I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
14
,
No
.
6
,
Dec
em
b
er
20
24
,
p
p
.
6
8
7
4
~
6
8
8
2
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/
ijece
.
v
14
i
6
.
pp
6
8
7
4
-
6
8
8
2
6874
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Dipteroca
rpa
ceae
trunk
t
e
x
t
ure cla
ss
ificatio
n using
t
wo
-
sta
g
e
co
nv
o
lutiona
l neural network
-
ba
se
d
trans
fe
r learnin
g
mo
del
M
a
s
na
Wa
t
i,
No
v
ia
nti
P
us
pi
t
a
s
a
ri,
Umm
ul H
a
ira
h,
J
o
a
n
Ang
elina
Widi
a
n
s
,
Ade
F
iqri
T
j
ik
o
a
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
m
a
t
i
c
s,
F
a
c
u
l
t
y
o
f
En
g
i
n
e
e
r
i
n
g
,
M
u
l
a
w
a
r
ma
n
U
n
i
v
e
r
si
t
y
,
S
a
mari
n
d
a
,
I
n
d
o
n
e
si
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
No
v
2
6
,
2
0
2
3
R
ev
is
ed
J
u
l 3
0
,
2
0
2
4
Acc
ep
ted
Au
g
6
,
2
0
2
4
Th
e
imp
o
rtan
c
e
o
f
p
lan
t
id
e
n
ti
fic
a
ti
o
n
h
a
s
b
e
e
n
re
c
o
g
n
ize
d
b
y
a
c
a
d
e
m
ia
a
n
d
in
d
u
stry
.
Th
e
re
h
a
v
e
b
e
e
n
se
v
e
ra
l
a
tt
e
m
p
ts
to
u
ti
li
z
e
lea
v
e
s
a
n
d
fl
o
we
rs
fo
r
id
e
n
ti
fica
ti
o
n
.
Ho
we
v
e
r,
t
h
e
tr
u
n
k
c
a
n
a
lso
b
e
h
e
l
p
fu
l,
e
sp
e
c
ial
ly
f
o
r
tall
tree
s.
In
Bo
r
n
e
o
,
th
e
Di
p
tero
c
a
rp
a
c
e
a
e
fa
m
il
y
a
re
th
e
m
a
in
c
o
n
s
ti
tu
e
n
ts
o
f
th
e
tro
p
ica
l
ra
in
fo
re
st
e
c
o
s
y
ste
m
.
Th
is
re
se
a
rc
h
fo
c
u
se
s
o
n
th
e
c
la
ss
ifi
c
a
ti
o
n
o
f
t
h
e
d
ip
tero
c
a
rp
fa
m
il
y
,
wh
ich
c
a
n
re
a
c
h
a
h
e
i
g
h
t
o
f
b
e
twe
e
n
7
0
a
n
d
8
5
m
.
Lev
e
ra
g
in
g
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
two
r
k
(CNN
)
m
o
d
e
ls,
t
h
is
re
se
a
rc
h
p
ro
p
o
se
s
a
two
-
sta
g
e
tran
sfe
r
lea
rn
in
g
stra
teg
y
.
In
t
h
e
first
sta
g
e
,
th
e
p
re
-
train
e
d
CNN
m
o
d
e
ls
a
re
re
fin
e
d
b
y
o
n
l
y
m
o
d
if
y
i
n
g
t
h
e
c
las
sifica
ti
o
n
lay
e
r
wh
il
e
k
e
e
p
in
g
th
e
fe
a
tu
re
lay
e
r
fro
z
e
n
.
Th
e
se
c
o
n
d
sta
g
e
i
n
v
o
lv
e
s
se
lec
ti
n
g
a
n
d
fre
e
z
in
g
se
v
e
ra
l
c
o
n
v
o
l
u
ti
o
n
a
l
lay
e
rs
to
a
d
a
p
t
th
e
m
o
d
e
l
t
o
c
las
sify
d
ip
ter
o
c
a
rp
ste
m
s.
Th
e
d
a
tas
e
t
c
o
n
sists
o
f
8
5
7
ima
g
e
s
o
f
d
iffere
n
t
d
ip
ter
o
c
a
rp
sp
e
c
ies
.
Ex
p
e
rime
n
ts
sh
o
w
t
h
a
t
th
e
VG
G
1
6
m
o
d
e
l
wi
th
a
two
-
sta
g
e
tran
sfe
r
lea
rn
i
n
g
stra
teg
y
a
c
h
iev
e
s
a
h
ig
h
a
c
c
u
ra
c
y
o
f
9
8
.
2
4
6
%
.
T
h
is
stu
d
y
a
ims
to
a
c
c
u
ra
tely
i
d
e
n
t
ify
sp
e
c
ies
,
b
e
n
e
fit
i
n
g
c
o
n
se
rv
a
ti
o
n
a
n
d
e
c
o
lo
g
ica
l
stu
d
ies
b
y
e
n
a
b
li
n
g
f
a
st
a
n
d
re
li
a
b
le
tree
sp
e
c
ies
c
la
ss
ifi
c
a
ti
o
n
b
a
se
d
o
n
ste
m
tex
tu
re
ima
g
e
s.
K
ey
w
o
r
d
s
:
B
o
r
n
eo
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Dip
ter
o
ca
r
p
ac
ea
e
I
m
ag
e
class
if
icatio
n
T
r
an
s
f
er
lear
n
i
n
g
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
:
Ma
s
n
a
W
ati
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atics,
Facu
lty
o
f
E
n
g
in
ee
r
i
n
g
,
M
u
lawa
r
m
an
Un
iv
e
r
s
ity
Sam
ar
in
d
a,
E
ast Ka
lim
an
tan
,
I
n
d
o
n
esia
E
m
ail:
m
asn
awa
ti@
f
k
ti.u
n
m
u
l.a
c.
id
1.
I
NT
RO
D
UCT
I
O
N
T
r
o
p
ical
r
ain
f
o
r
ests
ar
e
p
a
r
t
o
f
th
e
wo
r
ld
'
s
ec
o
s
y
s
tem
d
iv
e
r
s
ity
an
d
ar
e
h
o
m
e
to
a
d
iv
er
s
i
ty
o
f
f
lo
r
a
an
d
f
au
n
a
s
p
ec
ies.
I
n
d
o
n
esia
h
as
g
r
ea
t
b
io
d
iv
er
s
ity
.
B
io
d
iv
er
s
ity
is
a
p
r
o
m
in
en
t
n
atio
n
al
ass
et
to
m
ain
tain
it
s
s
u
s
tain
ab
ilit
y
an
d
u
tili
za
tio
n
.
On
e
ty
p
e
o
f
en
d
em
ic
p
la
n
t
th
a
t
d
o
m
in
ates
th
e
tr
o
p
ical
r
ai
n
f
o
r
ests
o
f
Kalim
an
tan
is
tr
ee
s
f
r
o
m
th
e
d
ip
te
r
o
ca
r
p
f
am
ily
.
T
h
is
f
am
ily
is
em
b
lem
atic
o
f
th
e
tr
o
p
ical
r
ain
f
o
r
e
s
ts
ac
r
o
s
s
So
u
th
ea
s
t
Asi
a,
as
well
a
s
n
u
m
er
o
u
s
s
ea
s
o
n
ally
d
r
y
f
o
r
ests
in
So
u
th
an
d
So
u
th
ea
s
t
Asi
a
[
1
]
.
T
h
e
Dip
ter
o
ca
r
p
ac
ea
e
f
am
ily
is
a
n
ativ
e
s
p
ec
ies
f
o
u
n
d
in
th
e
tr
o
p
ical
r
ain
f
o
r
est
s
o
f
Kalim
an
tan
.
T
h
is
p
lan
t
f
am
ily
ca
n
g
r
o
w
to
h
eig
h
ts
r
an
g
i
n
g
f
r
o
m
7
0
to
8
5
m
an
d
h
o
l
d
s
co
n
s
id
er
ab
le
i
m
p
o
r
tan
ce
.
Dip
ter
o
ca
r
p
ac
ea
e,
a
s
u
b
f
am
ily
with
in
th
is
g
r
o
u
p
,
co
m
p
r
is
es
n
u
m
er
o
u
s
s
p
ec
ies
with
a
r
elativ
ely
b
r
o
ad
g
eo
g
r
ap
h
ical
p
r
esen
ce
.
T
h
ese
tr
ee
s
,
k
n
o
wn
as
d
ip
ter
o
ca
r
p
s
,
ar
e
th
e
m
ain
s
o
u
r
ce
s
o
f
tim
b
er
in
t
r
o
p
ic
al
r
ain
f
o
r
ests
ac
r
o
s
s
r
eg
io
n
s
in
clu
d
in
g
wester
n
I
n
d
o
n
esia,
Ma
lay
s
ia,
B
r
u
n
ei,
an
d
th
e
Ph
ilip
p
in
es,
ex
te
n
d
in
g
ea
s
twar
d
to
Pap
u
a
an
d
Pap
u
a
New
Gu
in
ea
[
2
]
.
T
h
is
f
am
ily
is
a
p
lan
t
tr
ib
e
wh
o
s
e
m
em
b
er
s
ar
e
all
tr
ee
s
th
at
h
av
e
an
ess
en
tial
r
o
le,
b
o
th
f
r
o
m
an
ec
o
n
o
m
ic
an
d
ec
o
lo
g
ical
p
e
r
s
p
ec
tiv
e.
Fro
m
an
ec
o
n
o
m
ic
p
er
s
p
ec
ti
v
e,
m
o
s
t
o
f
th
ese
tr
ib
es
ar
e
co
m
m
er
cial
wo
o
d
p
r
o
d
u
ce
r
s
to
m
ee
t
v
ar
io
u
s
n
e
ed
s
.
Ap
ar
t
f
r
o
m
th
at,
s
ev
er
al
ty
p
es
o
f
d
ip
ter
o
ca
r
p
s
also
p
r
o
d
u
ce
o
il,
r
esin
an
d
f
r
u
it with
tr
ad
e
v
alu
e.
T
h
is
s
u
b
f
am
ily
h
as 1
3
g
e
n
er
a
an
d
4
7
0
s
p
ec
ies,
9
o
f
w
h
ich
ar
e
f
o
u
n
d
in
I
n
d
o
n
esia
[
3
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Dip
tero
ca
r
p
a
ce
a
e
tr
u
n
k
textu
r
e
cla
s
s
ifica
tio
n
u
s
in
g
tw
o
-
s
ta
g
e
…
(
Ma
s
n
a
Wa
ti)
6875
Dif
f
er
en
tiatio
n
o
f
d
i
p
ter
o
ca
r
p
s
p
ec
ies
h
as
tr
ad
itio
n
ally
b
ee
n
b
ased
o
n
v
a
r
io
u
s
m
o
r
p
h
o
lo
g
ical
ch
ar
ac
ter
is
tics
,
in
clu
d
in
g
s
tem
tex
tu
r
e,
leaf
p
atter
n
,
a
n
d
o
th
er
v
is
u
al
attr
ib
u
tes.
B
ased
o
n
s
ev
er
al
r
esear
ch
r
esu
lts
in
E
ast
Kalim
an
tan
,
th
er
e
ar
e
ar
o
u
n
d
eig
h
t
g
en
e
r
a
with
d
o
ze
n
s
o
f
ty
p
es
o
f
d
ip
te
r
o
ca
r
p
tr
ee
s
.
T
h
ese
g
en
er
a
in
cl
u
d
e
S
h
o
r
ea
,
Dryo
b
a
la
n
o
p
s
,
Dip
tero
ca
r
p
u
s
,
A
n
is
o
p
tera
,
P
a
r
a
s
h
o
r
ea
,
C
o
tylel
o
b
iu
m
,
V
a
tica
,
an
d
Ho
p
ea
[
4
]
.
Dip
ter
o
ca
r
p
s
p
ec
ies
ar
e
p
r
im
ar
ily
f
o
u
n
d
in
ar
e
as
with
wet
cl
im
ates,
h
ig
h
h
u
m
id
ity
,
altitu
d
es
o
f
0
–
8
0
0
m
a
b
o
v
e
s
ea
lev
el,
a
n
d
r
ai
n
f
all
ab
o
v
e
2
,
0
0
0
m
m
/y
ea
r
with
a
s
h
o
r
t
d
r
y
s
ea
s
o
n
.
Mo
s
t
ty
p
es
o
f
Dip
tero
ca
r
p
a
ce
a
e
also
g
r
o
w
in
d
r
y
,
ac
id
ic,
d
ee
p
s
o
lu
m
an
d
clay
s
o
ils
.
I
n
ac
id
s
o
il
co
n
d
itio
n
s
,
th
e
r
o
o
ts
o
f
d
ip
ter
o
ca
r
p
p
lan
ts
ass
o
ciate
with
ec
to
myc
o
r
r
h
iz
a
s
o
th
ey
ca
n
s
u
r
v
iv
e
an
d
d
ev
elo
p
[
5
]
.
C
o
n
s
er
v
atio
n
o
f
d
ip
ter
o
ca
r
p
f
o
r
ests
b
en
ef
its
th
e
f
o
r
estry
s
ec
to
r
b
y
p
r
o
te
ctin
g
b
io
d
iv
er
s
ity
,
p
r
ev
en
tin
g
lan
d
s
lid
es,
a
n
d
ab
s
o
r
b
in
g
atm
o
s
p
h
er
ic
ca
r
b
o
n
d
io
x
id
e.
