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
.
4
,
Aug
us
t
201
9
, p
p.
2386
~
2393
IS
S
N:
20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v
9
i
4
.
pp2386
-
23
93
2386
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Classific
atio
n
of
palm oil
fresh f
ruit bun
ch using mu
ltib
and
optical s
ensors
Agun
g
W
. S
e
t
iawan
,
Rich
ar
d M
e
ngko
, Ay
u P.H. P
ut
ri
,
Donn
y Danud
ir
dj
o,
Alfie R
. An
an
d
a
School
of El
ec
tr
i
ca
l
Engi
n
ee
r
ing
and
Inform
at
i
cs
,
Instit
ut Te
kno
lo
gi
Bandung
,
Ind
onesia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ja
n 5
, 201
8
Re
vised
Jan
19
, 201
9
Accepte
d
Ma
r
19
, 201
9
Thi
s
stud
y
inv
esti
gated
opt
ical
se
nsor
s
y
stem
con
sist
of
sixteen
l
i
ght
emitting
diode
(LE
D)
in
visibl
e/
n
ea
r
inf
rar
ed
reg
ion
to
det
e
ct
pal
m
oil
fre
sh
fruit
bunch
(FF
B)
qual
ity
.
Prac
ticall
y,
expe
r
ie
n
ce
gr
a
der
assess
ed
FF
B
qual
ity
b
y
it
s
ripe
n
ess
base
d
on
externa
l
fe
a
ture
s
such
as
colour
and
num
ber
of
det
a
che
d
fruit
l
et
s.
How
ev
er,
d
iffe
r
ent
s
eed
and
p
la
nt
at
io
n
m
ana
gement
result
ing
in
FF
B
qual
ity
v
ariati
on
.
Sam
e
ext
e
rna
l
fe
at
ure
s
not
li
ne
arly
cor
r
elat
e
with
FF
B
oil
content
th
at
cor
responding
with
industri
a
l
nee
ds.
Th
e
660
nm
LE
D
is
choose
n
to
be
used
to
esti
ma
te
the
oi
l
con
t
ent
of
FF
B.
Us
ing
li
nea
r
discri
m
ina
nt
an
a
l
y
sis
(LDA)
with
Maha
la
nobis
dista
nc
e,
the
a
cc
u
racy
of
th
e
s
y
stems
is
79.
8%
and
88.
2%.
F
rom
33
F
FB
oil
cont
en
t
m
ea
sure
m
ent
,
gra
der
m
iscl
assifie
d
4
out
of
17
FF
B
as
ripe
FF
B
but
with
low
oil
cont
en
t
(<17.
5%
)
and
m
isc
la
ss
ifie
d
7
out
o
f
16
FF
B
as
unripe
but
with
h
igh
oil
con
te
n
t
(>=
17.
5%)
.
Cl
a
ss
if
y
ing
m
odel
buil
d
from
FF
B
from
m
ai
n
pla
nta
ti
on
the
n
te
sted
to
ev
al
u
a
te
FF
B
from
s
m
al
lhol
der
.
Clas
sific
at
ion
m
od
el
gene
r
ated
from
FFB
oil
cont
ent
da
ta
show
ed
m
ore
ac
cur
ate
resul
t
compar
e
d
to
m
odel
gene
ra
te
d
from
visual
insp
ec
t
ion
66.
7%
co
m
par
ed
to
52
.
1%.
Model
ac
cur
acie
s
a
tt
a
in
ed
b
y
Discr
iminant
Ana
l
y
s
is
(DA
)
and
k
-
Ne
are
s
t
Neighbor
s
(k
-
NN
)
were
79.
8%
and
80.
7
%,
respe
c
ti
ve
l
y
base
d
on
gra
d
er
evalua
t
ion.
Model
a
cc
ura
cies
base
d
on
FF
B
oil
conte
nt
was
88.
2%
for
both
cl
assif
y
ing
al
gor
it
hm
s.
Ke
yw
or
d
s
:
Op
ti
cal
se
ns
in
g
Pal
m
o
il
content
Ri
pen
ess
estim
at
ion
Visible
-
near i
nfra
red sen
sor
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
:
Agu
ng
W.
Seti
awan,
School
of Elec
tric
al
Engineer
ing
a
nd
Inform
at
ic
s,
In
sti
tut Te
knol
og
i B
a
ndun
g,
St
. G
a
nesa
No.
10,
Ba
ndun
g 4
0132, I
ndonesi
a.
Em
a
il
: awseti
a
wan@ste
i.i
tb.a
c.id
1.
INTROD
U
CTION
Accor
ding
to
The
U
nited
St
at
es
De
par
tm
e
nt
of
A
gr
ic
ult
ure,
i
n
2016,
I
ndonesi
a
s
uppli
ed
m
or
e
tha
n
half
of
t
he
gl
obal
palm
oil
m
ark
et
[
1].
Pal
m
oil
Fr
esh
Fruit
Bun
c
h
(F
F
B)
qual
it
y
depends
on
qu
a
ntit
y
and
qu
al
it
y
of
oil
that
can
be
extr
act
ed
from
the
bunc
h.
Ri
pe
F
FB
has
m
or
e
oi
l
than
unripe
bunc
h
an
d
ha
v
e
le
ss
fr
ee
fatt
y
aci
d
com
par
ed
t
o
over
ripe
bu
nch
[2
]
.
Oil
c
on
te
nt
in
palm
oil
m
esocar
p
an
d
ke
rn
el
i
ncr
ease
s
al
ong
with
palm
oil
fr
esh
fruit
bunc
hes
(FFB
)
ripe
ning
proce
ss.
On
e
of
ripe
nes
s
ind
ic
at
io
n
is
wh
e
n
f
r
uitl
et
s
easy
to
detache
d
from
the
bunc
h
be
see
n
i
n
Fi
gure
1.
Bu
nches
w
it
h
50
-
200
lo
ose
fruit
s
had
oi
l/
bu
nc
h
1.9%
higher
than
a
bu
nch wi
th one loose
fr
uit [
3].
Ther
e
a
re
th
re
e
palm
oil
var
ie
ti
es,
that
are
nig
re
scen
s,
vi
recens
,
an
d
al
benscens
.
T
he
ripeness
of
virece
ns
an
d
al
benscens
var
ie
ti
es
can
be
seen
by
the
col
or
.
Fo
r
virecens,
the
colo
r
of
the
FFB
turn
to
oran
ge
wh
e
n
it
s
ripe.
Sabr
i,
et
al
.
has
de
velo
pe
d
cam
era
-
base
d
syst
e
m
t
o
detect
the
FF
B
ripen
ess
[4]
-
[12].
