Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
9
, No
.
6
,
Decem
ber
201
9
, p
p.
5636
~
5643
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v9
i
6
.
pp563
6
-
5
643
5636
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
An
a
ccurate
p
attern c
l
assification
for
e
mp
t
y
f
ruit b
un
ch (EFB)
b
ase
d
on
the
a
ge
p
ro
fil
e of
o
il
p
alm
t
ree
usin
g n
eu
ra
l
n
etw
or
k
Wafi
A
z
iz
1
,
‘Afif K
asno
2
,
N
u
rkamili
a Kam
arud
in
3
,
Z
aidi
Tumari
4
,
Sha
hrie
el
A
ra
s
5
,
Herdy
R
usn
andi
6
, Kam
al
Musa
7
1,2,4
Cent
re
for
R
oboti
cs
&
Indust
ria
l
Autom
at
ion
(CeRIA),
Fa
cul
t
y
of
Eng
ineeri
ng
Technol
og
y
,
Univer
siti
Te
kn
i
kal
M
a
lay
sia
Me
la
ka
,
Ma
lay
sia
3,6,7
Facul
t
y
of En
gine
er
ing
T
ec
hn
olog
y
,
Univ
ersit
i
Te
kn
ika
l
Mal
a
ysia
Mel
aka
,
Ma
l
a
y
si
a
5
CeRIA,
Fa
cul
t
y
of
E
lectr
i
ca
l
En
gine
er
ing, Univers
it
i Te
knik
al Mal
a
y
si
a
Mel
aka
,
Malay
s
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
a
n
11
, 2
01
9
Re
vised
Ju
l
2
3
,
201
9
Accepte
d
J
ul
30
, 2
01
9
Thi
s
pape
r
proposes
an
eff
ic
ie
n
t
m
et
hod
for
pat
te
rn
cl
assifi
ca
t
io
n
sy
st
em
of
empt
y
frui
t
bun
ch
(
EFB
)
b
y
us
ing
a
neur
al
ne
t
work
te
chni
qu
e.
The
m
ai
n
adva
nt
age
of
th
i
s
m
et
hod
is
the
ac
cur
acy
and
spee
d
of
a
lgori
thm
such
tha
t
i
t
ca
n
b
e
comput
e
d
rap
idly
with
th
e
proposed
s
y
ste
m
.
To
t
est
th
e
ef
fec
t
ive
ness
of
the
proposed
m
et
hod
,
120
of
EFB’s
dat
a
wi
t
h
diffe
r
ent
age
s
and
l
engt
h
tha
t
bee
n
obt
ained
from
Mal
a
y
sian
Palm
Oil
Board
(MP
OB)
are
use
in
the
patter
n
class
ifi
c
at
ion
proc
ess
.
In
addi
ti
on
,
there
a
re
thr
ee
class
es
of
EFB
in
thi
s
s
y
st
em,
which
are
C
la
ss
1
(le
ss
tha
n
7
y
e
ar
old
),
C
la
ss
2
(8
to
17
y
e
a
r
old
)
and
C
la
ss
3
(m
ore
tha
n
17
y
e
ar
o
ld
).
I
t
is
e
nvisage
d
th
at
th
e
proposed
m
et
hod
is
ver
y
useful
in
cl
as
sif
y
ing
the
EFB
and
90%
of
the
sam
ple
par
amete
rs
are
s
ucc
essful
l
y
class
ifi
ed
to its
cl
ass.
Ke
yw
or
d
s
:
Em
pty
f
ru
it
b
unche
s
Im
age
p
r
ocessi
ng
Neural
n
et
w
ork
Copyright
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Ab
W
a
fi Bi
n A
b Aziz
,
Faculty
of E
ngineerin
g
T
ech
nolo
gy
,
Un
i
ver
sit
i Te
knikal M
al
ay
sia
Mel
aka
,
75300 D
ur
ia
n Tu
nggal, Mel
a
ka
,
Ma
la
ysi
a
.
Em
a
il
: wafi@ut
e
m
.ed
u.
m
y
1.
INTROD
U
CTION
In
2017,
Ma
la
ysi
a
had
5.8
1
m
i
ll
ion
hectares
of
plante
d
area
with
oil
pal
m
trees
[1
]
.
The
cr
op
norm
al
l
y
bear
s
fr
uit
within
three
ye
ars
aft
er
plantin
g
in
the
fiel
d
[2
]
.
As
a
resu
lt
,
m
ixtur
es
of
FF
B
fr
om
diff
e
re
nt
ages
of
oil
palm
trees
were
ha
r
ves
te
d,
c
ollec
te
d
and
tran
sporte
d
to
palm
oil
m
ills
fo
r
proce
ssing.
