I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
3
8
,
No
.
1
,
A
p
r
il
20
2
5
,
p
p
.
447
~
458
I
SS
N:
2
502
-
4
7
52
,
DOI
: 1
0
.
1
1
5
9
1
/ijee
cs
.v
3
8
.
i
1
.
pp
447
-
4
5
8
447
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs
.
ia
esco
r
e.
co
m
Recog
nition o
f
pl
a
nt
lea
f
disea
ses
b
a
sed o
n deep
l
ea
r
ning
and
the
ch
emica
l reac
tion o
ptimiza
tion
a
lg
o
rithm
Ng
hien Ng
uy
en
B
a
1
,
Nhu
ng
Ng
uy
en
T
hi
1
,
Du
ng
Vuo
ng
Q
uo
c
2
,
Cuo
ng
Ng
uy
en
Co
ng
3
1
F
a
c
u
l
t
y
o
f
I
n
f
o
r
m
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
H
a
n
o
i
U
n
i
v
e
r
s
i
t
y
o
f
I
n
d
u
st
r
y
,
H
a
n
o
i
C
i
t
y
,
V
i
e
t
n
a
m
2
C
e
n
t
e
r
o
f
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
y
,
H
a
n
o
i
U
n
i
v
e
r
si
t
y
o
f
I
n
d
u
s
t
r
y
,
H
a
n
o
i
C
i
t
y
,
V
i
e
t
n
a
m
3
F
a
c
u
l
t
y
o
f
El
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
,
H
a
n
o
i
U
n
i
v
e
r
s
i
t
y
o
f
I
n
d
u
st
r
y
,
H
a
n
o
i
C
i
t
y
,
V
i
e
t
n
a
m
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
y
1
3
,
2
0
2
4
R
ev
is
ed
Oct
2
2
,
2
0
2
4
Acc
ep
ted
Oct
3
0
,
2
0
2
4
Ag
ricu
lt
u
re
p
lay
s
a
c
r
u
c
ial
r
o
le
in
d
e
v
e
lo
p
in
g
c
o
u
n
tri
e
s
s
u
c
h
a
s
Vie
tn
a
m
,
wh
e
re
7
0
p
e
rc
e
n
t
o
f
t
h
e
p
o
p
u
l
a
ti
o
n
is
e
m
p
lo
y
e
d
in
a
g
ricu
lt
u
r
e
,
a
n
d
5
7
p
e
rc
e
n
t
o
f
th
e
so
c
ial
lab
o
r
f
o
rc
e
wo
rk
s
in
t
h
e
a
g
ricu
lt
u
ra
l
se
c
to
r.
Th
e
re
fo
re
,
c
ro
p
p
r
o
d
u
c
ti
v
it
y
d
irec
tl
y
a
ffe
c
ts
th
e
li
v
e
s
o
f
m
a
n
y
p
e
o
p
le.
O
n
e
o
f
t
h
e
p
rima
ry
re
a
so
n
s
fo
r
re
d
u
c
e
d
c
ro
p
y
ield
s
is
p
lan
t
lea
f
d
ise
a
se
s
c
a
u
se
d
b
y
b
a
c
teria
,
fu
n
g
i,
a
n
d
v
ir
u
se
s.
He
n
c
e
,
th
e
re
is
a
n
e
e
d
f
o
r
a
m
e
th
o
d
t
o
h
e
l
p
fa
rm
e
rs i
d
e
n
ti
fy
lea
f
d
ise
a
se
s e
a
rl
y
to
ta
k
e
a
p
p
r
o
p
riate
a
c
ti
o
n
t
o
p
r
o
tec
t
c
ro
p
s
a
n
d
sh
ift
to
sm
a
rt
a
g
ricu
l
tu
ra
l
p
r
o
d
u
c
t
io
n
.
T
h
is
p
a
p
e
r
p
ro
p
o
se
s
li
g
h
twe
ig
h
t
d
e
e
p
lea
rn
in
g
(DL)
m
o
d
e
ls
c
o
m
b
i
n
e
d
with
a
su
p
p
o
r
t
v
e
c
to
r
m
a
c
h
in
e
(S
VM),
with
h
y
p
e
rp
a
r
a
m
e
ters
fin
e
-
tu
n
e
d
b
y
c
h
e
m
ica
l
re
a
c
ti
o
n
o
p
ti
m
iza
ti
o
n
(CRO)
,
fo
r
d
e
tec
ti
n
g
p
la
n
t
lea
f
d
ise
a
se
s.
Th
e
m
a
in
a
d
v
a
n
tag
e
o
f
th
e
m
e
th
o
d
is
th
e
sim
p
li
c
it
y
o
f
t
h
e
a
rc
h
it
e
c
tu
re
a
n
d
o
p
ti
m
iza
ti
o
n
o
f
t
h
e
D
L
m
o
d
e
l
’
s
h
y
p
e
r
p
a
ra
m
e
ters
,
m
a
k
in
g
it
e
a
sil
y
d
e
p
l
o
y
a
b
le
o
n
lo
w
h
a
r
d
wa
re
d
e
v
ice
s.
To
tes
t
th
e
p
e
rfo
rm
a
n
c
e
o
f
th
e
p
r
o
p
o
se
d
m
e
th
o
d
,
e
x
p
e
rime
n
ts
a
re
p
e
rfo
rm
e
d
o
n
th
e
P
lan
tVi
ll
a
g
e
d
a
tas
e
t
u
si
n
g
P
y
th
o
n
.
T
h
e
su
p
e
ri
o
rit
y
o
f
th
e
p
ro
p
o
se
d
m
e
th
o
d
o
v
e
r
th
e
we
ll
-
k
n
o
wn
v
isu
a
l
g
e
o
m
e
try
g
r
o
u
p
-
1
6
(VG
G
-
1
6
)
a
n
d
M
o
b
i
leN
e
tV2
m
o
d
e
ls
is
d
e
m
o
n
s
trate
d
b
y
a
1
0
%
in
c
re
a
se
in
a
c
c
u
ra
c
y
p
re
d
ictio
n
a
n
d
a
d
e
c
re
a
se
o
f
5
%
a
n
d
6
6
%
i
n
train
in
g
ti
m
e
,
re
sp
e
c
ti
v
e
ly
.
K
ey
w
o
r
d
s
:
Ag
r
icu
ltu
r
e
CRO
Dee
p
lear
n
in
g
m
o
d
el
Hy
p
er
p
ar
a
m
eter
s
o
p
tim
izatio
n
L
ea
f
d
is
ea
s
es
SVM
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
:
Ng
h
ien
Ng
u
y
en
B
a
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
E
n
g
i
n
ee
r
in
g
,
Facu
lty
o
f
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
Ha
n
o
i U
n
iv
e
r
s
ity
o
f
I
n
d
u
s
tr
y
2
9
8
C
au
Dien
Stre
et,
Han
o
i,
Vietn
am
E
m
ail: n
g
u
y
e
n
b
an
g
h
ien
_
c
n
tt
@
h
au
i.e
d
u
.
v
n
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
2
0
2
3
g
lo
b
al
r
ep
o
r
t
o
n
f
o
o
d
cr
is
es
s
h
o
ws
th
at
th
er
e
i
s
a
g
r
o
win
g
n
ee
d
f
o
r
f
o
o
d
an
d
liv
elih
o
o
d
ass
is
tan
ce
.
T
h
e
r
ep
o
r
t
r
ev
ea
ls
th
at
o
v
er
2
5
0
m
illi
o
n
in
d
iv
i
d
u
als
ar
e
f
ac
ed
with
ac
u
te
h
u
n
g
er
.
I
n
2
0
2
2
,
th
e
co
u
n
t
o
f
in
d
iv
id
u
als
ex
p
er
ien
cin
g
s
ev
er
e
f
o
o
d
in
s
ec
u
r
ity
at
cr
is
is
o
r
m
o
r
e
s
ev
er
e
lev
els
r
ea
ch
ed
2
5
8
m
illi
o
n
acr
o
s
s
5
8
co
u
n
tr
ies
an
d
ter
r
i
to
r
ies,
a
n
o
tab
le
in
cr
ea
s
e
f
r
o
m
1
9
3
m
illi
o
n
in
d
i
v
id
u
als
in
5
3
co
u
n
tr
ies
an
d
ter
r
ito
r
ies
in
2
0
2
1
.
On
e
o
f
t
h
e
p
r
im
a
r
y
r
ea
s
o
n
s
f
o
r
r
ed
u
c
ed
cr
o
p
y
ield
s
is
p
lan
t
leaf
d
is
ea
s
es
ca
u
s
ed
b
y
b
ac
ter
ia,
f
u
n
g
i,
an
d
v
ir
u
s
es.
B
ec
au
s
e,
p
lan
t
leaf
d
is
ea
s
es
o
f
ten
af
f
ec
t
th
e
ch
lo
r
o
p
h
y
ll
co
n
ten
t
an
d
s
tr
u
ctu
r
e
o
f
leav
es,
wh
ich
ar
e
cr
u
cial
f
o
r
p
h
o
to
s
y
n
t
h
esis
.
R
ed
u
ce
d
p
h
o
t
o
s
y
n
th
esis
lead
s
to
a
d
ec
r
ea
s
e
in
th
e
v
,
u
ltima
tely
af
f
ec
tin
g
its
g
r
o
wth
an
d
y
ield
.
I
n
a
d
d
itio
n
,
m
an
y
leaf
d
is
ea
s
es
ca
u
s
e
s
y
m
p
to
m
s
s
u
c
h
as
n
ec
r
o
s
is
,
wiltin
g
,
o
r
p
r
em
atu
r
e
d
ea
th
o
f
lea
v
es.
T
h
is
r
esu
lts
in
a
r
ed
u
ctio
n
o
f
th
e
to
tal
leaf
ar
ea
av
ailab
le
f
o
r
p
h
o
to
s
y
n
th
esis
.
T
h
e
lo
s
s
o
f
leav
es
ca
n
s
ig
n
if
ican
t
ly
im
p
ac
t
th
e
p
lan
t
’
s
a
b
ilit
y
to
ca
p
tu
r
e
s
u
n
li
g
h
t
a
n
d
co
n
v
er
t
it
in
to
en
e
r
g
y
.
T
h
er
ef
o
r
e
,
p
r
o
tectin
g
cr
o
p
s
a
n
d
in
cr
ea
s
in
g
cr
o
p
p
r
o
d
u
ctiv
i
ty
is
an
ex
tr
em
ely
im
p
o
r
tan
t
task
.
C
u
r
r
en
tly
in
Vietn
am
,
th
e
wid
esp
r
ea
d
u
s
e
o
f
p
esti
cid
es
lead
s
to
af
f
ec
tin
g
th
e
h
ea
lth
o
f
f
ar
m
e
r
s
,
p
o
llu
t
in
g
lan
d
an
d
wate
r
s
o
u
r
ce
s
,
an
d
r
e
d
u
cin
g
th
e
q
u
ality
o
f
ag
r
ic
u
ltu
r
al
p
r
o
d
u
cts.
I
n
ad
d
itio
n
,
th
e
ev
o
lu
tio
n
o
f
clim
ate
ch
an
g
e
is
v
er
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
8
,
No
.
1
,
Ap
r
il
20
2
5
:
4
4
7
-
4
5
8
448
co
m
p
licated
,
lead
in
g
to
in
cr
e
asin
g
ly
s
er
io
u
s
d
is
ea
s
es
r
elate
d
to
p
lan
t
leav
es.
I
n
p
ar
ticu
la
r
,
d
is
ea
s
es
o
n
cr
o
p
leav
es a
r
e
v
er
y
c
o
m
m
o
n
an
d
s
er
io
u
s
ly
af
f
ec
t c
r
o
p
p
r
o
d
u
ctiv
i
ty
[
1
]
.
B
ased
o
n
[
2
]
,
p
lan
t
leaf
d
is
ea
s
es
p
o
s
e
a
s
ig
n
if
ican
t
ch
allen
g
e
in
r
ice
p
r
o
d
u
ctio
n
,
with
th
e
p
o
ten
tial
to
ad
v
er
s
ely
af
f
ec
t
t
h
e
cr
o
p
an
d
lead
to
a
d
ec
r
ea
s
e
in
y
ield
.
Gen
ith
a
et
a
l.
