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.
15
,
No
.
5
,
Octo
b
er
20
25
,
p
p
.
4
7
2
3
~
4
7
3
1
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
5
.
pp
4
7
2
3
-
4
7
3
1
4723
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Co
mpa
ra
tive
a
na
ly
sis
of
co
nv
o
lutiona
l neural ne
twor
k
a
rchitec
t
ure
for
p
o
st
forest
fire
are
a
clas
sifica
tion ba
sed o
n
v
eg
etation
ima
g
e
Ahm
a
d B
inta
ng
Arif
1
,
I
m
a
s
Su
k
a
esih
Sita
ng
g
a
ng
1
,
H
a
ri
Ag
un
g
Adria
nto
1
,
L
a
ila
n Sy
a
ufina
2
1
S
c
h
o
o
l
o
f
D
a
t
a
S
c
i
e
n
c
e
,
M
a
t
h
e
ma
t
i
c
s
a
n
d
I
n
f
o
r
m
a
t
i
c
s,
I
P
B
U
n
i
v
e
r
s
i
t
y
,
B
o
g
o
r
,
I
n
d
o
n
e
s
i
a
2
D
e
p
a
r
t
me
n
t
o
f
S
i
l
v
i
c
u
l
t
u
r
e
,
F
a
c
u
l
t
y
o
f
F
o
r
e
st
r
y
a
n
d
En
v
i
r
o
n
me
n
t
,
I
P
B
U
n
i
v
e
r
si
t
y
,
B
o
g
o
r
,
I
n
d
o
n
e
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Au
g
2
6
,
2
0
2
4
R
ev
is
ed
Ap
r
1
7
,
2
0
2
5
Acc
ep
ted
J
u
l 3
,
2
0
2
5
Th
is
stu
d
y
p
re
se
n
ts
a
c
o
m
p
a
ra
ti
v
e
a
n
a
ly
sis
o
f
7
C
o
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
(CNN
)
a
rc
h
it
e
c
tu
re
s
—
M
o
b
i
leN
e
tV2
,
VG
G
1
6
,
VG
G
1
9
,
LeN
e
t5
,
Ale
x
Ne
t,
Re
sN
e
t5
0
,
a
n
d
In
c
e
p
ti
o
n
V3
—
fo
r
c
las
sify
in
g
p
o
st
-
f
o
re
st
fire
a
re
a
s
u
sin
g
field
-
b
a
se
d
v
e
g
e
tati
o
n
ima
g
e
ry
.
A
t
o
tal
o
f
5
6
m
o
d
e
ls
we
re
e
v
a
lu
a
ted
th
ro
u
g
h
c
o
m
b
i
n
a
ti
o
n
s
o
f
b
a
tch
s
ize
,
in
p
u
t
siz
e
,
a
n
d
o
p
t
imiz
e
r.
Th
e
re
su
lt
s
sh
o
w
th
a
t
M
o
b
il
e
Ne
tV2
,
VG
G
1
6
,
a
n
d
VG
G
1
9
o
u
tp
e
rf
o
rm
e
d
o
th
e
r
m
o
d
e
ls,
with
v
a
li
d
a
ti
o
n
a
c
c
u
ra
c
ies
e
x
c
e
e
d
in
g
8
8
%
.
M
o
b
il
e
Ne
tV2
e
m
e
rg
e
d
a
s
th
e
m
o
st
b
a
lan
c
e
d
m
o
d
e
l,
a
c
h
ie
v
in
g
9
6
%
a
c
c
u
ra
c
y
with
th
e
l
o
we
st
m
o
d
e
l
siz
e
a
n
d
train
i
n
g
ti
m
e
,
m
a
k
i
n
g
it
i
d
e
a
l
fo
r
re
so
u
rc
e
-
c
o
n
stra
in
e
d
a
p
p
li
c
a
ti
o
n
s.
T
h
is
stu
d
y
h
i
g
h
l
ig
h
ts
th
e
p
o
te
n
ti
a
l
o
f
CNN
-
b
a
se
d
c
las
sifica
ti
o
n
u
sin
g
m
o
b
il
e
field
ima
g
e
ry
,
o
ffe
rin
g
a
n
e
fficie
n
t
a
lt
e
rn
a
ti
v
e
t
o
c
o
stly
a
n
d
c
o
n
d
it
i
o
n
-
d
e
p
e
n
d
e
n
t
sa
telli
te
o
r
d
r
o
n
e
d
a
ta.
Th
e
fin
d
in
g
s
su
p
p
o
r
t
re
a
l
-
ti
m
e
,
lo
c
a
li
z
e
d
id
e
n
ti
fica
ti
o
n
o
f
b
u
rn
e
d
a
re
a
s
a
fter
fo
re
st
fires
,
p
ro
v
id
i
n
g
a
c
ti
o
n
a
b
le
in
sig
h
ts
fo
r
p
ri
o
rit
izin
g
re
c
o
v
e
r
y
a
re
a
s
a
n
d
g
u
i
d
in
g
e
c
o
lo
g
ica
l
re
sto
ra
ti
o
n
a
n
d
lan
d
re
h
a
b
il
it
a
ti
o
n
stra
teg
ies
.
K
ey
w
o
r
d
s
:
Ar
ch
itectu
r
e
co
m
p
ar
is
o
n
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Field
im
ag
er
y
Fo
r
est an
d
lan
d
f
ir
e
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
I
m
as Su
k
ae
s
ih
Sit
an
g
g
an
g
Sch
o
o
l o
f
Data
Scien
ce
,
Ma
t
h
em
atics a
n
d
I
n
f
o
r
m
atics
,
I
PB
Un
iv
er
s
ity
Me
r
an
ti
Stre
et
,
W
in
g
2
0
L
e
v
el
5
,
I
PB
C
am
p
u
s
Dar
m
ag
a
B
o
g
o
r
1
6
6
8
0
,
I
n
d
o
n
esia
E
m
ail:
im
as.si
tan
g
g
an
g
@
ap
p
s
.
ip
b
.
ac
.
id
1.
I
NT
RO
D
UCT
I
O
N
I
n
d
o
n
esia
h
as
a
lo
n
g
-
s
tan
d
in
g
h
is
to
r
y
o
f
r
ec
u
r
r
in
g
f
o
r
est
f
ir
es,
wh
ich
h
av
e
b
ec
o
m
e
an
an
n
u
a
l
o
cc
u
r
r
e
n
ce
,
s
ev
er
ely
af
f
ec
tin
g
th
e
en
v
ir
o
n
m
en
t
an
d
lo
ca
l
c
o
m
m
u
n
ities
.
Fo
r
est
f
ir
es
h
av
e
b
ee
n
a
s
ig
n
if
ican
t
is
s
u
e
as
ea
r
ly
as
th
e
1
9
8
0
s
[
1
]
.
T
h
ese
f
ir
es
h
av
e
c
o
n
tr
ib
u
te
d
to
air
p
o
llu
tio
n
,
h
ea
lth
is
s
u
es,
b
io
d
iv
er
s
ity
lo
s
s
,
an
d
th
e
d
e
g
r
ad
atio
n
o
f
ec
o
s
y
s
tem
s
[
2
]
.
An
aly
zin
g
p
o
s
t
-
f
ir
e
ar
ea
s
is
th
er
ef
o
r
e
cr
u
c
ial
f
o
r
m
itig
atio
n
,
r
esto
r
atio
n
,
an
d
s
u
s
tain
ab
le
lan
d
m
an
ag
em
e
n
t.
R
ec
en
t
s
tu
d
ies
h
av
e
in
tr
o
d
u
ce
d
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs)
as
a
p
r
o
m
is
in
g
s
o
lu
tio
n
f
o
r
wild
f
ir
e
d
am
ag
e
class
if
icatio
n
d
u
e
to
th
eir
p
o
wer
f
u
l
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
im
ag
e
r
ec
o
g
n
itio
n
ca
p
a
b
ilit
ies
[
3
]
.
Mo
s
t
ex
is
tin
g
wo
r
k
s
u
tili
ze
s
a
tellite
an
d
d
r
o
n
e
im
ag
e
r
y
t
o
cl
ass
if
y
p
o
s
t
-
f
o
r
est f
ir
e
ar
ea
s
u
s
in
g
C
NNs.
Fo
r
ex
am
p
le,
[
4
]
an
d
[
5
]
u
s
ed
d
r
o
n
e
im
ag
es
with
C
NN
-
b
ased
m
o
d
els
s
u
ch
as
VGG
-
1
6
an
d
FF
ir
eNe
t,
ac
h
iev
in
g
h
i
g
h
ac
cu
r
ac
y
lev
el
s
o
f
9
6
% a
n
d
9
8
.
4
2
%,
r
esp
ec
ti
v
ely
.
Oth
er
r
esear
c
h
er
u
tili
ze
d
Sen
tin
el
-
2
s
atellite
im
ag
es
to
class
if
y
im
ag
es
in
t
o
f
iv
e
ca
te
g
o
r
ies
(
‘
f
ield
,
’
‘
f
o
r
est,
’
‘
s
m
o
k
e
,
’
‘
u
r
b
a
n
,
’
an
d
‘
b
u
r
n
ed
’
)
u
s
in
g
R
esNet
an
d
Xce
p
tio
n
m
o
d
els,
with
Xce
p
tio
n
ac
h
iev
in
g
t
h
e
h
ig
h
e
s
t
ac
cu
r
ac
y
o
f
9
6
.
7
%
[
6
]
.
De
s
p
ite
th
eir
s
u
cc
ess
,
th
ese
ap
p
r
o
ac
h
es
s
till
d
ep
en
d
o
n
ex
p
en
s
iv
e
im
a
g
e
ac
q
u
is
itio
n
,
v
a
r
y
in
g
im
ag
e
r
eso
lu
tio
n
s
,
an
d
ar
e
lim
ited
b
y
s
p
atial
o
r
wea
th
er
-
r
elate
d
c
o
n
s
tr
ain
ts
[
7
]
.
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
2
3
-
4
7
3
1
4724
T
o
o
v
er
c
o
m
e
th
ese
ch
allen
g
es,
f
ield
-
b
ased
o
r
g
r
o
u
n
d
-
lev
el
im
ag
er
y
h
as
em
er
g
e
d
a
s
a
v
iab
le
alter
n
ativ
e.
Field
im
ag
es
ca
p
tu
r
ed
u
s
in
g
m
o
b
ile
p
h
o
n
es
p
r
o
v
id
e
d
etailed
v
eg
etatio
n
a
n
d
s
o
il
in
f
o
r
m
atio
n
a
n
d
ar
e
less
af
f
ec
ted
b
y
clo
u
d
co
v
er
o
r
atm
o
s
p
h
e
r
ic
d
is
to
r
tio
n
s
.
A
s
tu
d
y
b
y
[
8
]
ap
p
lied
M
o
b
ileNetV2
o
n
f
ield
im
ag
er
y
f
r
o
m
J
am
b
i
Pro
v
in
c
e,
ac
h
iev
in
g
7
7
.
7
%
ac
cu
r
ac
y
.
Ho
wev
er
,
r
esear
ch
in
th
is
d
ir
ec
tio
n
r
em
ain
s
lim
ited
,
p
ar
ticu
lar
ly
in
o
p
tim
iz
in
g
C
NN
ar
ch
itectu
r
es a
n
d
h
y
p
er
p
ar
am
eter
co
m
b
i
n
atio
n
s
f
o
r
s
u
ch
d
atasets
.
T
h
is
s
tu
d
y
ad
d
r
ess
es
th
is
g
a
p
b
y
s
y
s
tem
atica
lly
ev
al
u
atin
g
7
C
NN
ar
ch
itectu
r
es
(
Mo
b
ileNetV2
,
VGG1
6
,
VGG1
9
,
L
eNe
t5
,
Al
ex
Net,
R
esNet5
0
,
an
d
I
n
ce
p
ti
o
n
V3
)
o
n
a
d
ataset
o
f
p
o
s
t
-
f
ir
e
v
eg
etatio
n
im
ag
es
co
llected
in
th
e
f
ield
.
T
h
is
s
tu
d
y
ap
p
lied
h
y
p
er
p
a
r
am
ete
r
tu
n
in
g
in
v
o
lv
in
g
b
atch
s
ize,
in
p
u
t
s
ize,
an
d
o
p
tim
izer
to
d
eter
m
in
e
t
h
eir
in
f
lu
e
n
ce
o
n
m
o
d
el
p
er
f
o
r
m
an
ce
.
T
h
e
g
o
al
o
f
th
is
s
tu
d
y
is
to
id
e
n
tify
a
n
ar
ch
itectu
r
e
th
at
b
alan
ce
s
h
ig
h
class
if
icat
io
n
ac
cu
r
ac
y
with
co
m
p
u
tatio
n
al
ef
f
icien
cy
—
esp
ec
ially
r
elev
an
t
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
s
in
r
em
o
te
o
r
r
eso
u
r
ce
-
lim
ited
en
v
ir
o
n
m
en
ts
.
T
h
e
k
ey
co
n
tr
ib
u
tio
n
s
o
f
th
is
wo
r
k
in
clu
d
e
a
co
m
p
a
r
ativ
e
e
v
alu
atio
n
o
f
C
NN
ar
ch
itectu
r
es
s
p
ec
if
ically
tailo
r
ed
f
o
r
p
o
s
t
-
f
o
r
est
f
ir
e
cla
s
s
if
icatio
n
in
f
ield
co
n
d
itio
n
s
,
an
an
aly
s
is
o
f
h
o
w
b
atch
s
ize,
in
p
u
t
s
ize,
an
d
o
p
tim
izer
af
f
ec
t
p
er
f
o
r
m
an
ce
,
an
d
v
alu
ab
le
in
s
ig
h
ts
in
to
th
e
v
ia
b
ilit
y
o
f
u
s
in
g
m
o
b
ile
-
ca
p
tu
r
ed
f
ield
im
ag
es
as
a
co
s
t
-
ef
f
ec
tiv
e
a
n
d
p
r
a
ctica
l
alter
n
ativ
e
to
tr
ad
itio
n
al
r
em
o
te
s
en
s
in
g
m
et
h
o
d
s
.