Ap
a
rt
from
th
e
producti
on
of
palm
oil,
the
m
i
ll
s
al
so
gen
erate
li
gnocel
lu
losic
resid
ue
s
that
include
e
m
pty
fruit
bunc
hes
(
EFB)
an
d
m
es
ocarp
fib
ers
(
pa
l
m
pr
essed
fi
ber).
Fig
ure
1
il
lustrate
s
the
extracti
on
of
pa
l
m
oil,
sta
rting
from
the FFB
harvest
ing
i
n
the
f
ie
ld
to o
il
e
xtracti
on at
palm
o
il
m
il
l.
Ba
sed
on
t
he
dr
y
m
at
te
r
con
te
nt,
t
he
am
ou
nt
of
dr
y
E
F
B
would
be
5.5
m
i
ll
ion
tonn
es
per
ye
a
r.
Hen
ce
t
he
qu
a
ntit
ie
s
at
hand
cou
l
d
m
ake
a
ver
y
s
ubsta
ntial
con
tri
bu
ti
on
to
the
s
upply
of
raw
m
at
eria
ls
for
the
pr
oductio
n
of
bi
oco
m
po
sit
e
pro
du
c
ts
that
ha
ve
tradit
io
nally
bee
n
m
ade
of
woo
d
fi
b
ers
[
3].
Th
us
,
the
in
dustry
is
act
ively
loo
king
for
com
m
ercial
o
utlet
s
to
el
i
m
i
nate
po
s
sible
po
ll
utio
n
or
di
sp
osa
l
pro
blem
s
caused
by
these
res
idu
es
[4
,
5]
.
T
his
will
ind
irec
tl
y
help
to
incr
ease
the
value
of
EFB
f
or
the
pal
m
oil m
il
le
rs.
This
pa
per
pre
sents
a
m
et
ho
d
to
cl
assify
EF
B
based
on
it
s
age
s
by
us
in
g
a
neu
ral
net
work
te
ch
nique
.
Ther
e
will
be
t
hr
ee
cl
asses
in
total
w
hich
ar
e
Cl
ass
1,
Cl
as
s
2
an
d
cl
ass
3.
Detai
l
f
or
eac
h
cl
ass
is
sho
wn
in
Table
1.
T
he
a
ge
is
bein
g
ide
ntifie
d
by
m
ea
su
ri
ng
the
le
ngth
of
E
FB
sp
ik
el
et
[
6
,
7
]
.
T
his
is
do
ne
m
anu
al
ly
by
MPOB em
plo
ye
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
An acc
ur
ate
pa
tt
ern
cl
as
sif
ic
at
ion
fo
r e
mp
ty
f
ru
it
bu
nch (
EFB)
b
as
ed
on t
he
age
profi
le
o
f oi
l
...
(
Wafi
Az
iz
)
5637
Figure
1. Palm
o
il
pr
ocess fl
ow
Table
1.
Cl
ass
segr
e
gatio
n
Clas
s
Year
s
Ind
u
stries
1
2
to 7
Mattr
ess
,
Co
ir
2
8
to 1
7
So
f
t
b
o
ard, Pottin
g
,
Fib
re
m
at
3
1
8
to 2
5
Mon
etary
n
o
tes
Fig
ure
2
s
how
s
dr
ie
d
EFB
w
her
e
the
recta
ngle
represe
nts
EFB
sp
ikele
t.
The
cu
rr
e
nt
ut
il
iz
at
ion
of
the
EFB
is
inef
fici
ent
an
d
has
low
producti
vi
ty
du
e
to
the
a
ge
is
not
bee
n
determ
ined
aft
er
proce
sses
of
an
oil
palm
fr
uits
extracti
on
in
the
palm
oil
m
ill
ing
pr
ocess.
T
he
EFB
is
sai
d
to
be
the
final
pro
du
ct
ac
hiev
e
d
wh
e
re
by
it
is
a
vaila
ble
in
ab
unda
nce
afte
r
th
e
m
illi
ng
proc
ess
of
F
FBsuc
cessf
ully
do
ne
.
The
c
urre
nt
pr
act
ic
e
do
e
s
not
has
a
good
p
aram
eter
s
to
re
us
e
th
e
EFB
fo
r
m
a
king
goods
to
the
industry
.
T
his
is
du
e
to
la
ck
of
econom
ic
al
sy
stem
fo
r
handl
ing
a
nd
sto
rage
and
le
ss
de
sirable
qu
al
it
y
of
res
ulti
ng
pro
du
ct
s
.
By
pro
posin
g
the
rig
ht
m
et
h
od
t
o
cl
assify
the
EFB,
it
wi
ll
help
the
in
dustry
t
o
deter
m
ine
the
rig
ht
qu
al
it
y
of
EF
B
for
a
sp
eci
fic
ap
pl
ic
at
ion
.