[
3
]
c
o
m
m
en
t
f
ar
m
er
s
h
av
e
d
i
f
f
icu
lty
d
etec
tin
g
an
d
class
if
y
in
g
d
is
e
ases
o
n
leav
es.
R
ely
in
g
o
n
tr
ad
itio
n
al
p
h
y
s
ical
o
b
s
er
v
atio
n
m
eth
o
d
s
f
o
r
d
is
ea
s
e
id
en
tific
atio
n
an
d
ca
teg
o
r
izat
io
n
in
ag
r
icu
ltu
r
e
is
f
r
e
q
u
en
t
ly
u
n
r
eliab
le
an
d
co
u
ld
lea
d
to
a
co
n
s
id
er
a
b
le
d
ec
lin
e
d
u
r
i
n
g
p
r
o
d
u
ctio
n
.
T
h
er
ef
o
r
e
,
an
a
d
v
an
ce
d
ap
p
r
o
a
ch
is
n
ec
ess
ar
y
to
ass
is
t
f
ar
m
er
s
in
q
u
ick
l
y
an
d
ac
cu
r
ately
d
etec
tin
g
p
lan
t
lea
f
d
is
ea
s
es.
R
ec
en
tly
,
m
an
y
r
esear
ch
er
s
h
a
v
e
f
o
cu
s
ed
o
n
ap
p
l
y
in
g
d
ee
p
lear
n
in
g
(
DL
)
an
d
m
ac
h
in
e
lear
n
in
g
(
ML
)
to
cr
ea
te
m
o
d
els
to
r
ec
o
g
n
ize
h
ig
h
l
y
ac
cu
r
ate
id
en
tifi
ca
tio
n
o
f
p
lan
t
leaf
d
is
ea
s
es
[
4
]
.
R
am
esh
et
a
l.
[
5
]
u
s
e
th
e
r
an
d
o
m
f
o
r
est
(
R
F)
to
class
if
y
th
e
h
ea
lth
y
an
d
d
i
s
ea
s
ed
leav
es
f
r
o
m
th
eir
co
llected
d
ata
s
et
.
Firstl
y
,
th
ey
u
s
ed
th
e
h
is
to
g
r
am
o
f
an
o
r
ien
ted
g
r
ad
ien
t
to
e
x
tr
a
ct
im
ag
e
f
ea
tu
r
es.
T
h
en
,
th
e
y
tr
ain
ed
th
e
R
F m
o
d
el
with
th
e
im
ag
e
f
ea
tu
r
es.
Sh
ar
m
a
e
t
a
l.
[
6
]
p
r
esen
t
im
a
g
e
s
eg
m
en
tatio
n
co
n
d
u
cted
to
d
elin
ea
te
leav
es
f
r
o
m
th
e
b
ac
k
g
r
o
u
n
d
.
T
h
e
s
eg
m
en
tatio
n
p
r
o
ce
s
s
em
p
lo
y
s
K
-
m
ea
n
s
clu
s
ter
in
g
with
two
clu
s
ter
ce
n
ter
s
,
d
es
ig
n
atin
g
o
n
e
f
o
r
th
e
b
ac
k
g
r
o
u
n
d
an
d
th
e
o
t
h
er
f
o
r
th
e
f
o
r
eg
r
o
u
n
d
.
T
h
en
,
th
e
p
ix
els
o
f
t
h
e
b
ac
k
g
r
o
u
n
d
im
ag
e
a
r
e
ch
an
g
ed
to
b
lack
to
elim
in
ate
ir
r
elev
an
t
in
f
o
r
m
atio
n
an
d
en
h
an
ce
p
r
ed
ictio
n
ac
cu
r
ac
y
.
Fin
ally
,
s
o
m
e
ML
alg
o
r
ith
m
s
s
u
ch
as
lo
g
is
tic
r
eg
r
ess
io
n
,
K
-
n
ea
r
est
n
eig
h
b
o
r
s
(
KNN)
,
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM
)
ar
e
i
m
p
lem
en
ted
f
o
r
th
e
class
if
icatio
n
o
f
p
la
n
t
leaf
d
i
s
ea
s
es.
T
h
e
lo
g
is
tic
r
eg
r
ess
io
n
ac
h
iev
e
d
t
h
e
b
est
class
if
icatio
n
ac
cu
r
ac
y
o
f
6
6
.
4
%.
Gen
ith
a
et
a
l.
[
3
]
p
r
o
p
o
s
ed
a
f
u
s
io
n
m
eth
o
d
f
o
r
t
h
e
class
if
icatio
n
o
f
p
lan
t
leaf
d
is
ea
s
es.
Fir
s
t,
leaf
im
ag
es
ar
e
p
r
e
-
p
r
o
ce
s
s
ed
b
y
t
r
an
s
f
o
r
m
in
g
th
e
co
lo
r
im
ag
e
t
o
a
g
r
ay
s
ca
le
im
ag
e.
T
h
en
,
n
o
is
e
is
r
em
o
v
ed
b
y
th
e
m
ed
ian
f
ilter
,
ed
g
e,
a
n
d
d
ir
ec
tio
n
o
f
th
e
p
la
n
t
leaf
a
r
e
d
etec
ted
b
y
th
e
So
b
el
f
ilter
,
a
n
d
th
e
Ga
b
o
r
f
ilter
r
esp
ec
tiv
ely
.
Nex
t,
th
e
p
r
im
a
r
y
ch
a
r
ac
ter
is
tics
o
f
th
e
leaf
im
ag
e
ar
e
ex
tr
ac
ted
b
y
th
e
p
r
in
cip
al
co
m
p
o
n
e
n
t
an
aly
s
is
(
PC
A
)
tech
n
iq
u
e.
Fin
ally
,
th
e
p
r
in
ci
p
al
ch
ar
ac
ter
is
tics
f
ed
to
th
e
SVM
f
o
r
class
if
icatio
n
p
u
r
p
o
s
es.
T
h
e
ex
p
er
im
en
t
r
esu
lts
d
em
o
n
s
tr
ate
th
at
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
ac
h
iev
es
a
p
r
ed
ictio
n
ac
cu
r
ac
y
o
f
9
0
.
6
6
%.
Me
n
g
is
tu
et
a
l.
[
7
]
u
tili
ze
b
ac
k
p
r
o
p
ag
atio
n
n
e
u
r
al
n
etwo
r
k
s
an
d
d
ec
is
io
n
tr
ee
s
(
DT
)
to
d
etec
t
th
r
ee
p
r
im
ar
y
d
is
ea
s
es a
f
f
ec
tin
g
co
f
f
ee
tr
ee
s
.
T
h
e
ac
cu
r
ac
y
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
r
ea
ch
e
d
9
4
.
5
%.
Sh
ar
m
a
et
a
l.
[
8
]
u
tili
ze
d
a
b
len
d
o
f
im
a
g
e
p
r
o
ce
s
s
in
g
m
eth
o
d
s
an
d
v
ar
io
u
s
d
ata
m
in
in
g
alg
o
r
ith
m
s
s
u
ch
as
KNN,
SV
M,
R
F,
an
d
DT
,
f
o
r
th
e
tim
ely
id
en
tifi
ca
tio
n
o
f
r
ice
p
la
n
t
ailm
en
ts
.
T
h
e
r
esu
lts
o
f
t
h
e
ex
p
er
im
en
t
d
em
o
n
s
tr
ate
th
e
R
F
alg
o
r
ith
m
ac
h
iev
es
th
e
h
ig
h
es
t
ac
cu
r
ac
y
p
r
ed
ictio
n
,
r
ea
c
h
in
g
9
0
%.
M
o
h
an
t
y
et
a
l.
[
9
]
u
s
e
two
well
-
k
n
o
w
n
DL
ar
c
h
itectu
r
es,
n
am
ely
Ale
x
Net
an
d
Go
o
g
L
eNe
t,
t
o
id
e
n
tify
p
lan
t
leaf
d
is
ea
s
es.
T
h
ey
u
tili
ze
im
ag
es
f
r
o
m
th
e
Plan
tVillag
e
d
ataset
f
o
r
tr
ain
in
g
an
d
test
in
g
tr
an
s
f
er
m
o
d
els,
as
well
as
f
o
r
tr
ain
in
g
m
o
d
els
f
r
o
m
s
cr
atch
.
T
h
ey
test
ed
s
ce
n
ar
i
o
s
,
in
clu
d
i
n
g
co
l
o
r
im
a
g
e
s
,
g
r
a
y
s
ca
le
im
ag
es,
an
d
s
eg
m
en
ted
i
m
a
g
e
s
,
w
i
t
h
t
r
ai
n
i
n
g
a
n
d
t
e
s
ti
n
g
d
a
t
a
p
e
r
c
e
n
t
a
g
e
s
o
f
8
0
%
-
2
0
%
,
6
0
%
-
4
0
%
,
a
n
d
5
0
%
-
5
0
%
,
as
w
el
l
as
2
0
%
-
8
0
%
.
T
h
e
ex
p
er
im
en
tal
f
in
d
in
g
s
s
h
o
w
th
at
Go
o
g
L
eNe
t,
with
tr
an
s
f
er
lear
n
in
g
an
d
a
tr
ain
in
g
-
test
in
g
s
et
r
atio
o
f
8
0
%
-
2
0
%,
a
ch
iev
ed
th
e
h
ig
h
est
ac
cu
r
ac
y
p
r
ed
ictio
n
o
f
9
9
.
3
5
%.
An
d
r
ew
et
a
l.
[
1
0
]
co
n
d
u
cted
ex
p
er
im
en
ts
u
s
in
g
th
e
Plan
tVillag
e
im
ag
e
d
ataset
with
f
o
u
r
co
m
m
o
n
DL
m
o
d
els:
r
esid
u
al
n
et
wo
r
k
(
R
esNet)
-
5
0
,
I
n
ce
p
tio
n
V4
,
v
is
u
al
g
eo
m
etr
y
g
r
o
u
p
-
1
6
(
VGG
-
1
6
)
,
an
d
De
n
s
eNe
t
-
1
2
1
,
all
p
r
e
-
tr
ain
e
d
o
n
th
e
I
m
a
g
eNe
t
d
ataset.
E
x
p
er
im
en
tal
o
u
tco
m
es
s
h
o
w
th
at
th
e
p
r
e
-
tr
ain
ed
De
n
s
eNe
t
-
1
2
1
ac
h
ie
v
ed
th
e
h
ig
h
est
class
if
icatio
n
ac
cu
r
ac
y
,
r
ea
c
h
in
g
9
9
.
8
1
%.
J
u
n
g
et
a
l.
[
1
1
]
p
r
o
p
o
s
ed
a
m
eth
o
d
f
o
r
class
if
y
in
g
cr
o
p
s
,
d
etec
tin
g
d
is
ea
s
es,
an
d
ca
teg
o
r
ical
cr
o
p
ailm
en
ts
.
T
h
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
is
ex
ec
u
ted
in
th
r
ee
s
tag
es
.
Fo
r
th
e
in
itial
s
tag
e,
a
p
r
e
-
tr
a
in
ed
m
o
d
el
is
u
s
ed
to
class
if
y
cr
o
p
s
.
Fo
r
th
e
s
u
b
s
eq
u
en
t
s
tag
e,
s
ev
er
al
m
o
d
els,
ea
ch
d
ed
icate
d
to
a
s
p
ec
if
ic
c
r
o
p
,
a
r
e
em
p
l
o
y
ed
to
d
etec
t
p
lan
t
d
is
ea
s
es
f
r
o
m
im
a
g
es.
I
n
th
e
f
in
al
s
tep
,
a
s
et
o
f
p
r
e
-
tr
ain
ed
m
o
d
els
is
u
tili
ze
d
to
class
if
y
d
is
ea
s
es
f
o
r
ea
ch
cr
o
p
.
Fiv
e
p
r
e
-
tr
ain
e
d
m
o
d
els,
in
clu
d
i
n
g
R
esNet5
0
,
Alex
Net,
Go
o
g
leNe
t,
VGG1
9
,
an
d
E
f
f
icien
tNet,
ar
e
ex
p
er
im
en
ted
with
i
n
ea
ch
s
tep
.