2.
M
E
T
H
O
D
T
h
e
m
eth
o
d
o
l
o
g
y
o
f
th
is
r
ese
ar
ch
co
m
p
r
is
es
f
iv
e
m
ain
s
tag
es:
d
ata
co
llectio
n
,
p
r
ep
r
o
ce
s
s
in
g
,
d
ata
s
p
litt
in
g
,
m
o
d
elin
g
,
a
n
d
e
v
alu
atio
n
,
as
illu
s
tr
ated
in
Fig
u
r
e
1
.
T
h
ese
s
tag
es
a
r
e
ad
a
p
ted
f
r
o
m
wid
ely
ac
ce
p
ted
im
ag
e
class
if
icatio
n
wo
r
k
f
lo
w
s
[
9
]
,
an
d
ea
ch
is
ca
r
ef
u
lly
d
e
s
ig
n
ed
to
en
s
u
r
e
h
ig
h
-
q
u
ality
in
p
u
t
d
ata,
o
p
tim
al
m
o
d
el
p
e
r
f
o
r
m
an
ce
,
an
d
r
ig
o
r
o
u
s
ev
alu
atio
n
.
T
h
e
o
v
e
r
all
ap
p
r
o
ac
h
in
teg
r
ates
f
ield
-
co
llected
im
ag
e
d
ata
with
d
ee
p
lear
n
in
g
-
b
ased
class
if
icatio
n
to
ass
ess
p
o
s
t
-
f
ir
e
lan
d
co
n
d
itio
n
s
.
Fig
u
r
e
1
.
R
esear
ch
s
tag
es
2
.
1
.
Da
t
a
c
o
llect
io
n
T
h
is
s
tu
d
y
u
tili
ze
d
f
ield
im
ag
es
ca
p
tu
r
ed
u
s
in
g
m
o
b
ile
p
h
o
n
es
f
r
o
m
4
p
o
s
t
-
f
ir
e
lo
ca
tio
n
s
in
J
am
b
i
Pro
v
in
ce
,
n
am
el
y
Pem
atan
g
R
ah
im
,
Pem
atan
g
L
u
m
u
t,
Pelay
an
g
an
,
an
d
T
en
a
m
.
Field
im
ag
er
y
was
ch
o
s
en
d
u
e
to
its
h
ig
h
r
eso
lu
tio
n
,
c
lear
er
v
is
u
al
d
etails,
im
m
u
n
i
ty
to
atm
o
s
p
h
er
ic
d
is
tu
r
b
an
c
es,
an
d
lo
we
r
co
s
t
co
m
p
ar
ed
to
s
atellite
o
r
d
r
o
n
e
im
ag
er
y
.
T
h
e
co
llected
im
ag
es
wer
e
in
itially
ca
teg
o
r
ized
in
to
th
r
ee
f
ac
to
r
s
:
ar
ea
,
s
o
il,
an
d
v
eg
etatio
n
.
Ho
wev
er
,
o
n
ly
v
eg
etatio
n
-
r
elate
d
im
ag
es
wer
e
u
s
ed
f
o
r
th
e
class
if
icatio
n
m
o
d
el,
as
v
eg
etatio
n
p
lay
s
a
s
ig
n
if
ican
t
r
o
le
in
ass
ess
in
g
p
o
s
t
-
f
ir
e
ar
ea
s
[
1
0
]
.
I
n
to
tal,
2
3
9
im
ag
es
wer
e
u
s
ed
,
d
iv
id
e
d
in
to
two
class
es:
b
u
r
n
e
d
a
r
ea
(
1
3
9
im
ag
es)
a
n
d
u
n
b
u
r
n
e
d
a
r
ea
(
1
0
0
im
a
g
es).
Sam
p
le
im
a
g
es
f
r
o
m
ea
ch
class
ar
e
p
r
esen
ted
in
Fig
u
r
es 2
a
n
d
3
.
Fig
u
r
e
2
.
B
u
r
n
ed
ar
ea
im
ag
e
Fig
u
r
e
3
.
Un
b
u
r
n
ed
ar
ea
im
ag
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
C
o
mp
a
r
a
tive
a
n
a
lysi
s
o
f c
o
n
vo
lu
tio
n
a
l
n
eu
r
a
l n
etw
o
r
k
a
r
ch
itectu
r
e
fo
r
…
(
A
h
ma
d
B
i
n
ta
n
g
A
r
if
)
4725
2
.
2
.
P
re
pro
ce
s
s
ing
d
a
t
a
Pre
p
r
o
ce
s
s
in
g
is
a
cr
itical
s
te
p
in
d
ee
p
lea
r
n
in
g
p
ip
elin
es
t
o
im
p
r
o
v
e
m
o
d
el
lear
n
in
g
ef
f
i
cien
cy
an
d
o
u
tp
u
t
q
u
ality
[
1
1
]
.
I
n
th
is
s
tu
d
y
,
th
r
ee
k
e
y
p
r
ep
r
o
ce
s
s
in
g
o
p
e
r
atio
n
s
wer
e
ap
p
lied
.
First,
r
esizin
g
was
p
er
f
o
r
m
ed
b
y
s
tan
d
ar
d
izin
g
a
ll
im
ag
es
to
d
im
e
n
s
io
n
s
o
f
ei
th
er
1
9
2
×1
9
2
o
r
2
2
4
×
2
2
4
p
i
x
els.
T
h
is
n
o
t
o
n
ly
r
ed
u
ce
d
co
m
p
u
tatio
n
al
c
o
s
t
a
n
d
m
em
o
r
y
u
s
ag
e
b
u
t
also
en
s
u
r
ed
co
n
s
is
ten
t
in
p
u
t
d
i
m
en
s
io
n
s
ac
r
o
s
s
m
o
d
els.
W
h
ile
r
esizin
g
ca
n
in
tr
o
d
u
c
e
in
f
o
r
m
atio
n
lo
s
s
,
th
is
was
m
itig
ated
b
y
s
elec
tin
g
r
ela
tiv
ely
h
ig
h
tar
g
et
r
eso
lu
tio
n
s
an
d
p
r
eser
v
in
g
asp
ec
t
r
atio
s
wh
er
e
p
o
s
s
ib
le
[
1
2
]
.
Seco
n
d
,
n
o
r
m
aliza
tio
n
wa
s
ap
p
lied
b
y
s
ca
lin
g
p
ix
el
v
alu
es
to
th
e
[
0
,
1
]
r
a
n
g
e,
wh
ic
h
f
ac
ilit
ated
f
aster
an
d
m
o
r
e
s
tab
le
co
n
v
er
g
e
n
ce
d
u
r
i
n
g
tr
ai
n
in
g
,
p
ar
ticu
lar
ly
wh
e
n
u
s
in
g
g
r
ad
i
en
t
-
b
ased
o
p
tim
izer
s
[
1
3
]
.
T
h
i
r
d
,
d
ata
au
g
m
en
tatio
n
tech
n
iq
u
es
s
u
ch
as
r
a
n
d
o
m
r
o
tatio
n
,
f
lip
p
in
g
,
an
d
z
o
o
m
in
g
wer
e
em
p
lo
y
ed
to
in
cr
ea
s
e
d
ata
v
ar
iab
ilit
y
,
r
e
d
u
ce
o
v
er
f
i
ttin
g
,
an
d
im
p
r
o
v
e
m
o
d
el
g
e
n
er
aliza
tio
n
[
1
4
]
.
2
.
3
.
Da
t
a
pa
rt
i
t
io
n
B
ef
o
r
e
tr
ain
in
g
,
th
e
d
ataset
was
d
iv
id
ed
in
to
th
r
ee
s
u
b
s
ets:
tr
ain
in
g
,
v
alid
atio
n
,
an
d
te
s
t
s
et
s
.
T
h
e
tr
ain
in
g
s
et
was
u
s
ed
to
f
it
th
e
m
o
d
el,
t
h
e
v
alid
atio
n
s
et
was
u
s
ed
to
tu
n
e
th
e
m
o
d
el
an
d
p
r
ev
e
n
t
o
v
e
r
f
itti
n
g
d
u
r
in
g
tr
ain
in
g
,
an
d
th
e
test
s
et
was
r
eser
v
ed
f
o
r
f
in
al
p
e
r
f
o
r
m
a
n
ce
ev
alu
atio
n
o
n
u
n
s
ee
n
d
ata
[
1
5
]
.
T
h
is
p
ar
titi
o
n
in
g
e
n
s
u
r
es
th
at
th
e
m
o
d
el
h
as
s
u
f
f
icien
t
d
ata
to
lear
n
ef
f
ec
tiv
ely
wh
ile
also
b
ei
n
g
p
r
o
p
er
l
y
v
alid
ated
an
d
test
ed
.
T
h
e
d
ataset
was
s
p
lit
with
th
e
f
o
llo
win
g
p
r
o
p
o
r
ti
o
n
s
:
8
0
%
f
o
r
tr
ain
in
g
(
1
1
1
b
u
r
n
ed
im
ag
es
an
d
8
0
u
n
b
u
r
n
ed
im
ag
es),
1
0
%
f
o
r
v
alid
atio
n
(
1
4
b
u
r
n
ed
an
d
1
0
u
n
b
u
r
n
ed
)
,
an
d
1
0
%
f
o
r
test
in
g
(
1
4
b
u
r
n
e
d
a
n
d
1
0
u
n
b
u
r
n
ed
)
.
2
.
4
.
M
o
del dev
elo
pm
ent
T
h
is
s
tu
d
y
e
x
p
lo
r
e
d
two
ap
p
r
o
ac
h
es
to
C
NN
m
o
d
el
d
ev
elo
p
m
en
t:
tr
a
n
s
f
er
lear
n
in
g
u
s
in
g
p
r
etr
ain
e
d
m
o
d
els
an
d
tr
ain
in
g
f
r
o
m
s
cr
atch
.
T
h
e
p
r
etr
ai
n
ed
m
o
d
e
ls
in
clu
d
ed
Mo
b
ileNetV2
,
VGG
-
1
6
,
VGG
-
19,
R
esNet
-
50
,
an
d
I
n
ce
p
tio
n
,
all
o
f
wh
ich
wer
e
p
r
etr
ai
n
ed
o
n
t
h
e
I
m
a
g
eNe
t
d
ataset.
L
e
v
er
ag
i
n
g
tr
a
n
s
f
er
lear
n
in
g
allo
ws
f
o
r
im
p
r
o
v
ed
p
er
f
o
r
m
a
n
ce
o
n
s
m
all
d
atasets
an
d
f
ast
er
co
n
v
er
g
e
n
ce
d
u
e
t
o
th
e
r
eu
s
e
o
f
lear
n
ed
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
[
1
6
]
.
I
n
p
a
r
allel,
two
C
NN
m
o
d
els
—
L
eNe
t
-
5
an
d
Alex
Net
—
wer
e
tr
ain
e
d
f
r
o
m
s
cr
atch
.
T
h
ese
ar
ch
itectu
r
es
s
er
v
e
as
b
aselin
e
m
o
d
els
an
d
en
ab
le
co
m
p
a
r
is
o
n
o
f
s
h
allo
w
v
er
s
u
s
d
ee
p
f
ea
tu
r
e
ex
tr
ac
to
r
s
,
p
ar
ticu
lar
ly
in
t
h
e
co
n
te
x
t o
f
f
o
r
est f
ir
e
class
if
icatio
n
,
wh
ich
lack
s
d
ed
icate
d
p
r
etr
ain
ed
m
o
d
els.
T
o
o
p
tim
ize
m
o
d
el
p
er
f
o
r
m
an
ce
,
h
y
p
er
p
ar
am
eter
tu
n
i
n
g
was
co
n
d
u
cted
o
n
th
r
e
e
p
r
im
ar
y
p
ar
am
eter
s
:
b
atch
s
ize
(
1
6
an
d
3
2
)
,
in
p
u
t
s
ize
(
1
9
2
×1
9
2
a
n
d
2
2
4
×2
2
4
)
,
an
d
o
p
tim
izer
(
Ad
a
m
an
d
R
MSp
r
o
p
)
.
B
atch
s
ize
d
eter
m
in
es
h
o
w
m
an
y
s
am
p
les
ar
e
p
r
o
ce
s
s
ed
at
ea
ch
tr
ain
in
g
s
tep
[
1
7
]
,
wh
ile
in
p
u
t
s
ize
d
ef
in
es
th
e
im
ag
e
d
im
en
s
io
n
s
p
r
o
v
i
d
ed
to
th
e
C
NN
m
o
d
el.
O
p
tim
izer
s
,
wh
ich
h
elp
f
in
d
th
e
o
p
tim
al
m
o
d
el
p
ar
am
eter
s
b
y
m
i
n
im
izin
g
t
h
e
lo
s
s
f
u
n
ctio
n
th
r
o
u
g
h
g
r
ad
ien
t
co
m
p
u
tatio
n
,
s
ig
n
if
ican
tly
im
p
ac
t
tr
ain
in
g
s
p
ee
d
an
d
co
n
v
e
r
g
en
ce
[
1
8
]
.