A
h
igh
qual
it
y
of
E
FB
can
be
s
ol
d
at
higher
pr
i
ce
su
bse
que
ntly
gen
erate
a
ddit
ion
al
incom
e to o
il
pa
l
m
ind
us
try
.
Figure
2. Em
pt
y oil
pum
p
fruit
b
unc
h (EFB)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
6
,
Dece
m
ber
2
01
9
:
563
6
-
56
4
3
5638
The
ob
j
ect
ive
of
this
pa
pe
r
w
o
rk
is
to
dev
el
op
a
m
et
ho
d
to
cl
assify
the
E
FB
based
on
it
age
cl
asses
by
usi
ng
neura
l
netw
ork
.
Ne
ur
al
netw
ork
ha
s
capa
bili
ty
to
a
naly
ze
data
especial
ly
w
he
n
th
e
data
do
es
not
fo
ll
ow
the
sam
e
distribu
ti
on
patte
rn
[
8
,
9]
.
Ar
duin
o
m
i
cro
c
ontrolle
r
i
s
us
e
to
dem
on
strat
e
the
res
ult
from
the
neural
network.
Furthe
r
m
or
e,
if
the
ob
j
ect
ive
achiev
e
d,
the
inc
om
e
of
palm
oil
m
i
ll
ers
will
be
in
crease
d
du
e
to
t
he
inc
reasin
g
of
EF
B
util
iz
at
ion
in
oth
e
r
in
dustrie
s.
T
his
is
ve
ry
cr
ucial
in
the
palm
oil
industr
y
becaus
e
cl
assifi
cat
ion
of
EFB
by
it
s
age
s
hav
e
a
good
valu
e
fo
r
diff
e
ren
t
m
anu
fact
ur
i
ng
industries.
T
he
EFB
can
be
s
old
ba
sed
on
it
s
cl
ass
age
an
d
the
pr
ic
e
will
be
higher
c
om
par
e
to
EFB
that
no
t
be
en
cl
as
sifie
d
An
e
xam
ple
of
products
that
can
be
m
ade
from
the
EFB
are
high
qual
it
y
pr
i
nting
pa
per
,
photog
raphic
pap
e
r
,
m
on
et
ary no
te
s
,
fi
br
e
m
at
s,
fib
re
-
bo
a
r
d
, a
nd s
of
t
boar
d [
10
-
12
].
2.
RESEA
R
CH MET
HO
D
OL
OGY
2.1.
Neur
al ne
two
rk impl
emen
t
at
i
on
The
desig
ning
of
th
e
ne
ur
al
ne
twork
is
us
in
g
MATL
AB
s
of
t
war
e.
Fig
ure
3
sho
ws
the
f
low
c
har
t
f
or
the
ne
ur
al
net
work
im
ple
m
e
ntati
on
.
The
da
ta
le
ng
th
of
EFB
sp
i
kelet
giv
e
n
by
MPOB
was
im
po
rted
t
o
MATLAB
f
or
trai
ning
in
the
neural
netw
ork
.
Lear
ni
ng
Vec
tor
Q
uan
ti
zat
io
n
(LVQ
)
is
c
hose
n
as
t
he
ty
pe
of
neural
net
wo
r
k
in
this
rese
arch
[13
-
21]
.
Since
it
is
su
i
ta
ble
to
deal
with
a
com
plex
pa
ram
et
er
s
wh
e
re
the
operati
on
can
be
im
plem
ented
in
s
uper
vised
te
c
hniqu
e
or
unsup
erv
ise
d.
By
usi
ng
t
he
supe
rv
ise
d
te
chn
iq
ue
,
this
m
e
tho
d
can
reduce
t
he
nu
m
ber
of
m
isc
l
assifi
ed
si
nce
the
va
riable
s
will
be
re
cl
ust
ered
accor
ding t
o
cl
ass sp
e
ci
ficat
io
n [
22
-
24]
. F
i
g
ure
4
s
hows
t
he
netw
ork
a
rch
it
ect
ur
e
for
L
V
Q
.
Figure
3.
Flo
w
ch
a
rt
of
neural
n
et
w
ork
im
ple
m
entat
ion
Figure
4. LV
Q
n
e
ur
al
netw
ork
a
rch
it
ect
ure [
10
]
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
An acc
ur
ate
pa
tt
ern
cl
as
sif
ic
at
ion
fo
r e
mp
ty
f
ru
it
bu
nch (
EFB)
b
as
ed
on t
he
age
profi
le
o
f oi
l
...
(
Wafi
Az
iz
)
5639
Fo
r
the
set
ti
ng
of
LV
Q,
a
num
ber
of
first
-
la
ye
r
hid
de
n
ne
uro
ns
is
set
to
10,
the
le
arn
in
g
rate
is
set
to
(
0.33
0.3
3
0.34)
a
nd
no
le
arn
i
ng
f
un
ct
i
on
is
a
ppli
ed.