In
Ind
on
esi
a
,
the
virece
ns
a
nd
al
benscens
is
ver
y
ra
re.
The
m
os
t
var
ie
ty
that
is
gr
own
i
n
I
ndon
esi
a
is
nigrescen
s
var
i
et
y.
It
is
ver
y
diff
ic
ult
to
a
ss
ess
the
FFB
ri
pen
e
ss
of
this
var
ie
ty
.
T
her
e
is
no
c
olor
c
ha
ng
i
ng
wh
e
n
it
is
unde
r
-
ripe, ripe
, a
nd
ov
e
r
-
ripe
.
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
Cl
as
sif
ic
ation
of palm
o
il
fres
h
fr
uit b
un
c
h u
sing m
ulti
band
opti
cal se
ns
or
s
(
Ag
ung
W
. Se
ti
awa
n
)
2387
Figure
1.
Ri
pe palm
o
il
f
resh f
ru
it
bunc
hes w
it
h
detac
hed f
r
uit
In
pract
ic
al
,
industry
purc
hase
d
palm
oil
FFB
fro
m
s
m
al
lho
lder
ba
sed
on
FFB
ri
pen
e
ss.
Conve
ntion
al
l
y,
oil
pa
lm
FFB
ripe
ness
a
ss
essed
by
gr
i
de
r
by
co
unti
ng
the
num
ber
of
detach
ed
f
ru
it
s
pe
r
bunc
h
a
nd
dis
colo
rati
on.
Unripe
a
nd
ove
rri
pe
FFB
retu
r
n
to
sm
al
l
ho
lder
.
Be
cau
se
of
gra
der
only
look
at
exter
nal
featu
r
e
of
FFB,
sm
al
lho
lde
r
can
detache
d
f
r
uit
from
FFB
so
it
seen
li
ke
ri
pe
FFB
.
This
gr
a
de
r
appr
oach
is
so
su
bject
ive
.
S
o,
the
m
ai
n
probl
e
m
or
resea
rc
h
qu
est
io
n
is
how
t
o
cl
assifi
e
d
the
FFB
ripe
ness
of
the n
i
gr
esce
ns
var
ie
ty
ob
j
ect
ively
.
Ther
e
a
re
s
ome
non
-
dest
ru
ct
i
ve
te
ch
niques
to
cl
assifi
ed
F
FB
ripe
ness
ha
ve
bee
n
st
ud
i
ed
by
m
any
researc
hers.
Junkw
on,
et
al
.
i
ntr
oduce
hy
pe
rsp
ect
ral
-
ba
sed
to
gr
a
de
the
FFB
ripe
ness.
They
fou
nd
t
ha
t
the
cl
assifi
cat
ion
f
or
fou
r
le
vel
of
F
FB
r
i
pen
e
ss
can
be
do
ne
us
in
g
re
flect
ance
val
ue
of
Near
Infr
a
re
d
(NIR)
sp
ect
r
um
a
t
75
0
nm
and
dev
e
lop
the
m
od
el
for
oil
con
te
nt
pr
e
dicti
on
[
13]
.
But,
hype
rspect
ral
i
m
aging
te
nds
to
ha
ve
re
dundant
featu
re
a
nd
m
ake
dec
re
asi
ng
in
cl
as
sific
at
ion
ac
cu
ra
cy
[14].
In
2014,
Ma
kk
y
a
nd
S
oni
record
dif
f
us
e
ref
le
ct
ance
us
i
ng
fibe
r
ref
le
ct
ion
pro
bes
that
pro
vid
e
d
s
pec
tral
inf
or
m
at
ion
be
twee
n
25
0
an
d
1000
nm
.
They
su
ccessf
ully
cl
assifi
ed
seve
n
fr
act
ion
F
FB
r
ipeness
us
in
g
s
pectral
inf
or
m
at
ion
from
40
0,
540,
560,
590,
670,
800,
91
0,
940
and
10
00
nm
.
And
al
s
o,
th
e
y
dev
el
op
a
m
od
el
for
oil
co
ntent
pr
e
dicti
on
usi
ng
sp
ect
ral
data
from
440,
47
0,
480,
510,
610,
69
0,
72
0,
75
0,
76
0,
8
80,
900,
910,
940,
980,
990
a
nd
1000
nm
[1
5].
Chem
o
m
e
tric
analy
sis
us
in
g
visible
an
d
ne
ar
inf
rar
e
d
s
pe
ct
ru
m
wer
e
use
d
to
obta
in
in
te
rn
al
pro
per
ti
es
of
F
FB.
It
is
im
po
r
ta
nt
to
int
rod
uc
e
the
c
urren
t
t
echnolo
gy
in
pa
l
m
oil
ind
ust
r
y.
The
ulti
m
a
te
goal
s
are
to
in
crease
the
palm
oil
plantat
ion
a
nd
f
act
or
y
[
8].
I
n
pr
act
ic
al
ap
plica
ti
on
,
t
he
ra
pi
d
an
d
non
-
de
structiv
e
NI
R
te
ch
ni
qu
e
has
po
te
ntial
us
e
f
or
cl
assify
ing
oil
palm
fr
esh
f
r
uit
into
ri
pen
e
ss
gra
des.
Non
-
dest
ru
ct
iv
e
an
d
rap
i
d
NI
R
te
c
hn
i
qu
e
ca
n
be
us
ed
to
cl
a
ssi
fy
FFB
m
at
ur
it
y
le
vel
[1
6]
-
[
23
]
.
Fi
gure
2
sh
ows
the
pro
po
s
ed
syst
e
m
s that is consi
sts o
f
a
s
il
ic
on
bas
e
d p
hotod
et
ect
or a
nd
16 LE
Ds desig
ned in t
he
sam
e p
la
ne
.
Figure
2.
Pro
pose
d
syst
em
s
2.
RESEA
R
CH MET
HO
D
2.1.
S
amp
le
p
repar
at
i
on
and optic
al sens
or
This
stud
y
wa
s
carried
out
on
la
te
Ma
rch
to
early
April
2016
in
Ri
au
Pr
ovi
nce,
I
ndonesi
a.
I
n
this
stud
y
191
F
FB
colle
ct
ed
with
two
ripe
ness
c
at
egories,
ri
pe
and
unri
pe.
One
hundre
d
a
nd
twenty
FFB
sa
m
ple
wer
e
ta
ke
n
f
rom
m
ai
n
plantat
ion
a
nd
71
FF
B
from
s
m
all
ho
lde
rs.
Ri
pe
ne
ss
cl
ass
wa
s
de
te
rm
ined
by
gr
ader
by
visu
al
i
ns
pecti
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.
