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.
8
, No
.
6
,
Decem
ber
201
8
, p
p.
5389
~
5398
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v8
i
6
.
pp
5389
-
53
98
5389
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Identific
atio
n
of
Plant T
yp
es
by
L
eaf Tex
t
ures Bas
ed on th
e
Bac
kp
ropagati
on Ne
u
ra
l
Net
work
Tauf
ik
Hid
ayat
,
A
syar
oh
R
am
ad
ona
N
il
aw
at
i
Depa
rtment
o
f
I
nform
at
ic
s,
Gunada
rm
a
Univ
ersi
t
y
,
Indon
esia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Sep
30
, 201
7
Re
vised
Feb
2
1
, 2
01
8
Accepte
d
J
ul
22
, 2
01
8
The
num
ber
of
spec
i
es
of
pl
ant
s or
flora
in
Indon
esia
is
abunda
n
t.
The
wea
l
t
h
of
Indone
sia
'
s
flora
spec
ie
s
is
not
to
be
doubted.
Alm
ost
eve
r
y
reg
ion
in
Indone
sia
has
o
ne
or
som
e
disti
nct
iv
e
pla
nt(s)
which
m
a
y
not
exi
st
in
othe
r
count
ri
es.
In
en
hanc
ing
th
e
pot
ent
i
al
dive
rsi
t
y
of
tropi
cal
pla
nt
resourc
es,
good
m
ana
gement
and
uti
l
izati
o
n
of
biodi
ver
sit
y
is
req
uire
d.
B
ase
d
on
such
dive
rsit
y
,
pla
n
t
cl
assifi
ca
t
ion
be
comes
a
challe
n
ge
to
do.
The
m
ost
comm
on
wa
y
to
r
ec
ogni
z
e
be
twee
n
one
p
l
ant
and
ano
the
r
i
s
to
id
ent
if
y
the
l
ea
f
of
e
ac
h
pla
nt
.
L
ea
f
-
b
ase
d
cl
assifi
cation
i
s
an
alter
n
at
iv
e
and
the
m
ost
eff
ec
t
ive
w
a
y
to
do
bec
ause
leave
s
will
exi
st
all
the
ti
m
e,
while
fruit
s
and
flowe
rs
m
ay
o
n
l
y
exi
st
at
an
y
g
iv
en
ti
m
e.
In
thi
s
study
,
the
rese
arc
her
s
will
identif
y
p
la
nt
s
base
d
on
the
t
e
xture
s
of
the
l
e
ave
s.
L
ea
f
f
ea
t
ure
ext
r
ac
t
ion
i
s
done
b
y
ca
l
cul
a
ti
ng
th
e
a
rea
val
u
e,
p
eri
m
et
er
,
and
additio
nal
feature
s
of
l
ea
f
images
such
as
shape
r
oundness
and
sl
ende
rne
ss
.
The
r
esult
s
of
th
e
ex
t
rac
t
ion
will
the
n
be
sel
ec
t
ed
for
training
b
y
using
the
b
ac
kp
ropa
gation
neur
al
ne
twork.
The
result
of
th
e
tra
ini
ng
(th
e
for
m
at
ion
of
the
training
set)
will
be
ca
lc
u
la
t
ed
to
prod
uce
the
v
al
ue
of
rec
ogn
it
i
on
ac
cur
a
c
y
wit
h
which
the
fea
t
ure
val
ue
o
f
the
dataset
of
t
he
le
af
images
is
the
n
to
be
m
at
che
d.
Th
e
r
esult
of
the
ide
nti
f
icati
on
of plant
spec
ie
s ba
s
ed
on
le
af
t
ext
ur
e
cha
r
ac
t
eri
sti
cs
is e
xpec
t
ed
to
acce
l
erate
th
e
proc
ess
of
pla
nt
cl
assifi
ca
t
ion
ba
sed
on
the
ch
aract
er
isti
cs
of
the
le
av
es.
Ke
yw
or
d:
Feat
ur
e
Leaf
P
e
rim
et
er
Ra
ti
o
R
oundnes
s
Slend
e
r
ness
Copyright
©
201
8
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
:
Tauf
i
k Hidaya
t
,
Dep
a
rtm
ent o
f Info
rm
at
ic
s,
Gu
na
da
rm
a U
niv
ersit
y,
Ma
rgonda Ray
a Street
100, P
ondok C
ina,
Depok,
In
donesi
a.
Em
a
il
:
thidaya
t
@staff
.gu
nad
a
rm
a.ac.id
1.
INTROD
U
CTION
Ind
on
esi
a
is
one
of
the
c
ount
ries
with
high
plant
div
e
rsity
.
The
nu
m
ber
of
sp
eci
es
of
plants
or
fl
ora
in
I
ndonesi
a
is
abun
dan
t.