E
ac
h
o
p
tim
izer
also
af
f
ec
ts
th
e
lear
n
in
g
s
p
ee
d
an
d
co
n
v
er
g
en
ce
o
f
a
m
o
d
el
[
1
9
]
.
T
h
ese
h
y
p
er
p
ar
am
eter
co
m
b
i
n
atio
n
s
r
esu
lted
in
8
u
n
iq
u
e
s
ch
em
e
as
s
h
o
wn
in
T
ab
le
1
.
Hy
p
er
p
ar
am
eter
tu
n
in
g
is
cr
itical
to
f
in
d
in
g
t
h
e
m
o
s
t
ef
f
ec
tiv
e
co
n
f
ig
u
r
atio
n
f
o
r
ac
h
iev
in
g
ac
cu
r
ate
an
d
e
f
f
icien
t
cla
s
s
if
icatio
n
o
f
p
o
s
t
-
f
ir
e
ar
ea
s
[
2
0
]
.
All
m
o
d
els
wer
e
t
r
ain
ed
f
o
r
5
0
e
p
o
ch
s
a
n
d
o
th
er
tr
ain
in
g
p
ar
am
eter
s
,
s
u
ch
as
lear
n
in
g
r
ate
a
n
d
d
r
o
p
o
u
t
r
ate,
wer
e
k
ep
t
at
th
e
ir
d
ef
au
lt
v
alu
es
to
is
o
late
th
e
im
p
ac
t
o
f
th
e
s
elec
ted
h
y
p
er
p
ar
am
eter
s
.
Mo
d
el
tr
ain
in
g
an
d
ev
al
u
atio
n
wer
e
co
n
d
u
cte
d
o
n
a
lo
ca
l
m
ac
h
in
e
eq
u
ip
p
e
d
with
a
n
AM
D
R
y
ze
n
5
5
6
0
0
X
C
PU,
6
4
GB
R
AM
,
an
d
a
5
1
2
GB
SS
D.
T
h
e
s
o
f
twar
e
en
v
ir
o
n
m
en
t
in
clu
d
e
d
Py
th
o
n
3
.
8
a
lo
n
g
with
ess
en
tia
l
lib
r
ar
ies s
u
ch
as T
en
s
o
r
Flo
w
2
.
1
3
,
Scik
it
-
lear
n
,
Pil
lo
w,
Nu
m
Py
,
Seab
o
r
n
,
Pan
d
as,
an
d
Ma
tp
lo
tlib
.
T
ab
le
1
.
Hy
p
er
p
ar
a
m
eter
s
ch
e
m
e
in
b
u
ild
i
n
g
m
o
d
el
S
c
h
e
me
B
a
t
c
h
si
z
e
I
n
p
u
t
si
z
e
O
p
t
i
mi
z
e
r
S
c
h
e
me
B
a
t
c
h
si
z
e
I
n
p
u
t
si
z
e
O
p
t
i
mi
z
e
r
1
32
1
9
2
×
1
9
2
A
d
a
m
4
32
1
9
2
×
1
9
2
A
d
a
m
2
32
2
2
4
×
2
2
4
A
d
a
m
6
16
2
2
4
×
2
2
4
R
M
S
P
r
o
p
3
16
1
9
2
×
1
9
2
A
d
a
m
7
32
1
9
2
×
1
9
2
R
M
S
P
r
o
p
4
16
2
2
4
×
2
2
4
A
d
a
m
8
32
2
2
4
×
2
2
4
R
M
S
P
r
o
p
2
.
5
.
E
v
a
lua
t
i
o
n a
nd
co
m
pa
r
is
o
n
T
h
e
p
er
f
o
r
m
an
ce
o
f
ea
ch
m
o
d
el
was
ev
alu
ated
th
r
o
u
g
h
b
o
th
q
u
an
titativ
e
m
etr
ics,
tr
ain
in
g
m
etr
ics,
an
d
v
is
u
al
ass
ess
m
en
t
tech
n
iq
u
es.
Qu
an
titativ
e
m
etr
ics
in
v
o
lv
ed
th
e
ca
lc
u
latio
n
o
f
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
b
ased
o
n
th
e
co
n
f
u
s
io
n
m
atr
ix
.
T
h
ese
m
etr
ics
p
r
o
v
id
e
a
co
m
p
r
e
h
e
n
s
iv
e
ass
es
s
m
en
t
o
f
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
,
e
s
p
ec
ially
in
s
ce
n
ar
io
s
in
v
o
l
v
in
g
class
im
b
alan
ce
[
2
1
]
,
[
2
2
]
.
Acc
u
r
ac
y
,
in
p
ar
ticu
lar
,
was c
o
m
p
u
ted
u
s
in
g
in
(
1
)
:
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
2
3
-
4
7
3
1
4726
=
+
+
+
+
(
1
)
w
h
er
e
T
P,
T
N,
FP
,
an
d
FN
r
ep
r
esen
t
tr
u
e
p
o
s
itiv
es,
tr
u
e
n
eg
ativ
es,
f
alse
p
o
s
itiv
es,
a
n
d
f
alse
n
eg
ativ
es,
r
esp
ec
tiv
ely
.
T
o
ass
ess
tr
ain
in
g
d
y
n
am
ics
a
n
d
m
o
d
el
g
en
e
r
aliza
tio
n
,
tr
ain
in
g
m
etr
ics
s
u
ch
as
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
,
tr
ain
in
g
tim
e
(
in
s
ec
o
n
d
s
)
,
an
d
m
o
d
el
s
ize
(
in
m
eg
ab
y
tes)
wer
e
als
o
r
ec
o
r
d
e
d
.
Vis
u
al
an
aly
s
is
was c
o
n
d
u
cted
u
s
in
g
co
n
f
u
s
io
n
m
atr
ices to
o
b
s
er
v
e
m
is
class
if
icatio
n
tr
en
d
s
,
alo
n
g
with
ac
cu
r
ac
y
an
d
lo
s
s
p
lo
ts
th
r
o
u
g
h
o
u
t
tr
ain
in
g
ep
o
ch
s
to
e
v
alu
ate
lear
n
i
n
g
p
r
o
g
r
ess
an
d
d
etec
t
p
o
te
n
tial
o
v
er
f
itti
n
g
o
r
u
n
d
er
f
itti
n
g
.
T
h
e
o
v
er
all
ev
alu
atio
n
aim
ed
to
id
en
tif
y
n
o
t
o
n
ly
th
e
b
est
-
p
er
f
o
r
m
in
g
C
NN
ar
ch
itectu
r
e
b
u
t
also
th
e
o
p
tim
al
c
o
m
b
in
atio
n
o
f
h
y
p
er
p
a
r
am
eter
s
b
y
co
n
s
id
er
in
g
b
o
t
h
p
r
ed
ictiv
e
p
er
f
o
r
m
an
c
e
an
d
co
m
p
u
tatio
n
al
ef
f
icien
cy
.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
M
o
del
t
ra
ini
ng
re
s
ults
B
ased
o
n
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
a
to
tal
o
f
5
6
C
NN
m
o
d
el
s
wer
e
g
en
e
r
ated
f
r
o
m
8
ex
p
er
im
en
ts
co
n
d
u
cte
d
ac
r
o
s
s
7
d
if
f
e
r
en
t
a
r
ch
itectu
r
es.
T
h
e
p
er
f
o
r
m
a
n
ce
s
u
m
m
ar
y
o
f
ea
ch
ar
c
h
itectu
r
e
was
p
r
esen
ted
in
T
ab
le
2
,
wh
ich
in
clu
d
es
m
etr
i
cs
s
u
ch
as
tr
ain
in
g
ac
cu
r
ac
y
,
v
alid
atio
n
ac
cu
r
ac
y
,
tr
ain
in
g
ti
m
e,
m
o
d
el
s
ize,
a
n
d
th
e
co
r
r
esp
o
n
d
in
g
h
y
p
e
r
p
ar
am
eter
co
m
b
in
atio
n
s
.
E
ac
h
ar
ch
it
ec
tu
r
e
ex
h
ib
ited
v
ar
y
in
g
p
er
f
o
r
m
an
ce
d
ep
e
n
d
in
g
o
n
th
e
h
y
p
er
p
ar
am
eter
c
o
m
b
in
atio
n
s
ap
p
lied
.
T
ab
le
2
.
Su
m
m
a
r
y
o
f
m
o
d
el
p
er
f
o
r
m
a
n
ce
A
r
c
h
i
t
e
c
t
u
r
e
Tr
a
i
n
a
c
c
V
a
l
a
c
c
T
r
a
i
n
l
o
ss
V
a
l
l
o
ss
Tr
a
i
n
T
i
me
(
sec
o
n
d
)
M
o
d
e
l
s
i
z
e
(
m
b
)
S
c
h
e
me
M
o
b
i
l
e
N
e
t
V
2
0
.
9
6
0
.
9
6
0
.
1
3
0
.
1
6
5
6
3
9
6
V
G
G
1
6
0
.
9
8
0
.
8
8
0
.
0
7
0
.
3
4
5
6
9
1
2
9
.
7
2
V
G
G
1
9
0
.
9
7
0
.
9
6
0
.
0
6
0
.
1
7
5
6
7
1
5
0
.
0
4
Le
N
e
t
5
0
.
9
4
0
.
8
8
0
.
1
8
0
.
8
2
5
5
0
2
9
.
8
5
A
l
e
x
N
e
t
0
.
5
8
0
.
5
8
0
.
6
8
0
.
6
8
5
4
6
2
9
.
8
7
R
e
sN
e
t
5
0
0
.
6
6
0
.
6
3
0
.
6
5
0
.
6
8
5
6
0
9
0
.
3
6
I
n
c
e
p
t
i
o
n
V
3
0
.
9
4
0
.
9
2
0
.
2
3
0
.
3
2
5
5
2
8
3
.
7
7
Fro
m
T
ab
le
2
,
it
ca
n
b
e
o
b
s
e
r
v
ed
th
at
VGG1
6
,
VGG1
9
,
a
n
d
M
o
b
ileNetV2
d
e
m
o
n
s
tr
ate
d
th
e
b
est
p
er
f
o
r
m
an
ce
am
o
n
g
th
e
s
ev
e
n
C
NN
ar
ch
itectu
r
es
e
v
alu
ate
d
.
Mo
b
ileNetV2
ac
h
iev
e
d
th
e
h
ig
h
est
v
alid
atio
n
ac
cu
r
ac
y
o
f
9
6
%,
with
r
elativ
e
ly
f
ast tr
ain
in
g
tim
e
(
5
6
3
s
ec
o
n
d
s
)
an
d
th
e
s
m
allest m
o
d
el
s
ize
(
9
MB),
m
ak
in
g
it
h
ig
h
ly
ef
f
icien
t
f
o
r
d
ep
lo
y
m
en
t
o
n
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
d
ev
ices.
Me
an
wh
ile,
VGG1
6
r
ec
o
r
d
ed
th
e
h
ig
h
est
tr
ain
in
g
ac
cu
r
ac
y
at
9
8
%,
alt
h
o
u
g
h
its
v
alid
atio
n
ac
cu
r
ac
y
(
8
8
%)
was
lo
wer
th
an
th
at
o
f
Mo
b
ileNetV2
an
d
VGG1
9
.
VGG1
9
also
p
er
f
o
r
m
ed
well,
with
a
v
alid
atio
n
a
cc
u
r
ac
y
o
f
9
6
%
an
d
a
lo
w
v
alid
atio
n
lo
s
s
v
alu
e
(
0
.
1
7
)
,
d
esp
ite
its
r
elativ
ely
lar
g
e
m
o
d
el
s
ize
(
1
5
0
MB).
I
n
co
n
tr
ast,
ar
ch
itectu
r
es
s
u
ch
as
Alex
Net
an
d
R
esNet5
0
s
h
o
wed
p
o
o
r
p
er
f
o
r
m
an
ce
,
with
lo
w
v
alid
atio
n
ac
cu
r
ac
ies
o
f
5
8
%
an
d
6
3
%.
T
h
ese
r
esu
lts
s
u
g
g
est
th
at
d
esp
ite
th
ei
r
ar
c
h
itectu
r
al
d
ep
th
,
th
ese
m
o
d
els
(
Alex
Net
an
d
R
esNet5
0
)
m
a
y
b
e
less
e
f
f
ec
tiv
e
in
lear
n
i
n
g
f
r
o
m
th
e
lim
ited
,
g
r
o
u
n
d
-
le
v
el
im
ag
er
y
u
s
ed
in
t
h
is
s
tu
d
y
.
C
o
n
s
id
er
in
g
v
alid
atio
n
ac
c
u
r
ac
y
,
tr
ain
i
n
g
tim
e
ef
f
icien
cy
,
an
d
m
o
d
el
s
ize,
Mo
b
ileNetV2
is
th
e
m
o
s
t
b
alan
ce
d
m
o
d
el,
wh
ile
VGG1
9
an
d
VGG1
6
s
tan
d
o
u
t in
ter
m
s
o
f
ac
c
u
r
ac
y
.
T
o
f
u
r
th
er
an
aly
ze
m
o
d
el
p
er
f
o
r
m
a
n
ce
,
tr
ain
in
g
h
is
to
r
y
p
lo
ts
an
d
co
n
f
u
s
io
n
m
atr
ices
o
f
th
e
th
r
ee
m
o
d
els
ar
e
p
r
esen
t
ed
.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
Mo
b
ileNetV2
,
VGG1
9
,
an
d
VGG1
6
is
illu
s
tr
ated
in
Fig
u
r
es 5
,
6
,
a
n
d
7
.
