
r
p
h
o
lo
g
ically
s
im
ilar
tax
a.
Sp
ec
ies
with
d
is
tin
ctiv
e
s
tr
u
ctu
r
al
ch
ar
ac
ter
is
tics
ex
h
ib
it
h
ig
h
s
ep
ar
ab
ilit
y
,
wh
ile
m
ass
iv
e
an
d
en
cr
u
s
tin
g
co
r
als
with
co
m
p
ar
ab
le
g
r
o
wth
f
o
r
m
s
r
em
ain
m
o
r
e
d
if
f
icu
lt
to
d
is
tin
g
u
is
h
,
s
u
g
g
esti
n
g
th
at
ce
r
tain
am
b
ig
u
ities
ar
is
e
f
r
o
m
in
tr
in
s
ic
b
io
lo
g
ical
s
im
ilar
ity
r
ath
er
th
an
p
u
r
ely
alg
o
r
ith
m
ic
lim
itatio
n
s
.
Alth
o
u
g
h
th
e
d
atasets
u
s
ed
i
n
th
is
s
tu
d
y
co
v
er
m
u
ltip
le
c
o
r
al
tax
a
a
n
d
im
ag
in
g
c
o
n
d
itio
n
s
,
f
u
r
th
er
ev
alu
atio
n
o
n
ad
d
itio
n
al
r
ee
f
en
v
ir
o
n
m
en
ts
,
g
eo
g
r
ap
h
ic
r
eg
io
n
s
,
an
d
lar
g
e
r
-
s
ca
le
d
atasets
wo
u
ld
s
tr
en
g
th
e
n
th
e
g
en
e
r
aliza
b
ilit
y
o
f
t
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
.
Fu
tu
r
e
w
o
r
k
m
ay
also
in
v
esti
g
ate
m
o
r
e
ad
v
a
n
ce
d
f
ea
tu
r
e
ex
tr
ac
tio
n
a
n
d
f
u
s
io
n
s
tr
ateg
ies,
in
clu
d
in
g
m
u
lti
-
lev
el
wav
elet
d
ec
o
m
p
o
s
itio
n
,
lear
n
ab
le
wav
elet
f
ilter
s
,
an
d
ad
ap
tiv
e
f
ea
tu
r
e
f
u
s
io
n
m
ec
h
an
is
m
s
,
to
f
u
r
th
er
im
p
r
o
v
e
r
o
b
u
s
tn
ess
f
o
r
m
o
r
p
h
o
lo
g
icall
y
am
b
ig
u
o
u
s
co
r
al
s
p
ec
ies.
ACK
NO
WL
E
DG
M
E
N
T
S
T
h
e
au
th
o
r
s
g
r
atef
u
lly
ac
k
n
o
wled
g
e
th
e
Facu
lty
o
f
Scien
ce
,
C
h
u
lalo
n
g
k
o
r
n
Un
iv
er
s
ity
,
f
o
r
in
s
titu
tio
n
al
s
u
p
p
o
r
t
an
d
f
ac
ilit
ies
th
at
m
ad
e
th
is
r
esear
ch
p
o
s
s
ib
le.
T
h
e
a
u
th
o
r
s
also
th
an
k
co
lleag
u
es
an
d
s
taf
f
m
em
b
er
s
f
o
r
th
eir
ass
i
s
tan
ce
an
d
co
n
s
tr
u
ctiv
e
f
ee
d
b
ac
k
d
u
r
i
n
g
th
e
p
r
ep
ar
atio
n
o
f
th
is
wo
r
k
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
r
esear
ch
r
ec
eiv
ed
n
o
e
x
te
r
n
al
f
u
n
d
in
g
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.