4
,
A
ugust
201
9
:
23
86
-
2393
2388
Fo
r
eac
h
FFB,
the
sp
ect
ral
da
ta
are
colle
cted
from
two
side
of
FFB
.
O
utput
from
the
syst
e
m
s
is
vo
lt
age
va
lue,
descr
i
be
as
rel
at
ive
ref
le
ct
a
nc
e
value
of
FF
B.
Sixtee
n
LE
D
(li
gh
t
em
it
t
i
ng
di
od
e
)
from
615
nm
to
940
nm
with
about
20
n
m
i
nter
val
wer
e
ut
il
ise
d.
Com
bin
at
ion
visible
an
d
nea
r
inf
rar
e
d
sp
ect
ral
hav
e b
et
te
r
po
te
ntiat
e
for
FFB
ripe
ness
est
i
m
ation
[
24
]
.
Chloro
ph
yl
l,
on
e
of
F
FB
pi
gm
ent
that
chan
ge
duri
ng
m
at
ur
it
y
has
c
omm
on
absor
ban
ce
within
visible
s
pe
ct
ru
m
between
600
a
nd
700
nm
[2
5].
Mole
cules
interact
i
on
with
el
ect
ro
m
agn
et
ic
rad
ia
ti
on
within
N
IR
sp
ect
r
um
arise
fr
om
ov
e
rto
nes
an
d
com
bin
at
ion
bands
O
-
H
an
d
C
-
H
existe
d
i
n
m
oistur
e a
nd
oil [
26]
.
Figure
2
sho
w
s
the
pr
op
os
e
d
syst
e
m
s
tha
t
i
s
con
sist
s
of
a
s
il
ic
on
base
d
phot
od
et
ect
or
a
nd
16
LE
Ds
desig
ne
d
in
the
sa
m
e
plane.
A
600
nm
lon
gp
a
ss
filt
er
placed
in
fro
nt
of
photodete
ct
or
to
transm
i
t
re
flect
e
d
li
gh
t
great
er
th
an
60
0
nm
and
reduce
ef
fect
of
am
bient
li
gh
t
from
external
so
urce.
T
he
sy
stem
was
power
e
d
by
5
V
DC
an
d
c
onnecte
d
to
com
pu
te
r
f
or
data
ac
qu
isi
ti
on
a
nd
sto
ra
ge.
Data
c
ollec
te
d
from
two
si
de
FFB
,
aver
a
ge
of
t
hes
e
two
rea
dings f
r
om
each
FFB
us
e
d
f
or
f
ur
t
he
r
analy
sis.
T
ot
al
262
s
pectral
data
we
re
c
ollec
te
d
and
di
vid
e
i
nto
191
data
trai
ning
a
nd
71
da
ta
te
st.
FFB
spe
ct
ral
data
f
rom
m
ai
n
plantat
ion
us
ed
f
or
tr
ai
ning
data
and
te
ste
d
fo
r
cl
assify
in
g
FFB
ripen
es
s
from
s
m
al
lho
lders
.
Cl
assifi
cat
ion
base
d
on
vi
su
al
insp
ect
io
n
the
n
will
be
c
om
par
ed
to
FFB
oil
co
ntent
i
ns
pe
ct
ion
.
Twe
nty
on
e
sa
m
ple
fro
m
m
ai
n
planta
ti
on
us
e
d
as
tr
ai
nin
g
data
an
d
12
sa
m
ple
fr
om
s
m
al
l
ho
lde
r
us
ed
as
te
st
data.
C
la
ssific
at
ion
de
velo
ped
us
i
ng
Linear
Discri
m
inant
An
al
ysi
s
(
LD
A)
with
Ma
ha
la
nobis
dista
nc
e
an
d
k
-
N
N
al
gorithm
us
ing
E
uclidean
di
sta
nce
with
f
eat
ur
e
sel
ect
io
n.
Data
an
al
ysi
s w
a
s
done
usi
ng S
PS
S S
ta
ti
sti
cs 2
0.
FFB
oil
co
nten
t
are
determ
ined
thr
ough
la
bo
rator
y
c
hem
ic
a
l
analy
sis
wh
ic
h
are
ti
m
e
con
su
m
ing
and
la
bor
intensi
ve
.
Thirty
three
sam
ple
data
c
ollec
te
d
for
f
urt
her
a
naly
sis.
Fr
uitl
et
s
were
detache
d
fro
m
the
bunc
h
an
d
the
n
weig
ht.
T
hirt
y
fr
uitl
et
s
rand
om
l
y
cho
os
e
,
10
from
ou
te
r,
10
f
r
om
m
idd
le
,
and
10
fro
m
inn
er
fruit
le
ts
as
can
be
see
n
in
Figure
3.
The
n,
these
fruit
le
ts
are
chop
pe
d
to
se
par
at
e
the
fruit
’s
m
esocar
p.
Me
so
car
p
the
n
we
re
dri
ed
to
rem
ov
e
physi
cal
water
fr
om
the
m
esocarp.
The
oil
in
the
m
esocarp
was
extracte
d
us
i
ng
S
oxhlet
e
xtr
act
or
with
hexane
as
so
l
ven
t.
The
rem
ai
nin
g
fi
ber
an
d
s
ol
ven
t
were
dri
e
d
a
nd
coo
le
d
in
th
e
desiccat
or
.
Sa
m
ple
in
ever
y
ste
p
w
ere
weig
he
d.
Oil
in
the
m
esocarp
(
Oil
m
)
is
cal
cul
at
ed wit
h
(
1
)
.
Figure
3
.
Palm
o
il
s
pik
el
et
w
i
th fru
it
le
ts
%
Oil
m
=
W
2
−
W
3
W
1
×
100%
(1)
wh
e
re
W
1
is
m
esocarp
sam
ple
wei
ght
(
g),
W
2
is
m
esocarp
wei
gh
t
a
ft
er
dr
yi
ng
(g),
and
W
3
is
m
e
so
ca
r
p
weig
ht af
te
r
e
xt
racti
on
(g). T
he
cr
ud
e
p
al
m
o
il
r
eco
ver
y
fro
m
the sam
ple FFB
is
cal
culat
e
d usin
g
(
2
)
.
%
=
∑
×
%
×
%
(2)
wh
e
re
W
FFB
is
FFB
wei
gh
t
(
g),
W
f
is
fruit
le
ts
weigh
t
(g)
,
%
W
m
is
perce
ntage
of
m
e
so
car
p
weig
ht
from
fruit
le
ts (%)
,
a
nd %O
il
m
is pe
rcen
ta
ge of
m
esocar
p oil
(
%
).