Fig
u
r
e
5
.
Mo
b
ileNet
V2
p
er
f
o
r
m
an
ce
r
esu
lt
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
C
o
mp
a
r
a
tive
a
n
a
lysi
s
o
f c
o
n
vo
lu
tio
n
a
l
n
eu
r
a
l n
etw
o
r
k
a
r
ch
itectu
r
e
fo
r
…
(
A
h
ma
d
B
i
n
ta
n
g
A
r
if
)
4727
Fig
u
r
e
6
.
VGG
-
1
9
p
er
f
o
r
m
a
n
c
e
r
esu
lt
Fig
u
r
e
7
.
VGG
-
1
6
p
er
f
o
r
m
a
n
c
e
r
esu
lt
Fig
u
r
es
5
,
6
,
an
d
7
p
r
esen
t
th
e
tr
ain
in
g
h
is
to
r
y
an
d
co
n
f
u
s
io
n
m
atr
ices
o
f
Mo
b
ileNetV2
,
VGG1
6
,
an
d
VGG1
9
.
T
h
ese
v
is
u
aliza
t
io
n
s
in
d
icate
th
at
all
th
r
ee
m
o
d
els
ex
h
ib
ited
s
tead
ily
in
cr
e
asin
g
ac
cu
r
ac
y
o
v
er
ep
o
ch
s
an
d
d
ec
r
ea
s
in
g
lo
s
s
cu
r
v
es,
co
n
f
ir
m
in
g
s
tab
le
lear
n
in
g
b
eh
av
io
r
.
Ho
wev
er
,
Mo
b
ile
NetV2
s
h
o
ws
m
o
r
e
f
lu
ctu
atio
n
s
d
u
r
i
n
g
tr
ain
in
g
,
s
u
g
g
esti
n
g
th
at
its
p
er
f
o
r
m
an
c
e,
alth
o
u
g
h
ef
f
icien
t,
m
a
y
b
e
m
o
r
e
s
en
s
itiv
e
to
d
ata
v
ar
iatio
n
s
o
r
d
u
e
to
in
s
u
f
f
icien
t
d
ata
.
I
n
s
u
m
m
ar
y
,
VGG1
9
p
r
o
d
u
ce
d
th
e
m
o
s
t
ac
cu
r
ate
p
r
ed
ictio
n
s
,
wh
ile
Mo
b
ileNetV2
o
f
f
er
e
d
t
h
e
b
est
tr
ad
e
-
o
f
f
b
etwe
en
p
e
r
f
o
r
m
a
n
ce
an
d
ef
f
icien
c
y
.
T
h
is
r
ein
f
o
r
ce
s
p
r
io
r
f
in
d
in
g
s
[
2
3
]
w
h
ich
s
tate
th
at
Mo
b
ileNetV2
was
d
esig
n
e
d
to
b
alan
ce
ac
cu
r
ac
y
an
d
co
m
p
u
tatio
n
al
d
em
an
d
s
th
r
o
u
g
h
h
y
p
e
r
p
ar
am
eter
f
lex
i
b
ilit
y
,
m
ak
in
g
it
i
d
ea
l
f
o
r
f
ield
-
b
ased
,
l
o
w
-
r
eso
u
r
ce
s
ce
n
a
r
io
s
.
T
h
ese
r
esu
lts
in
d
icate
th
at
d
ee
p
lear
n
i
n
g
m
o
d
els
—
esp
ec
ially
lig
h
tweig
h
t
ar
ch
itectu
r
es
lik
e
M
o
b
ileNetV2
—
ca
n
ef
f
ec
tiv
el
y
d
is
tin
g
u
is
h
b
etwe
en
b
u
r
n
ed
an
d
u
n
b
u
r
n
ed
ar
ea
s
b
ased
o
n
f
ield
i
m
ag
er
y
,
s
u
p
p
o
r
tin
g
t
h
e
s
tu
d
ies
o
f
C
NN
ap
p
licab
ilit
y
in
p
o
s
t
-
f
ir
e
ass
es
s
m
en
t.
3.
2
.
T
he
ef
f
ec
t
o
f
hy
perpa
ra
m
et
er
s
o
n
mo
del
perf
o
rm
a
n
ce
I
n
th
is
s
tu
d
y
,
s
ev
er
al
h
y
p
er
p
ar
am
eter
s
wer
e
test
ed
o
n
C
NN
m
o
d
els
to
ev
alu
ate
th
eir
im
p
ac
t
o
n
p
er
f
o
r
m
an
ce
.
E
ac
h
ar
ch
itectu
r
e
was
ass
ig
n
ed
a
n
u
m
er
ical
lab
el
f
o
r
ea
s
e
o
f
r
ef
er
en
ce
(
1
=
Mo
b
ileNetV2
,
2
=
VGG1
6
,
3
=
VGG1
9
,
4
=
L
e
Net5
,
5
=
Alex
Net,
6
=
R
esNet5
0
,
an
d
7
=
I
n
ce
p
tio
n
V
3
)
.
T
h
e
h
y
p
er
p
ar
am
eter
s
test
ed
in
clu
d
e
b
atch
s
ize,
in
p
u
t
s
ize,
an
d
o
p
tim
izer
.
T
h
e
b
a
tch
s
izes
test
ed
wer
e
1
6
an
d
3
2
,
in
p
u
t
s
izes
wer
e
192
×
1
9
2
a
n
d
2
2
4
×
2
2
4
,
an
d
th
e
o
p
tim
izer
s
u
s
ed
wer
e
Ad
am
an
d
R
MSPr
o
p
.
T
h
e
f
o
llo
win
g
s
ec
tio
n
d
is
cu
s
s
es
ea
ch
h
y
p
er
p
ar
a
m
eter
in
d
etail.
3
.
2
.
1
.
B
a
t
ch
s
ize
B
atch
s
ize
af
f
ec
ts
s
ev
er
al
asp
ec
ts
o
f
tr
ain
in
g
,
in
clu
d
in
g
co
n
v
er
g
en
ce
tim
e,
tr
ain
in
g
s
tab
ilit
y
,
an
d
th
e
m
o
d
el
’
s
ab
ilit
y
to
g
e
n
er
alize
to
u
n
s
ee
n
d
ata.
Fo
r
in
s
tan
ce
,
s
m
aller
b
atch
s
izes
o
f
ten
allo
w
f
aster
co
m
p
u
tatio
n
s
b
u
t
m
ay
r
e
q
u
ir
e
m
o
r
e
iter
atio
n
s
to
co
n
v
er
g
e
c
o
m
p
ar
e
d
to
lar
g
er
b
atch
s
izes
[
2
4
]
.
T
o
f
u
r
th
e
r
ex
p
lo
r
e
t
h
e
ef
f
ec
t
o
f
b
atch
s
ize
o
n
m
o
d
el
p
er
f
o
r
m
an
ce
,
th
e
r
esu
lts
o
f
th
e
ex
p
er
im
en
ts
ar
e
p
r
esen
ted
in
T
ab
le
3
.
B
ased
o
n
T
ab
le
3
,
th
e
u
s
e
o
f
d
if
f
er
en
t
b
atc
h
s
izes
h
ad
a
s
ig
n
if
ican
t
im
p
ac
t
o
n
b
o
th
v
alid
atio
n
ac
cu
r
ac
y
an
d
tr
ain
i
n
g
tim
e
ac
r
o
s
s
v
ar
io
u
s
C
NN
ar
ch
itect
u
r
es.
W
ith
a
b
atch
s
ize
o
f
1
6
,
m
o
d
els
s
u
ch
as
Mo
b
ileNetV2
(
1
)
p
e
r
f
o
r
m
ed
ex
ce
p
tio
n
ally
well,
ac
h
iev
i
n
g
v
alid
atio
n
ac
c
u
r
ac
ies
u
p
to
1
.
0
0
i
n
ce
r
tain
ex
p
er
im
en
ts
.
H
o
wev
er
,
o
th
er
m
o
d
els
lik
e
VGG1
6
(
2
)
a
n
d
VGG1
9
(
3
)
s
h
o
we
d
m
o
r
e
v
ar
iab
ilit
y
,
with
v
alid
atio
n
ac
cu
r
ac
ies
ten
d
in
g
to
b
e
lo
wer
(
ar
o
u
n
d
0
.
7
9
)
.
C
o
n
v
er
s
ely
,
a
b
atch
s
ize
o
f
3
2
g
en
er
ally
p
r
o
d
u
ce
d
m
o
r
e
co
n
s
is
ten
t
v
alid
atio
n
ac
cu
r
ac
y
.
Fo
r
in
s
tan
ce
,
I
n
ce
p
tio
n
V3
(
7
)
ac
h
iev
ed
n
ea
r
-
p
er
f
ec
t
v
alid
atio
n
ac
cu
r
ac
y
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
2
3
-
4
7
3
1
4728
in
s
ev
er
al
ex
p
er
im
en
ts
.
Ho
we
v
er
,
VGG1
6
an
d
VGG1
9
s
h
o
wed
a
d
ec
r
ea
s
e
in
ac
cu
r
ac
y
w
h
en
u
s
in
g
th
e
lar
g
e
r
b
atch
s
ize.
I
n
ter
m
s
o
f
tr
ain
in
g
tim
e,
b
atch
s
ize
3
2
g
en
er
all
y
r
ed
u
ce
d
t
h
e
o
v
er
all
tr
ain
in
g
tim
e
co
m
p
ar
ed
to
b
atch
s
ize
1
6
,
as
th
e
m
o
d
el
c
o
u
ld
p
r
o
ce
s
s
m
o
r
e
d
ata
i
n
a
s
in
g
le
iter
atio
n
,
th
u
s
r
e
q
u
ir
in
g
f
ewe
r
iter
atio
n
s
t
o
co
m
p
lete
th
e
tr
ain
i
n
g
p
r
o
ce
s
s
.
Nev
er
th
eless
,
s
m
aller
b
atch
s
i
ze
s
(
e.
g
.
,
1
6
)
ten
d
ed
t
o
r
esu
lt
i
n
h
ig
h
er
v
alid
atio
n
ac
cu
r
ac
y
,
t
h
e
co
s
t
o
f
lo
n
g
er
tr
ain
i
n
g
tim
e.
Fo
r
ar
c
h
itectu
r
es
lik
e
Alex
Net
an
d
R
e
s
Net5
0
,
v
alid
atio
n
ac
cu
r
ac
y
r
em
ain
ed
r
elativ
ely
lo
w
u
n
d
er
b
o
th
b
atch
s
ize
s
ettin
g
s
,
s
u
g
g
esti
n
g
th
at
b
atch
s
ize
h
ad
less
in
f
lu
en
ce
o
n
th
ese
m
o
d
els
with
in
th
e
co
n
tex
t
o
f
t
h
is
d
ataset.
I
n
s
u
m
m
ar
y
,
s
m
aller
b
atch
s
izes
y
i
eld
ed
b
etter
ac
cu
r
ac
y
f
o
r
ce
r
t
ain
m
o
d
els,
wh
er
ea
s
lar
g
er
b
atch
s
izes
wer
e
m
o
r
e
ef
f
icien
t
in
ter
m
s
o
f
t
r
ain
in
g
t
im
e
b
u
t
co
u
l
d
p
o
ten
tially
c
o
m
p
r
o
m
is
e
ac
c
u
r
ac
y
.
T
h
is
tr
en
d
is
co
n
s
is
ten
t
wi
th
f
in
d
in
g
s
f
r
o
m
p
r
ev
io
u
s
r
esear
ch
[
2
5
]
,
wh
er
e
s
m
aller
b
atch
s
izes
(
e.
g
.
,
8
)
y
ield
ed
h
ig
h
er
ac
c
u
r
ac
y
c
o
m
p
ar
e
d
to
l
ar
g
er
b
atc
h
s
izes (
e.
g
.
,
1
2
8
)
.
T
ab
le
3.
Mo
d
el
p
er
f
o
r
m
a
n
ce
r
esu
lts
f
o
r
ea
ch
b
atch
s
ize
B
a
t
c
h
si
z
e
S
c
h
e
me
V
a
l
a
c
c
u
r
a
c
y
Tr
a
i
n
i
n
g
t
i
me
(
sec
o
n
d
)
1
2
3
4
5
6
7
1
2
3
4
5
6
7
16
3
0
.
8
8
0
.
8
3
0
.
8
3
0
.
7
5
0
.
5
8
0
.
6
3
0
.
9
6
5
4
7
5
5
3
5
4
4
5
4
5
5
4
9
5
5
3
5
5
0
4
0
.
9
2
0
.
8
8
0
.
9
6
0
.
7
9
0
.
5
8
0
.
6
3
0
.
8
8
5
6
8
5
6
8
5
6
7
5
6
2
5
6
1
5
6
7
5
5
8
5
0
.
9
6
0
.
9
2
0
.
8
8
0
.
8
8
0
.
5
8
0
.
5
8
0
.
8
8
5
4
7
5
4
9
5
4
9
5
5
0
5
4
8
5
5
4
5
5
4
6
0
.
9
6
0
.
8
8
0
.
9
2
0
.
7
1
0
.
5
8
0
.
6
3
0
.
9
6
5
6
3
5
6
8
5
6
8
5
6
2
5
5
9
5
6
0
5
7
1
32
1
0
.
8
8
0
.
7
9
0
.
9
2
0
.
7
5
0
.
5
8
0
.
5
8
0
.
8
8
5
6
1
5
6
0
5
5
0
5
4
6
5
4
3
5
6
1
5
6
1
2
1
.
0
0
0
.
8
8
0
.
9
2
0
.
7
9
0
.
5
8
0
.