In
t
he
al
gorith
m
s
se
tt
ing
,
trai
ning
is
set
to
r
andom
weig
ht/bias
ru
l
e
and
pe
rfor
m
ance
is
set
to
m
ean
sq
ua
red
error
(
m
se
)
[22].
The
num
ber
of
it
erati
ons
is
set
to
1000
[
21,
23,
25]
.
T
he
co
de
us
e
d
f
or
th
e
neural
netw
ork
f
or
this
pro
je
ct
is
sh
own
in
(
1)
.
Fi
gure
5
sh
ow
s
neural
netw
ork
lay
er s
yst
em
d
esi
gn u
si
ng the
Ma
tl
ab
softwa
re.
=
(
(
)
,
10
,
[
0
.
33
0
.
33
0
.
34
]
)
(1)
Figure
5. Ne
ural
n
et
w
ork
la
y
er s
yst
em
To
m
on
it
or
si
m
ula
ti
on
of
t
he
process
in
Ma
tl
ab,
grap
hi
cal
us
er
inter
f
ace
(
GUI)
is
c
on
st
ru
ct
e
d
as
sh
ow
n
in
Fi
gur
e 6
.
I
t i
nclu
des t
hr
ee
phases
w
hich
a
re i
nput, pr
ocesses
an
d
r
esult.
Figure
6. G
UI
for neu
ral
netw
ork
2.2.
Int
er
fa
ce
The
pur
pose
of
ha
rdwa
re
de
velo
pm
ent
is
to
s
how
the
resu
lt
s
of
the
neural
netw
ork
cl
assifi
e
r.
This
syst
em
c
an
be
us
e
d
as
a
ref
e
rence
in
or
der
t
o
rec
ognise
the
cl
a
ss
of
EFB.
Th
ree
LE
Ds
are
us
e
d
t
o
ind
ic
at
e
the
s
uc
cessf
ul
gro
up
age
cl
ass
in
th
e
neural
net
work
syst
em
.
Moreover,
an
A
rduin
o
m
ic
ro
con
t
ro
ll
e
r
is
us
ed
to
c
on
trol
the
L
EDs
an
d
as
the
li
nk
a
ge
betwee
n
MATL
AB
software
via
USB
port.
Yell
ow
LE
D
ind
ic
at
es
that
Cl
ass
1
is
dete
ct
ed,
re
d
LE
D
ind
ic
at
es
Cl
ass
2
is
detect
e
d
and
green
LE
D
in
dicat
es
Cl
ass
3
is
detect
ed. Fig
ure
7 sh
ows the
c
ircuit
d
ia
gr
am
f
or the
h
a
r
dw
a
re in
te
r
face
.
Fig
ure
8
sho
w
s
a
blo
c
k
diag
ram
fo
r
the
ha
rdwar
e
inter
fa
ce.
T
he
Ard
uin
o
is
c
onnecte
d
t
o
la
pt
op
instal
le
d
with
MATLAB
soft
war
e
via
US
B
cable.
M
ATL
AB
S
upport
P
ackag
e
f
or
A
r
du
i
no
is
instal
le
d
to
enab
le
c
omm
u
nicat
ion
betwe
en
MAT
LAB
and
A
rduin
o.
Re
su
lt
ob
ta
i
ne
d
f
ro
m
the
ne
ural
netw
or
k
is
sent
to
Ardu
i
no.
Yell
ow
LE
D,
re
d
LED
a
nd
gr
ee
n
LE
D
a
re
use
d
as
the
outp
ut
in
dicat
or
w
her
e
each
one
of
it
represe
nts
Cl
ass 1
,
Cl
ass
2
a
nd Cl
ass
3
,
res
pe
ct
ively
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
6
,
Dece
m
ber
2
01
9
:
563
6
-
56
4
3
5640
Figure
7. Ha
rdwar
e
d
em
on
str
at
ion
Figure
8. Bl
oc
k diag
ram
o
f
ha
rdwar
e
inter
fa
ce
3.
RESU
LT
S
A
ND AN
ALYSIS
Figure
9
s
how
s
the
pro
gr
es
s
of
neural
ne
twork
trai
ni
ng
and
the
res
ult
is
show
n
t
hro
ugh
nnto
ols
al
ong
with
the
trai
ning
perfor
m
ance.
T
he
num
ber
of
it
erati
on
can
be
cl
as
sifie
d
if
the
ne
ur
al
netw
ork
is
a
fast
le
arn
er
or
sl
ow
le
arn
er
wh
e
r
e
as
the
le
sser
th
e
nu
m
ber
of
it
erati
on,
the
le
s
ser
tim
e
ta
ken
the
m
achine
to
finish
the
trai
ning
[26].