Me
an
w
h
ile,
en
v
ir
o
n
m
en
tal
ch
an
g
es,
p
o
llu
tio
n
,
te
m
p
er
atu
r
e
in
c
r
ea
s
es,
clim
ate
ch
an
g
e,
f
o
r
est
f
ir
es,
an
d
ec
o
s
y
s
tem
p
o
llu
tio
n
ca
n
th
r
ea
ten
b
io
d
iv
e
r
s
ity
.
Ou
r
r
esear
ch
is
a
f
o
r
m
o
f
co
n
ce
r
n
f
o
r
t
h
e
p
r
eser
v
atio
n
o
f
d
ip
ter
o
ca
r
p
s
in
Kalim
an
tan
,
esp
ec
ially
in
E
ast Ka
lim
an
tan
.
I
d
en
tify
in
g
d
if
f
e
r
en
t
ty
p
es
o
f
tr
ee
tr
u
n
k
s
p
o
s
es
a
d
if
f
icu
lty
in
tex
tu
r
e
class
if
icatio
n
.
O
v
er
tim
e,
r
esear
ch
er
s
h
av
e
r
elied
o
n
m
an
u
ally
cr
ea
ted
f
ea
tu
r
es
to
ad
d
r
ess
th
is
ch
allen
g
e.
Fo
r
in
s
t
an
ce
,
s
o
m
e
s
tu
d
ies
h
av
e
d
e
v
elo
p
e
d
th
eir
ap
p
r
o
a
ch
es
u
s
in
g
lo
ca
l
b
in
ar
y
p
atte
r
n
s
(
L
B
P)
as
a
b
asis
[
6
]
.
Us
in
g
s
ca
le
-
in
v
ar
ia
n
t
f
ea
tu
r
e
tr
a
n
s
f
o
r
m
(
SIFT
)
d
escr
ip
to
r
s
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
es
(
SVM)
ac
h
iev
es
an
ac
cu
r
ac
y
o
f
ar
o
u
n
d
7
0
%
wh
en
a
p
p
lied
t
o
th
e
Au
s
tr
ian
Fed
er
al
Fo
r
est
d
ataset
.
Fo
u
r
s
tatis
tical
attr
ib
u
tes,
u
n
if
o
r
m
ity
,
en
tr
o
p
y
,
asy
m
m
etr
y
,
an
d
s
m
o
o
t
h
n
ess
ar
e
u
s
ed
i
n
tex
tu
r
e
class
if
icatio
n
f
o
r
b
ar
k
im
ag
es.
T
h
e
y
u
s
e
d
ec
is
io
n
tr
ee
alg
o
r
ith
m
s
f
o
r
class
if
icatio
n
task
s
[
7
]
.
T
h
e
s
u
p
p
o
r
t
v
e
cto
r
m
ac
h
in
e
(
SVM)
alg
o
r
ith
m
is
u
tili
ze
d
f
o
r
ca
teg
o
r
izin
g
v
a
r
io
u
s
tex
tu
r
e
m
ater
ials
,
as
s
h
o
wca
s
ed
in
ex
p
er
im
en
ts
co
n
d
u
cted
o
n
th
e
Fl
ick
r
m
ater
ial
d
ataset
[
8
]
.
C
o
n
v
o
lu
tio
n
al
n
e
u
r
al
n
et
wo
r
k
s
(
C
NN)
h
av
e
r
ec
en
tly
b
eg
u
n
to
o
u
tp
er
f
o
r
m
tr
a
d
itio
n
al
m
eth
o
d
s
in
a
v
ar
iety
o
f
f
ield
s
,
m
o
s
t
n
o
tab
ly
p
atter
n
r
ec
o
g
n
itio
n
.
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
u
s
e
s
u
p
er
v
is
ed
m
ac
h
i
n
e
lear
n
in
g
tech
n
iq
u
es,
an
d
th
er
e
ar
e
a
lo
t
o
f
C
NN
ar
ch
itectu
r
es
ca
lled
p
r
e
-
tr
ain
ed
C
NN
m
o
d
el.
C
NN
is
th
e
b
est
n
etwo
r
k
f
o
r
im
ag
e
r
ec
o
g
n
itio
n
task
s
b
ec
au
s
e
it
ca
n
ex
tr
ac
t
s
p
ec
ial
f
ea
tu
r
es
o
f
th
e
in
p
u
t
im
ag
es
w
ith
its
lar
g
e
n
u
m
b
er
o
f
f
ilter
s
.
T
h
e
d
iv
er
s
ity
in
th
ese
ar
ch
itectu
r
es
co
m
es
f
r
o
m
t
h
e
im
p
r
o
v
em
en
ts
o
r
m
o
d
if
icati
o
n
s
th
at
ar
e
a
p
p
lied
to
th
e
o
r
ig
in
al
C
NN
ar
ch
itectu
r
e
[
9
]
.
C
NN
m
o
d
els
th
at
ar
e
wid
ely
u
s
ed
an
d
h
av
e
h
ig
h
p
er
f
o
r
m
a
n
ce
in
clu
d
e
Alex
Net
[
1
0
]
,
VGG1
6
[
1
1
]
,
r
esid
u
al
n
etwo
r
k
(
R
esNet
)
m
o
d
el
[
1
2
]
,
d
en
s
e
n
etwo
r
k
(
Den
s
eNe
t
)
[
1
3
]
,
an
d
I
n
ce
p
tio
n
[
1
4
]
.
Nu
m
er
o
u
s
C
NN
m
o
d
els
h
a
v
e
b
ee
n
cr
ea
ted
,
d
em
o
n
s
tr
atin
g
ex
ce
p
tio
n
al
p
er
f
o
r
m
an
ce
t
h
r
o
u
g
h
th
e
u
tili
za
tio
n
o
f
tr
an
s
f
er
lear
n
i
n
g
.
T
r
an
s
f
er
lear
n
in
g
in
v
o
lv
es
r
ep
u
r
p
o
s
in
g
a
m
o
d
el
d
esig
n
ed
f
o
r
o
n
e
task
as
th
e
f
o
u
n
d
atio
n
f
o
r
an
o
th
er
task
.
T
h
is
m
eth
o
d
is
p
ar
ticu
lar
ly
f
av
o
u
r
ed
in
d
ee
p
lear
n
in
g
,
w
h
er
e
p
r
e
-
ex
is
tin
g
m
o
d
els
s
er
v
e
as
th
e
b
asis
f
o
r
co
m
p
u
ter
v
is
io
n
an
d
n
at
u
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
en
d
ea
v
o
u
r
s
.
I
t
ad
d
r
ess
es
th
e
s
ig
n
if
ican
t
co
m
p
u
tatio
n
al
r
es
o
u
r
ce
s
an
d
tim
e
ty
p
ically
n
ee
d
ed
to
co
n
s
tr
u
ct
n
eu
r
al
n
etwo
r
k
m
o
d
els
f
o
r
s
u
ch
task
s
.
[
1
5
]
.
R
esear
ch
th
at
h
a
s
u
tili
ze
d
th
e
u
s
e
o
f
tr
an
s
f
er
lear
n
in
g
in
tex
t
u
r
e
im
ag
e
r
e
co
g
n
itio
n
,
s
u
c
h
as
co
wh
id
e
tex
tu
r
e
d
etec
tio
n
u
s
in
g
th
e
VGG1
6
m
o
d
el
with
an
ac
cu
r
ac
y
p
er
f
o
r
m
an
ce
o
f
9
5
%
[
1
6
]
.
T
ex
tu
r
e
id
en
tific
atio
n
o
n
lea
v
es
h
as
also
b
ee
n
ca
r
r
ie
d
o
u
t
u
s
in
g
R
esNet5
0
with
an
ac
cu
r
a
cy
o
f
9
8
.
1
7
%
[
1
7
]
.
R
esear
ch
er
s
also
co
m
b
in
e
s
ev
er
al
p
r
e
-
t
r
ain
ed
m
o
d
els
to
f
u
r
th
er
in
c
r
ea
s
e
p
er
f
o
r
m
an
ce
.
T
h
e
T
ex
Fu
s
io
n
Net
m
o
d
el,
m
er
g
es
th
e
Alex
Net
an
d
VGG1
6
m
o
d
els
b
y
co
m
b
in
in
g
th
eir
f
in
al
r
e
p
r
esen
t
atio
n
lay
er
s
.
T
h
ey
ev
alu
ated
th
e
m
o
d
el
u
s
in
g
v
ar
io
u
s
tex
tu
r
e
d
ataset
s
in
clu
d
in
g
B
r
o
d
atz
,
C
o
lu
m
b
ia
-
Utr
ec
h
t
r
ef
lecta
n
ce
an
d
tex
tu
r
e
d
atab
a
s
e
(
C
UR
eT
)
,
an
d
tex
tu
r
es
u
n
d
e
r
v
ar
y
i
n
g
I
llu
m
in
atio
n
,
p
o
s
e
an
d
s
ca
le
(
KT
H
-
T
I
PS
)
.
T
h
e
r
esu
lts
o
f
th
eir
ex
p
er
im
en
ts
d
em
o
n
s
tr
ate
th
e
s
u
p
er
io
r
p
e
r
f
o
r
m
an
ce
o
f
th
e
T
ex
Fu
s
io
n
Net
ar
c
h
ite
ctu
r
e
in
tex
tu
r
e
class
if
icatio
n
[
1
8
]
.
No
wad
ay
s
,
th
e
ad
v
an
ce
m
en
t
o
f
d
ee
p
lear
n
in
g
tech
n
o
lo
g
y
in
d
ig
ita
l
im
ag
e
p
r
o
ce
s
s
in
g
allo
ws
u
s
to
au
t
o
m
ate
th
e
cla
s
s
if
icatio
n
o
f
d
i
p
ter
o
ca
r
p
s
te
m
im
ag
es
to
s
u
p
p
o
r
t
c
o
n
s
er
v
atio
n
in
itiativ
es
an
d
ec
o
lo
g
ical
r
esear
c
h
.
W
e
u
s
e
a
C
NN
-
b
ased
two
-
s
tag
e
tr
an
s
f
er
lear
n
i
n
g
ap
p
r
o
ac
h
to
o
b
tai
n
o
p
tim
al
r
esu
lts
in
ca
teg
o
r
izin
g
th
e
d
ip
ter
o
ca
r
p
b
ased
o
n
s
tem
tex
tu
r
e.
T
h
e
r
em
ain
d
er
o
f
t
h
is
p
ap
er
is
s
tr
u
ctu
r
ed
as
f
o
llo
ws:
T
h
e
s
ec
o
n
d
s
ec
tio
n
d
escr
ib
es
th
e
r
esear
ch
m
eth
o
d
.
R
esu
lts
an
d
d
is
cu
s
s
io
n
ar
e
p
r
esen
ted
in
th
e
th
ir
d
s
ec
tio
n
.
I
n
th
e
f
in
al
s
ec
tio
n
,
we
will d
r
aw
s
o
m
e
co
n
clu
s
io
n
s
an
d
s
u
g
g
est f
u
tu
r
e
r
esear
ch
.
2.
M
E
T
H
O
D
T
h
e
co
m
p
lete
s
tag
es
o
f
th
e
r
e
s
ea
r
ch
m
eth
o
d
o
lo
g
y
ar
e
in
Fig
u
r
e
1
.
T
h
is
r
esear
c
h
m
eth
o
d
o
lo
g
y
h
as
f
iv
e
m
ain
s
tag
es:
im
ag
e
ac
q
u
i
s
itio
n
,
p
r
e
-
p
r
o
ce
s
s
in
g
,
f
ir
s
t
tr
a
in
in
g
,
s
ec
o
n
d
tr
ain
in
g
a
n
d
e
v
a
lu
atio
n
.
T
h
e
i
m
a
g
e
ac
q
u
is
itio
n
is
ca
p
tu
r
in
g
o
r
s
ca
n
n
in
g
a
n
o
b
ject
to
o
b
tain
a
d
i
g
ital
o
n
e.
T
h
e
i
m
ag
e
ac
q
u
is
itio
n
u
s
es
im
ag
es
f
r
o
m
Dip
ter
o
ca
r
p
ac
ea
e
to
o
b
tain
im
ag
es
in
d
ig
ital
f
o
r
m
.
Dig
ital
im
ag
es
will
b
e
p
r
o
ce
s
s
ed
to
class
if
y
Dip
ter
o
ca
r
p
ac
ea
e.
Nex
t,
two
s
tag
es
o
f
tr
ain
in
g
ar
e
ca
r
r
ied
o
u
t.
T
h
e
f
ir
s
t
s
tag
e
o
f
tr
ain
i
n
g
is
ca
r
r
ied
o
u
t
f
o
r
tr
an
s
f
er
lear
n
i
n
g
u
n
til
th
e
m
o
d
el
ca
n
q
u
ick
l
y
ad
a
p
t
to
n
ew
t
ask
s
.
T
h
e
s
ec
o
n
d
s
tag
e
is
to
a
d
ju
s
t
an
d
r
ef
in
e
th
e
f
ea
tu
r
es it h
as lea
r
n
ed
in
r
esp
o
n
s
e
to
n
ew
task
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
e
r
20
24
:
6
8
7
4
-
6
8
8
2
6876
Fig
u
r
e
1
.