2.2. Cl
as
sific
ati
on c
ri
teria
The
ri
pen
e
ss
est
i
m
ation
bas
ic
al
ly
us
ed
to
get
FFB
wit
h
hi
gh
oil
co
ntent.
Mi
scl
as
sific
at
ion
is
cond
ucted
by
the
grade
r
co
ul
d
be
cl
assify
in
g
ripe
FFB
des
pite
of
lo
wer
oi
l
yi
e
ld,
vice
ve
rsa.
I
n
this
pa
per,
we
will
inv
est
igate
classi
ficat
ion
algorit
hm
r
ob
ust
ness
ba
sed o
n
exter
nal f
eat
ur
e e
valuate
d
by
g
rad
e
r
an
d
in
trinsic
crit
eria base
d o
n
F
FB o
il
yi
el
d. T
his f
l
ow
c
hart
o
f
the classi
fi
cat
ion
al
gorith
m
can
be
see
n i
n
Fig
ure
4.
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
Cl
as
sif
ic
ation
of palm
o
il
fres
h
fr
uit b
un
c
h u
sing m
ulti
band
opti
cal se
ns
or
s
(
Ag
ung
W
. Se
ti
awa
n
)
2389
Figure
4
.
Ef
fec
ts of sel
ect
ing
diff
e
re
nt sw
it
c
hing
unde
r dyn
a
m
ic
co
nd
it
io
n
In
t
his
stu
dy,
Eucli
dea
n
an
d
Ma
halan
ob
is
di
sta
nce
were
use
d
to
disti
ngui
sh
bet
ween
m
od
el
cl
asses
assessed
by
grader
a
nd
FFB
oil
yi
el
d.
Eucl
idean
distance
wer
e
us
e
d
to
cl
us
te
r
sim
i
la
r
data
points
i
nt
o
the
sam
e
c
lusters,
wh
il
e
Ma
hal
anobis
m
easure
dissim
i
la
r
or
dista
nt
data
p
oi
nts
are
pl
aced
into
dif
fer
e
nt
cl
us
te
rs
[27].
Be
tween
tw
o
da
ta
po
i
nt
x
a
nd
y,
Eucli
dea
n
di
sta
nce
m
easur
ed
by
(
3
)
a
nd
Ma
halan
ob
is
di
sta
nce
m
easur
ed
b
y
(
4
)
.
=
√
∑
(
−
)
2
=
1
(3)
=
√
(
−
)
S
−
1
(
−
)
(4)
wh
e
re
S is the
cov
a
riance
m
atr
ix
of the
datas
et
.
3.
RESU
LT
S
A
ND AN
ALYSIS
Av
e
ra
ge
relat
i
ve
re
flect
ance
vo
lt
age
value
f
ro
m
16
LE
D
s
is
us
ed
to
dif
fe
ren
ti
at
e
two
F
FB
ripe
ness
cat
egories, r
ipe
an
d
unripe
. U
s
ing
t
he
box
-
an
d
-
wh
is
ke
r
plo
t
s,
ou
tl
ie
r
sam
ples
ind
ic
at
ed
as
dot
m
ark
in
th
e
bo
x
plo
t
we
re
rem
ov
e
d
f
ro
m
the
dataset
.
Th
us
t
otal
nu
m
ber
of
sa
m
ples
fo
r
ripe
an
d
unri
pe
cl
ass
was
57
a
nd
52,
resp
ect
i
vely
.
T
rainin
g
dataset
f
or
FFB
oil
c
on
te
nt
w
ere
17
FFB
with
oil
con
te
nt
m
or
e
t
han
17.
5%
(
hi
gh
oil
con
te
nt)
a
nd
4
FFB
wit
h
oil
co
ntent
le
ss
t
han
17.
5%
(lo
w
oil
c
on
te
nt).
Four
hi
gh
oil
co
ntent
outl
ie
r
dat
a
rem
ov
ed
f
ro
m
dataset
.
V
oltage
dif
fer
e
ntiat
ion
betwee
n
two
cl
asses
of
ripe
ness
al
m
os
t
fo
un
d
in
al
l
LED.
T
he b
ox
-
an
d
-
w
hisk
e
r plots
of
615 nm
, 6
35
nm
, 6
60
nm
, and 830
nm
are
s
how
n
in
Fig
ure
5
.
The
diff
e
re
nce
s
of
volt
age
va
lue
in
visible
r
egio
n
c
orrelat
ed
wit
h
cha
nge
of
c
olor
du
e
to
chlo
ropyl
l
pr
ese
nce
t
hat
hav
e
abs
orba
nc
e
in
66
0
nm
[10].
Palm
oil
FFB
col
or
bec
om
e
m
or
e
reddish
duri
ng
rip
enin
g.
Accor
ding
to
the
Ikem
efu
na
and
Ad
am
so
n,
the
m
ajo
r
pigm
ent
chan
ge
s
associat
ed
wit
h
ripe
ning
pro
ses
were
decr
easi
ng the
chlo
rop
hyll
and the
m
assive accum
ulati
on
of
car
otenes
[
28
]
.
Mod
el
li
ng
of
FFB
ripe
ness
us
es
ste
pw
ise
appr
oach
f
or
var
ia
ble
sel
ect
ion
,
sta
rts
by
sel
ect
ing
th
e
m
os
t
discri
m
inati
ng
var
ia
ble
and
ad
d
ne
w
va
riable
in
ste
pwise
m
ann
er,
a
com
bin
at
ion
of
f
orward
sel
ect
ion
and
bac
kwar
d
el
i
m
inati
on
.
T
o
f
ound
t
he
m
os
t
dis
c
rim
inativ
e
va
riable,
F
-
te
st
is
pe
rform
ed
at
each
ste
p
of
t
he
process
.
In
t
his
stud
y,
f
or
the
sel
ect
ion
of
L
ED
wa
velen
gt
h
set
at
0.
05
f
or
the
i
nclusion
va
riable
an
d
0.1
f
or
rem
ov
al
var
ia
ble.
A
s
umm
a
ry
of
the
res
ults
from
ste
pw
ise
proce
dure
,
includi
ng
the
w
avele
ng
t
h
sel
ect
io
n
pr
ese
nted
in
T
able
1
.
T
his
w
avelen
gth
was
sel
ect
ed
as
th
e
m
os
t
i
m
po
rtant
wa
velen
gt
h
f
or
disc
rim
i
natin
g
betwee
n
treat
m
ent.
Dif
fer
e
nt
wav
el
e
ng
t
h
obta
ined
f
or
cl
a
ssific
at
ion
bas
ed
on
FFB
oil
con
te
nt.