6
3
0
.
9
6
5
4
8
5
6
9
5
5
3
5
3
4
5
4
3
5
5
1
5
4
6
7
0
.
9
2
0
.
7
9
0
.
7
9
0
.
8
8
0
.
5
8
0
.
5
8
0
.
9
2
5
5
7
5
5
8
5
5
5
5
4
7
5
4
6
5
5
7
5
5
2
8
1
.
0
0
0
.
7
9
0
.
6
7
0
.
7
9
0
.
5
8
0
.
5
8
0
.
9
6
5
5
6
5
6
0
5
4
6
5
3
9
5
2
6
5
4
2
5
4
1
3
.
2
.
2
.
I
np
ut
s
ize
I
n
p
u
t
s
ize,
o
r
im
a
g
e
r
eso
lu
tio
n
,
in
f
lu
e
n
ce
s
th
e
e
x
ten
t
to
w
h
ich
s
p
atial
f
ea
tu
r
es
with
in
a
n
im
ag
e
ca
n
b
e
ef
f
ec
tiv
ely
ex
tr
ac
ted
b
y
a
m
o
d
el.
T
o
ex
am
in
e
th
e
im
p
ac
t
o
f
in
p
u
t
s
ize
v
ar
iatio
n
s
o
n
m
o
d
el
p
er
f
o
r
m
an
ce
,
ex
p
er
im
en
ts
wer
e
co
n
d
u
cted
u
s
in
g
in
p
u
t
s
izes
o
f
1
9
2
×
1
9
2
an
d
2
2
4
×2
2
4
p
ix
els.
T
h
is
allo
wed
f
o
r
an
ass
es
s
m
en
t
o
f
h
o
w
in
p
u
t
s
ize
af
f
ec
ts
b
o
th
v
alid
atio
n
ac
c
u
r
ac
y
an
d
tr
ain
in
g
tim
e
ac
r
o
s
s
d
if
f
er
en
t
C
NN
ar
ch
itectu
r
es.
T
h
e
ex
p
er
im
en
t
al
r
esu
lts
ar
e
s
u
m
m
a
r
ized
in
T
ab
le
4
.
Ov
er
all,
T
ab
le
4
s
h
o
ws
th
at
an
in
p
u
t
s
ize
o
f
2
2
4
×2
2
4
ten
d
s
to
p
r
o
d
u
ce
h
ig
h
er
an
d
m
o
r
e
c
o
n
s
is
ten
t
v
alid
atio
n
ac
c
u
r
ac
y
co
m
p
ar
ed
to
1
9
2
×1
9
2
,
p
ar
ticu
lar
ly
in
ar
c
h
itectu
r
es
s
u
ch
as
Mo
b
ileNetV2
,
VGG1
6
,
an
d
VGG1
9
.
Fo
r
e
x
am
p
le
,
in
Mo
b
ileNetV2
(
m
o
d
el
1
)
,
th
e
h
ig
h
est
v
alid
atio
n
ac
cu
r
ac
y
o
f
1
.
0
0
was
ac
h
iev
ed
with
a
2
2
4
×
2
2
4
in
p
u
t
s
ize
d
u
r
in
g
ex
p
er
im
en
t
8
,
wh
er
ea
s
with
1
9
2
×1
9
2
in
p
u
t
s
ize,
th
e
h
ig
h
est
ac
cu
r
ac
y
r
ea
ch
ed
o
n
ly
0
.
9
6
in
ex
p
er
im
e
n
t
5
.
T
h
is
s
u
g
g
ests
th
at
in
cr
ea
s
in
g
im
ag
e
r
eso
lu
tio
n
en
ab
les
th
e
m
o
d
el
to
b
etter
r
ec
o
g
n
ize
p
atter
n
s
an
d
e
x
tr
ac
t
f
ea
tu
r
es.
T
h
is
is
in
lin
e
with
f
in
d
in
g
s
f
r
o
m
s
tu
d
ies
[
2
6
]
,
wh
ich
s
tate
th
at
la
r
g
er
im
a
g
e
r
eso
lu
tio
n
s
ten
d
to
im
p
r
o
v
e
m
o
d
el
p
er
f
o
r
m
an
ce
,
t
h
o
u
g
h
th
ey
als
o
in
cr
ea
s
e
co
m
p
u
tatio
n
al
tim
e
an
d
r
eso
u
r
ce
c
o
n
s
u
m
p
tio
n
,
le
ad
in
g
to
a
tr
a
d
e
-
o
f
f
b
etwe
en
co
m
p
u
tatio
n
al
ef
f
icie
n
cy
an
d
r
ec
o
g
n
itio
n
ac
c
u
r
ac
y
.
Ho
wev
er
,
o
th
er
s
tu
d
ies
h
av
e
s
h
o
wn
th
at
in
cr
ea
s
in
g
im
ag
e
s
ize
d
o
es
n
o
t
n
ec
ess
ar
ily
im
p
r
o
v
e
d
ee
p
lear
n
in
g
m
o
d
el
p
er
f
o
r
m
an
ce
,
as
it
h
ig
h
ly
d
ep
en
d
s
o
n
th
e
co
m
p
lex
ity
o
f
th
e
im
ag
es
an
d
th
e
p
r
o
b
lem
b
ein
g
s
o
lv
ed
.
I
n
g
e
n
er
al,
in
cr
ea
s
in
g
in
p
u
t size
d
o
es n
o
t g
u
a
r
an
tee
b
etter
ac
cu
r
ac
y
; in
s
o
m
e
ca
s
es,
s
m
aller
in
p
u
t sizes
ca
n
y
ield
b
etter
p
er
f
o
r
m
an
ce
a
n
d
v
ice
v
er
s
a.
T
h
is
is
b
ec
au
s
e
ea
ch
d
ataset
m
ay
h
av
e
an
o
p
tim
al
in
p
u
t
s
ize
th
at
y
ield
s
th
e
b
est r
esu
lts
,
an
d
ac
c
u
r
ac
y
ca
n
e
v
en
d
ec
r
ea
s
e
if
im
ag
e
s
ize
ex
ce
ed
s
a
ce
r
tain
th
r
e
s
h
o
ld
[
2
7
]
.
I
n
ter
m
s
o
f
tr
ain
in
g
tim
e,
lar
g
er
in
p
u
t
s
izes
ten
d
to
in
cr
ea
s
e
tr
ain
in
g
d
u
r
atio
n
.
T
h
is
is
o
b
s
er
v
ab
le
ac
r
o
s
s
s
ev
er
al
ar
ch
itectu
r
es,
wh
er
e
m
o
d
els
tr
ain
ed
o
n
2
2
4
×
2
2
4
in
p
u
ts
r
eq
u
ir
ed
m
o
r
e
tim
e
th
a
n
th
o
s
e
tr
ain
ed
o
n
1
9
2
×1
9
2
.
T
h
is
f
in
d
in
g
was
co
n
s
is
ten
t
with
[
2
8
]
,
wh
ich
n
o
tes
th
at
la
r
g
er
in
p
u
t
s
izes
in
v
a
r
iab
ly
lead
to
l
o
n
g
e
r
tr
ain
in
g
tim
es.
T
ab
l
e
4.
Mo
d
el
p
er
f
o
r
m
a
n
ce
r
esu
lts
f
o
r
ea
ch
in
p
u
t size
I
n
p
u
t
si
z
e
S
c
h
e
me
V
a
l
a
c
c
u
r
a
c
y
Tr
a
i
n
i
n
g
t
i
me
(
sec
o
n
d
)
1
2
3
4
5
6
7
1
2
3
4
5
6
7
1
9
2
×
1
9
2
1
0
.
8
8
0
.
7
9
0
.
9
2
0
.
7
5
0
.
5
8
0
.
5
8
0
.
8
8
5
6
1
5
6
0
5
5
0
5
4
6
5
4
3
5
6
1
5
6
1
3
0
.
8
8
0
.
8
3
0
.
8
3
0
.
7
5
0
.
5
8
0
.
6
3
0
.
9
6
5
4
7
5
5
3
5
4
4
5
4
5
5
4
9
5
5
3
5
5
0
5
0
.
9
6
0
.
9
2
0
.
8
8
0
.
8
8
0
.
5
8
0
.
5
8
0
.
8
8
5
4
7
5
4
9
5
4
9
5
5
0
5
4
8
5
5
4
5
5
4
7
0
.
9
2
0
.
7
9
0
.
7
9
0
.
8
8
0
.
5
8
0
.
5
8
0
.
9
2
5
5
7
5
5
8
5
5
5
5
4
7
5
4
6
5
5
7
5
5
2
2
2
4
×
2
2
4
2
1
.
0
0
0
.
8
8
0
.
9
2
0
.
7
9
0
.
5
8
0
.
6
3
0
.
9
6
5
4
8
5
6
9
5
5
3
5
3
4
5
4
3
5
5
1
5
4
6
4
0
.
9
2
0
.
8
8
0
.
9
6
0
.
7
9
0
.
5
8
0
.
6
3
0
.
8
8
5
6
8
5
6
8
5
6
7
5
6
2
5
6
1
5
6
7
5
5
8
6
0
.
9
6
0
.
8
8
0
.
9
2
0
.
7
1
0
.
5
8
0
.
6
3
0
.
9
6
5
6
3
5
6
8
5
6
8
5
6
2
5
5
9
5
6
0
5
7
1
8
1
.
0
0
0
.
7
9
0
.
6
7
0
.
7
9
0
.
5
8
0
.
5
8
0
.
9
6
5
5
6
5
6
0
5
4
6
5
3
9
5
2
6
5
4
2
5
4
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
C
o
mp
a
r
a
tive
a
n
a
lysi
s
o
f c
o
n
vo
lu
tio
n
a
l
n
eu
r
a
l n
etw
o
r
k
a
r
ch
itectu
r
e
fo
r
…
(
A
h
ma
d
B
i
n
ta
n
g
A
r
if
)
4729
3
.
2
.
3
.
O
ptim
izer
I
n
th
is
s
tu
d
y
,
two
ty
p
es
o
f
o
p
t
im
izer
s
wer
e
em
p
lo
y
ed
:
Ad
a
m
an
d
R
MSp
r
o
p
.
B
o
th
ar
e
wi
d
ely
u
s
ed
in
d
ee
p
lear
n
i
n
g
m
o
d
el
tr
ain
in
g
d
u
e
to
th
ei
r
ad
ap
tiv
e
lear
n
in
g
r
ate
ca
p
ab
ilit
ies.
T
h
e
im
p
ac
t
o
f
ea
ch
o
p
tim
izer
o
n
v
alid
atio
n
ac
cu
r
ac
y
ac
r
o
s
s
d
if
f
er
en
t
C
NN
ar
ch
itectu
r
es
was
ex
am
in
ed
to
ass
e
s
s
t
h
e
co
n
s
is
ten
cy
an
d
o
p
tim
izatio
n
ef
f
ec
tiv
e
n
ess
d
u
r
in
g
tr
ain
in
g
.
T
h
e
p
er
f
o
r
m
an
ce
r
esu
lts
b
ased
o
n
th
e
o
p
tim
izer
u
s
ed
ar
e
p
r
esen
ted
in
T
ab
le
5
.
Acc
o
r
d
in
g
to
T
ab
le
5
,
R
MSPr
o
p
g
e
n
er
ally
p
r
o
v
id
ed
h
ig
h
er
v
ali
d
atio
n
ac
c
u
r
ac
y
ac
r
o
s
s
s
ev
er
al
m
o
d
els,
m
o
s
t
n
o
tab
ly
M
o
b
ile
NetV2
an
d
I
n
ce
p
tio
n
V
3
.
Fo
r
i
n
s
tan
ce
,
Mo
b
ileNetV2
r
ea
c
h
e
d
a
p
e
r
f
ec
t
ac
cu
r
ac
y
o
f
1
.
0
u
n
d
er
R
MSPr
o
p
in
o
n
e
tr
ain
in
g
s
ch
em
e,
wh
ile
it
o
n
ly
r
ea
ch
ed
u
p
to
0
.
9
2
u
n
d
e
r
Ad
am
.
I
n
ce
p
tio
n
V3
also
p
er
f
o
r
m
e
d
co
n
s
is
ten
tly
well
with
b
o
th
o
p
tim
izer
s
,
o
f
t
en
ac
h
iev
in
g
v
alid
atio
n
ac
cu
r
ac
y
as
h
ig
h
as
0
.
9
6
.
On
th
e
o
th
er
h
an
d
,
o
th
er
a
r
c
h
itectu
r
es
s
u
ch
as
R
e
s
Net5
0
an
d
Alex
Net
d
eliv
er
ed
lo
wer
p
er
f
o
r
m
a
n
ce
,
with
v
alid
atio
n
ac
c
u
r
ac
y
ty
p
ically
r
an
g
i
n
g
b
etwe
en
0
.
5
8
an
d
0
.
63
,
i
n
d
icatin
g
th
eir
lim
itatio
n
s
in
h
a
n
d
lin
g
th
e
co
m
p
lex
tex
t
u
r
es a
n
d
f
ea
tu
r
es
p
r
esen
t in
b
u
r
n
ed
f
ield
im
ag
e
r
y
.
I
n
ad
d
itio
n
t
o
ac
cu
r
ac
y
,
tr
ai
n
in
g
tim
e
was
also
an
aly
ze
d
.
B
o
th
o
p
tim
izer
s
s
h
o
wed
c
o
m
p
ar
ab
le
tr
ain
in
g
d
u
r
atio
n
s
,
th
o
u
g
h
R
MSPr
o
p
o
cc
asio
n
ally
o
f
f
e
r
ed
s
lig
h
tly
s
h
o
r
te
r
tr
ain
i
n
g
tim
es
in
m
o
d
els
lik
e
R
esNet5
0
an
d
I
n
ce
p
tio
n
V3
.