R
esear
ch
m
eth
o
d
o
l
o
g
y
2
.
1
.
Da
t
a
co
llect
i
o
n a
nd
prep
ro
ce
s
s
ing
W
e
co
llected
im
ag
es
f
r
o
m
Sem
p
aja
Ar
b
o
r
et
u
m
an
d
B
B
PS
I
L
H
Sam
ar
in
d
a
wh
er
e
f
o
r
ester
s
h
ad
id
en
tifie
d
ce
r
tain
tr
ee
s
b
y
th
ei
r
wo
o
d
s
p
ec
ies.
T
h
ese
im
ag
es
wer
e
ca
p
tu
r
ed
u
s
in
g
an
iPh
o
n
e
1
1
u
n
d
er
v
a
r
io
u
s
wea
th
er
co
n
d
itio
n
s
,
f
r
o
m
s
u
n
n
y
to
clo
u
d
y
,
an
d
wer
e
tak
en
f
r
o
m
a
d
is
tan
ce
o
f
3
0
-
6
0
cm
awa
y
f
r
o
m
th
e
tr
ee
tr
u
n
k
.
T
h
e
ca
m
e
r
a
was
p
o
s
itio
n
ed
with
a
v
er
tical
ax
is
th
at
d
i
f
f
er
ed
f
r
o
m
th
e
im
ag
e
p
lan
e
.
An
ex
am
p
le
o
f
o
n
e
o
f
th
ese
tr
u
n
k
tex
tu
r
e
im
ag
es
ca
n
b
e
s
ee
n
in
Fig
u
r
e
2
.
W
e
co
llected
im
ag
es
o
f
5
d
if
f
er
en
t
Dip
tero
ca
r
p
a
ce
a
e
s
p
ec
ies:
Dip
tero
ca
r
p
u
s
a
la
tu
s
in
Fig
u
r
e
2
(
a)
,
Dip
tero
ca
r
p
u
s
co
n
fert
u
s
in
Fig
u
r
e
2
(
b
)
,
Dryo
b
a
la
n
o
p
s
b
ec
ca
r
ii
in
Fig
u
r
e
2
(
c)
,
Dryo
b
a
la
n
o
p
s
la
n
ce
o
la
ta
in
Fig
u
r
e
2
(
d
)
,
an
d
S
h
o
r
ea
s
mith
ia
n
a
i
n
Fig
u
r
e
2
(
e)
.
T
h
e
n
u
m
b
er
o
f
tr
ee
d
atasets
tak
en
f
o
r
ea
c
h
ty
p
e
is
ar
o
u
n
d
4
to
8
t
r
ee
s
.
T
h
is
v
ar
iab
ilit
y
p
r
e
v
en
ts
o
v
er
f
itti
n
g
in
th
e
f
u
tu
r
e
d
ee
p
-
lear
n
in
g
m
o
d
el
tr
ain
i
n
g
.
W
e
c
o
m
p
o
s
ed
th
e
im
ag
es
to
k
ee
p
t
h
e
d
ata
class
es
b
alan
ce
d
.
T
h
e
im
ag
es
ar
e
d
iv
id
e
d
in
to
th
r
ee
s
ep
a
r
ate
d
atasets
:
tr
ain
in
g
,
v
ali
d
atio
n
,
a
n
d
test
in
g
.
T
h
ese
d
atasets
ar
e
allo
ca
ted
i
n
a
r
atio
o
f
6
0
%
f
o
r
tr
ain
in
g
,
2
0
%
f
o
r
v
alid
atio
n
,
an
d
2
0
%
f
o
r
test
in
g
.
T
h
e
tr
ai
n
in
g
an
d
v
alid
atio
n
s
ets
ar
e
f
o
r
m
o
d
el
tr
ain
in
g
,
wh
ile
th
e
test
in
g
s
et
is
f
o
r
ass
ess
in
g
m
o
d
el
p
er
f
o
r
m
a
n
ce
.
(
a)
(
b
)
(
c)
(
d
)
(
e)
Fig
u
r
e
2
.
T
r
u
n
k
tex
tu
r
e
im
ag
e
(
a)
Dip
tero
ca
r
p
u
s
a
la
tu
s
(
b
)
D
ip
tero
ca
r
p
u
s
co
n
fert
u
s
(
c)
Dryo
b
a
la
n
o
p
s
b
ec
ca
r
ii
(
d
)
Dryo
b
a
l
a
n
o
p
s
la
n
ce
o
la
ta
,
an
d
(
e)
S
h
o
r
ea
s
mith
ia
n
a
T
ab
le
1
d
is
p
lay
s
th
e
q
u
a
n
tity
o
f
im
a
g
es
allo
ca
ted
eq
u
ally
am
o
n
g
d
if
f
er
e
n
t
s
p
ec
ies.
T
h
e
s
e
im
ag
es
h
av
e
b
ee
n
r
esized
f
r
o
m
th
ei
r
o
r
ig
in
al
d
im
en
s
io
n
s
o
f
3
0
2
4
×
4
0
3
2
to
7
5
0
×
1
0
0
0
.
Fo
llo
win
g
c
u
r
atio
n
,
th
er
e
ar
e
a
to
tal
o
f
8
5
7
im
ag
es.
No
r
m
aliza
tio
n
o
f
th
ese
im
ag
es
in
v
o
lv
es
ad
ju
s
tin
g
ea
ch
p
ix
el'
s
v
alu
e
b
y
s
u
b
tr
ac
tin
g
th
e
m
ea
n
an
d
d
iv
id
i
n
g
b
y
th
e
s
tan
d
ar
d
d
e
v
iatio
n
o
f
all
p
ix
el
v
alu
es.
T
h
is
n
o
r
m
aliza
tio
n
p
r
o
ce
s
s
i
s
in
ten
d
ed
to
en
h
an
ce
tr
ain
i
n
g
ef
f
icien
cy
a
n
d
m
in
im
ize
tr
ain
in
g
er
r
o
r
s
.
T
ab
le
1
.
Data
s
et
d
is
tr
ib
u
tio
n
S
p
e
c
i
e
s
Tr
a
i
n
i
n
g
se
t
V
a
l
i
d
a
t
i
o
n
se
t
Te
st
i
n
g
s
e
t
D
i
p
t
e
r
o
c
a
rp
u
s
a
l
a
t
u
s
91
32
30
D
i
p
t
e
r
o
c
a
rp
u
s
c
o
n
f
e
r
t
u
s
1
0
6
36
35
D
ryo
b
a
l
a
n
o
p
s
b
e
c
c
a
r
i
i
1
0
6
35
35
D
ryo
b
a
l
a
n
o
p
s
l
a
n
c
e
o
l
a
t
a
1
0
2
34
35
S
h
o
re
a
sm
i
t
h
i
a
n
a
1
0
8
36
36
To
t
a
l
5
1
3
1
7
3
1
7
1
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Dip
tero
ca
r
p
a
ce
a
e
tr
u
n
k
textu
r
e
cla
s
s
ifica
tio
n
u
s
in
g
tw
o
-
s
ta
g
e
…
(
Ma
s
n
a
Wa
ti)
6877
2
.
2
.
Co
nv
o
lutio
na
l
neura
l net
wo
rk
s
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
i
s
a
f
ee
d
f
o
r
war
d
n
eu
r
al
n
etwo
r
k
th
at
ca
n
ex
tr
ac
t
f
ea
tu
r
es
f
r
o
m
d
ata
with
a
co
n
v
o
lu
tio
n
al
s
tr
u
ctu
r
e.
C
NN
d
o
es
n
o
t
n
ee
d
to
ex
tr
ac
t
f
ea
tu
r
es
m
an
u
ally
.
I
n
ad
d
itio
n
,
C
NN
ca
n
lear
n
co
m
p
lex
p
r
o
b
lem
s
q
u
ic
k
ly
d
u
e
to
weig
h
t
s
h
ar
in
g
an
d
m
o
r
e
co
m
p
le
x
m
o
d
els,
allo
w
in
g
f
o
r
m
ass
iv
e
p
ar
alleliza
tio
n
[
1
9
]
.
T
h
e
s
tan
d
ar
d
C
NN
ar
ch
itectu
r
e
c
o
m
p
r
is
es
lay
er
s
to
ex
tr
ac
t
th
e
d
is
tin
g
u
is
h
in
g
f
ea
tu
r
es.
T
h
e
cr
itical
lay
er
s
o
f
C
NN
ar
e
th
e
in
p
u
t
lay
e
r
,
th
e
co
n
v
o
lu
tio
n
al
lay
er
,
th
e
p
o
o
lin
g
l
ay
er
,
an
d
th
e
f
u
lly
co
n
n
ec
ted
lay
er
[
2
0
]
.
C
NN'
s
f
ea
tu
r
e
ex
t
r
ac
tio
n
p
r
o
ce
s
s
in
v
o
lv
es
co
n
v
o
lu
tio
n
an
d
p
o
o
lin
g
lay
er
s
,
wh
ile
th
e
f
u
lly
co
n
n
ec
ted
lay
er
d
o
es th
e
f
in
al
class
if
icatio
n
.
T
h
e
co
n
v
o
lu
tio
n
lay
er
is
f
u
n
d
am
en
tal
in
C
NN;
it
in
v
o
lv
es a
s
er
ies
o
f
m
ath
em
atica
l
o
p
e
r
atio
n
s
,
a
s
p
ec
ial
f
o
r
m
o
f
lin
ea
r
o
p
e
r
atio
n
s
.
T
h
e
C
NN
ar
c
h
itectu
r
e
co
n
s
is
ts
o
f
s
ev
er
al
co
n
v
o
lu
tio
n
a
n
d
p
o
o
li
n
g
lay
er
s
r
e
p
etitio
n
s
,
f
o
llo
wed
b
y
o
n
e
o
r
m
o
r
e
f
u
lly
c
o
n
n
ec
te
d
lay
er
s
[
2
1
]
.
2
.
3
.
T
ra
ns
f
er
lea
rning
T
r
an
s
f
er
l
ea
r
n
in
g
in
v
o
lv
es
r
e
u
s
in
g
a
m
o
d
el
t
h
at
h
as
p
er
f
o
r
m
ed
well
in
a
p
ar
ticu
lar
task
to
s
o
lv
e
a
r
elate
d
task
.
[
2
2
]
.
T
r
an
s
f
er
l
ea
r
n
in
g
s
ee
k
s
to
e
n
h
an
ce
th
e
ef
f
ec
tiv
en
ess
o
f
tar
g
et
lear
n
er
s
with
in
s
p
ec
if
ic
d
o
m
ain
s
b
y
lev
e
r
ag
in
g
k
n
o
w
led
g
e
f
r
o
m
r
elate
d
b
u
t
d
is
tin
ct
s
o
u
r
ce
d
o
m
ain
s
.
T
h
is
ap
p
r
o
ac
h
r
ed
u
ce
s
th
e
r
elian
ce
o
n
ex
te
n
s
iv
e
tar
g
et
-
d
o
m
ain
d
ata
f
o
r
d
ev
el
o
p
in
g
ta
r
g
et
lear
n
er
s
[
2
3
]
.
T
r
an
s
f
e
r
lear
n
in
g
h
o
ld
s
p
o
ten
tial
to
en
h
an
ce
th
e
p
er
f
o
r
m
an
ce
o
f
tar
g
et
lear
n
er
s
b
y
tap
p
i
n
g
in
to
ex
ter
n
al
k
n
o
wled
g
e.
An
o
t
h
er
p
er
s
p
ec
tiv
e
with
in
tr
an
s
f
er
lear
n
in
g
is
to
co
n
s
i
d
er
tr
ai
n
in
g
an
d
tar
g
et
d
ata
as
o
r
i
g
in
atin
g
f
r
o
m
s
ep
a
r
ate
s
u
b
-
d
o
m
ain
s
b
u
t
co
n
n
ec
ted
t
h
r
o
u
g
h
a
s
h
ar
ed
o
v
er
ar
ch
in
g
d
o
m
ain
[
2
4
]
.
T
r
an
s
f
er
lear
n
in
g
b
ec
o
m
es
n
ec
ess
ar
y
wh
en
t
h
er
e'
s
a
s
h
o
r
tag
e
o
f
tar
g
et
tr
ain
in
g
d
at
a.
T
h
is
s
h
o
r
ta
g
e
co
u
ld
s
tem
f
r
o
m
lim
ited
d
ata
av
ailab
ilit
y
,
h
ig
h
co
s
ts
ass
o
ciate
d
with
co
llectin
g
an
d
lab
ellin
g
d
ata,
o
r
c
h
allen
g
es
in
ac
ce
s
s
in
g
th
e
d
ata.
As
b
ig
d
ata
r
ep
o
s
ito
r
ies
b
ec
o
m
e
m
o
r
e
co
m
m
o
n
,
lev
er
a
g
in
g
ex
is
tin
g
d
atasets
th
at
ar
e
r
elate
d
to
,
th
o
u
g
h
n
o
t
i
d
en
tical
with
,
th
e
tar
g
et
d
o
m
ain
b
ec
o
m
es
an
ap
p
ea
lin
g
s
tr
ateg
y
f
o
r
tr
a
n
s
f
er
lear
n
in
g
.