On
ly
660
nm
LED
inclu
de
to
the
m
od
el
.
U
sin
g
li
near
di
scri
m
inant
analy
sis
(LD
A)
with
Ma
halan
ob
is
dista
nce,
m
od
el
accuracy
79.
8%
an
d
88.2
%
was
ob
ta
ine
d
f
or
m
od
el
that
gen
e
rated
base
d
on
grade
r
e
va
luati
on
a
nd
F
FB
oil
con
te
nt,
resp
e
ct
ively
.
This
m
od
el
was
use
d
f
or
e
valuat
ing
FFB
qu
al
i
ty
fr
om
sm
a
llh
ol
der
an
d
ac
cur
acy
ob
ta
ine
d
by
52.
1%
an
d
66.
7%
for
exter
nal
featur
e
c
rite
ria
and
F
FB
oil
con
te
nt,
res
pe
ct
ively
.
Mod
el
that
gen
e
rated
b
as
e
d on FFB
oil c
on
te
nt s
howe
d hig
her rate
of a
ccur
acy
.
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.
4
,
A
ugust
201
9
:
23
86
-
2393
2390
Figure
5. Bo
x
-
and
Wh
is
ker
pl
ots r
e
presenti
ng
relat
ive r
e
flect
an
ce v
oltage
value o
f
ri
pe
and un
ripe pal
m
o
il
f
resh
frui
t bunc
hes
Table
1.
Var
ia
bles in
the a
nal
ysi
s
LE
D
Wav
elen
g
th
s (n
m
)
Tolerance
Sig
.
o
f
F
to
Re
m
o
v
e
Min. D
Squ
ared
LE
D
76
5
n
m
1
.00
0
.00
0
LE
D
76
5
n
m
.61
9
.00
0
.07
8
LE
D
68
0
n
m
.61
9
.00
0
.88
3
LE
D
76
5
nm
.15
2
.00
0
.95
4
LE
D
68
0
n
m
.55
8
.00
0
.93
0
LE
D
89
0
n
m
.20
1
.01
6
2
.07
6
LE
D
76
5
n
m
.15
0
.00
0
1
.06
1
LE
D
68
0
n
m
.03
6
.00
1
2
.03
9
LE
D
89
0
n
m
.20
1
.01
4
2
.29
3
LE
D
66
0
n
m
.04
0
.04
6
2
.41
9
Ri
pe
FFB
c
orr
el
at
ed
with
higher
oil
yi
el
d
th
an
unripe
FFB
,
but
in
t
he
Fi
g
ure
6
s
howe
d
that
grade
r
te
nd
s
to
unde
r
est
i
m
at
e
m
ai
n
plantat
io
n
FF
B
oil
yi
el
d
and
overe
stim
at
e
s
m
al
lho
lder
F
FB
oil
yi
el
d.
Gr
a
der
m
isc
la
ssifie
d
4
ou
t
of
17
FF
B
as
ripe
FFB
bu
t
with
lo
w
oil
con
te
nt
an
d
m
isc
la
ssifie
d
7
ou
t
of
16
F
FB
as
unripe
but
with
high
oil co
nte
nt. Inter
nal
qua
li
ty
o
f
FFB
ha
r
dly determ
ined
only
f
r
om
v
isu
al
insp
ect
io
n.
Diff
e
re
nt
cl
assifi
cat
ion
al
gori
thm
was
us
ed
to
get
bette
r
ac
cur
acy
.
K
-
NN
is
a
distance
ba
sed
m
et
ho
d
in
w
hich
t
he
tr
ai
nin
g
dataset
is
store
d
s
o
th
at
a
new
rec
ord
m
ay
be
cl
assifi
ed
by
c
om
par
in
g
it
with
the
m
os
t
si
m
il
ar
reco
rds
in
the
trai
ning
set
.
The
num
ber
of
near
est
neig
hbors
te
ste
d
ra
ng
e
d
f
r
om
3
to
5.
The
k
value
us
e
d
f
or
analy
s
is
was
set
to
a
m
axi
m
u
m
5
w
hich
was
t
he
num
ber
that
ap
pe
ared
m
or
e
fr
e
qu
e
ntly
in
the
cro
ss
-
v
al
idati
on
te
sts.
Eucli
dea
n
dis
ta
nce
us
e
d
to
m
easur
e
sim
il
a
rity
.
All
pr
edic
tors
are
sta
nda
rd
iz
e
d
to
guar
antee
m
easur
ed
on
the
sam
e
scal
e.
Fo
r
k
-
N
N
m
od
el
base
d
on
grade
r
eval
uatio
n,
3
sel
ect
ed
predict
ors
was
LED
660,
615,
a
nd
830
nm
with
k
=
5.
For
buil
t
m
od
el
based
on
FFB
oil
con
te
nt
crit
er
ia
,
2
pr
e
dictor
s
were
sel
ect
ed,
635
a
nd 61
5 nm
. The accu
racy f
or
bo
t
h
m
od
el
w
e
re
80.7
%
and
88.
2%, res
pecti
vely
.
Table
2
disp
la
ys
the
ov
e
rall
cl
assifi
cat
ion
a
ccur
aci
es
ac
hieved
by
the
two
cl
assifi
cat
i
on
m
et
ho
ds
us
e
d
in
this
stu
dy.
Trai
ning
da
ta
set
was
buil
t
fr
om
FFB
fro
m
m
a
in
plantat
ion
a
nd
te
ste
d
to
FFB
sam
ple
from
sm
a
ll
ho
lders
.
Cl
assifi
cat
ion
m
od
el
gen
erat
ed
from
FFB
oil
con
te
nt
data
showe
d
m
or
e
accurate
resu
lt
com
par
ed
t
o
m
od
el
ge
ner
at
e
d
from
visu
al
ins
pecti
on
66.7
%
com
pa
red
t
o
52.
1%.
Th
us
in
di
cat
ing
that
FF
B
oil
con
te
nt
was
m
or
e
r
obus
t
va
riable
f
or
e
valua
ti
ng
var
ia
nt
of
palm
oil
FFB
than
e
xtern
al
fe
at
ur
e
su
c
h
as
c
olour
and num
ber
of
detach
fruit
.
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
Cl
as
sif
ic
ation
of palm
o
il
fres
h
fr
uit b
un
c
h u
sing m
ulti
band
opti
cal se
ns
or
s
(
Ag
ung
W
. Se
ti
awa
n
)
2391
Figure
6
.
Com
par
is
on of
ripe
ness
est
im
at
ion
and
fr
es
h fruit
bun
c
hes
o
il
c
onte
nt
Table
2
.