L
ea
f
im
ag
es
o
f
th
r
ee
cr
o
p
s
b
ell
p
ep
p
er
,
to
m
ato
,
an
d
p
o
tato
ar
e
ex
tr
ac
te
d
f
r
o
m
th
e
Plan
tVillag
e
d
ataset
u
tili
ze
d
in
th
e
ex
p
er
im
en
t.
T
h
e
r
esu
lts
o
f
th
e
ex
p
er
im
en
t
in
d
icate
th
at
th
e
p
r
e
-
tr
ain
ed
E
f
f
icien
tNet
m
o
d
el
ac
h
iev
ed
th
e
h
i
g
h
est
ac
cu
r
ac
y
o
f
9
9
.
3
3
%
an
d
9
9
.
4
0
%
in
th
e
f
ir
s
t
an
d
last
s
tep
s
,
r
esp
ec
tiv
ely
.
Ad
d
itio
n
ally
,
th
e
p
r
e
-
tr
ain
ed
Go
o
g
L
eNe
t
attain
ed
th
e
h
ig
h
est
ac
cu
r
ac
y
o
f
1
0
0
%
f
o
r
th
e
b
ell
p
ep
p
er
c
r
o
p
,
VGG1
9
y
ield
ed
th
e
b
est
ac
cu
r
ac
y
,
r
ea
ch
in
g
1
0
0
%
f
o
r
th
e
p
o
tato
c
r
o
p
,
a
n
d
R
esNet5
0
attain
ed
th
e
u
tm
o
s
t a
cc
u
r
ac
y
o
f
9
9
.
7
5
% f
o
r
th
e
t
o
m
ato
cr
o
p
in
th
e
s
ec
o
n
d
s
tep
.
Han
g
et
a
l.
[
1
2
]
s
u
g
g
ested
an
ap
p
r
o
ac
h
th
at
r
ep
lace
d
th
e
f
u
lly
co
n
n
ec
ted
lay
er
o
f
th
e
VGG1
6
m
o
d
el
with
th
e
i
n
ce
p
tio
n
an
d
s
q
u
ee
ze
-
an
d
-
e
x
citatio
n
m
o
d
u
les.
T
h
e
e
x
p
er
im
e
n
tal
f
i
n
d
in
g
s
illu
s
tr
ate
t
h
e
s
u
p
e
r
io
r
p
er
f
o
r
m
an
ce
o
f
th
eir
p
r
o
p
o
s
ed
m
o
d
el
co
m
p
ar
ed
to
o
th
er
co
m
m
o
n
s
tr
u
ctu
r
es
in
clu
d
in
g
Alex
Net,
Go
o
g
L
eN
et,
VGG1
6
,
VGG1
9
,
R
esN
et
-
5
0
,
I
n
c
ep
tio
n
v
2
,
I
n
ce
p
tio
n
v
3
,
I
n
ce
p
tio
n
v
4
,
an
d
SENe
t.
Sh
ar
m
a
et
a
l.
[
6
]
g
ath
er
ed
2
0
,
0
0
0
im
ag
es
f
ea
tu
r
i
n
g
b
o
th
h
ea
lth
y
an
d
d
is
ea
s
ed
leav
es
ac
r
o
s
s
1
9
d
if
f
er
e
n
t
class
es.
T
h
e
d
ataset
en
co
m
p
ass
es
p
r
ev
alen
t
lea
f
d
is
ea
s
es
lik
e
b
lack
r
o
t,
r
u
s
t,
b
ac
ter
ial
s
p
o
ts
,
an
d
o
th
er
s
,
af
f
ec
tin
g
v
ar
io
u
s
cr
o
p
s
s
u
ch
as
co
r
n
,
ap
p
le,
p
o
tato
,
an
d
to
m
ato
.
Su
b
s
eq
u
e
n
tly
,
th
e
im
ag
es
u
n
d
e
r
wen
t
s
eg
m
en
tatio
n
u
s
in
g
th
e
K
-
m
e
an
s
alg
o
r
ith
m
to
i
s
o
late
th
e
l
ea
v
es
an
d
b
ac
k
g
r
o
u
n
d
.
Fin
ally
,
th
e
r
esear
ch
er
s
tr
ain
ed
an
d
test
ed
s
ev
er
al
ML
m
o
d
els
an
d
a
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NNs)
u
s
in
g
th
e
s
eg
m
en
ted
d
ataset.
T
h
ey
r
ep
o
r
ted
t
h
at
th
e
C
NN
ar
ch
itectu
r
e
attain
ed
th
e
u
tm
o
s
t c
lass
if
icatio
n
ac
cu
r
ac
y
o
f
9
8
%.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
R
ec
o
g
n
itio
n
o
f p
la
n
t le
a
f d
is
ea
s
es b
a
s
ed
o
n
d
ee
p
lea
r
n
in
g
…
(
N
g
h
ien
N
g
u
ye
n
B
a
)
449
Sar
d
o
g
an
et
a
l.
[
1
3
]
s
u
g
g
este
d
an
ap
p
r
o
ac
h
th
at
c
o
m
b
in
es
C
NN
an
d
lear
n
in
g
v
ec
to
r
q
u
an
tizatio
n
(
L
VQ)
.
T
h
e
leaf
im
ag
es
’
f
e
atu
r
es
ar
e
e
x
tr
ac
ted
u
s
in
g
C
NN
an
d
t
h
en
class
if
ied
th
r
o
u
g
h
L
VQ.
Fo
r
th
e
ex
p
er
im
en
t,
5
0
0
to
m
ato
leaf
im
ag
es
wer
e
u
tili
ze
d
,
ca
teg
o
r
ize
d
in
to
h
ea
lth
y
,
b
ac
ter
ial
s
p
o
t,
late
b
r
ig
h
t,
s
ep
to
r
ia
s
p
o
t,
an
d
y
ell
o
w
cu
r
v
ed
f
r
o
m
th
e
Plan
tVillag
e
d
ataset.
T
h
e
d
ataset
u
n
d
er
we
n
t
d
iv
i
s
io
n
in
to
a
t
r
ain
in
g
s
et
(
8
0
%)
an
d
a
test
s
et
(
2
0
%)
.
T
h
e
e
x
p
er
im
e
n
t
d
ata
i
n
d
icate
th
at
th
eir
p
r
o
p
o
s
ed
m
et
h
o
d
a
ch
iev
es
an
a
v
er
ag
e
class
if
icatio
n
ac
cu
r
ac
y
o
f
8
6
%.
Nik
ith
et
a
l.
[
1
4
]
em
p
l
o
y
e
d
th
e
C
NN
m
o
d
el
to
class
if
y
leaf
d
is
ea
s
es
in
th
e
So
y
ab
ea
n
leaf
d
ataset.
T
h
ey
r
ep
o
r
ted
a
9
6
%
class
if
icatio
n
ac
cu
r
ac
y
d
u
r
in
g
th
e
tr
ain
i
n
g
p
h
ase
an
d
8
4
%
d
u
r
in
g
th
e
test
in
g
p
h
ase.
Alg
an
i
et
a
l.
[
1
5
]
p
r
o
p
o
s
ed
a
m
eth
o
d
th
at
in
teg
r
ates
an
t
c
o
lo
n
y
o
p
tim
izatio
n
(
AC
O)
an
d
C
NN
f
o
r
th
e
d
etec
tio
n
an
d
cl
ass
if
icatio
n
o
f
p
lan
t
leaf
d
is
ea
s
es.
I
n
th
is
ap
p
r
o
ac
h
,
AC
O
is
u
tili
ze
d
f
o
r
f
ea
t
u
r
e
ex
tr
ac
tio
n
,
w
h
ile
C
NN
is
em
p
lo
y
ed
to
d
if
f
er
en
tiate
b
etwe
en
h
ea
lth
y
an
d
d
is
ea
s
ed
leav
es
.
T
h
e
e
x
p
er
im
e
n
tal
d
ata
d
em
o
n
s
tr
ate
th
at
th
e
s
u
g
g
ested
ap
p
r
o
ac
h
attain
e
d
th
e
h
ig
h
est
ac
cu
r
ac
y
in
clas
s
if
icatio
n
,
r
ea
ch
in
g
9
9
.
9
8
%,
s
u
r
p
ass
in
g
p
o
p
u
lar
m
o
d
els
lik
e
g
en
er
ativ
e
ad
v
e
r
s
ar
ial
n
etwo
r
k
(
GAN
)
a
n
d
C
N
N.
Naw
az
et
a
l.
[
1
6
]
em
p
lo
y
ed
a
c
o
m
b
in
atio
n
o
f
R
esNet
-
3
4
,
th
e
co
n
v
o
lu
tio
n
b
lo
ck
atten
tio
n
m
o
d
u
le
(
C
B
AM
)
,
an
d
f
aster
r
eg
io
n
-
b
ased
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
e
two
r
k
(
f
aster
-
R
C
NN
)
to
lo
ca
te
th
e
r
e
g
io
n
an
d
class
if
y
leaf
d
is
ea
s
es
in
to
m
ato
cr
o
p
s
.
R
esNet
-
3
4
,
in
co
n
ju
n
c
tio
n
with
C
B
AM
,
i
s
r
esp
o
n
s
i
b
le
f
o
r
e
x
tr
ac
tin
g
f
ea
tu
r
es
f
r
o
m
th
e
leaf
d
is
ea
s
e
i
m
a
g
e
s
.
M
e
an
w
h
i
l
e,
f
a
s
t
e
r
-
R
C
N
N
i
s
p
i
v
o
t
a
l
i
n
l
o
ca
l
i
z
i
n
g
a
n
d
ca
t
e
g
o
r
i
z
in
g
th
e
id
e
n
t
if
i
e
d
l
ea
f
d
i
s
e
a
s
e
s
.
T
h
e
ex
p
er
i
m
en
t
a
l
o
u
t
co
m
e
s
h
i
g
h
l
i
g
h
t
t
h
a
t
th
e
i
r
s
u
g
g
e
s
t
ed
a
p
p
r
o
a
c
h
a
c
h
i
ev
ed
a
n
im
p
r
e
s
s
i
v
e
p
r
e
d
i
ct
i
o
n
a
c
c
u
r
a
c
y
o
f
9
9
.
7
%
.
Alb
attah
et
a
l.
[
1
7
]
s
u
g
g
ested
a
m
o
d
if
ied
C
en
ter
Net
m
o
d
el
b
y
r
ep
lacin
g
R
esNet
-
1
0
1
in
th
e
o
r
ig
i
n
al
m
o
d
el
with
Den
s
eNe
t
-
7
7
.
T
h
ey
u
tili
ze
d
th
e
m
o
d
if
ied
C
en
ter
Net
to
d
etec
t
r
eg
io
n
s
an
d
class
if
y
v
ar
io
u
s
leaf
d
is
ea
s
es.
E
x
p
er
im
en
tal
o
u
tco
m
es
r
e
p
o
r
te
d
th
at
th
e
s
u
g
g
ested
m
o
d
el
ac
h
iev
e
d
an
m
ea
n
av
er
ag
e
p
r
ec
is
io
n
(
m
AP
)
o
f
0
.
9
9
an
d
a
m
ea
n
in
ter
s
ec
tio
n
o
v
er
u
n
io
n
(
I
OU
)
o
f
0
.
9
9
3
.
A
n
ar
i
et
a
l
.
[
1
8
]
,
m
o
d
if
ied
th
e
o
r
ig
in
al
r
esid
u
al
b
lo
ck
b
y
i
n
co
r
p
o
r
atin
g
weig
h
ts
f
o
r
th
e
f
ee
d
b
ac
k
b
l
o
ck
.
Ad
d
itio
n
ally
,
th
e
a
u
th
o
r
em
p
lo
y
ed
m
u
ltip
le
SVM
m
o
d
els
f
o
r
class
if
icati
o
n
p
u
r
p
o
s
es,
u
s
in
g
a
p
o
r
tio
n
o
f
th
e
Plan
tVillag
e
im
ag
es
d
ataset
f
o
r
th
e
ex
p
er
im
en
t.
T
h
e
r
ep
o
r
ted
r
esu
lts
in
d
icate
th
at
th
e
p
r
o
p
o
s
ed
m
o
d
el
ac
h
iev
e
d
a
p
r
ed
ictio
n
a
cc
u
r
ac
y
o
f
9
9
.
1
%.
B
alaji
e
t
a
l.