T
h
is
a
p
p
r
o
ac
h
o
f
f
er
s
s
ev
e
r
al
b
e
n
ef
it
s
,
in
clu
d
in
g
s
h
o
r
ter
tr
ain
in
g
tim
es,
en
h
an
ce
d
n
e
u
r
al
n
etwo
r
k
p
er
f
o
r
m
an
ce
ac
r
o
s
s
v
ar
io
u
s
s
ce
n
ar
io
s
,
an
d
th
e
ab
ilit
y
to
ac
h
iev
e
g
o
o
d
r
esu
lts
with
o
u
t
r
eq
u
ir
in
g
ex
ten
s
iv
e
am
o
u
n
ts
o
f
d
ata.
[
2
5
]
.
T
y
p
ically
,
d
ata
is
n
ee
d
ed
to
tr
ain
a
n
eu
r
al
m
o
d
el
f
r
o
m
s
cr
atch
,
b
u
t
o
b
tai
n
in
g
s
u
ch
d
ata
is
o
n
ly
s
o
m
etim
es
p
o
s
s
ib
le.
T
r
an
s
f
er
lear
n
in
g
b
ec
o
m
es
v
alu
ab
le
in
s
u
ch
ca
s
es.
B
y
tr
ain
in
g
a
p
r
e
-
ex
is
tin
g
m
o
d
el,
tr
an
s
f
e
r
l
ea
r
n
in
g
en
ab
les
t
h
e
cr
e
atio
n
o
f
p
r
ac
tical
m
ac
h
i
n
e
lear
n
in
g
m
o
d
els with
r
elativ
el
y
s
m
all
tr
ain
in
g
d
ata
[
2
6
]
.
T
h
er
e
a
r
e
a
n
u
m
b
er
o
f
p
o
p
u
la
r
p
r
e
-
tr
ain
ed
C
NN
-
b
ased
m
o
d
els
av
ailab
le,
s
o
m
e
o
f
th
em
ar
e
VGG1
6
[
1
1
]
,
R
esNet
m
o
d
el
[
1
2
]
,
Den
s
eNe
t
[
1
3
]
,
an
d
I
n
ce
p
tio
n
[
1
4
]
.
T
h
o
s
e
m
o
d
els
ar
e
tr
ain
ed
o
n
I
m
ag
eNe
t
d
ataset,
th
is
d
ataset
s
p
an
s
1
,
0
0
0
clas
s
es
an
d
co
n
tain
s
1
,
2
8
1
,
1
6
7
t
r
ain
in
g
im
a
g
es,
5
0
,
0
0
0
v
alid
atio
n
im
ag
es
a
n
d
1
0
0
,
0
0
0
test
im
ag
es
[
2
7
]
.
T
h
ese
p
r
e
-
tr
ain
e
d
m
o
d
els
h
av
e
ac
h
iev
ed
s
tate
-
of
-
th
e
-
a
r
t
p
e
r
f
o
r
m
an
ce
o
n
v
a
r
io
u
s
co
m
p
u
ter
v
is
io
n
task
s
,
s
u
ch
as
im
ag
e
class
if
icatio
n
an
d
o
b
j
ec
t
d
etec
tio
n
.
T
r
an
s
f
er
lear
n
in
g
o
f
f
er
s
a
p
r
ac
tical
an
d
ef
f
icien
t
way
to
lev
er
ag
e
p
r
e
-
ex
is
tin
g
k
n
o
wled
g
e
f
r
o
m
r
elate
d
d
o
m
ain
s
to
im
p
r
o
v
e
th
e
p
er
f
o
r
m
an
ce
o
f
m
ac
h
in
e
lear
n
in
g
m
o
d
els.
I
t
r
ed
u
ce
s
th
e
n
ee
d
f
o
r
lar
g
e
am
o
u
n
ts
o
f
tar
g
et
d
ata
an
d
tr
ain
i
n
g
tim
e,
m
ak
i
n
g
it
p
ar
ticu
lar
ly
u
s
ef
u
l in
s
ce
n
ar
io
s
wh
er
e
d
ata
is
lim
ited
o
r
ex
p
e
n
s
iv
e
to
o
b
tain
[
2
8
]
.
2
.
4
.
T
wo
-
s
t
a
g
e
t
ra
ns
f
er
lea
r
nin
g
a
pp
ro
a
ch
T
h
is
r
esear
ch
e
m
p
lo
y
s
a
tw
o
-
s
tep
tr
an
s
f
er
lear
n
i
n
g
a
p
p
r
o
ac
h
u
tili
zin
g
p
r
e
-
tr
ain
ed
C
NN
-
b
ased
m
o
d
els
s
u
ch
as
VGG1
6
[
1
1
]
,
R
es
Net1
2
1
[
1
2
]
,
Den
s
eNe
t5
0
[
1
3
]
,
I
n
ce
p
tio
n
V3
[
1
4
]
,
Mo
b
ileNetV2
[
2
9
]
,
an
d
E
f
f
icien
tNetV2
-
B0
[
3
0
]
.
T
h
es
e
m
o
d
els
h
av
e
b
ee
n
tr
ain
e
d
o
n
th
e
I
m
a
g
eNe
t
d
ataset,
wh
ich
in
clu
d
es
1
,
0
0
0
class
es.
T
h
e
m
o
d
if
icatio
n
s
will
b
e
m
ad
e
to
t
h
e
ar
ch
itectu
r
e
o
f
th
ese
m
o
d
els,
ex
p
licitly
i
m
p
lem
en
tin
g
a
n
ew
class
if
icatio
n
lay
er
to
ad
d
r
ess
d
ip
ter
o
ca
r
p
class
if
icatio
n
.
T
wo
lear
n
in
g
s
tag
es
h
a
v
e
b
ee
n
in
clu
d
ed
t
o
en
s
u
r
e
s
u
itab
ilit
y
f
o
r
th
e
n
ew
class
if
icatio
n
task
.
I
n
th
e
in
itial
s
tag
e,
th
e
p
r
e
v
io
u
s
ly
t
r
ain
ed
f
e
atu
r
e
lay
er
s
o
f
th
e
m
o
d
el
ar
e
f
r
o
ze
n
to
p
r
e
v
en
t
f
u
r
th
er
u
p
d
ates
o
r
tr
ain
in
g
.
T
h
i
s
ap
p
r
o
ac
h
aim
s
to
r
ec
o
n
f
i
g
u
r
e
th
e
m
o
d
el
o
u
tp
u
t
ex
clu
s
iv
ely
an
d
tr
ain
t
h
e
n
ew
ly
ad
d
ed
class
if
icatio
n
lay
er
.
T
h
e
m
o
d
el
q
u
ick
ly
a
d
ap
ts
its
o
u
tp
u
t
to
th
e
n
ew
task
wh
ile
s
till
r
etain
in
g
th
e
f
e
atu
r
es lea
r
n
ed
f
r
o
m
I
m
ag
eNe
t.
I
n
th
e
s
ec
o
n
d
s
tag
e,
s
elec
tiv
ely
u
n
f
r
ee
ze
s
o
m
e
o
r
all
o
f
th
e
co
n
v
o
l
u
tio
n
al
lay
e
r
s
o
f
th
e
m
o
d
el.
T
h
is
s
tep
aim
s
to
r
ein
tr
o
d
u
ce
t
h
ese
s
p
ec
if
ic
lay
er
s
in
to
th
e
tr
ai
n
in
g
p
r
o
ce
s
s
,
allo
win
g
th
em
to
r
ef
in
e
a
n
d
ad
ap
t
th
e
ac
q
u
ir
ed
f
ea
t
u
r
es
to
s
u
it
th
e
d
em
an
d
s
o
f
th
e
n
ew
task
.
W
e
also
in
ten
tio
n
ally
u
n
f
r
ee
ze
ce
r
tain
co
n
v
o
lu
tio
n
al
lay
er
s
to
cr
ea
te
b
alan
ce
.
T
h
is
ap
p
r
o
ac
h
allo
ws
th
e
m
o
d
el
to
ad
ap
t
to
th
e
co
m
p
lex
task
o
f
d
ip
ter
o
ca
r
p
u
s
s
tem
ca
teg
o
r
izatio
n
wh
ile
r
etain
in
g
th
e
im
p
o
r
tan
t
f
ea
tu
r
es
o
r
ig
in
a
lly
lear
n
ed
.
T
h
is
m
eth
o
d
em
p
o
wer
s
th
e
m
o
d
el
to
p
r
o
b
e
d
ee
p
er
wh
ile
lev
e
r
ag
in
g
its
k
n
o
wled
g
e
f
r
o
m
p
r
ev
io
u
s
tr
ain
in
g
.
T
h
is
ap
p
r
o
ac
h
g
iv
e
s
th
e
m
o
d
el
s
tr
o
n
g
f
ea
tu
r
e
e
x
tr
ac
tio
n
ca
p
ab
ilit
ies
f
r
o
m
p
r
e
-
tr
ain
e
d
m
o
d
els
an
d
ef
f
ec
tiv
el
y
a
d
ap
ts
th
em
to
d
ip
ter
o
ca
r
p
tr
u
n
k
class
if
icatio
n
.
2
.
5
.
M
o
del
e
v
a
lua
t
i
o
n
T
h
e
ev
al
u
atio
n
m
etr
ics
we
u
s
ed
wer
e
av
er
a
g
e
ac
c
u
r
ac
y
an
d
Ma
cr
o
F1
-
s
co
r
e.
T
h
e
a
v
e
r
ag
e
ac
cu
r
ac
y
ca
lcu
lates
th
e
ac
cu
r
ac
y
f
o
r
e
ac
h
class
s
ep
ar
ately
an
d
th
e
n
tak
es
th
e
av
er
ag
e.
Av
er
ag
e
ac
cu
r
ac
y
o
v
er
all
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
e
r
20
24
:
6
8
7
4
-
6
8
8
2
6878
m
ea
s
u
r
es
h
o
w
well
a
class
if
icatio
n
m
o
d
el
p
er
f
o
r
m
s
ac
r
o
s
s
all
class
es
[
3
1
]
.
I
n
co
n
tr
ast,
t
h
e
Ma
cr
o
F1
-
s
co
r
e
p
r
o
v
id
es
a
b
alan
ce
d
m
ea
s
u
r
e
o
f
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
to
ac
h
iev
e
g
o
o
d
p
r
ec
is
io
n
a
n
d
r
ec
all
ac
r
o
s
s
all
class
es
wi
th
o
u
t
co
n
s
id
er
i
n
g
class
im
b
alan
ce
.
Ma
cr
o
F1
-
s
co
r
e
is
p
ar
ticu
lar
ly
u
s
ef
u
l
wh
en
d
ea
lin
g
with
im
b
alan
ce
d
d
atasets
,
as
it
en
s
u
r
es
th
at
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
is
ev
alu
ated
ac
r
o
s
s
all
c
lass
es,
p
r
o
v
id
in
g
a
b
alan
ce
d
ass
ess
m
en
t
o
f
its
a
b
ilit
y
to
h
an
d
le
d
if
f
er
en
t
ty
p
es
o
f
in
s
tan
ce
s
[
3
2
]
.
T
h
e
a
v
er
ag
e
ac
cu
r
ac
y
an
d
Ma
cr
o
F1
-
s
co
r
e
ar
e
ca
lcu
lated
as
(
1
)
an
d
(
2
)
:
=
∑
(
∗
)
=
1
(
1
)
1
−
=
∑
1
=
1
(
2
)
w
h
er
e
is
th
e
ac
cu
r
ac
y
f
o
r
th
e
class
ca
lcu
lated
as th
e
n
u
m
b
er
o
f
tr
u
e
p
o
s
itiv
e
p
r
ed
ictio
n
s
f
o
r
class
d
iv
id
e
d
b
y
t
h
e
to
tal
n
u
m
b
er
o
f
in
s
tan
ce
s
p
r
e
d
icted
as
class
.
is
th
e
n
u
m
b
er
o
f
ac
tu
al
in
s
tan
ce
s
o
f
class
in
t
h
e
test
s
et,
a
n
d
to
tal
s
am
p
les
a
r
e
th
e
to
tal
n
u
m
b
er
o
f
s
am
p
les
in
th
e
test
s
et.
1
−
is
th
e
h
ar
m
o
n
ic
m
ea
n
o
f
p
r
e
cisi
o
n
an
d
r
ec
all
f
o
r
class
,
w
h
er
e
p
r
ec
is
io
n
is
th
e
n
u
m
b
er
o
f
tr
u
e
p
o
s
itiv
e
p
r
e
d
ictio
n
s
f
o
r
class
d
iv
id
ed
b
y
th
e
to
tal
n
u
m
b
er
o
f
p
r
e
d
icted
i
n
s
tan
ce
s
f
o
r
class
,
an
d
r
ec
all
is
th
e
n
u
m
b
er
o
f
tr
u
e
p
o
s
itiv
e
p
r
ed
i
ctio
n
s
f
o
r
class
d
iv
id
ed
b
y
t
h
e
to
tal
n
u
m
b
er
o
f
ac
tu
al
p
o
s
itiv
e
in
s
tan
ce
s
f
o
r
class
.
r
ep
r
esen
ts
th
e
to
tal
n
u
m
b
er
o
f
class
es in
th
e
class
if
ic
atio
n
p
r
o
b
lem
[
3
3
]
.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
Ou
r
ex
p
er
im
e
n
ts
u
tili
ze
d
Py
th
o
n
T
en
s
o
r
Flo
w
2
.