Figure
10
sh
ows
trai
ni
ng
perform
ance
gr
ap
h
of
epo
c
h
ver
s
us
m
ean
s
qu
a
re
d
error
(
MSE)
.
Fr
om
the
gr
a
ph,
the
best
trai
ning
pe
rfor
m
ance
is
at
epo
c
h
14,
w
hich
is
0.0
27778
m
se
.
Me
ans,
afte
r
it
reaches
the
ep
och
14,
t
he
trai
ni
ng
perform
ance
becam
e
con
sist
ent
du
e
t
o
patte
r
n
recog
niti
on
al
r
eady
able
to
id
entify
the p
at
te
r
n.
Figure
9.
The
pro
gr
ess
of
ne
ur
al
netw
ork
t
r
ai
nin
g
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
An acc
ur
ate
pa
tt
ern
cl
as
sif
ic
at
ion
fo
r e
mp
ty
f
ru
it
bu
nch (
EFB)
b
as
ed
on t
he
age
profi
le
o
f oi
l
...
(
Wafi
Az
iz
)
5641
Figure
10. T
rai
ning
perform
a
nce
of ep
oc
h v
ersu
s
m
se
Figure
11
de
m
on
strat
es
the
conf
us
io
n
m
at
rix
for
th
e
trai
ning
re
su
lt
.
G
reen
c
olor
sho
ws
the
pe
rcen
ta
ge
and
num
ber
of
su
cc
essf
ul
da
ta
ou
t
pu
t
for
e
ach
ta
r
get
cl
ass.
Wh
er
eas
the
red
c
ol
or
in
di
cat
es
the
per
ce
ntage
and
nu
m
ber
of
wrongly
cl
assifi
cat
ion
data.
F
or
ta
r
get
Cl
ass
1,
33
out
of
40
data
for
Cl
ass
1
is
su
ccess
fu
ll
y
be
en
cl
assifi
e
d
wh
ic
h
giv
e
82.
5%
of
s
uccess
fu
l
rate.
Re
m
ain
in
g
7
data
of
Cl
ass
1
a
re
wrongl
y
cl
assifi
ed
int
o
Cl
ass
2
a
nd
Cl
ass
3
w
he
re
ea
ch
of
it
ha
s
a
num
ber
of
4
an
d
3
data
res
pec
ti
vely
.
The
fail
ur
e
of
the
cl
assifi
cat
ion
is
due
t
o
th
e
insuffici
e
nt
of
data
sam
ple
data
a
nd
inc
onsist
ency
valu
e
of
the
data
Cl
ass
1
durin
g
t
he
trai
ning
of the
ne
ural
net
work.
Figure
11. Ta
bl
e o
f
conf
us
i
on for
outp
ut clas
s
Fo
r
Cl
ass 2
,
t
he
su
ccess
f
ul
rat
e
of
cl
assifi
cat
ion
is
10
0%
where
al
l
40 d
at
a are
s
uccess
fu
ll
y
cl
assifi
ed
into
Cl
ass
2.
This
achie
vem
ent
is
con
ti
nu
ed
by
Cl
ass
3
w
her
e
al
l
da
ta
are
al
so
su
ccess
fu
l
cl
as
sifie
d.
The
pe
rf
ect
cl
assifi
cat
io
n
f
or
Cl
ass
2
and
Cl
ass
3
is
du
e
to
the
co
ns
ist
en
cy
of
sam
ple
d
at
a
fo
r
both
cl
asses
.
Ou
t
of
12
0
sa
m
ple
data
fr
om
al
l
cl
asses,
the
su
ccess
fu
l
rate
is
94
.
2%
.
Table
2
s
umm
arizes
the
resu
lt
of
the sim
ulati
on
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
6
,
Dece
m
ber
2
01
9
:
563
6
-
56
4
3
5642
Table
2.
Su
m
m
ary o
f
resu
lt
Tar
g
et Class
Su
ccessf
u
l data
d
etected
Percentag
e per
target class
Accurac
y
per
tar
g
e
t class
(ou
t of
120 sa
m
p
le
s)
1
33
8
2
.5%
2
7
.5%
2
40
100%
3
3
.3%
3
40
100%
3
3
.3%
Total
9
4
.2%
The
receive
r
operati
ng
c
ha
ra
ct
erist
ic
(ROC
)
is
plo
tt
ed
to
s
how
t
he
tr
ue
posit
ive
rate.
T
he
pe
rcen
ta
ge
per
ta
r
get
cl
as
s
is
show
n
in
Figure
12
belo
w.