A
ver
a
ge
cl
assifi
cat
io
n
acc
ur
aci
es
u
s
ing
discrim
inant
an
d K
NN
cl
assifi
er
for pal
m
o
il
f
re
sh
fruit
bu
nch
qu
al
it
y p
red
ic
ti
on b
a
sed
on gr
ader eval
uation an
d oil
conten
t
Clas
s
Linear
Discri
m
in
a
n
t Analysis
k
Near
est
Neig
h
b
o
u
r
Calib
ration
Valid
atio
n
Calib
ration
Valid
atio
n
Clas
sif
icatio
n
bas
e
d
on
Gr
ad
er
Evalu
atio
n
Ov
erall
7
9
.8
5
2
.1
8
0
.7
62
Un
ripe
6
7
.3
51
6
9
.2
68
Rip
e
9
1
.2
54
9
1
.2
50
Clas
sif
icatio
n
bas
e
d
on
FFB Oil Co
n
ten
t
Ov
erall
8
8
.2
6
6
.7
8
8
.2
50
<1
7
.5%
100
5
5
.6
75
3
3
.3
>=1
7
.5%
8
4
.6
100
9
2
.3
100
4.
CONCL
US
I
O
N
The
com
bin
at
ion
of
visible
a
nd
near
i
nfrar
e
d
sp
ect
ru
m
in
this
sens
or
sys
tem
can
po
te
nt
ia
ll
y
us
e
to
determ
ining
pa
l
m
oil
fr
es
h
f
r
ui
t
bunch
qual
it
y.
Lear
ning
al
gorithm
can
be
buil
t
base
d
on
exter
nal
fe
at
ure
an
d
internal
qual
it
y
of
palm
oil
FFB.
Lea
rn
i
ng
al
gorithm
based
on
FFB
oil
con
te
nt
ha
ve
be
tt
er
perf
or
m
a
nce
f
or
cl
assifi
cat
ion
va
riet
y
of
palm
oil
FFB
an
d
re
pr
ese
nt
bette
r
qu
al
it
y
as
pote
ntial
oil
extracti
on
rate.
Th
e
660
nm
LED
is
ch
oose
n
to
be
us
e
d
to
est
i
m
at
e
the
oi
l
con
te
nt
of
F
FB.
Using
li
ne
ar
discrim
inant
analy
sis
(
LD
A)
wit
h
Ma
halan
ob
is
distance,
the
accuracy
of
t
he
syst
em
s
is
79.
8%
a
nd
88.2%.
From
33
FFB
oil
co
ntent
m
easur
em
ent,
gr
a
der
m
isc
la
s
sifie
d
4
ou
t
of
17
FFB
as
ripe
F
FB
but
w
it
h
low
oil
c
onte
nt
(<17.
5%
)
an
d
m
isc
la
ss
ifie
d
7
ou
t
of
16
FF
B
as
un
ri
pe
bu
t
with
high
oil
con
te
nt
(>=17
.5
%
).
Cl
assify
ing
m
od
el
buil
d
from
FFB f
r
om
m
a
i
n
pla
ntati
on
th
en
te
ste
d
to e
va
luate
FF
B f
rom
s
m
allho
lder
.
Cl
assifi
cat
ion
m
od
el
g
ener
at
e
d
f
ro
m
FFB
oil
c
on
te
nt
data
s
howe
d
m
or
e
accu
ra
te
resu
lt
c
o
m
par
ed
t
o
m
od
el
ge
ner
at
ed
f
rom
visu
al
inspe
ct
ion
66.7%
com
pared
to
52.1
%
.
Mod
el
accu
r
aci
es
at
ta
ined
by
Discrim
inant
A
naly
sis
(DA)
a
nd
k
-
Near
est
Neig
hbors
(
k
-
NN)
we
re
79.8%
an
d
80.
7%,
resp
ect
ively
ba
sed
on
gr
a
de
r
evaluati
on.
M
odel
accu
racies
base
d
on FF
B
oi
l c
onte
nt w
as
88.2%
for
both classi
fyi
ng alg
or
it
hm
s.
REFERE
NCE
S
[1]
Unite
d
Sta
t
es
Depa
rtm
ent
of
Agric
ult
ur
e.
(2016,
June
9
).
Re
tri
ev
ed
from
htt
p://apps.
f
as.
usda.
gov/psdonl
in
e/
c
irc
ul
ars/oi
lse
e
ds.pdf
[2]
M.
Makk
y
,
e
t
al
.
,
“
Autom
at
ic
no
n
-
destruc
t
ive
qu
al
ity
inspec
ti
on
s
y
stem
for
oil
p
a
lm
fruit
s
,”
In
t.
A
grophys
vol.
28
,
pp.
319
-
329
,
20
14.
[3]
R.
Corl
e
y
and
P.
Ti
nk
er,
“
The
Oi
l
Palm
,”
4th
ed
ition. Be
rlin:
B
lackwe
ll Sc
ie
n
ce
,
2
003.
[4]
R.
Singh,
et
a
l.
,
“
The
o
il
p
al
m
VIRESCENS
gene
cont
rols
fr
uit
co
lour
and
enc
odes
a
R2R
3
-
MY
B,
”
Nature
Comm
unic
ati
ons
,
vol
.
5
,
pp
.
1
-
8
,
2014.
[5]
K.K.
Loo
,
e
t
al
.
,
“
Vi
resce
ns
co
lo
ur
change
and
o
il
ac
cumulation
in
correlation
w
it
h
ripen
ess
in
o
il
palm
fruit
s,”
2nd
Inte
rn
at
ion
a
l
Confer
ence
on
Plant
Sci
ence &
Ph
y
siolog
y
,
201
7.
[6]
N.
Sabri,
et
a
l.
,
“
Eva
lua
t
ion
of
Color
Models
for
Palm
Oil
Fresh
Fruit
Bunch
Ripe
ness
Cla
ss
ifi
c
at
ion
,”
Indone
si
a
n
Journal
of
Elec
t
rical
Engi
ne
erin
g
and
Computer
Sci
en
ce
(
IJE
ECS
)
,
vol.
11
,
no
.
2
,
pp.
549
-
557
,
20
18.
[7]
M.
Makk
y
and
P.
Soni,
“
Deve
lopment
of
an
aut
om
at
ic
gra
d
ing
m
ac
hine
for
oil
pal
m
fre
sh
fruit
s
bunche
s
(FF
B
s)
base
d
on
m
a
chi
n
e
vision
,
”
Computers and
e
lectro
nic
s in
agricultu
re
,
vol
.
93
,
pp
.
1
29
-
139
,
2013
.
[8]
B.
Pam
ornna
k,
e
t
al
.