9
.
2
with
a
NVI
DI
A
P5
0
0
0
GPU.
T
h
e
p
r
e
-
tr
ain
ed
m
o
d
el
was
tr
ain
ed
with
a
b
atch
s
ize
o
f
1
6
an
d
a
m
a
x
im
u
m
o
f
5
0
tr
ain
in
g
ep
o
c
h
s
.
E
ar
ly
s
to
p
p
in
g
was
u
s
ed
,
s
to
p
p
in
g
tr
ain
in
g
if
th
e
v
alid
a
tio
n
ac
cu
r
ac
y
r
em
ai
n
ed
co
n
s
t
an
t
an
d
u
n
c
h
an
g
e
d
f
o
r
f
iv
e
c
o
n
s
ec
u
tiv
e
ep
o
ch
s
.
T
h
e
Ad
am
o
p
tim
izer
was
in
iti
ally
u
s
ed
with
a
lear
n
in
g
r
ate
o
f
0
.
0
0
1
,
s
witch
in
g
to
0
.
0
0
0
1
in
th
e
s
ec
o
n
d
s
tag
e
to
f
ac
ilit
ate
g
r
ad
u
al
ch
an
g
es
an
d
r
etain
p
r
ev
io
u
s
ly
lear
n
e
d
f
ea
tu
r
es,
im
p
r
o
v
in
g
m
o
d
el
g
en
er
aliza
tio
n
f
o
r
Dip
ter
o
ca
r
p
ac
ea
e
s
tem
class
i
f
icatio
n
.
T
h
e
ca
teg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
f
u
n
ctio
n
m
ea
s
u
r
es
th
e
d
if
f
er
e
n
ce
b
etwe
en
th
e
t
r
u
e
la
b
els
an
d
p
r
ed
icted
p
r
o
b
a
b
ilit
ies.
C
o
n
v
er
s
ely
,
ac
cu
r
ac
y
ass
ess
es
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
d
u
r
in
g
tr
ain
in
g
,
in
d
icatin
g
th
e
p
r
o
p
o
r
tio
n
o
f
co
r
r
ec
tly
p
r
ed
ict
ed
ev
e
n
ts
.
Data
au
g
m
en
tatio
n
tech
n
iq
u
es
s
u
ch
as
r
an
d
o
m
zo
o
m
in
g
,
b
r
ig
h
tn
ess
ad
ju
s
tm
en
t,
s
ca
lin
g
,
cr
o
p
p
i
n
g
,
an
d
s
h
if
tin
g
wer
e
u
s
ed
to
a
u
g
m
en
t
th
e
lim
ited
Dip
ter
o
ca
r
p
ac
ea
e
im
a
g
e
d
atas
et
an
d
im
p
r
o
v
e
m
o
d
el
g
e
n
er
ali
za
tio
n
.
A
co
m
p
ar
ativ
e
an
aly
s
is
ex
p
lo
r
es
th
e
im
p
ac
t
o
f
d
if
f
er
en
t
f
ea
tu
r
e
lay
er
s
tr
ateg
ies
o
n
th
e
VGG1
6
m
o
d
el.
T
h
e
an
al
y
s
is
in
v
o
lv
es
f
r
ee
zin
g
a
n
d
t
h
awin
g
f
e
atu
r
e
lay
er
s
at
d
if
f
er
e
n
t
s
ca
les.
Per
f
o
r
m
an
c
e
co
m
p
ar
is
o
n
s
ar
e
p
er
f
o
r
m
e
d
u
n
d
er
v
ar
io
u
s
s
ce
n
ar
io
s
,
in
clu
d
i
n
g
n
o
lay
er
s
f
r
o
ze
n
,
th
awin
g
a
ll
lay
er
s
ex
ce
p
t
th
e
in
itial
lay
er
,
th
awin
g
all
lay
er
s
co
m
p
letely
,
an
d
th
awin
g
all
lay
er
s
ex
ce
p
t
th
e
last
lay
er
.
E
v
alu
atio
n
m
etr
ics
ass
es
s
th
e
ef
f
ec
tiv
en
ess
an
d
g
en
er
aliza
tio
n
ab
ilit
y
o
f
ea
ch
s
tr
ateg
y
.
T
ab
le
2
illu
s
tr
ates
th
e
ef
f
ec
ts
o
f
d
if
f
e
r
en
t
f
r
ee
ze
o
n
th
e
VGG1
6
m
o
d
el.
T
ab
le
2
.
VGG1
6
p
er
f
o
r
m
an
ce
co
m
p
ar
is
o
n
o
n
d
i
f
f
er
en
t stra
te
g
ies
S
t
r
a
t
e
g
y
A
v
e
r
a
g
e
a
c
c
u
r
a
c
y
M
a
c
r
o
F
1
-
s
c
o
r
e
No
f
r
e
e
z
e
t
h
e
f
e
a
t
u
r
e
l
a
y
e
r
9
4
.
1
1
8
%
9
4
.
0
2
4
%
U
n
f
r
e
e
z
e
a
l
l
f
e
a
t
u
r
e
l
a
y
e
r
s
e
x
c
e
p
t
t
h
e
1
st
b
l
o
c
k
l
a
y
e
r
9
7
.
7
9
4
%
9
7
.
7
2
3
%
U
n
f
r
e
e
z
e
a
l
l
f
e
a
t
u
r
e
l
a
y
e
r
s w
i
t
h
o
u
t
e
x
c
e
p
t
i
o
n
9
8
.
2
4
6
%
9
8
.
1
8
3
%
U
n
f
r
e
e
z
e
o
n
l
y
t
h
e
l
a
st
b
l
o
c
k
l
a
y
e
r
9
5
.
5
8
8
%
9
5
.
5
2
5
%
As
a
co
m
p
ar
is
o
n
,
we
co
n
d
u
cted
th
e
tr
ain
in
g
o
n
ea
ch
p
r
e
-
tr
ain
ed
m
o
d
el:
VGG1
6
,
I
n
ce
p
tio
n
V3
,
R
esNet5
0
,
Den
s
eNe
t1
2
1
,
Mo
b
ileNetV2
,
an
d
E
f
f
icien
tNet
V2
-
B
0
.
All
m
o
d
els
f
o
llo
w
th
e
two
-
s
tag
e
tr
an
s
f
er
lear
n
in
g
ap
p
r
o
ac
h
,
wh
er
e
all
co
n
v
o
l
u
tio
n
al
lay
er
s
ar
e
f
r
o
ze
n
in
th
e
f
ir
s
t
s
tag
e.
I
n
th
e
o
th
er
h
an
d
,
all
m
o
d
el
f
ea
tu
r
e
lay
e
r
s
ar
e
u
n
f
r
o
ze
n
.
W
e
u
n
f
r
ee
ze
all
th
e
f
ea
tu
r
e
lay
er
s
in
th
e
s
ec
o
n
d
s
tag
e
f
o
r
all
th
e
m
o
d
els
b
ased
o
n
th
e
p
r
e
v
io
u
s
s
tr
ateg
ies
e
x
p
er
im
en
t
.
Per
f
o
r
m
an
ce
co
m
p
ar
is
o
n
ca
n
b
e
s
ee
n
in
T
a
b
le
3
.
I
n
th
e
f
ir
s
t
s
tag
e
,
VGG1
6
h
ad
th
e
lo
west
ac
cu
r
a
cy
,
9
3
.
9
0
7
%.
Ho
wev
er
,
af
ter
t
h
e
s
ec
o
n
d
s
tag
e,
VGG1
6
b
ec
a
m
e
th
e
m
o
d
el
with
th
e
h
ig
h
est
ac
cu
r
ac
y
am
o
n
g
s
ix
o
th
er
m
o
d
els,
n
am
ely
an
ac
cu
r
ac
y
o
f
9
8
.
2
4
6
%
an
d
a
m
ac
r
o
F1
-
s
co
r
e
o
f
9
8
.
1
8
3
%.
R
esNet5
0
,
Den
s
eNe
t1
2
1
,
an
d
Mo
b
ileNetV2
h
av
e
s
im
ilar
ac
cu
r
ac
y
in
t
h
e
f
ir
s
t
s
t
ag
e
o
f
~9
6
%,
with
an
in
cr
ea
s
e
in
ac
cu
r
a
cy
in
th
e
s
ec
o
n
d
s
tag
e
o
f
1
%
f
o
r
t
h
e
R
esNet5
0
an
d
Den
s
eNe
t1
2
1
m
o
d
els.
Ho
wev
er
,
Mo
b
ileNetV2
ac
tu
ally
ex
p
er
ie
n
ce
d
a
r
ea
s
o
n
a
b
ly
h
ig
h
d
ec
r
ea
s
e
in
p
e
r
f
o
r
m
an
ce
,
a
d
ec
r
ea
s
e
i
n
ac
c
u
r
ac
y
o
f
u
p
to
8
.
2
7
9
%,
an
d
th
e
E
f
f
icien
t
NetV2
-
B
0
m
o
d
el
ex
p
er
ien
ce
d
a
d
ec
r
ea
s
e
in
ac
cu
r
ac
y
o
f
5
.
4
0
1
%.
T
h
ese
two
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Dip
tero
ca
r
p
a
ce
a
e
tr
u
n
k
textu
r
e
cla
s
s
ifica
tio
n
u
s
in
g
tw
o
-
s
ta
g
e
…
(
Ma
s
n
a
Wa
ti)
6879
m
o
d
els
ex
p
e
r
ien
ce
d
a
d
ec
r
ea
s
e
in
p
er
f
o
r
m
an
ce
,
p
o
s
s
ib
ly
b
ec
au
s
e
th
ey
wer
e
s
m
all
m
o
d
els
th
at
o
n
ly
h
ad
p
ar
am
eter
s
o
f
3
.
5
M
an
d
7
.
2
M,
th
u
s
as
th
e
s
ec
o
n
d
tr
ain
in
g
was
ca
r
r
ied
o
u
t,
it
r
esu
lted
in
o
v
er
f
itti
n
g
.
On
th
e
o
th
er
h
an
d
,
in
th
e
I
n
ce
p
tio
n
V3
m
o
d
el,
th
e
r
e
is
n
o
s
ig
n
if
ican
t
ch
an
g
e
in
p
e
r
f
o
r
m
an
ce
f
r
o
m
s
tag
e
o
n
e
to
s
tag
e
2
.
VGG1
6
ex
p
er
ien
ce
d
q
u
ite
a
s
ig
n
if
ican
t
in
cr
ea
s
e
in
p
er
f
o
r
m
an
ce
d
u
e
to
th
e
h
i
g
h
p
ar
am
eter
s
,
th
er
e
is
s
till
r
o
o
m
f
o
r
im
p
r
o
v
em
en
t d
u
r
in
g
s
ec
o
n
d
tr
ain
in
g
.
Fig
u
r
e
3
s
h
o
ws
th
e
a
cc
u
r
ac
y
an
d
lo
s
s
o
f
th
e
m
o
d
el
f
o
r
u
n
f
r
ee
zin
g
all
f
ea
tu
r
e
lay
er
s
with
o
u
t
ex
clu
s
io
n
.
I
n
t
h
e
f
ir
s
t
s
tag
e,
th
e
m
o
d
el
lea
r
n
s
f
o
r
1
3
ep
o
ch
s
,
ac
h
iev
i
n
g
tr
ain
in
g
ac
c
u
r
ac
y
o
f
~9
8
%
a
n
d
v
alid
atio
n
ac
c
u
r
ac
y
o
f
~9
4
%
ca
n
b
e
s
ee
n
in
Fig
u
r
e
3
(
a)
an
d
lo
s
s
at
f
ir
s
t
s
tag
e
ca
n
b
e
s
ee
n
in
Fig
u
r
e
3
(
b
)
.
I
n
th
e
s
ec
o
n
d
s
tag
e
as
s
h
o
wn
i
n
Fig
u
r
e
s
3
(
c)
an
d
3
(
d
)
,
th
e
r
e
is
in
s
tab
ilit
y
in
tr
ain
in
g
a
cc
u
r
ac
y
an
d
lo
s
s
.
T
h
er
ef
o
r
e,
ea
r
ly
s
to
p
p
in
g
is
ap
p
r
o
p
r
iate
to
p
r
ev
en
t
o
v
e
r
f
itti
n
g
/u
n
d
e
r
f
itti
n
g
d
u
e
to
t
r
ain
in
g
in
s
tab
ilit
y
.
Fig
u
r
e
3
(
c
)
d
escr
ib
es
th
e
m
o
d
el
in
th
e
s
ec
o
n
d
s
tag
e
r
ea
ch
es
1
0
0
%
v
alid
atio
n
ac
cu
r
ac
y
at
ep
o
ch
8
,
t
h
er
ef
o
r
e
th
e
m
o
d
el
s
to
p
t
r
ain
in
g
at
e
p
o
ch
1
3
.