T
rainin
g
100%
a
ble
to
pr
oc
ess,
the
valid
at
ion
is
0
a
s
w
el
l
as
the
te
st
ROC
is
li
near
zero
.
The
validat
io
n
a
nd
posit
ive
rate
are
li
near
and
this
is
cl
early
pr
ove
that
no
validat
io
n
s
houl
d
is
occured
.
Figure
12.
Re
c
ei
ver
operati
ng
ch
a
racteri
sti
c (RO
C)
r
es
ult
4.
CONCL
US
I
O
N
T
he
m
ai
n
obj
e
ct
ive
of
the
propose
d
m
et
ho
d
is
su
ccess
f
ully
achieve
d
wher
e
the
sam
ple
of
EFB
from
diff
e
re
nt
age
gro
ups
ar
e
cl
ass
ifie
d
int
o
it
s
s
pe
ci
fied
group
by
us
in
g
a
ne
ural
netw
ork.
F
urt
her
m
or
e,
an
im
age
processi
ng syst
e
m
is recomm
e
nd
e
d f
or
a
uto
m
at
ic
m
easur
e
m
ent of
t
he
EFB
’s
le
ngth
.
ACKN
OWLE
DGE
MENTS
The
aut
hors
w
ou
l
d
li
ke
to
thank
UTeM
for
s
ponsori
ng
this
work
unde
r
the
sh
ort
-
te
rm
gr
ant,
UTeM,
PJP/2
017/FT
K
-
CER
I
A/S01
555.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
An acc
ur
ate
pa
tt
ern
cl
as
sif
ic
at
ion
fo
r e
mp
ty
f
ru
it
bu
nch (
EFB)
b
as
ed
on t
he
age
profi
le
o
f oi
l
...
(
Wafi
Az
iz
)
5643
REFERE
NCE
S
[1]
Malay
s
ia
n
Pal
m
Oil
Board
2018
,
“
Overvi
e
w
of
m
al
a
y
s
ian
oil
pal
m
industr
y
,
”
2018,
[
Online
]
,
Avail
a
ble
:
htt
p://bepi.m
pob.
gov.
m
y
/i
ndex
.
p
hp/e
n/
.
[2]
A.
Kus
hai
ri,
et
a
l
.,
“
Chap
te
r
3
.
O
il
pa
lm
bre
edi
ng
and
see
d
produ
ct
ion
,”
Furthe
rn
Adv
anc
es
in
Oil
Pal
m
Re
s
earc
h
(
2000
to
2010)
,
MP
OB,
Bangi
,
v
ol.
1
,
p
p
.
47
-
101
,
2011
.
[3]
H.
Kam
aru
din,
e
t
al
.,
“
Oil
pa
lm
fibre
s
for
m
edi
u
m
density
fibr
eb
oar
d:
a
cl
assifi
cation
of
oil
pa
lm
biomass
for
m
ore
uniform
MDF
-
q
ual
ity
,
”
Proc
ee
d
ings
of
the
8th
Pac
ific
Ri
m
Bi
o
-
based
Compos
i
te
s
Symposium
,
Kuala
Lum
pur,
Malay
sia
,
p
p.
30
-
41
,
2006
.
[4]
A.
L
ewi
cki,
et
a
l
.
,
“
Potent
i
al
o
f
Bioga
s
Product
i
on
from
Palm
Oil
Empt
y
Fruit
B
unch
(EFB)
in
S
outh
-
Ea
st
As
ia
,
”
2018
2nd
Int
ernati
onal
Con
fe
ren
ce
on
Gr
ee
n
Ener
gy
and
App
licat
ions (
ICGEA)
,
Singapore
,
2018
,
pp.
1
-
4.
[5]
N.
N.
Kasim
,
et
al
.
,
“
Dem
ine
ralizati
on
of
oil
pa
lm
empt
y
fruit
bunch
(EFB)
intended
as
a
h
igh
qual
i
t
y
bio
-
oil
fee
dstock
,
”
4th
I
ET
Cle
an
Ene
rg
y
and
Te
chnol
og
y
Conf
ere
nc
e
(
CEA
T 2016)
,
Kual
a
Lumpur
,
2016
,
pp.
1
-
6
.
[6]
P.
B.
Tomlinson
and
C.
J.
Quinn,
“
Palms
do
not
under
go
sec
ond
ar
y
st
em
le
ngth
ening:
A
respon
se
t
o
Renni
nger
and
Phill
ips (
Am
eri
c
an
Journal
of
Bo
ta
n
y
99
:
607
-
61
3),
”
Am. J
.
Bo
t.
,
vol
.
100
,
pp
.
461
-
464,
2013
.
[7]
M.
Ninaz
unt
a,
e
t
al
.