Desp
ite
th
e
s
m
all
d
if
f
er
en
ce
s
,
th
ese
tim
e
s
av
in
g
s
co
u
ld
b
e
b
en
ef
icial
wh
e
n
s
ca
lin
g
u
p
to
lar
g
e
d
atasets
o
r
d
ep
lo
y
i
n
g
m
o
d
els
in
r
eso
u
r
c
e
-
co
n
s
tr
ain
ed
e
n
v
ir
o
n
m
en
ts
.
Am
o
n
g
all
s
ch
em
es,
th
e
co
m
b
in
atio
n
o
f
R
MSPr
o
p
an
d
Mo
b
ileNetV2
p
r
o
v
e
d
to
b
e
th
e
m
o
s
t
ef
f
ec
tiv
e,
ac
h
iev
in
g
p
er
f
ec
t
class
if
icatio
n
ac
cu
r
ac
y
in
ju
s
t
5
5
6
s
ec
o
n
d
s
o
f
tr
ain
i
n
g
,
s
u
g
g
esti
n
g
a
n
id
ea
l
b
alan
ce
o
f
p
er
f
o
r
m
a
n
ce
an
d
ef
f
icien
cy
.
T
h
is
alig
n
s
with
f
i
n
d
in
g
s
f
r
o
m
s
tu
d
y
[
2
9
]
wh
ich
h
ig
h
lig
h
ts
R
MSPr
o
p
as
o
n
e
o
f
th
e
b
est
d
ef
au
lt
o
p
tim
izer
s
d
u
e
to
its
u
s
e
o
f
d
e
ca
y
an
d
m
o
m
en
t
u
m
v
ar
ia
b
les to
o
p
tim
ize
im
ag
e
class
if
icatio
n
ac
cu
r
ac
y
.
T
ab
l
e
5
.
Mo
d
el
p
er
f
o
r
m
a
n
ce
r
esu
lts
f
o
r
ea
ch
o
p
tim
izer
O
p
t
i
mze
r
S
c
h
e
me
V
a
l
a
c
c
u
r
a
c
y
Tr
a
i
n
i
n
g
t
i
me
(
sec
o
n
d
)
1
2
3
4
5
6
7
1
2
3
4
5
6
7
A
d
a
m
1
0
.
8
8
0
.
7
9
0
.
9
2
0
.
7
5
0
.
5
8
0
.
5
8
0
.
8
8
5
6
1
5
6
0
5
5
0
5
4
6
5
4
3
5
6
1
5
6
1
2
1
.
0
0
0
.
8
8
0
.
9
2
0
.
7
9
0
.
5
8
0
.
6
3
0
.
9
6
5
4
8
5
6
9
5
5
3
5
3
4
5
4
3
5
5
1
5
4
6
3
0
.
8
8
0
.
8
3
0
.
8
3
0
.
7
5
0
.
5
8
0
.
6
3
0
.
9
6
5
4
7
5
5
3
5
4
4
5
4
5
5
4
9
5
5
3
5
5
0
4
0
.
9
2
0
.
8
8
0
.
9
6
0
.
7
9
0
.
5
8
0
.
6
3
0
.
8
8
5
6
8
5
6
8
5
6
7
5
6
2
5
6
1
5
6
7
5
5
8
R
M
S
P
r
o
p
5
0
.
9
6
0
.
9
2
0
.
8
8
0
.
8
8
0
.
5
8
0
.
5
8
0
.
8
8
5
4
7
5
4
9
5
4
9
5
5
0
5
4
8
5
5
4
5
5
4
6
0
.
9
6
0
.
8
8
0
.
9
2
0
.
7
1
0
.
5
8
0
.
6
3
0
.
9
6
5
6
3
5
6
8
5
6
8
5
6
2
5
5
9
5
6
0
5
7
1
7
0
.
9
2
0
.
7
9
0
.
7
9
0
.
8
8
0
.
5
8
0
.
5
8
0
.
9
2
5
5
7
5
5
8
5
5
5
5
4
7
5
4
6
5
5
7
5
5
2
8
1
.
0
0
0
.
7
9
0
.
6
7
0
.
7
9
0
.
5
8
0
.
5
8
0
.
9
6
5
5
6
5
6
0
5
4
6
5
3
9
5
2
6
5
4
2
5
4
1
3.
3
.
Dis
cus
s
io
n
T
h
e
f
in
d
in
g
s
h
ig
h
lig
h
t
th
at
lig
h
tweig
h
t
C
NNs
ca
n
b
e
h
ig
h
ly
ef
f
ec
tiv
e
f
o
r
im
ag
e
class
if
icatio
n
in
co
n
s
tr
ain
ed
e
n
v
ir
o
n
m
en
ts
,
s
u
ch
as
p
o
s
t
-
f
ir
e
f
ield
s
ettin
g
s
.
Un
lik
e
s
atellite
im
ag
er
y
,
f
ie
ld
im
ag
es
ca
p
tu
r
ed
with
m
o
b
ile
d
ev
ices
a
r
e
f
lex
i
b
le
an
d
co
s
t
-
ef
f
ec
tiv
e,
y
et
r
e
m
ain
u
n
d
er
u
tili
ze
d
in
wild
f
ir
e
d
am
ag
e
ass
ess
m
en
t.
T
h
is
s
tu
d
y
d
em
o
n
s
tr
ates
th
at,
with
p
r
o
p
er
p
r
ep
r
o
ce
s
s
in
g
an
d
m
o
d
el
s
elec
tio
n
,
C
NNs
ca
n
ac
h
iev
e
co
m
p
etitiv
e
r
esu
lts
ev
en
with
s
u
ch
g
r
o
u
n
d
-
lev
el
d
ata.
No
ta
b
ly
,
o
u
r
b
est
m
o
d
els
r
i
v
aled
o
r
ex
ce
ed
ed
r
ep
o
r
ted
p
er
f
o
r
m
an
ce
s
f
r
o
m
p
r
ev
io
u
s
d
r
o
n
e
-
b
ased
s
tu
d
ies.
Fo
r
in
s
tan
ce
,
VGG1
9
’
s
p
er
f
o
r
m
an
ce
(
9
6
%)
is
co
m
p
ar
ab
le
to
VGG1
6
in
[
4
]
,
wh
ich
u
s
ed
d
r
o
n
e
im
ag
er
y
with
s
im
ilar
task
s
.
Mo
r
e
im
p
o
r
ta
n
tly
,
M
o
b
ileNetV2
m
o
d
el
ac
h
iev
ed
9
6
%
v
al
ac
cu
r
ac
y
—
s
ig
n
if
ican
tly
o
u
tp
er
f
o
r
m
in
g
th
e
7
7
.
7
%
r
ep
o
r
ted
in
[
8
]
,
wh
ich
also
u
s
ed
Mo
b
ileNetV2
o
n
f
ield
im
ag
e
r
y
f
r
o
m
J
am
b
i
Pr
o
v
in
ce
.
T
h
i
s
im
p
r
o
v
e
m
en
t
m
ay
b
e
attr
ib
u
ted
to
th
e
u
s
e
o
f
en
h
an
ce
d
p
r
e
p
r
o
ce
s
s
in
g
,
o
p
ti
m
ized
h
y
p
er
p
a
r
am
eter
s
,
an
d
m
o
r
e
s
y
s
tem
atic
tr
ain
in
g
s
ch
e
m
es.
T
h
ese
f
in
d
i
n
g
s
r
ein
f
o
r
ce
t
h
e
p
r
ac
tical
v
alu
e
o
f
u
s
in
g
m
o
b
ile
im
ag
er
y
f
o
r
p
o
s
t
-
f
ir
e
class
if
icatio
n
an
d
d
em
o
n
s
tr
ate
th
at
ca
r
ef
u
l
m
o
d
el
tu
n
in
g
ca
n
y
ield
co
m
p
etitiv
e,
ev
en
s
u
p
er
io
r
,
r
esu
lt
s
co
m
p
ar
ed
to
p
r
i
o
r
ap
p
r
o
ac
h
es
u
s
in
g
th
e
s
am
e
m
o
d
el
ar
ch
itectu
r
e.
4.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
c
o
m
p
a
r
ed
s
ev
en
C
NN
ar
ch
itectu
r
es
f
o
r
class
if
y
in
g
p
o
s
t
-
f
o
r
est
f
ir
e
ar
ea
s
u
s
i
n
g
m
o
b
ile
-
ca
p
tu
r
ed
f
ield
im
ag
er
y
an
d
f
o
u
n
d
th
at
Mo
b
ileNetV2
,
VGG1
6
,
a
n
d
VGG1
9
ac
h
iev
ed
th
e
b
est
p
er
f
o
r
m
a
n
ce
,
with
Mo
b
ileNetV2
o
f
f
e
r
in
g
th
e
b
est
b
alan
ce
b
etwe
en
ac
c
u
r
ac
y
,
t
r
ain
in
g
tim
e,
an
d
m
o
d
el
s
ize.
T
h
e
r
esu
lts
h
ig
h
lig
h
t
th
e
p
o
ten
tial
o
f
f
iel
d
im
ag
e
r
y
as
a
l
o
w
-
co
s
t
an
d
f
lex
ib
le
alter
n
ativ
e
to
s
atellite
o
r
d
r
o
n
e
d
ata
f
o
r
wild
f
ir
e
d
am
ag
e
ass
ess
m
en
t.
T
h
e
s
tu
d
y
’
s
n
o
v
elty
lies
i
n
its
f
o
cu
s
o
n
m
o
b
ile
im
ag
er
y
an
d
s
y
s
tem
atic
ev
alu
atio
n
o
f
h
y
p
er
p
ar
a
m
eter
s
ac
r
o
s
s
m
u
ltip
le
a
r
ch
itectu
r
es.
T
h
ese
f
in
d
in
g
s
s
u
p
p
o
r
t
t
h
e
d
e
v
elo
p
m
e
n
t
o
f
lig
h
tweig
h
t,
r
ea
l
-
tim
e
to
o
ls
f
o
r
p
o
s
t
-
f
ir
e
ass
ess
m
en
t
an
d
s
u
g
g
est
f
u
t
u
r
e
r
esear
c
h
s
h
o
u
ld
ex
p
lo
r
e
lar
g
er
d
atasets
,
m
o
r
e
h
y
p
e
r
p
ar
am
ete
r
tu
n
in
g
,
an
d
o
n
-
d
ev
ice
d
e
p
lo
y
m
en
t to
en
h
an
ce
p
r
ac
tical
ap
p
l
icatio
n
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
2
3
-
4
7
3
1
4730
RE
F
E
R
E
NC
E
S
[
1
]
A
mi
n
a
h
,
C
.
K
r
a
h
,
a
n
d
P
e
r
d
i
n
a
n
,
“
F
o
r
e
st
f
i
r
e
s
a
n
d
ma
n
a
g
e
m
e
n
t
e
f
f
o
r
t
s
i
n
I
n
d
o
n
e
s
i
a
(
a
r
e
v
i
e
w
)
,
”
in
I
O
P
C
o
n
f
e
re
n
c
e
S
e
r
i
e
s:
E
a
rt
h
a
n
d
En
v
i
r
o
n
m
e
n
t
a
l
S
c
i
e
n
c
e
,
v
o
l
.
5
0
4
,
n
o
.
1
,
p
.
0
1
2
0
1
3
,
M
a
y
2
0
2
0
,
d
o
i
:
1
0
.
1
0
8
8
/
1
7
5
5
-
1
3
1
5
/
5
0
4
/
1
/
0
1
2
0
1
3
.
[
2
]
E.
N
u
r
a
d
i
e
t
a
l
.
,
“
D
e
v
e
l
o
p
m
e
n
t
o
f
a
m
o
b
i
l
e
a
p
p
l
i
c
a
t
i
o
n
f
o
r
f
o
r
e
s
t
a
n
d
l
a
n
d
f
i
r
e
p
a
t
r
o
l
u
si
n
g
p
r
o
t
o
t
y
p
e
m
e
t
h
o
d
,
”
in
I
O
P
C
o
n
f
e
r
e
n
c
e
S
e
r
i
e
s:
E
a
r
t
h
a
n
d
E
n
v
i
r
o
n
m
e
n
t
a
l
S
c
i
e
n
c
e
,
v
o
l
.
1
1
7
7
,
n
o
.
1
,
p
.
0
1
2
0
0
5
,
M
a
y
2
0
2
3
,
d
o
i
:
1
0
.
1
0
8
8
/
1
7
5
5
-
1
3
1
5
/
1
1
7
7
/
1
/
0
1
2
0
0
5
.
[
3
]
G
.
Z
h
a
n
g
,
M
.
W
a
n
g
,
a
n
d
K
.
L
i
u
,
“
F
o
r
e
s
t
f
i
r
e
s
u
s
c
e
p
t
i
b
i
l
i
t
y
m
o
d
e
l
i
n
g
u
s
i
n
g
a
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
f
o
r
Y
u
n
n
a
n
P
r
o
v
i
n
c
e
o
f
C
h
i
n
a
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
D
i
s
a
s
t
e
r
R
i
s
k
S
c
i
e
n
c
e
,
v
o
l
.
1
0
,
n
o
.
3
,
p
p
.
3
8
6
–
4
0
3
,
S
e
p
.
2
0
1
9
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
3
7
5
3
-
019
-
0
0
2
3
3
-
1.
[
4
]
D
.