T
ab
le
3
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
n
d
if
f
er
en
t
m
o
d
els
M
o
d
e
l
A
v
e
r
a
g
e
a
c
c
u
r
a
c
y
F
i
r
st
s
t
a
g
e
A
v
e
r
a
g
e
a
c
c
u
r
a
c
y
S
e
c
o
n
d
s
t
a
g
e
M
a
c
r
o
F
1
-
sc
o
r
e
F
i
r
st
s
t
a
g
e
M
a
c
r
o
F
1
-
sc
o
r
e
S
e
c
o
n
d
s
t
a
g
e
P
a
r
a
me
t
e
r
s
V
G
G
1
6
9
3
.
9
0
7
%
9
8
.
2
4
6
%
9
3
.
5
9
0
%
9
8
.
1
8
3
%
1
3
8
.
4
M
I
n
c
e
p
t
i
o
n
V
3
9
4
.
7
3
7
%
9
4
.
8
8
0
%
9
4
%
9
4
.
6
3
5
%
2
3
.
9
M
R
e
sN
e
t
5
0
9
6
.
6
4
0
%
9
7
.
6
6
1
%
9
6
.
5
4
4
%
9
7
.
6
2
8
%
2
5
.
6
M
D
e
n
seN
e
t
1
2
1
9
6
.
4
9
1
%
9
7
.
6
6
1
%
9
6
.
6
0
5
%
9
7
.
6
5
2
%
8
.
1
M
M
o
b
i
l
e
N
e
t
V
2
9
6
.
5
8
3
%
8
8
.
3
0
4
%
9
6
.
4
3
1
%
8
6
.
4
1
1
%
3
.
5
M
Ef
f
i
c
i
e
n
t
N
e
t
V
2
-
B0
9
5
.
4
5
9
%
9
0
.
0
5
8
%
9
5
.
3
2
0
%
8
9
.
5
8
1
%
7
.
2
M
(
a)
(
b
)
(
c)
(
d
)
Fig
u
r
e
3
.
Acc
u
r
ac
y
a
n
d
lo
s
s
f
o
r
two
-
s
tag
e
tr
an
s
f
er
lear
n
in
g
o
f
V
GG1
6
(
a)
f
ir
s
t stag
e
ac
cu
r
a
cy
(
b
)
l
o
s
s
at
f
ir
s
t
s
tag
e
(
c)
s
ec
o
n
d
s
tag
e
ac
c
u
r
ac
y
(
d
)
l
o
s
s
at
s
ec
o
n
d
s
tag
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
e
r
20
24
:
6
8
7
4
-
6
8
8
2
6880
4.
CO
NCLU
SI
O
N
Ma
n
ag
em
en
t
o
f
e
n
d
em
ic
p
l
an
t
class
if
icatio
n
in
tr
o
p
ical
r
ain
f
o
r
ests
as
an
ef
f
o
r
t
to
p
r
o
tect
b
io
d
iv
er
s
ity
.
T
h
is
s
tu
d
y
p
r
es
en
ts
a
two
-
s
tag
e
tr
an
s
f
er
lea
r
n
in
g
m
eth
o
d
to
im
p
r
o
v
e
th
e
class
if
icat
io
n
o
f
Dip
ter
o
ca
r
p
ac
ea
e
s
p
ec
ies
b
ased
o
n
s
tem
tex
tu
r
e
im
ag
es.
T
h
is
r
esear
ch
ass
i
s
ts
co
n
s
er
v
atio
n
ef
f
o
r
ts
b
y
en
ab
lin
g
r
a
p
id
an
d
r
eliab
le
tr
e
e
s
p
ec
ies
id
en
tific
atio
n
b
ased
o
n
tr
u
n
k
tex
tu
r
e
im
ag
es.
T
h
e
r
esu
lts
o
f
th
e
two
-
s
tag
e
tr
an
s
f
er
lear
n
in
g
VGG1
6
s
h
o
wed
an
in
c
r
ea
s
e
in
p
er
f
o
r
m
an
ce
o
f
9
8
.
2
4
6
%.
Ho
wev
e
r
,
n
o
t
all
m
o
d
els
ar
e
s
u
itab
le
f
o
r
s
ec
o
n
d
t
r
ain
in
g
;
m
o
d
els
with
s
m
all
p
ar
am
eter
s
,
s
u
ch
as
E
f
f
icien
tNetV2
-
B
0
an
d
Mo
b
ileNetV2
,
ac
tu
ally
ex
p
er
ien
ce
a
d
ec
r
ea
s
e
in
p
er
f
o
r
m
an
ce
o
f
u
p
to
8
%.
T
h
is
m
eth
o
d
is
s
u
itab
le
f
o
r
m
o
d
els
th
at
h
av
e
s
ig
n
if
ican
t
p
ar
am
eter
s
wh
er
e
th
er
e
is
s
til
l
a
lo
t
o
f
r
o
o
m
f
o
r
im
p
r
o
v
e
m
en
t.
New
m
eth
o
d
s
n
ee
d
to
b
e
d
ev
elo
p
e
d
to
im
p
r
o
v
e
m
o
d
el
p
er
f
o
r
m
a
n
c
e
with
s
m
all
p
ar
am
eter
s
.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
e
ex
p
r
ess
g
r
atitu
d
e
f
o
r
th
e
f
u
n
d
in
g
s
u
p
p
o
r
t
p
r
o
v
id
ed
in
th
e
r
esear
ch
g
r
a
n
t
to
th
e
Dir
e
cto
r
ate
o
f
R
esear
ch
,
T
ec
h
n
o
lo
g
y
,
an
d
C
o
m
m
u
n
ity
Ser
v
ice
(
DR
PM)
o
f
th
e
Min
is
tr
y
o
f
E
d
u
ca
tio
n
,
C
u
ltu
r
e,
R
esear
ch
,
an
d
T
ec
h
n
o
lo
g
y
o
f
th
e
R
ep
u
b
lic
o
f
I
n
d
o
n
esia
a
n
d
t
o
th
e
I
n
s
titu
te
f
o
r
R
esear
ch
a
n
d
C
o
m
m
u
n
ity
Ser
v
ice
(
L
PP
M)
o
f
Mu
lawa
r
m
a
n
U
n
iv
er
s
ity
(
C
o
n
tr
ac
t
Nu
m
b
er
2
9
0
/UN1
7
.
L
1
/HK/2
0
2
3
)
.
T
h
e
au
th
o
r
s
th
an
k
B
B
PS
I
L
H
Sam
ar
in
d
a
an
d
Ar
b
o
r
etu
m
Sam
ar
in
d
a
f
o
r
s
u
p
p
o
r
t
in
g
d
ata
co
llectio
n
.
RE
F
E
R
E
NC
E
S
[
1
]
F
.
Q
.
B
r
e
a
r
l
e
y
,
L.
F
.
B
a
n
i
n
,
a
n
d
P
.
S
a
n
e
r
,
“
Th
e
e
c
o
l
o
g
y
o
f
t
h
e
A
si
a
n
d
i
p
t
e
r
o
c
a
r
p
s
,
”
Pl
a
n
t
Ec
o
l
o
g
y
&
D
i
v
e
rs
i
t
y
,
v
o
l
.
9
,
n
o
.
5
–
6
,
p
p
.
4
2
9
–
4
3
6
,
2
0
1
6
.
[
2
]
M
.
W
a
t
i
,
N
.
P
u
s
p
i
t
a
s
a
r
i
,
U
.
H
a
i
r
a
h
,
J
.
A
.
W
i
d
i
a
n
s
,
A
.
S
e
p
t
i
a
r
i
n
i
,
a
n
d
A
.
F
.
T
j
i
k
o
a
,
“
B
o
r
n
e
o
s
mart
f
o
r
e
st
i
n
f
o
r
m
a
t
i
o
n
s
y
st
e
m
f
o
r
man
a
g
e
me
n
t
o
f
d
i
p
t
e
r
o
c
a
r
p
p
l
a
n
t
s
i
n
K
a
l
i
ma
n
t
a
n
r
a
i
n
f
o
r
e
s
t
,
”
i
n
2
0
2
2
I
E
EE
8
t
h
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
I
n
t
e
rn
a
t
i
o
n
a
l
S
e
m
i
n
a
r
(
I
T
I
S
)
,
I
EEE,
2
0
2
2
,
p
p
.
1
3
4
–
1
3
9
.
[
3
]
G
.
M
a
u
r
y
-
Le
c
h
o
n
a
n
d
L
.
C
u
r
t
e
t
,
“
B
i
o
g
e
o
g
r
a
p
h
y
a
n
d
e
v
o
l
u
t
i
o
n
a
r
y
s
y
st
e
m
a
t
i
c
s
o
f
D
i
p
t
e
r
o
c
a
r
p
a
c
e
a
e
,
”
A
R
e
v
i
e
w
o
f
D
i
p
t
e
ro
c
a
r
p
s:
T
a
x
o
n
o
m
y
,
E
c
o
l
o
g
y
a
n
d
S
i
l
v
i
c
u
l
t
u
r
e
.
C
e
n
t
e
r f
o
r I
n
t
e
rn
a
t
i
o
n
a
l
F
o
res
t
ry
Re
s
e
a
r
c
h
,
I
n
d
o
n
e
si
a
,
p
p
.
5
–
4
4
,
1
9
9
8
.
[
4
]
T.
C
v
e
t
k
o
v
i
ć
,
D
.
D
.
H
i
n
si
n
g
e
r
,
D
.
C
.
Th
o
mas
,
J.
J
.
W
i
e
r
i
n
g
a
,
E
.
V
e
l
a
u
t
h
a
m,
a
n
d
J.
S
.
S
t
r
i
j
k
,
“
P
h
y
l
o
g
e
n
o
m
i
c
s
a
n
d
a
r
e
v
i
s
e
d
t
r
i
b
a
l
c
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
s
u
b
f
a
m
i
l
y
D
i
p
t
e
r
o
c
a
r
p
o
i
d
e
a
e
(
D
i
p
t
e
r
o
c
a
r
p
a
c
e
a
e
)
,
”
T
a
x
o
n
,
v
o
l
.
7
1
,
n
o
.
1
,
p
p
.
8
5
–
1
0
2
,
2
0
2
2
.
[
5
]
R
.
K
a
e
w
t
h
o
n
g
r
a
c
h
,
Y
.
V
i
t
a
s
se
,
T.
La
mj
i
a
k
,
a
n
d
A
.
C
h
i
d
t
h
a
i
so
n
g
,
“
I
mp
a
c
t
o
f
s
e
v
e
r
e
d
r
o
u
g
h
t
d
u
r
i
n
g
t
h
e
st
r
o
n
g
2
0
1
5
/
2
0
1
6
El
N
i
n
o
o
n
t
h
e
p
h
e
n
o
l
o
g
y
a
n
d
su
r
v
i
v
a
l
o
f
se
c
o
n
d
a
r
y
d
r
y
d
i
p
t
e
r
o
c
a
r
p
sp
e
c
i
e
s
i
n
W
e
s
t
e
r
n
Th
a
i
l
a
n
d
,
”
Fo
re
st
s
,
v
o
l
.
1
0
,
n
o
.
1
1
,
p
.
9
6
7
,
2
0
1
9
.
[
6
]
M
.
H
e
i
k
k
i
l
ä
,
M
.
P
i
e
t
i
k
ä
i
n
e
n
,
a
n
d
C
.
S
c
h
m
i
d
,
“
D
e
scr
i
p
t
i
o
n
o
f
i
n
t
e
r
e
st
r
e
g
i
o
n
s
w
i
t
h
l
o
c
a
l
b
i
n
a
r
y
p
a
t
t
e
r
n
s,”
P
a
t
t
e
rn
R
e
c
o
g
n
i
t
i
o
n
,
v
o
l
.
4
2
,
n
o
.
3
,
p
p
.
4
2
5
–
4
3
6
,
2
0
0
9
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
a
t
c
o
g
.
2
0
0
8
.
0
8
.
0
1
4
.
[
7
]
V
.
R
e
m
e
š
a
n
d
M
.
H
a
i
n
d
l
,
“
B
a
r
k
r
e
c
o
g
n
i
t
i
o
n
u
si
n
g
n
o
v
e
l
r
o
t
a
t
i
o
n
a
l
l
y
i
n
v
a
r
i
a
n
t
m
u
l
t
i
s
p
e
c
t
r
a
l
t
e
x
t
u
r
a
l
f
e
a
t
u
r
e
s,”
P
a
t
t
e
r
n
re
c
o
g
n
i
t
i
o
n
l
e
t
t
e
rs
,
v
o
l
.
1
2
5
,
p
p
.
6
1
2
–
6
1
7
,
2
0
1
9
.
[
8
]
M
.
C
i
m
p
o
i
,
S
.
M
a
j
i
,
a
n
d
A
.
V
e
d
a
l
d
i
,
“
D
e
e
p
f
i
l
t
e
r
b
a
n
k
s
f
o
r
t
e
x
t
u
r
e
r
e
c
o
g
n
i
t
i
o
n
a
n
d
se
g
me
n
t
a
t
i
o
n
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EEE
c
o
n
f
e
re
n
c
e
o
n
c
o
m
p
u
t
e
r v
i
si
o
n
a
n
d
p
a
t
t
e
r
n
r
e
c
o
g
n
i
t
i
o
n
,
2
0
1
5
,
p
p
.
3
8
2
8
–
3
8
3
6
.
[
9
]
A
.
Y
.
N
a
w
a
f
a
n
d
W
.