,
“
An
au
tomatic
and
rap
id
s
ystem
for
gra
ding
p
al
m
bunch
usi
ng
a
Kin
ec
t
c
amera
,
”
Comput
ers
and
Elec
troni
cs
in
Agri
cul
ture
,
v
ol.
143
,
pp
.
227
-
237
,
2017
.
[9]
A.
Jaffa
r,
et
a
l.
,
“
Photogram
m
et
ric
gra
d
ing
of
oil
pal
m
fre
sh
fruit
bunche
s
,”
Int.
J
.
Me
ch
.
Me
chat
r
on.
Eng
,
vo
l.
9
,
pp.
18
-
24
,
2009
.
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.
4
,
A
ugust
201
9
:
23
86
-
2393
2392
[10]
J.
Rosele
en
a,
e
t
al.,
“
As
sessm
ent
of
p
al
m
oil
fre
sh
fruit
bu
nche
s
using
ph
otogra
m
m
et
ric
gra
ding
s
y
s
te
m
,
”
Inte
rnational
Fo
od
Re
search
Jou
rnal
,
vol
.
18
,
no
.
3
,
2011
.
[11]
N.
Sabri,
et
a
l.
,
“
Palm
oil
fre
sh
fruit
bunch
ri
pene
ss
gra
ding
ide
n
ti
f
icati
on
us
ing
col
or
f
eatur
es,
”
Journal
of
Fundament
al
an
d
Applied
S
cienc
es
,
vol
.
9
,
no
.
4S
,
p
p
.
563
-
579
,
2
017.
[12]
S.
Utom
,
et
al
.
,
“
Non
-
Destruc
ti
ve
Oil
Palm
Fresh
Fruit
Bunch
(FF
B)
Gradi
ng
Te
chni
qu
e
Us
ing
Optic
a
l
Sensor
,”
Inte
rnational
Jo
urnal
of
In
te
grat
ed
Eng
ine
ering
,
vol.
10
,
no
.
1
,
pp
.
35
-
39
,
2018
.
[13]
P.
Junkw
on,
et
al.
,
“
H
y
p
erspe
c
t
ral
imaging
for
nondestruc
ti
v
e
det
ermina
ti
on
o
f
int
ern
al
qua
li
t
ie
s
for
oil
pa
lm
(El
a
ei
s gui
n
ee
ns
is Ja
cq. va
r
.
te
n
e
ra)
,
”
Agri
cul
tura
l
Information
Res
earc
h
,
vol
.
18
,
pp.
130
-
141
,
20
09.
[14]
B.
Pam
ornna
k
,
et
al
.
,
“
Oil
con
t
ent
de
te
rm
ina
t
io
n
sche
m
e
of
postharve
st
o
il
pa
l
m
for
m
obil
e
devi
ce
s
,”
Bios
ystem
s
Engi
ne
ering
,
vol
.
134
,
pp
.
8
-
19
,
2015
.
[15]
A.
Ta
p
aru
gss
an
agor
n,
et
al
.
,
“
A
non
-
destruc
t
i
ve
oi
l
pa
lm
rip
ene
ss
rec
ogn
it
io
n
s
y
stem
using
rel
a
ti
ve
en
trop
y
,
”
Computers and Electronic
s
in A
gricul
ture
,
vol
.
1
18,
pp
.
340
-
349
,
2015
.
[16]
N.
Fadil
ah
,
e
t
al
.
,
“
Inte
ll
ig
ent
color
vision
sy
st
em
for
ripe
ness
cl
assific
a
ti
on
of
oil
pal
m
fre
sh
fruit
bunc
h,
”
Sensors
,
vol.
12,
no.
10,
p
p.
14179
-
14195
,
2012.
[17]
M.R.
Putta
sw
a
m
y
and
P.
B
al
a
m
uruga
n,
“
Neura
l
Network
App
roa
ch
to
Ide
nt
if
y
H
y
per
spe
ct
r
al
Im
age
Conte
n
t,”
Inte
rnational
Jo
urnal
of El
e
ct
ri
c
al
and
Comput
er
Engi
n
ee
ring
(
IJE
CE)
,
vol
.
8
,
no
.
4,
pp.
2115
-
212
5
, 2
018
.
[18]
M.
Makk
y
and
P.
Soni,
“
In
si
tu
qualit
y
assess
me
nt
of
intac
t
oi
l
pal
m
fre
sh
fru
it
bunche
s
using
r
api
d
por
ta
bl
e
no
n
-
cont
a
ct
and
non
-
destruc
t
ive
appr
oac
h,
”
J
ournal
o
f
Food
Engi
n
ee
r
ing
,
vo
l. 120, pp.
248
-
259
,
2014.
[19]
K.C.
Goh,
et
al
.
,
“
Evol
ut
ion
of
Prec
ision
Agric
ult
ur
e
Com
puti
ng
towar
d
s
S
ustai
nable
Oil
Palm
Industry
,
”
Indone
sian J
our
nal
of
Elec
tric
al
Engi
ne
ering
and
Computer
Sc
ie
n
ce
(
IJE
ECS)
,
vol
11
,
no
.
2
,
pp
.
72
5
-
732
,
2018
.
[20]
D.D.
Silalahi,
et
al.
,
“
Nea
r
infr
ar
ed
spec
troscop
y
:
a
r
api
d
and
non
-
destruc
t
ive
te
ch
nique
to
assess
t
he
rip
ene
ss
of
oi
l
pal
m
(El
a
ei
s
gu
ine
ensis
Jac
q
.
)
f
resh
fruit
,”
Journal
of
Near
In
frared
Spec
trosco
py
,
vol
.
24
,
no.
2
,
pp.
179
-
190
,
2016
.
[21]
S.
Kasem
su
m
ran
,
et
al
.
,
“
A
fea
sibil
i
t
y
stud
y
o
n
non
-
destruc
ti
v
e
det
ermina
ti
on
of
oil
cont
ent
i
n
pal
m
fruit
s
b
y
vis
ibl
e
–
ne
ar infr
are
d
spe
ct
roscop
y
,
”
Journal
o
f
N
ear
Infrared
Sp
e
ct
ros
copy
,
vol
20
,
no
.
6
,
pp
.
687
-
694
,
2012
.
[22]
K.N.
Basri
,
e
t
al.
,
“
Cla
ss
ifi
c
at
i
on
and
quan
ti
f
i
ca
t
ion
of
p
al
m
oil
adu
lt
e
ration
via
port
able
NI
R
spec
troscop
y
,”
Spec
troch
imic
a
Ac
ta
Part A:
Mo
le
cu
lar an
d
Bi
o
molec
ular S
p
ect
ros
copy
,
vol
.