H
i
n
d
a
r
t
o
,
“
C
o
m
p
a
r
i
s
o
n
a
c
c
u
r
a
c
y
o
f
C
N
N
a
n
d
V
G
G
1
6
i
n
f
o
r
e
s
t
f
i
r
e
i
d
e
n
t
i
f
i
c
a
t
i
o
n
:
a
c
a
s
e
s
t
u
d
y
,
”
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
e
r
N
e
t
w
o
r
k
s
,
A
r
c
h
i
t
e
c
t
u
r
e
a
n
d
H
i
g
h
P
e
r
f
o
r
m
a
n
c
e
C
o
m
p
u
t
i
n
g
,
v
o
l
.
6
,
n
o
.
1
,
p
p
.
1
3
7
–
1
4
8
,
D
e
c
.
2
0
2
3
,
d
o
i
:
1
0
.
4
7
7
0
9
/
c
n
a
h
p
c
.
v
6
i
1
.
3
3
7
1
.
[
5
]
S
.
K
h
a
n
a
n
d
A
.
K
h
a
n
,
“
F
F
i
r
e
N
e
t
:
d
e
e
p
l
e
a
r
n
i
n
g
b
a
se
d
f
o
r
e
st
f
i
r
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
a
n
d
d
e
t
e
c
t
i
o
n
i
n
sm
a
r
t
c
i
t
i
e
s,”
S
y
m
m
e
t
ry
,
v
o
l
.
1
4
,
n
o
.
1
0
,
p
.
2
1
5
5
,
O
c
t
.
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
s
y
m1
4
1
0
2
1
5
5
.
[
6
]
V
.
H
n
a
t
u
s
h
e
n
k
o
,
V
.
H
n
a
t
u
s
h
e
n
k
o
,
a
n
d
D
.
S
o
l
d
a
t
e
n
k
o
,
“
N
e
u
r
a
l
n
e
t
w
o
r
k
-
b
a
sed
a
n
a
l
y
si
s
o
f
f
o
r
e
s
t
f
i
r
e
a
f
t
e
r
m
a
t
h
i
n
c
l
a
ss
-
i
mb
a
l
a
n
c
e
d
r
e
mo
t
e
se
n
si
n
g
e
a
r
t
h
i
ma
g
e
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
T
h
e
I
n
t
e
rn
a
t
i
o
n
a
l
Ar
c
h
i
v
e
s
o
f
t
h
e
P
h
o
t
o
g
r
a
m
m
e
t
ry,
R
e
m
o
t
e
S
e
n
s
i
n
g
a
n
d
S
p
a
t
i
a
l
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
s
,
v
o
l
.
X
LV
I
I
I
-
3
–
2
,
p
p
.
2
2
3
–
2
2
9
,
N
o
v
.
2
0
2
4
,
d
o
i
:
1
0
.
5
1
9
4
/
i
s
p
r
s
-
a
r
c
h
i
v
e
s
-
X
LV
I
I
I
-
3
-
2
0
2
4
-
2
2
3
-
2
0
2
4
.
[
7
]
S
.
G
h
o
u
z
l
a
n
e
,
“
W
i
l
d
f
i
r
e
r
e
m
o
t
e
se
n
s
i
n
g
a
p
p
l
i
c
a
t
i
o
n
s,
”
i
n
6
th
I
n
t
e
rn
a
t
i
o
n
a
l
S
t
u
d
e
n
t
s
S
c
i
e
n
c
e
C
o
n
g
r
e
ss
Pr
o
c
e
e
d
i
n
g
s
Bo
o
k
,
I
z
mi
r
I
n
t
e
r
n
a
t
i
o
n
a
l
G
u
e
st
S
t
u
d
e
n
t
A
ss
o
c
i
a
t
i
o
n
,
S
e
p
.
2
0
2
2
,
d
o
i
:
1
0
.
5
2
4
6
0
/
i
ssc.
2
0
2
2
.
0
2
7
.
[
8
]
A
.
H
i
d
a
y
a
t
,
I
.
S
.
S
i
t
a
n
g
g
a
n
g
,
a
n
d
L.
S
y
a
u
f
i
n
a
,
“
C
l
a
ssi
f
i
c
a
t
i
o
n
o
f
f
o
r
e
st
a
n
d
l
a
n
d
f
i
r
e
se
v
e
r
i
t
y
l
e
v
e
l
s
u
si
n
g
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
i
n
BI
O
We
b
o
f
C
o
n
f
e
r
e
n
c
e
s
,
Ja
k
a
r
i
a
,
Ed
.
,
A
u
g
.
2
0
2
4
,
p
.
0
1
0
3
0
,
d
o
i
:
1
0
.
1
0
5
1
/
b
i
o
c
o
n
f
/
2
0
2
4
1
2
3
0
1
0
3
0
.
[
9
]
Y
.
H
u
a
n
g
a
n
d
A
.
R
e
n
a
u
d
,
“
T
h
e
a
p
p
l
i
c
a
t
i
o
n
o
f
C
N
N
-
b
a
s
e
d
i
m
a
g
e
c
l
a
ssi
f
i
c
a
t
i
o
n
t
o
w
i
l
d
f
i
r
e
e
a
r
l
y
d
e
t
e
c
t
i
o
n
,
”
J
o
u
r
n
a
l
o
f
S
t
u
d
e
n
t
Re
se
a
rc
h
,
v
o
l
.
1
2
,
n
o
.
4
,
N
o
v
.
2
0
2
3
,
d
o
i
:
1
0
.
4
7
6
1
1
/
j
sr
h
s
.
v
1
2
i
4
.
5
8
6
5
.
[
1
0
]
L.
S
y
a
u
f
i
n
a
a
n
d
A
.
A
b
i
H
a
mz
a
h
,
“
C
h
a
n
g
e
s
o
f
t
r
e
e
s
p
e
c
i
e
s
d
i
v
e
r
si
t
y
i
n
p
e
a
t
l
a
n
d
i
m
p
a
c
t
e
d
b
y
m
o
d
e
r
a
t
e
f
i
r
e
sev
e
r
i
t
y
a
t
Te
l
u
k
M
e
r
a
n
t
i
,
P
e
l
a
l
a
w
a
n
,
R
i
a
u
P
r
o
v
i
n
c
e
,
I
n
d
o
n
e
si
a
,
”
Bi
o
d
i
v
e
rsi
t
a
s
J
o
u
r
n
a
l
o
f
B
i
o
l
o
g
i
c
a
l
D
i
v
e
rsi
t
y
,
v
o
l
.
2
2
,
n
o
.
5
,
M
a
y
2
0
2
1
,
d
o
i
:
1
0
.
1
3
0
5
7
/
b
i
o
d
i
v
/
d
2
2
0
5
5
5
.
[
1
1
]
K
.
Z
h
o
u
,
S
.
-
K
.
O
h
,
W
.
P
e
d
r
y
c
z
,
a
n
d
J.
Q
i
u
,
“
D
a
t
a
p
r
e
p
r
o
c
e
ss
i
n
g
st
r
a
t
e
g
y
i
n
c
o
n
s
t
r
u
c
t
i
n
g
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
c
l
a
ssi
f
i
e
r
b
a
s
e
d
o
n
c
o
n
s
t
r
a
i
n
e
d
p
a
r
t
i
c
l
e
sw
a
r
m
o
p
t
i
mi
z
a
t
i
o
n
w
i
t
h
f
u
z
z
y
p
e
n
a
l
t
y
f
u
n
c
t
i
o
n
,
”
En
g
i
n
e
e
ri
n
g
Ap
p
l
i
c
a
t
i
o
n
s
o
f
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
1
1
7
,
p
.
1
0
5
5
8
0
,
J
a
n
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
n
g
a
p
p
a
i
.
2
0
2
2
.
1
0
5
5
8
0
.
[
1
2
]
P
.
R
.
To
g
a
t
o
r
o
p
,
Y
.
P
r
a
t
a
ma
,
A
.
M
o
n
i
c
a
S
i
a
n
t
u
r
i
,
M
.
S
a
r
i
P
a
sari
b
u
,
a
n
d
P
.
S
a
n
g
m
a
j
a
d
i
S
i
n
a
g
a
,
“
I
mag
e
p
r
e
p
r
o
c
e
ss
i
n
g
a
n
d
h
y
p
e
r
p
a
r
a
me
t
e
r
o
p
t
i
mi
z
a
t
i
o
n
o
n
p
r
e
t
r
a
i
n
e
d
m
o
d
e
l
M
o
b
i
l
e
N
e
t
V
2
i
n
w
h
i
t
e
b
l
o
o
d
c
e
l
l
i
ma
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
I
A
ES
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
I
J
-
A
I
)
,
v
o
l
.
1
2
,
n
o
.
3
,
p
.
1
2
1
0
,
S
e
p
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
1
2
.
i
3
.
p
p
1
2
1
0
-
1
2
2
3
.
[
1
3
]
J.
H
.
Y
o
u
s
i
f
a
n
d
H
.
A
.
K
a
z
e
m,
“
P
r
e
d
i
c
t
i
o
n
a
n
d
e
v
a
l
u
a
t
i
o
n
o
f
p
h
o
t
o
v
o
l
t
a
i
c
-
t
h
e
r
ma
l
e
n
e
r
g
y
sy
s
t
e
m
s
p
r
o
d
u
c
t
i
o
n
u
s
i
n
g
a
r
t
i
f
i
c
i
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
a
n
d
e
x
p
e
r
i
m
e
n
t
a
l
d
a
t
a
s
e
t
,
”
C
a
se
S
t
u
d
i
e
s
i
n
T
h
e
r
m
a
l
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
2
7
,
p
.
1
0
1
2
9
7
,
O
c
t
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
si
t
e
.
2
0
2
1
.
1
0
1
2
9
7
.
[
1
4
]
M
.
A
.
I
.
F
a
h
i
m
a
n
d
S
.
A
.
Tu
mp
a
,
“
I
mag
e
a
u
g
m
e
n
t
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s
:
e
n
h
a
n
c
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
p
e
r
f
o
r
ma
n
c
e
,
”
R
e
se
a
rc
h
G
a
t
e
,
p
p
.
1
–
7
,
2
0
2
3
.
[
1
5
]
R
.
A
.
d
e
O
l
i
v
e
i
r
a
a
n
d
M
.
H
.
J.
B
o
l
l
e
n
,
“
D
e
e
p
l
e
a
r
n
i
n
g
f
o
r
p
o
w
e
r
q
u
a
l
i
t
y
,
”
El
e
c
t
r
i
c
Po
w
e
r
S
y
st
e
m
s
Re
se
a
rc
h
,
v
o
l
.
2
1
4
,
p
.
1
0
8
8
8
7
,
Jan
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
p
sr
.
2
0
2
2
.
1
0
8
8
8
7
.
[
1
6
]
M
.
I
mad
,
O
.
D
o
u
k
h
i
,
a
n
d
D
.
-
J.
Le
e
,
“
Tr
a
n
sf
e
r
l
e
a
r
n
i
n
g
b
a
s
e
d
s
e
ma
n
t
i
c
se
g
me
n
t
a
t
i
o
n
f
o
r
3
D
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
f
r
o
m
p
o
i
n
t
c
l
o
u
d
,
”
S
e
n
so
rs
,
v
o
l
.
2
1
,
n
o
.
1
2
,
p
.
3
9
6
4
,
J
u
n
.
2
0
2
1
,
d
o
i
:
1
0
.
3
3
9
0
/
s2
1
1
2
3
9
6
4
.
[
1
7
]
I
.
K
a
n
d
e
l
a
n
d
M
.
C
a
s
t
e
l
l
i
,
“
T
h
e
e
f
f
e
c
t
o
f
b
a
t
c
h
si
z
e
o
n
t
h
e
g
e
n
e
r
a
l
i
z
a
b
i
l
i
t
y
o
f
t
h
e
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
o
n
a
h
i
s
t
o
p
a
t
h
o
l
o
g
y
d
a
t
a
se
t
,
”
I
C
T
Ex
p
ress
,
v
o
l
.
6
,
n
o
.
4
,
p
p
.
3
1
2
–
3
1
5
,
D
e
c
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
c
t
e
.
2
0
2
0
.
0
4
.
0
1
0
.
[
1
8
]
R
.
Z
a
h
e
e
r
a
n
d
H
.
S
h
a
z
i
y
a
,
“
A
S
t
u
d
y
o
f
t
h
e
o
p
t
i
m
i
z
a
t
i
o
n
a
l
g
o
r
i
t
h
ms
i
n
d
e
e
p
l
e
a
r
n
i
n
g
,
”
i
n
2
0
1
9
T
h
i
r
d
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
v
e
n
t
i
v
e
S
y
s
t
e
m
s
a
n
d
C
o
n
t
r
o
l
(
I
C
I
S
C
)
,
I
EEE,
Ja
n
.
2
0
1
9
,
p
p
.
5
3
6
–
5
3
9
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
S
C
4
4
3
5
5
.
2
0
1
9
.
9
0
3
6
4
4
2
.
[
1
9
]
S
.
H
.
H
a
j
i
a
n
d
A
.
M
.
A
b
d
u
l
a
z
e
e
z
,
“
C
o
mp
a
r
i
s
o
n
o
f
o
p
t
i
m
i
z
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s
b
a
se
d
o
n
g
r
a
d
i
e
n
t
d
e
sce
n
t
a
l
g
o
r
i
t
h
m:
a
r
e
v
i
e
w
,
”
Pa
l
Arc
h
’
s J
o
u
rn
a
l
o
f
Arc
h
a
e
o
l
o
g
y
o
f
Eg
y
p
t
/
E
g
y
p
t
o
l
o
g
y
,
v
o
l
.