,
“
Grow
th
strat
eg
ie
s
of
the
ar
bore
sce
nt
p
al
m
Iriartea
del
to
ide
a
in
a
weste
rn
Am
az
onia
n
fore
st
,
”
Bot
.
J
.
Linn
.
Soc
.
,
vol. 1
82,
pp.
4
11
-
424,
2016
.
[8]
Hendra
wan
Y.
,
e
t
al
.,
“
Im
age
An
aly
s
is using
Col
or
Co
-
occ
urr
ence
Matri
x
Te
x
tural
Feat
ur
es
for
Predicting
Nitrog
e
n
Conte
nt
in
Spi
nac
h
,”
TEL
KOMNIKA
Tele
co
mm
unic
ati
on
Computing
El
e
ctr
onic
s
and
Co
ntrol
,
vol.
16
(
6)
,
pp.
2712
-
2724
,
2018.
[9]
V.
Nedic
,
et
al
.,
“
Com
par
ison
of
Cla
ss
ic
al
St
a
ti
stical
Methods
and
Artifi
c
ia
l
Neura
l
Netw
ork
in
Tra
ffi
c
Noise
Predic
ti
on
,”
En
v
ironmental
Impa
ct
Assess
ment
R
ev
i
ew
,
vol.
49
,
p
p.
24
-
30
,
2014
.
[10]
N.
Abdulla
h
and
F.
Sulai
m
an
,
“
The
Oil
Palm
Wa
stes
in
Malay
s
i
a,
”
Biom
ass
No
w
-
Sus
ta
ina
ble
Grow
th
and
U
se,
Dr.
Miodrag
Darko
Matovi
c
(Ed.),
InTech,
2013
, [
Online
]
,
Avail
a
ble
:
htt
ps:
//
ww
w.i
nt
ec
hopen
.
co
m
/books/bi
om
ass
-
now
-
sus
ta
ina
ble
-
growth
-
and
-
use
/t
he
-
oi
l
-
pa
lm
-
waste
s
-
in
-
m
al
a
y
sia
[11]
Hus
in
M
.
,
et
a
l
.,
“
Resea
rch
and
Deve
lopment
of
Oil
Pa
lm
Biom
ass
in
W
ood
Ba
sed
Industrie
s,
”
Malay
sian
Pal
m
Oil
Board
Bul
l
etin
,
Pal
m Oi
l
De
v
el
opment
,
vo
l.
36
,
2005
.
[12]
S.
Sum
ant
hi,
et
al
.,
“
Util
i
zatio
n
of
Oil
Pa
lm
as
a
Sourc
e
of
Rene
wabl
e
En
er
g
y
in
Mal
a
y
s
ia,
”
R
ene
wabl
e
an
d
Sustainabl
e
Ener
gy
Revi
ews
,
vol
.
12
,
pp
.
2404
-
21
,
2008
.
[13]
H.
He,
et
a
l
.
,
“
A
Method
for
Ide
nti
ficat
ion
of
Dri
ving
Patt
ern
s
in
H
y
brid
E
lectr
i
c
Vehic
l
es
Based
on
a
LVQ
Neura
l
Network,
”
Ene
r
gie
s
,
pp
.
3363
-
3
380,
2012
.
[14]
P.
Meli
n,
et
a
l
.
,
“
A
new
neur
al
net
work
m
odel
base
d
on
th
e
LVQ
al
gorit
hm
for
m
ult
i
-
class
cl
assifi
ca
t
ion
o
f
arr
h
y
thmias,
”
In
f.
S
ci.
(
Ny)
.
,
vol
.
279,
pp
.
483
-
49
7,
2014
.
[15]
K.
J.
Devi,
et
al
.
,
“
Prediction
of
m
edi
c
ine
s
usin
g
LVQ
m
et
hodolog
y
,
”
2017
In
ter
nati
onal
Confer
enc
e
on
En
ergy
,
Comm
unic
ati
on,
Data
Ana
ly
t
ic
s
and
Soft Comput
ing
(
ICECDS)
,
Chenna
i
,
pp
.
38
8
-
391
,
2017
.
[16]
J.
Am
ez
cua,
et
al
.
,
“
A
new
c
la
s
sific
a
ti
on
m
et
ho
d
base
d
on
LVQ
n
eur
a
l
n
et
works
and
Fuz
z
y
Logic,
”
2015
Annual
Confe
renc
e
of
t
he
North
Ame
rican
Fuzzy
Information
Proce
ss
ing
Soci
ety
(
NAF
IPS)
hel
d
joi
nt
ly
wit
h
2015
5t
h
World
Confe
ren
ce
on
Sof
t
Comp
uti
ng
(
WConSC)
,
Redmond,
W
A
,
2015
,
pp.
1
-
5
.
[17]
E.
Ha
era
ni
,
e
t
a
l
.