M
.
Jas
i
m,
“
A
p
r
e
-
t
r
a
i
n
e
d
m
o
d
e
l
v
s
d
e
d
i
c
a
t
e
d
c
o
n
v
o
l
u
t
i
o
n
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
e
mo
t
i
o
n
r
e
c
o
g
n
i
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
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
3
,
n
o
.
1
,
p
p
.
1
1
2
3
–
1
1
3
3
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
c
e
.
v
1
3
i
1
.
p
p
1
1
2
3
-
1
1
3
3
.
[
1
0
]
A
.
K
r
i
z
h
e
v
s
k
y
,
I
.
S
u
t
s
k
e
v
e
r
,
a
n
d
G
.
E.
H
i
n
t
o
n
,
“
I
mag
e
N
e
t
c
l
a
ss
i
f
i
c
a
t
i
o
n
w
i
t
h
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
i
n
N
e
u
ra
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
st
e
m
s
,
2
0
1
2
,
p
p
.
1
1
0
6
–
1
1
1
4
.
[
1
1
]
K
.
S
i
mo
n
y
a
n
a
n
d
A
.
Z
i
sserm
a
n
,
“
V
e
r
y
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
s
f
o
r
l
a
r
g
e
-
sca
l
e
i
ma
g
e
r
e
c
o
g
n
i
t
i
o
n
,
”
a
rX
i
v
p
re
p
ri
n
t
a
rXi
v
:
1
4
0
9
.
1
5
5
6
,
2
0
1
4
.
[
1
2
]
K
.
H
e
,
X
.
Z
h
a
n
g
,
S
.
R
e
n
,
a
n
d
J.
S
u
n
,
“
D
e
e
p
r
e
si
d
u
a
l
l
e
a
r
n
i
n
g
f
o
r
i
mag
e
r
e
c
o
g
n
i
t
i
o
n
,
”
i
n
P
ro
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EE
E
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
C
o
n
f
e
ren
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
,
I
E
E
E
,
Ju
n
.
2
0
1
6
,
p
p
.
7
7
0
–
7
7
8
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
6
.
9
0
.
[
1
3
]
G
.
H
u
a
n
g
,
Z.
Li
u
,
L.
V
a
n
D
e
r
M
a
a
t
e
n
,
a
n
d
K
.
Q
.
W
e
i
n
b
e
r
g
e
r
,
“
D
e
n
se
l
y
c
o
n
n
e
c
t
e
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
s,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EEE
c
o
n
f
e
re
n
c
e
o
n
c
o
m
p
u
t
e
r v
i
si
o
n
a
n
d
p
a
t
t
e
r
n
r
e
c
o
g
n
i
t
i
o
n
,
2
0
1
7
,
p
p
.
4
7
0
0
–
4
7
0
8
.
[
1
4
]
C
.
S
z
e
g
e
d
y
,
V
.
V
a
n
h
o
u
c
k
e
,
S
.
I
o
f
f
e
,
J.
S
h
l
e
n
s,
a
n
d
Z.
W
o
j
n
a
,
“
R
e
t
h
i
n
k
i
n
g
t
h
e
i
n
c
e
p
t
i
o
n
a
r
c
h
i
t
e
c
t
u
r
e
f
o
r
c
o
mp
u
t
e
r
v
i
s
i
o
n
,
”
i
n
2
0
1
6
I
EEE
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
si
o
n
a
n
d
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
(
C
VPR),
La
s
V
e
g
a
s,
N
V
,
U
S
A
,
2
0
1
6
,
p
p
.
2
8
1
8
-
2
8
2
6
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
6
.
3
0
8
.
[
1
5
]
J.
G
u
p
t
a
,
S
.
P
a
t
h
a
k
,
a
n
d
G
.
K
u
mar
,
“
D
e
e
p
l
e
a
r
n
i
n
g
(
C
N
N
)
a
n
d
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
:
a
r
e
v
i
e
w
,
”
i
n
J
o
u
rn
a
l
o
f
P
h
y
si
c
s:
C
o
n
f
e
re
n
c
e
S
e
r
i
e
s
,
I
O
P
P
u
b
l
i
s
h
i
n
g
,
2
0
2
2
,
p
.
1
2
0
2
9
,
d
o
i
:
1
0
.
1
0
8
8
/
1
7
4
2
-
6
5
9
6
/
2
2
7
3
/
1
/
0
1
2
0
2
9
.
[
1
6
]
Z.
C
h
e
n
-
M
c
C
a
i
g
,
R
.
H
o
s
e
i
n
n
e
z
h
a
d
,
a
n
d
A
.
B
a
b
-
H
a
d
i
a
s
h
a
r
,
“
C
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
t
e
x
t
u
r
e
r
e
c
o
g
n
i
t
i
o
n
u
s
i
n
g
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
,
”
i
n
2
0
1
7
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
n
t
r
o
l
,
Au
t
o
m
a
t
i
o
n
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
s
,
I
C
C
AI
S
2
0
1
7
,
I
EEE,
2
0
1
7
,
p
p
.
1
8
7
–
192
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
A
I
S
.
2
0
1
7
.
8
2
1
7
5
7
3
.
[
1
7
]
R
.
Z
h
a
n
g
,
Y
.
Zh
u
,
Z
.
G
e
,
H
.
M
u
,
D
.
Q
i
,
a
n
d
H
.
N
i
,
“
Tr
a
n
sf
e
r
l
e
a
r
n
i
n
g
f
o
r
l
e
a
f
sma
l
l
d
a
t
a
se
t
u
s
i
n
g
i
m
p
r
o
v
e
d
R
e
sN
e
t
5
0
n
e
t
w
o
r
k
w
i
t
h
mi
x
e
d
a
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
s,
”
F
o
r
e
st
s
,
v
o
l
.
1
3
,
n
o
.
1
2
,
p
.
2
0
7
2
,
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
f
1
3
1
2
2
0
7
2
.
[
1
8
]
S
.
K
.
R
o
y
,
S
.
R
.
D
u
b
e
y
,
B
.
C
h
a
n
d
a
,
B
.
B
.
C
h
a
u
d
h
u
r
i
,
a
n
d
D
.
K
.
G
h
o
s
h
,
“
T
e
x
f
u
s
i
o
n
n
e
t
:
a
n
e
n
sem
b
l
e
o
f
d
e
e
p
c
n
n
f
e
a
t
u
r
e
f
o
r
t
e
x
t
u
r
e
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
3
rd
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
V
i
si
o
n
a
n
d
I
m
a
g
e
Pr
o
c
e
ss
i
n
g
:
C
VI
P
2
0
1
8
,
v
o
l
.
2
,
S
p
r
i
n
g
e
r
,
2
0
2
0
,
p
p
.
2
7
1
–
2
8
3
.
[
1
9
]
Z.
L
i
,
F
.
Li
u
,
W
.
Y
a
n
g
,
S
.
P
e
n
g
,
a
n
d
J.
Z
h
o
u
,
“
A
s
u
r
v
e
y
o
f
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
:
a
n
a
l
y
si
s,
a
p
p
l
i
c
a
t
i
o
n
s,
a
n
d
p
r
o
sp
e
c
t
s,”
I
EEE
t
r
a
n
s
a
c
t
i
o
n
s
o
n
n
e
u
r
a
l
n
e
t
w
o
r
k
s
a
n
d
l
e
a
r
n
i
n
g
s
y
st
e
m
s
,
2
0
2
1
.
[
2
0
]
S
.
N
a
d
k
a
r
n
i
a
n
d
K
.
N
o
r
o
n
h
a
,
“
B
r
e
a
s
t
c
a
n
c
e
r
d
e
t
e
c
t
i
o
n
w
i
t
h
m
u
l
t
i
p
a
r
a
me
t
r
i
c
ma
g
n
e
t
i
c
r
e
s
o
n
a
n
c
e
i
ma
g
i
n
g
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,
”
i
n
2
0
2
3
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
S
m
a
rt
S
y
st
e
m
s
f
o
r
a
p
p
l
i
c
a
t
i
o
n
s
i
n
El
e
c
t
r
i
c
a
l
S
c
i
e
n
c
e
s
(
I
C
S
S
E
S
)
,
T
u
ma
k
u
r
u
,
I
n
d
i
a
,
2
0
2
3
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Dip
tero
ca
r
p
a
ce
a
e
tr
u
n
k
textu
r
e
cla
s
s
ifica
tio
n
u
s
in
g
tw
o
-
s
ta
g
e
…
(
Ma
s
n
a
Wa
ti)
6881
p
p
.
1
-
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
S
S
ES5
8
2
9
9
.
2
0
2
3
.
1
0
1
9
9
4
7
0
.
[
2
1
]
R
.
Y
a
m
a
s
h
i
t
a
,
M
.
N
i
s
h
i
o
,
R
.
K
.
G
.
D
o
,
a
n
d
K
.
To
g
a
s
h
i
,
“
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
:
a
n
o
v
e
r
v
i
e
w
a
n
d
a
p
p
l
i
c
a
t
i
o
n
i
n
r
a
d
i
o
l
o
g
y
,
”
I
n
s
i
g
h
t
s
i
n
t
o
I
m
a
g
i
n
g
,
v
o
l
.
9
,
n
o
.
4
,
p
p
.
6
1
1
–
6
2
9
,
Ju
n
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
3
2
4
4
-
018
-
0
6
3
9
-
9.
[
2
2
]
P
.
H
r
i
d
a
y
a
m
i
,
I
.
K
.
G
.
D
.
P
u
t
r
a
,
a
n
d
K
.
S
.
W
i
b
a
w
a
,
“
F
i
s
h
s
p
e
c
i
e
s
r
e
c
o
g
n
i
t
i
o
n
u
s
i
n
g
V
G
G
1
6
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
J
o
u
rn
a
l
o
f
C
o
m
p
u
t
i
n
g
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
1
3
,
n
o
.
3
,
p
p
.
1
2
4
–
1
3
0
,
2
0
1
9
.
[
2
3
]
F
.
Z
h
u
a
n
g
e
t
a
l
.
,
“
A
C
o
mp
r
e
h
e
n
s
i
v
e
s
u
r
v
e
y
o
n
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
,
”
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
I
E
EE
,
v
o
l
.
1
0
9
,
n
o
.
1
,
p
p
.
43
–
7
6
,
Ja
n
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
JP
R
O
C
.
2
0
2
0
.
3
0
0
4
5
5
5
.
[
2
4
]
K
.
W
e
i
ss,
T.
M
.
K
h
o
s
h
g
o
f
t
a
a
r
,
a
n
d
D
.
W
a
n
g
,
“
A
s
u
r
v
e
y
o
f
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
,
”
J
o
u
r
n
a
l
o
f
Bi
g
d
a
t
a
,
v
o
l
.
3
,
n
o
.
1
,
p
p
.
1
–
4
0
,
2
0
1
6
.
[
2
5
]
C
.
Ta
n
,
F
.
S
u
n
,
T
.
K
o
n
g
,
W
.
Zh
a
n
g
,
C
.
Y
a
n
g
,
a
n
d
C
.
L
i
u
,
“
A
su
r
v
e
y
o
n
d
e
e
p
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
,
”
i
n
Art
i
f
i
c
i
a
l
N
e
u
ra
l
N
e
t
w
o
r
k
s a
n
d
Ma
c
h
i
n
e
L
e
a
r
n
i
n
g
–
I
C
AN
N
2
0
1
8
:
2
7
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
Ar
t
i
f
i
c
i
a
l
N
e
u
ra
l
N
e
t
w
o
rks
,
R
h
o
d
e
s
,
G
re
e
c
e
,
O
c
t
o
b
e
r
4
-
7
,
2
0
1
8
,
Pro
c
e
e
d
i
n
g
s,
P
a
rt
I
I
I
2
7
,
S
p
r
i
n
g
e
r
,
2
0
1
8
,
p
p
.
2
7
0
–
2
7
9
.
[
2
6
]
C
.
C
.
A
g
g
a
r
w
a
l
,
“
N
e
u
r
a
l
n
e
t
w
o
r
k
s
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
,
”
S
p
r
i
n
g
e
r
,
v
o
l
.
1
0
,
n
o
.
9
7
8
,
p
.
3
,
2
0
1
8
.
[
2
7
]
O
.
R
u
ssa
k
o
v
s
k
y
e
t
a
l
.
,
“
I
ma
g
e
N
e
t
l
a
r
g
e
sca
l
e
v
i
s
u
a
l
r
e
c
o
g
n
i
t
i
o
n
c
h
a
l
l
e
n
g
e
,
”
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
1
5
,
n
o
.
3
,
p
p
.
2
1
1
–
2
5
2
,
D
e
c
.
2
0
1
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
2
6
3
-
0
1
5
-
0
8
1
6
-
y.
[
2
8
]
T.
H
a
n
,
C
.
L
i
u
,
W
.
Y
a
n
g
,
a
n
d
D
.
Ji
a
n
g
,
“
Le
a
r
n
i
n
g
t
r
a
n
sf
e
r
a
b
l
e
f
e
a
t
u
r
e
s
i
n
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
d
i
a
g
n
o
si
n
g
u
n
s
e
e
n
m
a
c
h
i
n
e
c
o
n
d
i
t
i
o
n
s,
”
I
S
A
t
ra
n
sa
c
t
i
o
n
s
,
v
o
l
.