1
8
,
n
o
.
4
,
p
p
.
2
7
1
5
–
2
7
4
3
,
2
0
2
1
.
[
2
0
]
F
.
U
c
a
r
a
n
d
D
.
K
o
r
k
ma
z
,
“
C
O
V
I
D
i
a
g
n
o
si
s
-
N
e
t
:
d
e
e
p
B
a
y
e
s
-
S
q
u
e
e
z
e
N
e
t
b
a
se
d
d
i
a
g
n
o
si
s
o
f
t
h
e
c
o
r
o
n
a
v
i
r
u
s
d
i
s
e
a
s
e
2
0
1
9
(
C
O
V
I
D
-
1
9
)
f
r
o
m
X
-
r
a
y
i
m
a
g
e
s,”
M
e
d
i
c
a
l
H
y
p
o
t
h
e
s
e
s
,
v
o
l
.
1
4
0
,
p
.
1
0
9
7
6
1
,
Ju
l
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
m
e
h
y
.
2
0
2
0
.
1
0
9
7
6
1
.
[
2
1
]
S
.
Y
a
n
g
a
n
d
G
.
B
e
r
d
i
n
e
,
“
C
o
n
f
u
s
i
o
n
m
a
t
r
i
x
,
”
T
h
e
S
o
u
t
h
w
e
s
t
Re
s
p
i
r
a
t
o
ry
a
n
d
C
r
i
t
i
c
a
l
C
a
r
e
C
h
r
o
n
i
c
l
e
s
,
v
o
l
.
1
2
,
n
o
.
5
3
,
p
p
.
7
5
–
7
9
,
O
c
t
.
2
0
2
4
,
d
o
i
:
1
0
.
1
2
7
4
6
/
s
w
r
c
c
c
.
v
1
2
i
5
3
.
1
3
9
1
.
[
2
2
]
L.
A
l
z
u
b
a
i
d
i
e
t
a
l
.
,
“
R
e
v
i
e
w
o
f
d
e
e
p
l
e
a
r
n
i
n
g
:
c
o
n
c
e
p
t
s,
C
N
N
a
r
c
h
i
t
e
c
t
u
r
e
s
,
c
h
a
l
l
e
n
g
e
s,
a
p
p
l
i
c
a
t
i
o
n
s
,
f
u
t
u
r
e
d
i
r
e
c
t
i
o
n
s,
”
J
o
u
rn
a
l
o
f
Bi
g
D
a
t
a
,
v
o
l
.
8
,
n
o
.
1
,
p
.
5
3
,
M
a
r
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
8
6
/
s
4
0
5
3
7
-
021
-
0
0
4
4
4
-
8.
[
2
3
]
K
.
D
e
y
,
M
.
M
.
H
a
ssa
n
,
M
.
M
.
R
a
n
a
,
a
n
d
M
.
H
.
H
e
n
a
,
“
B
a
n
g
l
a
d
e
sh
i
i
n
d
i
g
e
n
o
u
s
f
i
s
h
c
l
a
ss
i
f
i
c
a
t
i
o
n
u
si
n
g
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
2
0
2
1
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
(
I
C
I
T
)
,
I
EE
E
,
Ju
l
.
2
0
2
1
,
p
p
.
8
9
9
–
9
0
4
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
T5
2
6
8
2
.
2
0
2
1
.
9
4
9
1
6
8
1
.
[
2
4
]
P
.
M
.
R
a
d
i
u
k
,
“
I
mp
a
c
t
o
f
t
r
a
i
n
i
n
g
s
e
t
b
a
t
c
h
si
z
e
o
n
t
h
e
p
e
r
f
o
r
ma
n
c
e
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
f
o
r
d
i
v
e
r
se
d
a
t
a
se
t
s,”
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
a
n
d
M
a
n
a
g
e
m
e
n
t
S
c
i
e
n
c
e
,
v
o
l
.
2
0
,
n
o
.
1
,
Ja
n
.
2
0
1
7
,
d
o
i
:
1
0
.
1
5
1
5
/
i
t
ms
-
2
0
1
7
-
0
0
0
3
.
[
2
5
]
K
.
K
u
r
n
i
a
w
a
n
,
A
.
P
e
r
d
a
n
a
W
i
n
d
a
r
t
o
,
a
n
d
S
.
S
o
l
i
k
h
u
n
,
“
R
e
f
i
n
i
n
g
C
N
N
a
r
c
h
i
t
e
c
t
u
r
e
f
o
r
f
o
r
e
s
t
f
i
r
e
d
e
t
e
c
t
i
o
n
:
i
m
p
r
o
v
i
n
g
a
c
c
u
r
a
c
y
t
h
r
o
u
g
h
e
f
f
i
c
i
e
n
t
h
y
p
e
r
p
a
r
a
m
e
t
e
r
t
u
n
i
n
g
,
”
B
u
l
l
e
t
i
n
o
f
El
e
c
t
ri
c
a
l
E
n
g
i
n
e
e
r
i
n
g
a
n
d
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
1
4
,
n
o
.
2
,
p
p
.
1
2
0
2
–
1
2
1
1
,
A
p
r
.
2
0
2
5
,
d
o
i
:
1
0
.
1
1
5
9
1
/
e
e
i
.
v
1
4
i
2
.
8
8
0
5
.
[
2
6
]
S
.
Ta
n
g
e
t
a
l
.
,
“
Th
e
e
f
f
e
c
t
o
f
i
ma
g
e
r
e
so
l
u
t
i
o
n
o
n
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
i
n
b
r
e
a
st
u
l
t
r
a
so
u
n
d
,
”
H
e
l
i
y
o
n
,
v
o
l
.
9
,
n
o
.
8
,
p
.
e
1
9
2
5
3
,
A
u
g
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
h
e
l
i
y
o
n
.
2
0
2
3
.
e
1
9
2
5
3
.
[
2
7
]
M
.
B
a
l
a
j
i
a
n
d
R
.
Jo
s
e
p
h
,
“
I
mp
a
c
t
o
f
i
mag
e
s
i
z
e
o
n
a
c
c
u
r
a
c
y
a
n
d
g
e
n
e
r
a
l
i
z
a
t
i
o
n
o
f
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s,”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
Re
s
e
a
r
c
h
a
n
d
A
n
a
l
y
t
i
c
a
l
R
e
v
i
e
w
s
,
v
o
l
.
6
,
n
o
.
1
,
p
p
.
7
0
–
8
0
,
2
0
1
9
.
[
2
8
]
S
.
S
a
p
o
n
a
r
a
a
n
d
A
.
E
l
h
a
n
a
s
h
i
,
“
I
mp
a
c
t
o
f
i
ma
g
e
r
e
si
z
i
n
g
o
n
d
e
e
p
l
e
a
r
n
i
n
g
d
e
t
e
c
t
o
r
s
f
o
r
t
r
a
i
n
i
n
g
t
i
m
e
a
n
d
m
o
d
e
l
p
e
r
f
o
r
ma
n
c
e
,
”
2
0
2
2
,
p
p
.
1
0
–
17
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
9
5
4
9
8
-
7
_
2
.
[
2
9
]
P
.
V
e
r
m
a
,
V
.
Tr
i
p
a
t
h
i
,
a
n
d
B
.
P
a
n
t
,
“
C
o
mp
a
r
i
s
o
n
o
f
d
i
f
f
e
r
e
n
t
o
p
t
i
mi
z
e
r
s
i
m
p
l
e
m
e
n
t
e
d
o
n
t
h
e
d
e
e
p
l
e
a
r
n
i
n
g
a
r
c
h
i
t
e
c
t
u
r
e
s
f
o
r
C
O
V
I
D
-
1
9
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
M
a
t
e
ri
a
l
s
T
o
d
a
y
:
Pro
c
e
e
d
i
n
g
s
,
v
o
l
.
4
6
,
p
p
.
1
1
0
9
8
–
1
1
1
0
2
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
ma
t
p
r
.
2
0
2
1
.
0
2
.
2
4
4
.
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
C
o
mp
a
r
a
tive
a
n
a
lysi
s
o
f c
o
n
vo
lu
tio
n
a
l
n
eu
r
a
l n
etw
o
r
k
a
r
ch
itectu
r
e
fo
r
…
(
A
h
ma
d
B
i
n
ta
n
g
A
r
if
)
4731
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Ahm
a
d
Bin
t
a
n
g
Ar
if
is
c
u
rre
n
tl
y
p
u
rs
u
in
g
h
is
m
a
ste
r
’
s
d
e
g
re
e
in
c
o
m
p
u
ter
sc
ien
c
e
a
t
IP
B
Un
iv
e
rsity
,
b
u
i
ld
i
n
g
u
p
o
n
h
is
b
a
c
h
e
lo
r
’
s
d
e
g
re
e
fro
m
th
e
sa
m
e
in
stit
u
ti
o
n
.
W
it
h
a
fo
c
u
s
o
n
d
a
ta
m
i
n
in
g
a
n
d
m
a
c
h
in
e
lea
rn
i
n
g
,
h
e
is
a
c
ti
v
e
l
y
e
n
g
a
g
e
d
a
s
a
lab
a
ss
istan
t
f
o
r
t
h
e
d
a
ta
m
in
in
g
c
o
u
rse
,
c
o
n
tri
b
u
ti
n
g
to
b
o
th
h
is
o
wn
lea
rn
in
g
a
n
d
t
h
a
t
o
f
h
is
p
e
e
rs
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
h
m
a
d
b
in
tan
g
a
rif@ap
p
s.ip
b
.
a
c
.
id
.
Im
a
s
S
u
k
a
e
sih
S
it
a
n
g
g
a
n
g
r
e
c
e
iv
e
d
a
P
h
.
D.
d
e
g
re
e
in
c
o
m
p
u
te
r
sc
ien
c
e
fro
m
th
e
F
a
c
u
lt
y
o
f
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
In
fo
rm
a
ti
o
n
Tec
h
n
o
lo
g
y
,
Un
i
v
e
rsiti
P
u
tra
M
a
lay
sia
,
i
n
2
0
1
3
.
S
h
e
is
a
lec
tu
re
r
in
S
c
h
o
o
l
o
f
Da
ta
S
c
ien
c
e
,
M
a
th
e
m
a
ti
c
s
a
n
d
In
fo
rm
a
ti
c
s
,
IP
B
Un
iv
e
rsit
y
,
In
d
o
n
e
sia
.
He
r
m
a
in
re
se
a
rc
h
in
t
e
re
sts
in
c
lu
d
e
s
p
a
ti
a
l
d
a
ta
m
in
i
n
g
a
n
d
sm
a
rt
a
g
ric
u
lt
u
re
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
ima
s.sitan
g
g
a
n
g
@a
p
p
s.ip
b
.
a
c
.
id
.
H
a
r
i
Ag
u
n
g
Adri
a
n
to
i
s
a
p
r
o
fe
ss
o
r
a
t
IP
B
Un
i
v
e
rsity
.
He
h
o
l
d
s
a
P
h
.
D.
s
t
u
d
e
n
t
in
S
c
h
o
o
l
o
f
Earth
a
n
d
E
n
v
ir
o
n
m
e
n
t
a
t
Un
iv
e
rsity
o
f
Lee
d
s,
a
m
a
ste
r
’
s
d
e
g
re
e
in
c
o
m
p
u
te
r
sc
ien
c
e
a
t
th
e
I
P
B
U
n
iv
e
rsit
y
a
n
d
a
b
a
c
h
e
lo
r
’
s
d
e
g
re
e
in
c
o
m
p
u
ter
sc
ien
c
e
a
t
IP
B
Un
i
v
e
rsity
.
His
re
se
a
r
c
h
a
re
a
s
a
re
in
th
e
field
s
o
f
d
a
ta
m
in
in
g
,
d
a
ta
wa
re
h
o
u
si
n
g
a
n
d
sp
a
ti
a
l
d
a
ta
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
g
u
n
g
@a
p
p
s.ip
b
.
a
c
.
id
.
La
il
a
n
S
y
a
u
fi
n
a
is
a
p
ro
fe
ss
o
r
in
fo
re
st
p
r
o
tec
ti
o
n
with
f
o
re
st
fi
re
a
s
h
e
r
m
a
jo
r
a
t
De
p
a
rtme
n
t
o
f
S
il
v
icu
lt
u
re
F
a
c
u
lt
y
o
f
F
o
re
stry
a
n
d
En
v
iro
n
m
e
n
t
I
P
B.
He
r
u
n
d
e
r
g
ra
d
u
a
te
f
ro
m
Bo
g
o
r
A
g
ricu
lt
u
ra
l
Un
i
v
e
rsity
(I
P
B),
m
a
ste
r
d
e
g
re
e
fro
m
G
e
o
rg
Au
g
u
st
Un
iv
e
rsity
G
e
rm
a
n
y
a
n
d
P
h
.
D.
fr
o
m
Un
i
v
e
rsiti
P
u
tra
M
a
lay
sia
.
He
r
field
o
f
i
n
tere
st
in
c
lu
d
in
g
fo
re
st
fire
in
t
h
e
a
sp
e
c
ts
o
f
fire
se
v
e
rit
y
a
ss
e
ss
m
e
n
t,
fire
m
a
n
a
g
e
m
e
n
t,
p
e
a
tl
a
n
d
fire,
f
ire
-
b
io
d
i
v
e
rsity
,
fire
-
c
li
m
a
te,
a
n
d
fire
-
e
m
issio
n
.
Sh
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
laila
n
s
@a
p
p
s.i
p
b
.
a
c
.
i
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.