[
1
9
]
em
p
lo
y
e
d
a
g
en
etic
alg
o
r
ith
m
(
GA)
an
d
a
p
r
e
-
tr
ain
ed
C
NN
f
o
r
th
e
class
if
icatio
n
o
f
p
lan
t
leaf
d
is
ea
s
es.
T
h
e
GA
was
u
tili
ze
d
f
o
r
d
en
o
is
in
g
a
n
d
f
ea
t
u
r
e
ex
tr
ac
tio
n
,
wh
ile
t
h
e
p
r
e
-
tr
ain
ed
R
esNet
-
5
0
an
d
VGG
-
1
6
m
o
d
els
aid
e
d
in
cla
s
s
if
y
in
g
th
e
id
en
tifie
d
leaf
d
i
s
ea
s
es.
A
cc
o
r
d
in
g
to
th
eir
r
e
p
o
r
t,
th
e
p
r
o
p
o
s
ed
m
eth
o
d
attain
ed
a
class
if
icati
o
n
ac
cu
r
ac
y
o
f
9
5
%.
Ag
ar
wal
et
a
l.
[
2
0
]
,
p
r
o
p
o
s
ed
th
e
u
tili
za
tio
n
o
f
C
NN
f
o
r
th
e
class
if
icatio
n
o
f
t
o
m
ato
leaf
d
is
ea
s
es.
T
h
e
r
ep
o
r
ted
r
e
s
u
lts
in
d
icate
a
clas
s
if
icatio
n
ac
cu
r
ac
y
o
f
9
1
.
2
%.
E
zh
ilar
asan
et
a
l.
[
2
1
]
p
r
o
p
o
s
ed
a
m
et
h
o
d
u
s
in
g
a
DL
R
esNet
m
o
d
el
with
p
ar
am
eter
s
o
p
tim
ized
b
y
t
h
e
jelly
f
is
h
o
p
tim
izatio
n
alg
o
r
ith
m
f
o
r
th
e
class
if
icatio
n
o
f
to
m
ato
leaf
d
is
ea
s
e.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
ev
al
u
ated
b
ased
o
n
th
e
class
if
icatio
n
o
f
to
m
ato
p
lan
t
leaf
d
is
ea
s
e.
T
h
e
e
x
p
er
im
en
tal
r
esu
lts
s
h
o
w
th
at
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
ac
h
iev
es
a
class
if
icatio
n
ac
cu
r
ac
y
o
f
9
7
.
3
%.
Pu
jar
et
a
l
.
[2
2
]
,
p
r
o
p
o
s
ed
th
e
en
h
an
ce
d
KNN
f
o
r
th
e
class
if
icatio
n
o
f
co
r
n
leaf
d
is
ea
s
es.
T
h
e
r
e
p
o
r
ted
r
esu
lts
in
d
ic
ate
a
c
lass
if
icatio
n
ac
cu
r
ac
y
o
f
9
9
.
8
6
%.
Ho
wev
er
,
alm
o
s
t
all
r
esear
ch
e
r
s
h
av
e
em
p
lo
y
ed
th
e
c
o
m
p
le
x
ar
ch
itectu
r
e
o
f
C
NNs,
wh
ich
in
v
o
lv
e
a
lar
g
e
n
u
m
b
e
r
o
f
p
ar
am
eter
s
.
C
o
n
s
eq
u
en
tly
,
p
r
ev
i
o
u
s
ly
p
r
o
p
o
s
ed
s
o
lu
tio
n
s
ar
e
ch
allen
g
i
n
g
to
im
p
lem
en
t
o
n
h
ar
d
war
e
with
lim
ited
p
er
f
o
r
m
an
ce
,
s
u
ch
as
m
o
b
ile
a
n
d
i
n
ter
n
et
o
f
th
i
n
g
s
(
I
o
T
)
d
ev
ice
s
.
Ad
d
itio
n
ally
,
th
e
h
y
p
er
p
ar
am
eter
s
o
f
C
NNs
ar
e
o
f
ten
ch
o
s
en
b
ased
o
n
ex
p
er
ien
ce
,
m
ak
in
g
it
d
if
f
icu
lt
t
o
ac
h
iev
e
o
p
tim
al
m
o
d
els.
I
n
th
is
s
tu
d
y
,
a
lig
h
tweig
h
t
C
NN
m
o
d
el
is
d
esig
n
ed
f
o
r
ea
s
y
in
t
e
g
r
atio
n
in
t
o
lo
w
-
p
er
f
o
r
m
an
ce
h
ar
d
war
e
d
e
v
ices
an
d
h
ig
h
-
a
cc
u
r
ac
y
p
r
e
d
ictio
n
.
I
n
ad
d
iti
o
n
,
a
n
o
v
el
m
eth
o
d
th
at
u
tili
ze
s
th
e
ch
em
ical
r
ea
ctio
n
o
p
tim
izatio
n
(
C
R
O)
alg
o
r
ith
m
to
o
p
tim
ize
t
h
e
tu
n
ab
le
p
ar
a
m
eter
s
o
f
th
e
C
NN
an
d
SVM
m
o
d
el
is
in
tr
o
d
u
ce
d
.
Fu
r
t
h
er
m
o
r
e
,
to
e
n
h
an
ce
class
if
icatio
n
ac
cu
r
ac
y
,
th
e
last
lay
er
o
f
th
e
tr
ain
e
d
C
NN
is
r
ep
lace
d
b
y
an
SVM
m
o
d
el.
I
n
itially
,
im
ag
es
d
ep
ictin
g
p
lan
t
leaf
d
is
e
ases
ex
tr
ac
ted
f
r
o
m
t
h
e
Plan
tVillag
e
d
ataset
ar
e
em
p
lo
y
ed
t
o
tr
ain
an
d
e
v
alu
at
e
th
e
p
r
o
p
o
s
ed
DL
m
o
d
el.
Su
b
s
eq
u
en
tly
,
th
e
f
in
al
lay
e
r
o
f
th
e
tr
ain
ed
DL
m
o
d
el
is
s
u
b
s
titu
ted
with
an
SVM
f
o
r
en
h
a
n
ce
d
class
if
icatio
n
ac
c
u
r
ac
y
.
T
h
e
h
y
p
er
p
a
r
am
eter
s
o
f
b
o
th
th
e
DL
an
d
SVM
m
o
d
els ar
e
f
in
e
-
tu
n
ed
u
s
in
g
th
e
C
R
O
alg
o
r
ith
m
.
T
h
e
p
r
i
n
cip
le
o
f
th
e
p
r
o
p
o
s
e
d
m
eth
o
d
is
p
r
esen
ted
in
s
ec
tio
n
2
.
T
h
e
e
x
p
er
im
e
n
tal
r
e
s
u
lts
an
d
d
is
cu
s
s
io
n
ar
e
p
r
esen
ted
in
s
ec
tio
n
3
.
T
h
e
g
o
al
o
f
th
e
s
tu
d
y
is
r
eiter
ated
,
th
e
f
in
d
in
g
s
ar
e
s
u
m
m
ar
ized
,
th
e
s
ig
n
if
ican
ce
o
f
th
e
f
in
d
in
g
s
is
d
is
cu
s
s
ed
,
an
d
f
u
tu
r
e
wo
r
k
is
o
u
tlin
ed
in
th
e
c
o
n
clu
s
io
n
s
ec
tio
n
.
2.
M
E
T
H
O
D
2
.
1
.
Co
nv
o
lutio
n neura
l net
wo
rk
s
T
h
e
s
tr
u
ctu
r
e
o
f
th
e
DL
n
etwo
r
k
illu
s
tr
ated
in
Fig
u
r
e
1
in
c
lu
d
es
co
n
v
o
lu
tio
n
al,
p
o
o
lin
g
,
an
d
f
u
lly
co
n
n
ec
ted
la
y
er
s
.
As
th
e
C
NN
p
r
o
g
r
ess
es
th
r
o
u
g
h
its
lay
er
s
,
its
co
m
p
lex
ity
g
r
o
ws,
allo
win
g
it
to
d
etec
t
lar
g
er
ar
ea
s
o
f
th
e
im
ag
e.
T
h
e
in
itial
lay
er
s
co
n
ce
n
tr
ate
o
n
b
asic
ch
ar
ac
ter
is
tics
s
u
ch
as
co
lo
r
s
a
n
d
ed
g
es,
g
r
a
d
u
ally
ad
v
an
cin
g
to
r
ec
o
g
n
ize
m
o
r
e
s
u
b
s
tan
tial
elem
en
ts
o
r
s
h
ap
es
o
f
th
e
o
b
ject.
Ultim
ately
,
th
e
C
NN
r
ea
ch
es
a
p
o
in
t
wh
er
e
it
s
u
cc
ess
f
u
lly
id
en
tifie
s
th
e
in
ten
d
ed
o
b
ject.
T
h
e
co
n
v
o
l
u
tio
n
al
lay
er
s
s
er
v
e
as
r
em
o
v
in
g
n
o
is
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
8
,
No
.
1
,
Ap
r
il
20
2
5
:
4
4
7
-
4
5
8
450
an
d
b
o
u
n
d
ar
y
d
etec
to
r
s
,
wh
er
ea
s
th
e
p
o
o
lin
g
lay
e
r
s
p
er
f
o
r
m
c
o
m
p
u
tatio
n
s
to
o
b
tain
eith
er
av
e
r
ag
e
o
r
m
ax
im
u
m
lo
ca
l
v
alu
es,
r
ed
u
ci
n
g
th
e
s
ize
o
f
th
e
im
ag
e.
Fu
lly
co
n
n
ec
ted
lay
er
s
p
er
f
o
r
m
clas
s
if
icatio
n
task
s
.
T
h
e
co
n
v
o
lu
tio
n
al
la
y
er
p
er
f
o
r
m
s
a
s
u
m
m
ar
izatio
n
o
f
th
e
elem
en
t
-
wis
e
p
r
o
d
u
ct
o
f
tw
o
m
atr
ices.
T
h
e
f
ir
s
t
m
atr
ix
r
ep
r
esen
ts
a
p
o
r
tio
n
o
f
th
e
in
p
u
t,
wh
ile
th
e
s
ec
o
n
d
m
atr
ix
co
r
r
esp
o
n
d
s
to
t
h
e
k
er
n
el.
Fig
u
r
e
2
illu
s
tr
ates
th
e
ca
lcu
latio
n
p
r
o
c
ess
o
f
th
e
co
n
v
o
lu
tio
n
al
lay
e
r
.
T
h
e
p
o
o
lin
g
ca
lcu
latio
n
is
illu
s
tr
ated
in
Fig
u
r
e
3
,
wh
er
e
th
e
s
tr
id
e
d
eter
m
in
es
h
o
w
q
u
ick
ly
th
e
k
er
n
el
m
o
v
es
b
o
th
h
o
r
izo
n
tally
a
n
d
v
er
ticall
y
ac
r
o
s
s
th
e
p
ix
els
o
f
th
e
in
p
u
t im
ag
e
d
u
r
i
n
g
co
n
v
o
lu
tio
n
a
n
d
p
o
o
lin
g.
Fig
u
r
e
1
.
T
h
e
g
e
n
er
al
C
NN
s
tr
u
ctu
r
es
Fig
u
r
e
2
.
T
h
e
p
r
i
n
cip
le
o
f
th
e
co
n
v
o
l
u
tio
n
o
p
er
atio
n
Fig
u
r
e
3
.
T
h
e
p
r
i
n
cip
le
o
f
th
e
p
o
o
lin
g
p
r
o
ce
s
s
2
.
2
.
Su
pp
o
rt
v
ec
t
o
r
ma
chine
SVM
ar
e
wid
ely
u
tili
ze
d
in
th
e
d
o
m
ain
o
f
DL
f
o
r
class
if
icatio
n
p
u
r
p
o
s
es.
SVM
aim
s
t
o
f
in
d
th
e
o
p
tim
al
h
y
p
er
p
lan
e
t
h
at
ef
f
e
ctiv
ely
s
ep
ar
ates
d
ata
in
to
t
wo
ca
teg
o
r
ies,
with
th
e
m
a
r
g
in
r
e
p
r
esen
tin
g
th
e
m
ax
im
u
m
d
is
tan
ce
b
etwe
en
t
h
e
h
y
p
e
r
p
lan
e
an
d
th
e
clo
s
est
s
am
p
le.