17
3,
pp
.
335
-
342
,
2017.
[23]
H.A.
Rahi
m
and
S.N.N.
A.
Rani
,
“
Cla
ss
ifi
ca
t
ion
of
Fatty
Fat
Acid
in
Palm
Oil
U
sing
Nea
r
Infra
red
Spect
roscop
y
,”
Sensors
&
T
ran
s
duce
rs
,
vol
.
156
,
no.
9,
pp.
247
-
2
50,
2013
.
[24]
O.M.B.
Saee
d
,
e
t
al
.
,
“
Cla
ss
ifi
cation
of
oil
pa
lm
fre
sh
fruit
bunc
hes
base
d
on
the
ir
m
at
urity
usin
g
p
orta
ble
four
-
band
sensor s
y
s
t
em,”
Comput
ers and
Elec
troni
cs
in
Agr
i
cul
ture
,
v
ol.
82
,
pp
.
55
-
60
,
2012
.
[25]
H
.
K.
Lichte
n
tha
l
er
and
C
.
Busch
m
ann,
“
Chloroph
y
l
ls
and
Caro
tenoids:
Mea
sure
m
ent
and
Charact
e
r
iz
a
ti
on
b
y
U
V
-
VIS
Spect
roscop
y
,
”
Curr
ent P
rot
ocol
s in
Food
A
naly
tical
Chemis
try
,
pp
.
F4.3
.
1
-
F
4.
3.
8
,
2001
.
[26]
D
-
W
.
Sun
,
“
H
y
p
erspe
ct
r
al Im
aging for
Food Qua
li
t
y
An
aly
sis a
nd
Control
”
.
San
D
ie
go:
Aca
d
emic P
ress
,
2010.
[27]
A.S.
Shirkhorshidi
,
e
t
al
.
,
“
A Com
par
ison Stud
y
on
Sim
il
ari
t
y
an
d
Diss
imila
rity
Mea
sure
s in
Clu
steri
ng
Conti
nuo
us
Data
,
”
PLOS
O
NE
,
vol
.
10
,
no
.
12
,
2015
.
[28]
J
.
Ike
m
efu
na
an
d
I
.
Adam
son
,
“
Chloroph
y
ll
An
d
Carot
enoi
d
Ch
ange
s
In
Ripe
ni
ng
Palm
Fruit,
El
aeis
Guinee
nsis,
”
Phy
to
che
mistry
,
v
ol
.
23
,
n
o
.
7
,
pp
.
1413
-
1415
,
198
4
.
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Agu
ng
Wahy
u
Setia
w
an
was
born
in
1982.
He
rec
e
ive
s
th
e
bachel
or,
m
aste
r
,
an
d
doct
ora
l
degr
e
e
in
el
e
ct
r
i
ca
l
engi
n
e
eri
ng
from
Instit
ut
Te
kno
logi
B
andung
(IT
B),
I
ndonesia
,
in
2005,
2008,
and
2
013,
respe
ctively
.
He
is
cur
ren
tly
fac
u
l
t
y
m
ember
of
Sc
hool
of
El
e
ct
r
ic
a
l
Eng
ine
er
ing
an
d
Inform
at
ic
s
,
I
TB.
His
rese
arc
h
in
t
ere
sts
include
m
edi
c
al
signa
l
&
i
m
age
proc
essing
for
e
-
health
app
li
c
at
ion
,
biomed
ic
a
l
spec
troscop
y
.
Dr.
Setiawan
is
a
m
ember
of
t
he
IE
EE
,
profe
ss
iona
l
engi
n
ee
r
(IPM
)
and
AS
EAN
Engi
ne
er
Reg
ist
ere
d.
Richard
Me
ng
ko
gra
duated
fr
om
Instit
ut
T
ek
nologi
Bandung
(IT
B)
in
1978
with
ba
che
lo
r
of
engi
ne
eri
ng
d
eg
ree
.
He
the
n
a
dvanc
ed
and
re
ce
iv
ed
his
doc
t
ora
l
degr
ee
fro
m
I
nstit
ut
National
Pol
y
t
ec
niqu
e
de
Grenobl
e
–
Franc
e
in
1985
.
He
is
fac
ulty
m
embe
r
of
School
of
El
ec
tr
ic
a
l
Engi
ne
e
ring
and
Inform
at
i
cs.
His re
sea
r
ch
in
terests
i
nc
lude
m
edi
cal
imag
e
and
signal
pro
ce
ss
in
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
Cl
as
sif
ic
ation
of palm
o
il
fres
h
fr
uit b
un
c
h u
sing m
ulti
band
opti
cal se
ns
or
s
(
Ag
ung
W
. Se
ti
awa
n
)
2393
Ay
u P
.
H. P
utr
i
rec
e
ive
s
the
b
achel
or
d
egr
e
e
in
p
har
m
acy
and
ph
armac
ist
f
rom
th
e
Insti
tut
Te
kno
l
ogi
Bandung
(IT
B)
,
Indone
sia, i
n
20
10
and
2012
,
res
pec
t
ive
l
y
.
Sh
e
th
en
adv
anced and
recei
v
ed
his
m
aste
r
of
eng
in
ee
r
ing
d
egr
ee (Bi
om
edic
al
Engi
ne
eri
ng)
f
rom
Instit
ut Te
k
nologi
B
andung in
2017.
Don
ny
Danud
i
rdjo
rec
ei
v
es
th
e
bac
he
lor
and
m
aste
r
degr
ee
i
n
el
ec
tr
ical
enginee
ring
from
Instit
ut
Te
knologi
B
and
ung
(IT
B),
Indo
nesia
,
in
20
05
a
nd
2007
.
He
th
e
n
adva
n
ce
d
and
rec
e
ive
d
his
doc
toral
degr
ee
from
Th
e
Univer
sit
y
of
Tok
y
o
–
Japa
n
i
n
2013
.
He
is
c
urre
ntly
f
ac
u
lty
m
ember
of
School
of
El
e
ct
ri
ca
l
Engi
n
ee
ring
and
Info
rm
at
ic
s.
His
r
ese
arc
h
intere
sts
i
ncl
ude
m
edi
cal
imaging
and
sign
a
l
proc
essing.
Alfie
R.
An
anda
rec
e
ive
s
the
b
ac
he
lor
degr
e
e
i
n
el
e
ct
ri
ca
l
enginee
ring
from
th
e
Instit
ut
Te
kno
logi
Nasiona
l
(
I
TE
N
AS
),
Bandung,
I
ndonesia
,
in
20
15
,
resp
ec
t
ive
l
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.