,
“
Appli
ca
t
ion
of
unsupervise
d
K
Nea
rest
Nei
ghb
or
(UN
N)
and
Le
arn
ing
Ve
ct
or
Quant
izati
o
n
(LVQ)
m
et
hods
in
pr
edi
c
ti
ng
Rupia
h
to
Doll
a
r,
”
2016
4th
In
t
ernati
onal
Con
f
ere
nce
on
Cyb
e
r
and
IT
Serv
i
c
e
Manage
ment
,
B
andung
,
2016
,
p
p.
1
-
4
.
[18]
M.
Sardoga
n,
et
al
.
,
“
Plant
Le
a
f
Disea
se
De
te
c
ti
on
and
Cla
ss
if
ic
a
ti
on
Bas
ed
o
n
CNN
with
L
VQ
Algorit
hm
,
”
2018
3rd Int
ern
ati
onal
Con
fe
ren
ce
on
Computer
Sci
en
ce and
Eng
ine
ering
(
UBMK)
,
Sara
je
vo
,
201
8
,
pp
.
382
-
385
.
[19]
F.
Yan,
et
al
.
,
“
LVQ
Neura
l
Network
appr
oach
for
fau
lt
lo
ca
t
ion
of
distri
bution
net
work,
”
2
011
Inte
rnation
al
Confe
ren
c
e
on
E
le
c
tric
al
and
Co
ntrol
Enginee
rin
g
,
Yich
ang
,
201
1
,
pp
.
237
-
240
.
[20]
X.
Li
and
Y.
Zh
ang,
“
Digit
a
l
image
edg
e
detec
ti
on
base
d
on
LVQ
neur
al
ne
twork,
”
2016
IE
EE
1
1th
Confe
renc
e
on
Industrial
E
le
c
tronic
s and Applic
ati
ons (
ICIEA
)
,
Hefe
i
,
2016
,
pp.
1251
-
1255
.
[21]
P.
Sharm
a
and
S
.
Ugale,
“
El
ectr
onic
tongue
bas
ed
li
quid
id
ent
if
i
ca
t
ion
using
LVQ
al
gorit
hm
,
”
2
015
Inte
rnationa
l
Confe
renc
e
on
C
omputer,
Comm
unic
ati
on
and
C
ontrol
(
IC4)
,
Indore
,
2015
,
p
p.
1
-
6
.
[22]
Jopri
M.
H.,
et
al
.,
“
An
Acc
ura
te
Cla
ss
ifica
t
ion
Method
of
Harm
o
nic
Signal
s
in
Pow
er
Distrib
uti
on
S
y
stem
b
y
Util
ising
S
-
Trans
form
,”
TEL
K
OMNIKA
Tele
c
omm
unic
ati
on
Computing
El
ectronics
and
Control
,
vol.
15
(1
)
,
pp.
62
-
70
,
2017
.
[23]
H
abba
n,
et
a
l
.
,
“
An
Eva
lu
at
ion
o
f
li
ne
ar
t
ime
fre
quency
distr
ibut
i
on
Anal
y
s
is
for
VS
I
sw
it
ch
fau
lts
ide
nti
f
icati
o
n
,”
Inte
rnational
Jo
urnal
of Powe
r
El
e
ct
ronics
and
Dr
iv
e
Syst
ems (
IJP
EDS)
,
vol
.
8
(
1)
,
pp
.
1
-
9
,
2017
.
[24]
A.
R.
Abdulla
h,
et
al
.
,
“
Loc
aliz
at
ion
of
Multi
p
l
e
Harm
onic
Source
s
for
Inve
r
ter
Loa
ds
Util
izing
Periodogra
m
,”
Journal
of
Telec
omm
unic
ati
on,
El
e
ct
ronic
and
Computer
Engi
n
ee
ring (
JTEC)
,
v
ol.
8
,
pp
.
87
-
91
,
2016
.
[25]
T.
Sutikno,
et
al
.,
“
Dete
c
ti
on
of
Ship
Im
ag
e
Proc
essing
and
Neura
l
Network
,”
TE
LK
OMNIKA
Tel
ec
omm
unic
at
ion
Computing
E
le
c
t
ronics
and
Cont
rol
,
vol
.
16
(1)
,
p
p.
259
-
264
,
201
8
.
[26]
T.
Sutikno
,
et
a
l
.,
“
Cla
ss
ifi
c
at
ion
of
Motorc
y
c
li
sts
not
W
ea
r
Helmet
on
Dig
it
a
l
Im
age
with
Ba
ckpr
opaga
t
ion
Neura
l
Network
,”
TEL
KOMNIKA
Telecomm
unic
ati
on
Computing
E
le
c
t
ronics
and
Cont
rol
,
vol
.
14
(3)
,
p
p.
1128
-
33
,
201
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.