Fig
u
r
e
4
illu
s
tr
ates
a
h
y
p
er
p
lan
e
alo
n
g
with
its
ass
o
ciate
d
m
ar
g
in
.
L
et
’
s
co
n
s
id
er
a
tr
ain
in
g
s
et
co
m
p
r
is
in
g
n
s
am
p
les
=
{
(
1
,
1
)
,
(
2
,
2
)
,
…
,
(
,
)
}
,
wh
er
e
ea
ch
r
ep
r
esen
ts
a
v
ec
to
r
in
a
d
-
d
im
en
s
io
n
al
s
p
ac
e,
an
d
∈
{
−
1
,
1
}
d
en
o
tes
th
e
co
r
r
esp
o
n
d
in
g
lab
els.
A
h
y
p
er
p
lan
e
th
at
p
a
r
titi
o
n
s
X
in
to
two
r
eg
i
o
n
s
is
ex
p
r
ess
ed
b
y
th
e
eq
u
atio
n
.
+
=
0
.
T
h
e
o
b
jectiv
e
o
f
th
e
SVM
alg
o
r
ith
m
is
to
d
eter
m
i
n
e
th
e
v
alu
es
o
f
w
an
d
b
to
m
ax
im
ize
th
e
m
ar
g
in
.
T
h
is
r
eq
u
ir
es
s
o
lv
in
g
th
e
s
u
b
s
eq
u
en
t o
p
tim
izatio
n
p
r
o
b
l
e
m
[
2
3
]:
min
,
,
{
1
2
‖
‖
2
2
+
∑
=
1
}
(
1
)
to
m
ee
t th
e
r
eq
u
ir
em
en
t
(
2
)
:
:
{
(
,
,
)
∈
×
×
+
(
〈
,
〉
+
)
≥
1
−
,
∀
1
≤
≤
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
R
ec
o
g
n
itio
n
o
f p
la
n
t le
a
f d
is
ea
s
es b
a
s
ed
o
n
d
ee
p
lea
r
n
in
g
…
(
N
g
h
ien
N
g
u
ye
n
B
a
)
451
wh
er
e,
th
e
s
lack
v
ar
iab
les,
d
e
n
o
ted
as
ξ
ᵢ
,
ar
e
i
n
tr
o
d
u
ce
d
to
r
elax
th
e
ca
teg
o
r
izatio
n
c
r
iter
io
n
,
w
h
ile
C
s
er
v
es
as
a
tu
n
i
n
g
f
ac
to
r
.
I
t
r
eg
u
lates
th
e
b
alan
ce
b
etwe
en
m
a
x
im
izin
g
th
e
m
a
r
g
in
an
d
m
in
im
izi
n
g
th
e
tr
ain
in
g
er
r
o
r
.
R
ath
er
th
an
tack
lin
g
th
e
p
r
im
ar
y
p
r
o
b
le
m
d
ir
ec
tly
,
it
is
o
f
ten
p
r
ef
er
ab
le
to
ad
d
r
ess
its
d
u
al
co
u
n
ter
p
a
r
t,
wh
ich
p
r
o
ce
ed
s
as f
o
llo
ws:
min
1
2
−
l
⃗
(
3
)
with
co
n
s
tr
ain
ts
:
∆
:
{
=
0
0
≤
≤
,
=
1
,
…
(
4
)
w
h
er
e
=
(
1
,
2
,
…
,
)
,
l
⃗
is
a
u
n
it v
ec
to
r
,
a
n
d
H
is
a
s
y
m
m
etr
ic
m
atr
ix
s
p
ec
if
i
ed
b
y
:
,
=
〈
Ф
(
)
,
Ф
(
)
〉
=
(
,
)
(
5
)
h
er
e,
〈
⋯
〉
p
r
esen
ts
a
d
o
t
p
r
o
d
u
ct,
a
n
d
(
.
)
d
en
o
tes
a
tr
an
s
f
o
r
m
atio
n
f
r
o
m
th
e
in
p
u
t
s
p
ac
e
to
a
f
ea
tu
r
e
s
p
ac
e
o
f
g
r
ea
ter
d
im
en
s
io
n
s
,
ad
d
r
ess
in
g
s
itu
atio
n
s
wh
er
e
s
am
p
les
a
r
e
n
o
t
lin
ea
r
ly
s
ep
ar
a
b
le.
(
∙
)
is
r
ef
er
r
ed
t
o
as
th
e
k
er
n
el
f
u
n
ctio
n
an
d
is
s
p
ec
if
ied
in
ex
p
r
ess
io
n
(
6
)
.
(
,
)
=
〈
Ф
(
)
,
Ф
(
)
〉
(
6
)
Fig
u
r
e
4
.
Sk
etch
p
r
in
ci
p
le
o
f
t
h
e
SVM
2
.
3
.
H
y
perpa
ra
m
e
t
er
o
ptim
i
za
t
io
n f
o
r
m
ula
t
io
n
Du
r
in
g
th
e
d
esig
n
p
h
ase
o
f
ML
m
o
d
els,
ef
f
icien
t
ex
p
lo
r
at
io
n
o
f
th
e
h
y
p
er
p
ar
am
eter
s
p
ac
e
th
r
o
u
g
h
o
p
tim
izatio
n
tec
h
n
iq
u
es
ca
n
id
en
tify
th
e
o
p
tim
al
h
y
p
e
r
p
ar
a
m
eter
s
(
HPO)
f
o
r
t
h
ese
m
o
d
els.
As
o
u
tlin
ed
in
th
e
ar
ticle
[
2
4
]
,
HPO
co
m
p
r
is
es
f
o
u
r
ess
en
tial
co
m
p
o
n
en
t
s
:
a
n
esti
m
ato
r
co
m
p
r
is
in
g
its
o
b
jectiv
e
f
u
n
ctio
n
,
a
s
ea
r
ch
s
p
ac
e
(
also
k
n
o
w
n
as
co
n
f
ig
u
r
atio
n
s
p
ac
e)
,
a
s
ea
r
ch
alg
o
r
ith
m
u
s
ed
t
o
u
n
c
o
v
er
t
u
n
in
g
p
ar
am
eter
s
,
a
n
d
a
s
co
r
in
g
f
u
n
ctio
n
f
o
r
co
m
p
ar
in
g
th
e
ef
f
ec
tiv
e
n
ess
o
f
v
ar
io
u
s
tu
n
in
g
p
ar
am
ete
r
s
etu
p
s
.
T
y
p
icall
y
,
th
e
g
o
al
o
f
th
e
HPO
p
r
o
b
lem
is
to
p
i
n
p
o
i
n
t a
s
p
ec
if
ic
p
o
in
t w
ith
in
th
e
s
ea
r
ch
s
p
ac
e
d
ef
i
n
ed
b
y
(
7
)
[
7
]
.
∗
=
a
r
g
min
∈
(
)
(
7
)
h
er
e,
(
)
p
r
esen
ts
th
e
o
b
jectiv
e
f
u
n
ctio
n
to
b
e
m
in
im
ized
,
∗
is
an
o
p
tim
al
p
o
in
t i
n
s
ea
r
ch
s
p
ac
e
S.
Fo
r
DL
m
o
d
els,
t
h
e
s
ea
r
ch
s
p
ac
e
ca
n
in
clu
d
e
th
e
n
u
m
b
e
r
an
d
s
ize
o
f
f
ilter
s
i
n
th
e
co
n
v
o
lu
tio
n
al
lay
er
s
,
th
e
ac
tiv
atio
n
f
u
n
ctio
n
,
th
e
n
e
u
r
o
n
co
u
n
t
with
in
t
h
e
f
u
lly
co
n
n
ec
ted
lay
er
,
a
n
d
t
h
e
i
n
itial
lear
n
in
g
r
ate.
Su
p
p
o
s
e
we
n
ee
d
to
f
in
d
n
o
p
tim
al
tu
n
in
g
p
a
r
am
eter
s
f
o
r
a
DL
m
o
d
el.
E
ac
h
tu
n
in
g
p
ar
a
m
eter
h
as
a
d
is
cr
ete
o
r
ca
teg
o
r
ical
v
alu
e
d
o
m
ain
with
s
ev
er
al
o
p
tio
n
s
in
th
e
co
r
r
esp
o
n
d
in
g
s
ea
r
c
h
s
p
ac
e
.
T
h
er
ef
o
r
e,
th
e
s
ea
r
ch
s
p
ac
e
ca
n
b
e
r
ep
r
esen
te
d
as
(
8
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
8
,
No
.
1
,
Ap
r
il
20
2
5
:
4
4
7
-
4
5
8
452
=
1
2
⋮
=
1
,
1
1
,
2
…
1
,
1
2
,
1
2
,
2
…
2
,
2
⋮
,
1
⋮
⋯
,
2
…
⋮
,
(
8
)
Hen
ce
,
∗
=
[
1
,
2
,
…
,
]
an
d
∈
.
T
o
f
in
d
th
e
∗
,
an
ap
p
r
o
p
r
iate
o
p
tim
izatio
n
al
g
o
r
ith
m
n
ee
d
s
to
b
e
u
s
ed
.
2
.
4
.
Chem
ic
a
l r
ea
ct
io
n o
pti
m
iza
t
io
n a
l
g
o
rit
hm
T
h
e
C
R
O
alg
o
r
ith
m
is
a
co
n
t
em
p
o
r
ar
y
r
a
n
d
o
m
s
ea
r
ch
em
p
lo
y
ed
f
o
r
th
e
p
r
o
b
lem
o
f
o
p
tim
izatio
n
,
m
im
ick
in
g
t
h
e
d
y
n
am
ics
o
f
lo
o
s
ely
co
u
p
led
ch
em
ical
tr
an
s
f
o
r
m
atio
n
s
with
in
o
p
tim
izati
o
n
p
r
o
ce
s
s
es.
I
n
a
ch
em
ical
r
ea
ctio
n
s
y
s
tem
,
v
a
r
io
u
s
ch
em
ical
s
u
b
s
tan
ce
s
in
t
er
ac
t
with
in
th
eir
e
n
v
ir
o
n
m
e
n
t,
ea
ch
p
o
s
s
ess
in
g
b
o
th
p
o
te
n
tial
(
PE)
an
d
k
in
eti
c
en
er
g
ies
(
KE
)
,
wh
ile
th
e
en
v
ir
o
n
m
e
n
t
its
elf
is
r
ep
r
esen
ted
b
y
a
ce
n
tr
al
en
er
g
y
b
u
f
f
er
[
2
5
]
.
As
th
e
ch
em
ical
r
ea
ctio
n
r
ea
ch
es
eq
u
ilib
r
iu
m
,
all
s
u
b
s
tan
ce
s
s
tab
ilize
with
m
in
im
al
p
o
ten
tial
en
er
g
y
.
C
R
O
em
u
lates
th
i
s
eq
u
ilib
r
iu
m
p
r
o
ce
s
s
b
y
tr
an
s
f
o
r
m
in
g
p
o
ten
tial
en
er
g
y
in
to
k
i
n
etic
en
er
g
y
,
s
lo
wly
d
is
ch
ar
g
in
g
th
e
en
er
g
y
s
to
r
ed
in
ch
em
ical
m
o
lecu
les
in
to
t
h
e
s
u
r
r
o
u
n
d
in
g
s
.
C
R
O
is
f
o
u
n
d
ed
u
p
o
n
f
o
u
r
b
asic
tr
an
s
f
o
r
m
s
:
on
−
wa
l
l
in
e
ff
e
c
tive
c
ol
l
ision
,
de
c
om
positi
on
,
in
ter
-
m
o
lec
u
lar
in
e
ff
e
c
tive
c
ol
l
ision
,
an
d
s
yn
the
s
is
.
W
h
ile
two
in
e
ff
e
c
tive
c
ol
l
isions
f
ac
ilit
ate
lo
ca
l
s
ea
r
ch
es,
th
e
o
th
er
s
f
ac
ilit
ate
g
lo
b
al
s
ea
r
ch
es.
C
o
n
s
eq
u
en
tly
,
C
R
O
ef
f
ec
tiv
el
y
in
teg
r
ates
th
ese
two
s
ea
r
ch
ty
p
es
to
ex
p
lo
r
e
th
e
g
lo
b
al
m
in
im
u
m
with
in
th
e
f
ea
s
ib
le
r
eg
io
n
.