9
3
,
p
p
.
3
4
1
–
3
5
3
,
2
0
1
9
.
[
2
9
]
M
.
S
a
n
d
l
e
r
,
A
.
H
o
w
a
r
d
,
M
.
Zh
u
,
A
.
Zh
m
o
g
i
n
o
v
,
a
n
d
L.
-
C
.
C
h
e
n
,
“
M
o
b
i
l
e
N
e
t
V
2
:
I
n
v
e
r
t
e
d
r
e
si
d
u
a
l
s
a
n
d
l
i
n
e
a
r
b
o
t
t
l
e
n
e
c
k
s,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
I
E
EE
c
o
n
f
e
re
n
c
e
o
n
c
o
m
p
u
t
e
r
v
i
s
i
o
n
a
n
d
p
a
t
t
e
rn
re
c
o
g
n
i
t
i
o
n
,
2
0
1
8
,
p
p
.
4
5
1
0
–
4
5
2
0
.
[
3
0
]
M
.
Ta
n
a
n
d
Q
.
Le
,
“
Ef
f
i
c
i
e
n
t
n
e
t
v
2
:
S
mal
l
e
r
m
o
d
e
l
s
a
n
d
f
a
s
t
e
r
t
r
a
i
n
i
n
g
,
”
i
n
I
n
t
e
r
n
a
t
i
o
n
a
l
c
o
n
f
e
r
e
n
c
e
o
n
m
a
c
h
i
n
e
l
e
a
rn
i
n
g
,
P
M
LR
,
2
0
2
1
,
p
p
.
1
0
0
9
6
–
1
0
1
0
6
.
[
3
1
]
G
.
F
o
r
ma
n
,
“
C
o
u
n
t
i
n
g
p
o
si
t
i
v
e
s
a
c
c
u
r
a
t
e
l
y
d
e
s
p
i
t
e
i
n
a
c
c
u
r
a
t
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
i
n
E
u
ro
p
e
a
n
c
o
n
f
e
r
e
n
c
e
o
n
m
a
c
h
i
n
e
l
e
a
rn
i
n
g
,
S
p
r
i
n
g
e
r
,
2
0
0
5
,
p
p
.
5
6
4
–
5
7
5
.
[
3
2
]
M
.
G
r
a
n
d
i
n
i
,
E.
B
a
g
l
i
,
a
n
d
G
.
V
i
sa
n
i
,
“
M
e
t
r
i
c
s
f
o
r
m
u
l
t
i
-
c
l
a
ss
c
l
a
ss
i
f
i
c
a
t
i
o
n
:
a
n
o
v
e
r
v
i
e
w
,
”
a
rX
i
v
p
r
e
p
ri
n
t
a
r
Xi
v
:
2
0
0
8
.
0
5
7
5
6
,
2
0
2
0
.
[
3
3
]
A
.
Th
a
r
w
a
t
,
“
C
l
a
ssi
f
i
c
a
t
i
o
n
a
ss
e
ssm
e
n
t
m
e
t
h
o
d
s
,
”
Ap
p
l
i
e
d
C
o
m
p
u
t
i
n
g
a
n
d
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
1
7
,
n
o
.
1
,
p
p
.
1
6
8
–
1
9
2
,
J
a
n
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
a
c
i
.
2
0
1
8
.
0
8
.
0
0
3
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Ma
sna
Wa
t
i
h
o
l
d
s
a
Ba
c
h
e
lo
r
o
f
S
c
ie
n
c
e
(S
.
S
i)
i
n
m
a
th
e
m
a
t
ics
sc
ien
c
e
a
n
d
a
m
a
ste
r
o
f
e
n
g
in
e
e
rin
g
(M
.
T
.
)
in
in
f
o
rm
a
ti
c
s
e
n
g
in
e
e
ri
n
g
fr
o
m
Ha
sa
n
u
d
d
in
U
n
iv
e
rsit
y
,
In
d
o
n
e
sia
.
S
h
e
h
a
s
lec
tu
re
d
in
th
e
In
f
o
rm
a
ti
c
s
De
p
a
rtme
n
t
a
t
M
u
law
a
rm
a
n
Un
i
v
e
rsity
,
In
d
o
n
e
sia
sin
c
e
2
0
1
4
.
S
h
e
is
a
m
e
m
b
e
r
o
f
th
e
In
d
o
n
e
sia
S
o
c
iet
y
o
f
En
g
i
n
e
e
rs
a
n
d
th
e
I
n
stit
u
t
e
o
f
El
e
c
tri
c
a
l
a
n
d
El
e
c
tro
n
ics
En
g
i
n
e
e
rs
(IE
EE
).
He
r
re
se
a
rc
h
a
re
a
s
o
f
in
tere
st
i
n
c
lu
d
e
m
a
c
h
in
e
lea
rn
in
g
,
sm
a
rt
sy
ste
m
s,
a
n
d
d
a
ta
sc
ien
c
e
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
m
a
sn
a
wa
ti
@fk
ti
.
u
n
m
u
l.
a
c
.
id
.
No
v
ia
n
ti
P
u
spi
ta
s
a
r
i
r
e
c
e
i
v
e
d
t
h
e
B
.
S
c
.
d
e
g
r
e
e
i
n
i
n
f
o
r
m
a
t
ic
s
e
n
g
i
n
e
e
r
i
n
g
f
r
o
m
t
h
e
U
n
i
v
e
r
s
i
t
a
s
Is
l
a
m
I
n
d
o
n
e
s
i
a
,
a
n
d
t
h
e
M
.
E
n
g
.
d
e
g
r
e
e
i
n
i
n
f
o
r
m
a
t
i
o
n
t
e
c
h
n
o
l
o
g
y
f
r
o
m
t
h
e
G
a
d
j
a
h
M
a
d
a
U
n
i
v
e
r
s
i
t
y
,
I
n
d
o
n
e
s
i
a
.
S
h
e
i
s
c
u
r
r
e
n
t
l
y
a
l
e
c
t
u
r
e
r
a
t
t
h
e
D
e
p
a
r
tm
e
n
t
o
f
I
n
f
o
r
m
a
t
i
c
s
,
M
u
l
a
w
a
r
m
a
n
U
n
i
v
e
r
s
i
t
y
,
I
n
d
o
n
e
s
ia
.
S
h
e
i
s
a
m
e
m
b
e
r
o
f
t
h
e
I
n
s
t
i
t
u
t
e
o
f
E
l
e
c
t
r
ic
a
l
a
n
d
E
l
e
c
t
r
o
n
i
c
s
E
n
g
i
n
e
e
r
s
(
I
E
E
E
)
,
I
n
d
o
n
e
s
i
a
n
C
o
m
p
u
t
e
r
,
E
l
e
c
t
r
o
n
i
c
s
,
I
n
s
t
r
u
m
e
n
t
a
t
i
o
n
S
u
p
p
o
r
t
S
o
c
i
e
t
y
(
I
n
d
o
C
E
I
S
S
)
,
A
s
s
o
c
i
a
t
i
o
n
o
f
C
o
m
p
u
t
i
n
g
a
n
d
I
n
f
o
r
m
a
t
i
c
s
I
n
s
t
i
t
u
t
i
o
n
s
I
n
d
o
n
e
s
i
a
(
A
P
T
I
K
O
M
)
s
o
c
i
e
t
i
e
s
a
n
d
t
h
e
I
n
s
t
i
t
u
t
i
o
n
o
f
E
n
g
i
n
e
e
r
s
I
n
d
o
n
e
s
i
a
(
P
I
I
)
.
S
h
e
h
a
s
a
u
t
h
o
r
e
d
o
r
c
o
-
a
u
t
h
o
r
e
d
m
o
r
e
t
h
a
n
7
0
p
u
b
l
i
c
a
t
i
o
n
s
w
i
t
h
5
H
-
i
n
d
e
x
.
H
e
r
r
e
s
e
a
rc
h
i
n
t
e
r
e
s
t
i
s
i
n
a
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
,
d
a
t
a
s
c
i
e
n
c
e
a
n
d
a
n
a
l
y
t
i
c
s
.
S
h
e
c
a
n
b
e
c
o
n
t
a
c
t
e
d
a
t
e
m
a
i
l
:
n
o
v
i
a
.
f
t
i
k
.
u
n
m
u
l
@
g
m
a
i
l
.
c
o
m
.
Um
m
u
l
H
a
ira
h
is
m
e
m
b
e
r
o
f
In
stit
u
te
o
f
El
e
c
tri
c
a
l
a
n
d
El
e
c
t
ro
n
ics
En
g
in
e
e
rs
(IE
EE
),
a
n
d
m
e
m
b
e
r
o
f
As
so
c
i
a
ti
o
n
o
f
C
o
m
p
u
ti
n
g
a
n
d
I
n
fo
rm
a
ti
c
s
In
stit
u
t
io
n
s
In
d
o
n
e
sia
(APT
IKO
M
)
so
c
ieties
.
C
u
rre
n
tl
y
,
sh
e
is
a
c
ti
v
e
ly
tea
c
h
in
g
a
n
d
re
s
e
a
rc
h
in
g
a
t
th
e
De
p
a
rtme
n
t
o
f
In
f
o
rm
a
ti
c
s,
M
u
law
a
rm
a
n
Un
i
v
e
rsity
.
As
a
writer
o
n
se
v
e
ra
l
j
o
u
rn
a
ls
a
n
d
c
o
n
fe
re
n
c
e
s
with
m
o
re
th
a
n
4
0
p
u
b
li
c
a
ti
o
n
s,
s
h
e
fo
c
u
se
s
h
e
r
re
se
a
rc
h
o
n
d
a
tab
a
s
e
,
in
f
o
rm
a
ti
o
n
sy
ste
m
a
n
d
a
rti
ficia
l
in
telli
g
e
n
c
e
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
u
m
m
u
l
.
h
a
irah
@fk
ti
.
u
n
m
u
l
.
a
c
.
id
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
e
r
20
24
:
6
8
7
4
-
6
8
8
2
6882
J
o
a
n
Ang
e
li
n
a
Wid
ia
n
s
is a d
o
c
to
ra
l
stu
d
e
n
t
i
n
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter S
c
ien
c
e
a
n
d
El
e
c
tro
n
ics
,
G
a
d
jah
M
a
d
a
U
n
iv
e
rsit
y
,
Yo
g
y
a
k
a
rta,
In
d
o
n
e
sia
.
S
h
e
re
c
e
iv
e
d
h
e
r
m
a
ste
r’s
d
e
g
re
e
in
De
p
a
rtme
n
t
o
f
Co
m
p
u
t
e
r
S
c
ien
c
e
,
G
a
d
jah
M
a
d
a
Un
iv
e
r
sity
,
Yo
g
y
a
k
a
rta,
I
n
d
o
n
e
sia
(2
0
0
8
).
S
h
e
re
c
e
iv
e
d
b
a
c
h
e
lo
r’s
d
e
g
re
e
i
n
i
n
fo
rm
a
ti
o
n
sy
ste
m
f
ro
m
S
TM
IK
Wi
d
y
a
Ci
p
ta
Dh
a
rm
a
,
Eas
t
Ka
li
m
a
n
tan
,
I
n
d
o
n
e
sia
(2
0
0
0
)
.
S
h
e
wo
r
k
s
a
s
a
lec
tu
re
r
in
De
p
a
rtme
n
t
o
f
In
fo
rm
a
ti
c
s,
M
u
law
a
rm
a
n
Un
iv
e
r
sity
,
I
n
d
o
n
e
sia
si
n
c
e
2
0
0
8
.
S
h
e
i
s
a
m
e
m
b
e
r
o
f
t
h
e
In
sti
tu
t
e
o
f
El
e
c
tri
c
a
l
a
n
d
El
e
c
tro
n
ics
E
n
g
i
n
e
e
rs
(IE
EE
).
He
r
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
m
a
c
h
in
e
lea
rn
in
g
,
tex
t
m
in
i
n
g
,
a
n
d
s
wa
rm
in
telli
g
e
n
c
e
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
n
g
e
lwid
ian
s@
u
n
m
u
l
.
a
c
.
id
.
Ade
Fi
q
r
i
Tjik
o
a
is
a
n
u
n
d
e
rg
ra
d
u
a
te
stu
d
e
n
t
i
n
In
f
o
rm
a
ti
c
s
a
t
M
u
law
a
rm
a
n
Un
iv
e
rsity
,
In
d
o
n
e
sia
.
Du
rin
g
h
i
s
stu
d
y
,
h
e
a
c
ti
v
e
ly
su
p
p
o
rts
lec
tu
re
rs
in
th
e
ir
re
se
a
rc
h
a
n
d
se
rv
e
d
a
s a
lab
o
ra
to
ry
a
ss
istan
t
fo
r
o
n
e
a
n
d
a
h
a
lf
y
e
a
rs.
His res
e
a
rc
h
in
tere
sts in
c
lu
d
e
m
a
c
h
in
e
lea
rn
in
g
,
d
e
e
p
lea
rn
i
n
g
,
c
o
m
p
u
ter
v
isio
n
,
a
n
d
n
a
tu
ra
l
lan
g
u
a
g
e
p
ro
c
e
ss
in
g
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
d
e
.
fi
q
ri1
@
g
m
a
il
.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.