B
y
am
alg
am
atin
g
th
e
b
en
e
f
ic
ial
asp
ec
ts
o
f
b
o
t
h
s
im
u
lated
an
n
ea
lin
g
(
SA)
an
d
GA,
C
R
O
m
ain
tain
s
en
er
g
y
c
o
n
s
er
v
atio
n
ak
in
to
th
e
m
etr
o
p
o
lis
alg
o
r
ith
m
u
s
ed
in
SA,
wh
ile
its
d
ec
o
m
p
o
s
itio
n
an
d
s
y
n
th
esis
o
p
er
atio
n
s
r
esem
b
le
th
e
cr
o
s
s
o
v
er
an
d
m
u
tatio
n
o
p
er
atio
n
s
o
f
GA.
I
n
C
R
O,
ev
er
y
m
o
lecu
l
e
is
d
ef
in
ed
b
y
a
m
o
lecu
lar
s
tr
u
ct
u
r
e
(
ω
)
,
s
er
v
in
g
as
an
an
s
wer
to
t
h
e
is
s
u
e,
alo
n
g
with
two
t
y
p
es
o
f
en
e
r
g
y
:
PE
an
d
KE
.
PE
co
r
r
esp
o
n
d
s
to
th
e
v
alu
e
o
f
th
e
f
itn
ess
f
u
n
ctio
n
,
wh
ile
KE
d
en
o
tes
a
m
o
lecu
le
’
s
to
ler
an
ce
to
an
in
cr
ea
s
e
in
its
en
er
g
y
s
tate
.
T
o
s
im
u
late
f
o
u
r
b
asic
r
ea
ctio
n
s
b
ased
o
n
th
r
ee
o
p
er
ato
r
s
n
eig
h
b
o
r
,
d
ec
o
m
p
o
s
itio
n
,
an
d
s
y
n
th
esis
.
T
h
e
n
eig
h
b
o
r
o
p
er
ato
r
is
u
s
e
d
in
th
e
co
llis
io
n
r
ea
ctio
n
to
c
r
ea
te
a
n
ew
s
o
l
u
tio
n
f
r
o
m
a
p
ar
ticu
lar
o
n
e.
T
h
e
n
e
w
s
o
lu
tio
n
is
cr
ea
ted
b
y
c
h
an
g
in
g
r
a
n
d
o
m
l
y
s
elec
ted
el
em
en
ts
in
th
e
cu
r
r
e
n
t
s
o
lu
tio
n
.
T
h
e
g
o
al
o
f
th
is
o
p
e
r
ato
r
is
to
co
n
d
u
ct
a
lo
ca
l
ex
p
lo
r
atio
n
f
o
r
a
b
etter
s
o
lu
tio
n
.
T
h
e
p
s
eu
d
o
co
d
e
o
f
th
is
o
p
er
ato
r
is
s
h
o
wn
b
y
Alg
o
r
ith
m
1
.
Alg
o
r
ith
m
1
.
Neig
h
b
o
r
(
ω
)
1.
Copy
to
′
2.
Generate a random integer i smaller than t
he total number of elements in
3.
Assign
′
(
)
=
,
∈
T
h
e
d
ec
o
m
p
o
s
itio
n
o
p
er
ato
r
i
s
u
s
ed
in
th
e
d
ec
o
m
p
o
s
itio
n
r
ea
ctio
n
.
T
h
is
o
p
e
r
ato
r
cr
ea
tes
two
n
ew
s
o
lu
tio
n
s
1
′
,
2
′
f
r
o
m
th
e
s
p
ec
if
ied
s
o
lu
tio
n
.
T
h
is
o
p
e
r
ato
r
h
elp
s
to
escap
e
l
o
ca
l
m
in
im
a
b
y
h
alf
th
e
to
tal
ch
an
g
e.
T
h
e
p
s
eu
d
o
co
d
e
o
f
t
h
is
o
p
er
ato
r
is
d
em
o
n
s
tr
ated
b
y
Alg
o
r
ith
m
2
.
Alg
o
r
ith
m
2
.
Dec
o
m
p
o
s
itio
n
(
ω
)
1.
Copy
to
1
′
2.
Randomly change 50% of the elements of
1
′
by selecting randomly
from set
3.
Repeat steps 1 and 2 for
2
′
in a similar manner
T
h
e
s
y
n
th
esis
o
p
er
ato
r
is
u
s
e
d
in
th
e
s
y
n
th
esi
s
r
ea
ctio
n
.
T
h
is
o
p
er
ato
r
cr
ea
tes
a
n
ew
s
o
lu
tio
n
′
f
r
o
m
t
h
e
two
g
iv
e
n
s
o
lu
tio
n
s
1
,
2
.
T
h
e
p
r
o
ce
s
s
in
v
o
lv
es
r
a
n
d
o
m
ly
s
elec
tin
g
co
m
p
o
n
e
n
ts
o
f
two
m
o
lec
u
les
with
s
im
ilar
ch
an
ce
s
to
cr
ea
te
a
n
ew
m
o
lecu
le.
T
h
e
p
s
eu
d
o
c
o
d
e
f
o
r
th
is
o
p
e
r
atio
n
is
p
r
o
v
i
d
ed
in
Alg
o
r
ith
m
3
.
T
h
e
C
R
O
alg
o
r
ith
m
is
s
tar
te
d
b
y
an
in
itial
p
o
p
u
latio
n
.
T
h
e
p
s
eu
d
o
co
d
e
f
o
r
cr
ea
tin
g
th
is
p
o
p
u
latio
n
is
p
r
esen
ted
in
Alg
o
r
ith
m
4
.
Alg
o
r
ith
m
3
.
C
r
ea
te
s
y
n
th
esis
(
1
,
2
)
1.
FOR i = 1 TO m DO
2.
Generate a random number r between 0 and 1
3.
IF r>0.5 DO
4.
′
(
)
=
1
(
)
5.
6.
′
(
)
=
2
(
)
7.
ENDIF
8.
ENDFOR
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
R
ec
o
g
n
itio
n
o
f p
la
n
t le
a
f d
is
ea
s
es b
a
s
ed
o
n
d
ee
p
lea
r
n
in
g
…
(
N
g
h
ien
N
g
u
ye
n
B
a
)
453
Alg
o
r
ith
m
4
.
C
r
ea
te
th
e
in
itial p
o
p
u
latio
n
1.
FOR i = 1 TO PopSize DO
2.
FOR j = 1 TO m DO
3.
Se
le
ct
ra
nd
om
ly
∈
fo
r
di
sc
re
te
or
ca
te
go
ri
ca
l
do
ma
in
s.
Ge
ne
ra
te
a
ra
nd
om
nu
m
be
r
∈
[
,
]
for the continuous domain.
4.
(
)
=
5.
ENDFOR
6.
Train the CNN or SVM model with the given
7.
Calculate the objective function
(
)
=
(
)
=
1
−
(
)
8.
ENDFOR
2
.
5
.
P
r
o
po
s
ed
m
et
ho
d
T
h
e
p
r
i
n
cip
le
o
f
th
e
p
r
o
p
o
s
e
d
ap
p
r
o
ac
h
in
v
o
lv
es
a
c
o
m
b
i
n
atio
n
o
f
a
DL
n
etwo
r
k
an
d
an
SVM
m
o
d
el.
T
h
e
DL
n
etwo
r
k
is
r
es
p
o
n
s
ib
le
f
o
r
ex
tr
ac
tin
g
f
ea
tu
r
e
s
f
r
o
m
im
ag
es
o
f
d
is
ea
s
ed
leav
es,
wh
ile
th
e
SVM
s
er
v
es
as
th
e
class
if
ier
.
I
n
itially
,
im
ag
es
o
f
d
is
ea
s
ed
leav
e
s
ar
e
ex
tr
ac
ted
f
r
o
m
th
e
Plan
tVillag
e
d
atab
ase.
Su
b
s
eq
u
en
tly
,
th
ese
im
ag
es
ar
e
au
g
m
en
ted
b
y
r
o
tatin
g
th
em
at
s
p
ec
if
ied
an
g
les.
T
h
e
au
g
m
en
ted
im
ag
es
th
e
n
u
n
d
er
g
o
f
ea
t
u
r
e
ex
tr
ac
tio
n
th
r
o
u
g
h
th
e
p
r
o
p
o
s
ed
li
g
h
tweig
h
t
C
NN
n
etwo
r
k
.
Fin
ally
,
th
e
im
ag
e
f
ea
tu
r
es
ar
e
class
if
ied
u
s
in
g
t
h
e
SVM
m
o
d
el.
T
h
e
p
er
f
o
r
m
an
ce
o
f
DL
n
etwo
r
k
s
m
ain
ly
d
ep
en
d
s
o
n
th
eir
ar
ch
itectu
r
e
a
n
d
h
y
p
e
r
p
ar
am
et
er
s
.
I
n
th
is
p
ap
e
r
,
th
e
C
NN
s
tr
u
ctu
r
e
is
d
esig
n
ed
t
o
b
ala
n
ce
s
im
p
licity
with
th
e
ab
ilit
y
to
ex
tr
ac
t
all
n
ec
ess
ar
y
f
ea
tu
r
es.
T
h
is
ar
ch
itectu
r
e
is
s
u
itab
le
to
b
e
d
ep
lo
y
e
d
o
n
d
ev
ices
with
lo
w
h
ar
d
war
e
c
o
n
f
ig
u
r
atio
n
s
,
s
u
c
h
as
em
b
ed
d
ed
c
o
m
p
u
te
r
s
o
r
m
o
b
ile
p
h
o
n
es.
A
d
d
itio
n
ally
,
th
e
o
p
tim
al
t
u
n
in
g
p
ar
am
eter
s
f
o
r
b
o
th
th
e
C
NN
n
etwo
r
k
an
d
SVM
m
o
d
el
ar
e
d
eter
m
in
ed
b
y
u
s
in
g
th
e
C
R
O
alg
o
r
ith
m
.
T
h
e
s
tr
u
ctu
r
e
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
in
cl
u
d
es
f
iv
e
c
o
n
v
o
lu
tio
n
al
lay
er
s
an
d
two
f
u
lly
c
o
n
n
ec
ted
la
y
er
s
,
as
il
lu
s
tr
ated
in
Fig
u
r
e
5
,
h
er
e,
k
i,
m
i,
f
i,
a
n
d
n
r
ep
r
es
en
t
th
e
n
u
m
b
er
o
f
f
ilter
s
,
s
ize
o
f
th
e
f
ilter
,
ac
tiv
atio
n
f
u
n
ctio
n
,
an
d
th
e
n
u
m
b
er
o
f
n
e
u
r
o
n
s
,
r
esp
ec
tiv
ely
.
T
h
ese
ar
e
tu
n
in
g
p
ar
a
m
eter
s
th
at
r
eq
u
ir
e
o
p
tim
izatio
n
.
T
o
en
h
a
n
ce
p
e
r
f
o
r
m
an
ce
in
class
if
icati
o
n
p
r
o
b
lem
s
,
A
f
u
s
io
n
o
f
C
NN
an
d
SVM
is
p
r
o
p
o
s
ed
.
I
n
itially
,
C
NN
f
u
n
ctio
n
s
as
th
e
f
ea
tu
r
e
ex
tr
ac
tio
n
co
m
p
o
n
e
n
t,
af
ter
th
at
th
e
f
ea
tu
r
e
v
ec
to
r
s
ex
tr
ac
ted
b
y
th
e
C
NN
s
er
v
e
as
in
p
u
ts
f
o
r
th
e
SVM.
T
h
is
f
u
s
io
n
is
ex
p
ec
ted
to
tak
e
th
e
s
tr
en
g
th
s
o
f
b
o
th
m
o
d
els:
C
NN
ef
f
icien
tly
ex
tr
ac
ts
f
ea
tu
r
es
f
r
o
m
im
ag
es,
wh
ile
SVM
d
em
o
n
s
tr
ates
h
ig
h
ca
teg
o
r
ical
ac
cu
r
ac
y
wh
en
th
e
in
p
u
t
d
ata
is
ef
f
ec
tiv
ely
p
r
e
p
r
o
ce
s
s
ed
.
Fig
u
r
e
6
d
ep
icts
th
e
in
te
g
r
at
io
n
o
f
t
h
ese
two
m
o
d
els.
Fig
u
r
e
5
T
h
e
ar
ch
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
C
NN
Fig
u
r
e
6
.
T
h
e
co
m
b
in
atio
n
o
f
t
h
e
p
r
o
p
o
s
ed
C
NN
an
d
SVM
T
h
e
tu
n
in
g
p
ar
am
eter
s
o
f
th
e
p
r
o
p
o
s
ed
C
NN
s
u
ch
as
1
,
2
,
3
,
4
,
5
,
1
,
2
,
3
,
4
,
5
,
1
,
2
,
3
,
4
,
5
,
6
,
a
n
d
n
o
r
C
,
an
d
k
e
r
n
el
ty
p
e
o
f
t
h
e
SVM
m
o
d
el
ar
e
e
n
co
d
e
d
as
a
m
o
lec
u
le
s
tr
u
ctu
r
e
(
s
o
lu
tio
n
)
,
a
n
d
t
h
en
th
e
C
R
O
alg
o
r
ith
m
is
a
p
p
li
ed
to
f
in
d
t
h
e
g
l
o
b
al
m
i
n
im
u
m
o
f
th
e
o
b
jectiv
e
f
u
n
ctio
n
.
I
n
th
is
ca
s
e,
th
e
o
b
jectiv
e
f
u
n
ctio
n
is
d
ef
i
n
ed
as
1
m
in
u
s
ac
cu
r
ac
y
o
r
1
d
i
v
id
ed
b
y
ac
c
u
r
ac
y
.
T
h
e
p
r
o
p
o
s
ed
alg
o
r
it
h
m
f
o
r
f
in
d
in
g
th
e
b
est
tu
n
in
g
p
ar
a
m
eter
s
o
f
th
e
C
NN
an
d
SVM
m
o
d
els
ar
e
p
r
esen
ted
in
Alg
o
r
ith
m
s
5
an
d
6
,
r
esp
ec
tiv
ely
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
8
,
No
.
1
,
Ap
r
il
20
2
5
:
4
4
7
-
4
5
8
454
Alg
o
r
ith
m
5
.
T
h
e
p
r
o
p
o
s
ed
al
g
o
r
ith
m
f
o
r
a
C
NN
’
s
h
y
p
e
r
p
ar
am
eter
o
p
tim
izatio
n
Input
:
Th
e
ob
je
ct
i
ve
fu
nc
ti
on
f
=
1
–
ac
cu
ra
c
y(
te
st
da
ta
se
t)
,
se
ar
ch
sp
ac
e
S,
im
ag
e
dataset, stopping criterion.
Output:
Optimal hyperparameters
1.
I
n
i
t
i
a
l
p
a
r
a
m
e
t
e
r
s
o
f
t
h
e
C
R
O
a
l
g
o
r
i
t
h
m
s
u
c
h
a
s
P
o
p
S
i
z
e
,
K
E
L
o
s
s
R
a
t
e
,
M
o
l
e
C
o
l
l
,
I
n
i
t
i
a
l
K
E
,
δ
,
θ
,
a
n
d
b
u
f
f
e
r
.
2.
C
r
e
a
t
e
t
h
e
i
n
i
t
i
a
l
p
o
p
u
l
a
t
i
o
n
a
c
c
o
r
d
i
n
g
t
o
A
l
g
o
r
i
t
h
m
4
a
n
d
f
i
n
d
t
h
e
c
u
r
r
e
n
t
b
e
s
t
s
o
l
u
t
i
o
n
∗
.
3.
R
E
P
E
A
T
3
.
1
.
G
e
n
e
r
a
t
e
a
r
a
n
d
o
m
n
u
m
b
e
r
b
∈
[
0
,
1
]
3
.
2
.
IF
b
>
M
o
l
e
C
o
l
l
T
H
E
N
3
.
3
.
R
a
n
d
o
m
s
e
l
e
c
t
o
n
e
s
o
l
u
t
i
o
n
f
r
o
m
t
h
e
p
o
p
u
l
a
t
i
o
n
d
e
n
o
t
e
d
b
y
s
3
.
4
.
IF
s
.
n
u
m
B
i
t
–
s
.
m
i
n
H
i
t
>
δ
T
H
E
N
3
.
5
.
P
e
r
f
o
r
m
d
e
c
o
m
p
o
s
i
t
i
o
n
r
e
a
c
t
i
o
n
a
c
c
o
r
d
i
n
g
t
o
t
h
e
o
p
e
r
a
t
i
o
n
i
l
l
u
s
t
r
a
t
e
d
b
y
a
l
g
o
r
i
t
h
m
2
3
.
6
.
E
L
S
E
3
.
7
.
P
e
r
f
o
r
m
o
n
W
a
l
l
I
n
e
f
f
e
c
t
i
v
e
C
o
l
l
i
s
i
o
n
r
e
a
c
t
i
o
n
a
c
c
o
r
d
i
n
g
t
o
t
h
e
o
p
e
r
a
t
i
o
n
i
l
l
u
s
t
r
a
t
e
d
b
y
a
l
g
o
r
i
t
h
m
1
3
.
8
.
E
N
D
I
F
3
.
9
.
E
L
S
E
3
.
1
0
.
S
e
l
e
c
t
r
a
n
d
o
m
l
y
t
w
o
s
o
l
u
t
i
o
n
s
f
r
o
m
t
h
e
p
o
p
u
l
a
t
i
o
n
d
e
n
o
t
e
d
b
y
s
1
a
n
d
s
2
3
.
1
1
.
IF
1
.
<
A
N
D
2
.
<
T
H
E
N
3
.
1
2
.
P
e
r
f
o
r
m
S
y
n
t
h
e
s
i
s
r
e
a
c
t
i
o
n
a
c
c
o
r
d
i
n
g
t
o
A
l
g
o
r
i
t
h
m
3
3
.
1
3
.
E
L
S
E
3
.
1
4
.
P
e
r
f
o
r
m
I
n
t
e
r
M
o
l
e
c
u
l
a
r
I
n
e
f
f
e
c
t
i
v
e
C
o
l
l
i
s
i
o
n
r
e
a
c
t
i
o
n
a
c
c
o
r
d
i
n
g
t
o
t
h
e
o
p
e
r
a
t
i
o
n
i
l
l
u
s
t
r
a
t
e
d
b
y
a
l
g
o
r
i
t
h
m
1
3
.
1
5
.
E
N
D
I
F
3
.
1
6
.
T
r
a
i
n
a
n
d
t
e
s
t
t
h
e
D
L
m
o
d
e
l
w
i
t
h
a
n
e
w
s
o
l
u
t
i
o
n
b
y
u
s
i
n
g
t
h
e
5
-
f
o
l
d
C
V
t
e
c
h
n
i
q
u
e
f
o
r
c
a
l
c
u
l
a
t
i
n
g
t
h
e
o
b
j
e
c
t
i
v
e
f
u
n
c
t
i
o
n
3
.
1
7
.
IF
a
b
e
t
t
e
r
s
o
l
u
t
i
o
n
i
s
f
o
u
n
d
T
H
E
N
3
.
1
8
.
U
p
d
a
t
e
t
h
e
b
e
s
t
s
o
l
u
t
i
o
n
3
.
1
9
.
E
N
D
I
F
4.
U
L
T
I
L
S
t
o
p
c
r
i
t
e
r
i
o
n
i
s
m
e
t
5.
O
b
t
a
i
n
∗
f
r
o
m
a
m
o
l
e
c
u
l
e
o
f
t
h
e
b
e
s
t
s
o
l
u
t
i
o
n
6.
T
r
a
i
n
a
n
d
t
e
s
t
t
h
e
D
L
m
o
d
e
l
w
i
t
h
t
h
e
∗
Alg
o
r
ith
m
6
.
T
h
e
p
r
o
p
o
s
ed
al
g
o
r
ith
m
f
o
r
th
e
SVM
’
s
h
y
p
er
p
ar
am
eter
o
p
tim
izatio
n
Input
: Trained DL, search space S, image dataset, the objective function f.
Output:
SVM
’
s
Optimal hyperparameters
1.
Initial
parameters
of
the
CRO
algorithm
such
as
PopSize,
KELossRate,
MoleColl,
Initial
KE, δ, θ, and buffer.
2.
Cr
ea
te
th
e
in
it
ia
l
po
pu
la
ti
on
ac
co
rd
in
g
to
Al
g
or
it
hm
4
an
d
fi
nd
th
e
cu
rr
en
t
be
st
solution
∗
.
3.
featureSet = {}
4.
FOREACH
img IN imageSet DO
4.1.
Calcul
at
e
th
e
ou
tp
ut
of
th
e
FC
(n
,f
6)
la
ye
r
of
th
e
tr
ai
ne
d
mo
de
l
wi
th
th
e
im
g
input as a feature vector.
4.2.
Append feature vector into featureSet
5.
ENDFOR
6.
REPEAT
6.1.
Generate a random number b
∈
[
0
,
1
]
6.2.
IF
b>MoleColl
THEN
6.3.
Random select one solution from the population denoted
by s
6.4.
IF
s.numBit
–
s.minHit > δ
THEN
6.5.
Perform
decomposition
reaction
according
to
the
operation
illustrated
by
algorithm 2.
6.6.
ELSE
6.7.
Perform
onWallIneffectiveCollision
reaction
according
to
the
op
eration
illustrated by algorithm 1
6.8.
ENDIF
6.9.
ELSE
6.10.
Select
randomly two solutions from the population denoted by s
1
and s
2
6.11.
IF
1
.
<
AND
2
.
<
THEN
6.12.
Perform Synthesis reaction according to Algorithm 3
6.13.
ELSE
6.14.
Perform
InterMolecularIneffectiveCollision
reaction
according
to
the
oper
ation
illustrated by algorithm 1
6.15.
ENDIF
6.16.
Tr
ai
n
an
d
te
st
th
e
SV
M
m
od
el
on
th
e
fe
at
ur
eS
et
wi
th
a
ne
w
so
lu
ti
on
by
us
in
g
th
e
5
-
fold CV technique for calculating the objective function
6.17.
IF
a better solution is found
THEN
6.18.
Update the
best solution
6.19.
ENDIF
7.
ULTIL
Stop criterion is met
8.
Obtain
∗
from a molecule of the best solution
9.
Train and test the SVM model with the
∗
on the featureSet.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
R
ec
o
g
n
itio
n
o
f p
la
n
t le
a
f d
is
ea
s
es b
a
s
ed
o
n
d
ee
p
lea
r
n
in
g
…
(
N
g
h
ien
N
g
u
ye
n
B
a
)
455
3.
E
XP
E
R
I
M
E
N
T
R
E
SU
L
T
S AN
D
DIS
CUSS
I
O
N
T
h
e
p
r
o
p
o
s
ed
s
o
lu
tio
n
h
as
b
e
en
ex
p
er
im
e
n
ted
o
n
th
e
Plan
t
Villag
e
d
ataset,
co
llected
b
y
Hu
g
h
es
an
d
Salath
e
[
2
6
]
.
T
h
is
d
ataset
is
h
ig
h
ly
r
en
o
wn
ed
,
en
co
m
p
a
s
s
in
g
m
o
r
e
th
a
n
5
0
,
0
0
0
im
a
g
es,
an
d
h
as
b
ee
n
ex
ten
s
iv
ely
u
tili
ze
d
b
y
ex
p
e
r
ts
f
o
r
p
lan
t
d
is
ea
s
e
d
iag
n
o
s
is
.
T
h
e
Plan
tVillag
e
co
m
p
r
is
es
3
8
im
ag
e
class
e
s
d
ep
ictin
g
d
is
ea
s
e
s
y
m
p
to
m
s
o
n
leav
es
a
n
d
1
im
a
g
e
class
with
o
u
t
leav
es.
Du
e
to
h
ar
d
w
ar
e
lim
itatio
n
s
,
th
e
au
th
o
r
s
r
estricte
d
th
eir
ex
p
er
i
m
en
t
to
u
s
in
g
o
n
ly
1
0
0
im
ag
e
s
p
er
class
.
So
m
e
im
ag
es
o
f
p
lan
t
leaf
d
is
ea
s
es
ar
e
illu
s
tr
ated
in
Fig
u
r
e
7
.
Deta
ils
o
n
th
e
d
ataset
u
s
ed
in
th
e
ex
p
e
r
im
en
t a
r
e
s
u
m
m
a
r
ized
in
T
a
b
le
1
.
Fig
u
r
e
7
.
So
m
e
im
ag
es o
f
p
lan
t le
af
d
is
ea
s
es
T
ab
le
1
.
Su
m
m
a
r
y
o
f
th
e
e
x
p
e
r
im
en
ted
d
ataset
N
u
mb
e
r
o
f
c
r
o
p
t
y
p
e
N
u
mb
e
r
o
f
p
l
a
n
t
l
e
a
f
d
i
s
e
a
s
e
s
N
u
mb
e
r
o
f
sam
p
l
e
s
N
u
mb
e
r
o
f
c
l
a
sse
s
I
mag
e
si
z
e
14
26
3
8
9
9
39
1
2
8
×
1
2
8
×
3
T
h
e
s
ea
r
ch
s
p
ac
es,
S1
f
o
r
th
e
h
y
p
e
r
p
ar
am
eter
s
o
f
t
h
e
p
r
o
p
o
s
ed
C
NN
an
d
S2
f
o
r
th
o
s
e
o
f
th
e
SVM
m
o
d
el
in
th
e
ex
p
er
im
en
t,
ar
e
s
et
u
p
as f
o
llo
ws
:
1
=
1
2
3
4
5
=
32
64
128
256
64
128
256
512
128
256
512
1
128
256
512
1024
3
256
512
1024
1024
2048
2048
4096
5
7
ℎ
512
1024
(
9
)
2
=
=
[
1
,
10
]
[
0
.
0001
,
1
.
0
]
(
1
0
)
T
h
e
ac
cu
r
ac
y
p
r
e
d
ictio
n
m
etr
i
c,
d
ef
in
ed
in
(
1
1
)
,
is
u
s
ed
t
o
e
v
alu
ate
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
.
=
(
1
1
)
Hy
p
er
p
ar
a
m
eter
s
∈
,
∈
,
∈
,
∈
,
∈
,
∈
,
∈
ar
e
f
o
u
n
d
b
y
th
e
C
R
O
alg
o
r
ith
m
.
T
h
e
in
itial
p
ar
a
m
eter
s
o
f
th
e
C
R
O
alg
o
r
ith
m
ar
e
s
et
as
Po
p
Size=
1
0
,
KE
L
o
s
s
R
ate=
0
.
1
,
Mo
leC
o
ll=0
.
1
,
I
n
itial KE
=1
0
0
,
δ
=5
0
0
,
θ
=1
0
,
an
d
b
u
f
f
er
=0
.
T
h
e
s
u
g
g
ested
alg
o
r
ith
m
is
i
m
p
lem
en
ted
u
s
in
g
Py
t
h
o
n
p
r
o
g
r
am
m
in
g
lan
g
u
ag
e
o
n
Go
o
g
le
C
o
lab
.
T
h
e
p
r
o
p
o
s
ed
DL
is
im
p
lem
e
n
ted
b
y
T
en
s
o
r
Flo
w
a
n
d
Ker
a
s
f
latf
o
r
m
.
T
h
e
SMV
m
o
d
el
is
im
p
lem
en
ted
u
s
in
g
th
e
Sk
lear
n
f
r
am
ew
o
r
k
.
T
h
e
au
th
o
r
s
im
p
lem
en
ted
th
e
C
R
O
alg
o
r
ith
m
.
T
h
e
p
r
o
p
o
s
ed
C
NN
is
tr
ain
ed
u
s
in
g
E
ar
ly
Sto
p
in
g
,
R
ed
u
ce
L
R
On
P
latea
u
,
an
d
th
e
5
-
f
o
ld
C
V
tech
n
iq
u
es.
T
h
e
SVM
m
o
d
el
i
s
tr
ain
ed
u
s
in
g
th
e
5
-
f
o
ld
C
V
tech
n
i
q
u
e.
T
h
e
tu
n
in
g
p
a
r
am
eter
s
o
f
t
h
e
C
NN
m
o
d
el
wer
e
f
o
u
n
d
b
y
r
u
n
n
in
g
t
h
e
C
R
O
alg
o
r
ith
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
8
,
No
.
1
,
Ap
r
il
20
2
5
:
4
4
7
-
4
5
8
456
im
p
lem
en
ted
b
y
alg
o
r
ith
m
5
a
r
e
m
1
=
m
2
=
m
3
=
m
4
=
m
5
=
1
,
k
1
=
3
2
,
k
2
=
6
4
,
k
3
=
1
2
8
,
k
4
=
2
5
6
,
k
5
=
512
,
n
=
5
1
2
,
f
1
=
f
2
=
f
3
=
f
4
=
f
5
=
f
6
=
R
eL
U
(
r
ec
tifie
d
li
n
ea
r
u
n
it
)
.
T
h
i
s
f
i
n
d
i
n
g
i
s
r
e
a
s
o
n
a
b
l
e
b
e
c
a
u
s
e
i
t
i
s
w
e
l
l
-
s
u
i
t
e
d
t
o
n
u
m
e
r
o
u
s
D
L
f
r
a
m
e
w
o
r
k
s
a
n
d
h
a
r
d
w
a
r
e
a
c
c
e
l
e
r
a
t
o
r
s
,
w
h
i
c
h
a
r
e
t
a
i
l
o
r
e
d
f
o
r
o
p
t
i
m
i
z
i
n
g
p
o
w
e
r
-
of
-
t
w
o
d
i
m
e
n
s
i
o
n
s
.
I
n
a
d
d
i
t
i
o
n
,
g
r
a
d
u
a
l
l
y
i
n
c
r
e
a
s
i
n
g
t
h
e
n
u
m
b
e
r
o
f
f
i
l
t
e
r
s
i
n
d
e
e
p
e
r
l
a
y
e
r
s
e
n
a
b
l
e
s
t
h
e
n
e
t
w
o
r
k
t
o
c
a
p
t
u
r
e
i
n
c
r
e
a
s
i
n
g
l
y
i
n
t
r
i
c
a
t
e
f
e
a
t
u
r
e
s
a
n
d
p
a
t
t
e
r
n
s
.
F
u
r
t
h
e
r
m
o
r
e
,
t
h
e
p
r
o
p
o
s
e
d
C
N
N
m
o
d
e
l
,
w
i
t
h
t
h
e
i
d
e
n
t
i
f
i
e
d
c
o
n
f
i
g
u
r
a
t
i
o
n
,
c
o
n
v
e
r
g
e
d
q
u
i
c
k
l
y
,
a
c
h
i
e
v
i
n
g
1
0
0
%
a
c
c
u
r
a
c
y
i
n
t
h
e
t
r
a
i
n
i
n
g
p
h
a
s
e
a
f
t
e
r
a
p
p
r
o
x
i
m
a
t
e
l
y
3
0
e
p
o
c
h
s
.
T
h
i
s
i
s
d
e
m
o
n
s
t
r
a
t
e
d
i
n
F
i
g
u
r
e
s
8
a
n
d
9
,
w
h
i
c
h
i
l
l
u
s
t
r
a
t
e
t
h
e
a
c
c
u
r
a
c
y
a
n
d
l
o
s
s
f
u
n
c
t
i
o
n
v
a
l
u
e
s
o
v
e
r
e
a
c
h
e
p
o
c
h
d
u
r
i
n
g
t
h
e
t
r
a
i
n
i
n
g
p
h
a
s
e
.
T
h
e
tu
n
i
n
g
p
ar
am
eter
s
o
f
t
h
e
SVM
wer
e
d
eter
m
in
e
d
b
y
ap
p
ly
in
g
t
h
e
C
R
O
alg
o
r
ith
m
im
p
lem
en
ted
b
y
th
e
Alg
o
r
ith
m
6
a
r
e
C
=1
0
,
0
.
7
6
0
6
5
,
an
d
th
e
k
er
n
el
is
lin
ea
r
.
T
o
ass
ess
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
e
d
m
o
d
el
in
co
m
p
ar
is
o
n
to
o
th
er
s
,
th
e
au
th
o
r
s
co
n
d
u
cted
ex
p
e
r
im
en
ts
u
s
in
g
two
wid
ely
r
ec
o
g
n
ized
m
o
d
els
f
o
r
co
m
p
u
ter
v
is
io
n
:
VGG1
6
a
n
d
Mo
b
ileNetV2
,
b
o
t
h
p
r
e
-
tr
ain
ed
o
n
t
h
e
I
m
a
g
eNe
t
d
ataset
f
o
r
tr
an
s
f
er
lear
n
in
g
.
T
ab
le
2
p
r
esen
ts
th
e
co
m
p
ar
is
o
n
r
esu
lts
f
o
r
t
h
e
ac
cu
r
ac
y
p
r
ed
ictio
n
m
etr
ic
in
t
h
e
tr
ain
in
g
an
d
test
in
g
p
h
ases
,
an
d
Fig
u
r
e
1
0
g
r
a
p
h
ically
d
is
p
lay
s
th
is
p
er
f
o
r
m
a
n
ce
m
etr
ic
.
T
h
e
r
esu
lts
s
h
o
w
th
at
th
e
co
m
b
in
atio
n
o
f
C
NN
an
d
SVM
in
th
e
p
r
o
p
o
s
ed
m
o
d
el
h
as
im
p
r
o
v
ed
th
e
class
if
icatio
n
ac
cu
r
ac
y
f
r
o
m
9
2
.
8
1
%
t
o
9
9
.
1
8
%
co
m
p
a
r
ed
to
u
s
in
g
o
n
ly
th
e
C
NN
m
o
d
el.
Fu
r
th
er
m
o
r
e,
th
e
s
im
p
le
C
NN
ar
ch
itectu
r
e
with
h
y
p
er
p
a
r
a
m
eter
s
is
o
p
tim
ized
to
attain
b
etter
r
esu
lts
th
an
co
m
p
lex
C
NN
ar
ch
itectu
r
e
s
u
c
h
as
VGG1
6
an
d
Mo
b
ileNet
V2
.
I
n
ad
d
itio
n
,
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
a
cu
s
to
m
-
b
u
ilt
lig
h
tweig
h
t
C
NN,
r
esu
ltin
g
in
s
h
o
r
ter
tr
ain
i
n
g
tim
es
co
m
p
ar
ed
t
o
VGG1
6
an
d
M
o
b
ileNetV2
.
T
h
e
tr
ai
n
in
g
tim
e
o
f
th
e
p
r
o
p
o
s
ed
C
NN
m
o
d
e
l
is
1
3
9
7
.
0
4
s
ec
o
n
d
s
,
w
h
er
ea
s
th
e
VGG1
6
an
d
Mo
b
ileNetV2
m
o
d
els to
o
k
1
4
6
9
.
1
3
an
d
2
3
2
9
.
7
5
s
ec
o
n
d
s
,
r
esp
ec
tiv
ely
.
Fig
u
r
e
8
.
Acc
u
r
ac
y
class
if
icatio
n
o
v
e
r
ep
o
ch
Fig
u
r
e
9
.
E
r
r
o
r
o
v
er
ep
o
c
h
T
ab
le
2
.
T
h
e
m
o
d
el
’
s
class
if
ic
atio
n
ac
cu
r
ac
y
S
t
r
u
c
t
u
r
e
Tr
a
i
n
i
n
g
sam
p
l
e
(
%)
Te
st
sam
p
l
e
(
%)
Th
e
p
r
o
p
o
s
e
d
C
N
N
1
0
0
.
0
0
9
2
.
8
1
Th
e
c
o
m
b
i
n
a
t
i
o
n
o
f
t
h
e
p
r
o
p
o
se
d
C
N
N
a
n
d
S
V
M
1
0
0
.
0
0
9
9
.
1
8
V
G
G
1
6
9
9
.
6
3
8
9
.
7
6
M
o
b
i
l
e
N
e
t
V
2
9
8
.
2
7
9
0
.
5
3
Fig
u
r
e
1
0
.
T
h
e
ac
c
u
r
ac
y
p
r
ed
i
ctio
n
in
th
e
tr
ain
i
n
g
a
n
d
test
in
g
p
h
ase
Evaluation Warning : The document was created with Spire.PDF for Python.