The
wealt
h
of
the
sp
eci
es
of
I
nd
on
e
sia
’s
fl
or
a
i
s
unquest
io
nabl
e
[1
]
.
Alm
os
t
every
reg
i
on
in
Ind
onesi
a
has
one
or
s
om
e
disti
nctive
plant(s
)
wh
ic
h
m
ay
no
t
exist
in
oth
e
r
countries.
N
ot
on
ly
the
div
e
rsity
bu
t
s
om
e
t
ypes
of
pl
ants
in
I
ndonesi
a
ha
ve
m
a
ny
be
nef
it
s
f
or
healt
h.
I
n
en
han
ci
ng
t
he
pote
ntial
div
e
rsity
of
tr
opic
al
plant
res
ources,
good
m
anag
em
ent
a
nd
util
iz
at
ion
of
biodive
rsity
is
required.
G
eran
i
um
plants
i
n
Ind
onesi
a,
as
a
n
e
xam
ple,
ha
ve
so
m
e
sp
eci
es.
Ger
a
niu
m
flo
wer
pla
nts
are
al
so
know
n
a
s
he
rb
al
plants
cal
le
d
ta
pak
dar
a
.
Th
e
ger
a
niu
m
fl
ow
e
r
bel
ongs
to
the
plant
pro
duci
ng
the
essenti
al
oil
and
i
s
cat
egorized
as
a
fa
m
ily
of
Ger
aniace
a.
T
his
plant,
in
the
I
ndonesi
a’
s
herb
al
fa
m
ily,
is
kn
ow
n
as
ta
pak
dar
a
flo
wer
w
hile
i
n
Lat
in
it
is
na
m
ed
as
Palr
go
niu
m
gr
a
veo
le
ns
[2
]
.
The
ot
he
r
plant
s
uch
a
s
j
a
bon
or
in
L
at
in
is
ref
e
rr
e
d
to
as A
nt
ho
ce
phal
us
Ca
dam
ba
wh
ic
h
gro
ws
wild in
the
f
or
est
a
nd
w
hich
has
be
com
e
the
po
pu
la
r
on
e
of
t
he
al
te
r
native
herbs
i
n
I
ndonesi
a
in
rece
nt
ye
ars.
C
urre
ntly
,
m
any
Jabo
n
pla
nts
are
c
ulti
vated
beca
us
e
of
their
ad
va
ntag
es
com
par
ed
t
o
ot
her
plants
[3
]
.
T
he
pla
nt
belongs
to
t
he
one
that
is
ver
y
easy
to
grow
i
n
Ind
on
esi
a,
bec
ause
the
cl
im
ate
supports
t
he
perfect
gro
wth
of
t
his
Ge
ran
i
um
flow
er.
It
is
est
i
m
at
ed
that
there
are
2
m
illi
on
s
of
pla
nt
s
pecies
w
or
l
dw
i
de
that
ha
ve
been
rec
ognized
a
nd
60%
of
the
m
are
in
Ind
onesi
a;
howe
ver,
up
to
the
pr
ese
nt,
the
exact
num
ber
of
how
m
any
plant
s
pec
ie
s
hav
i
ng
bee
n
gro
w
n
in
I
ndonesi
a
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5389
-
53
98
5390
cou
l
d
not
ha
ve
been
e
xactl
y
determ
ined
.
Cu
r
ren
tl
y
there
are
on
ly
8,0
00
s
pe
ci
es
that
hav
e b
een
i
den
ti
fie
d.
The
a
m
ou
nt is
esti
m
at
ed
on
ly
20
per
ce
nt
of
t
he
t
otal flo
ra t
hat e
xists in
I
ndone
sia
[4
]
.
Ba
sed
on
su
c
h
div
ersit
y,
pla
nt
cl
assifi
cat
ion
beco
m
es
a
chall
eng
e
to
do.
The
m
os
t
co
m
m
on
way
to
reco
gniz
e
bet
ween
one
pla
nt
and
an
ot
her
i
s
to
identify
th
e
le
af
of
eac
h
plant.
Lea
f
-
bas
ed
cl
assifi
cat
io
n
is
an
al
te
rn
at
ive
a
nd
the
m
os
t
eff
e
ct
ive
way
to
do
bec
ause
le
a
ve
s
will
exist
al
l
the
tim
e,
wh
i
le
fr
uits
a
nd
fl
ow
e
rs
m
ay
on
ly
exist
at
any g
ive
n
ti
m
e.
Cl
ass
ific
at
ion
of f
r
uit
pla
nts b
ased
o
n
le
aves
ca
n
be
do
ne
on
the
basis
of
the
m
or
phologica
l
char
act
e
risti
cs
of
te
xtu
re
s
that
can
be
obser
ve
d
or
m
easur
e
d
f
ro
m
the
le
aves
or
the
im
a
ges
of
the
le
aves
[5
]
.
So
m
e
researches
relat
ed
to
t
he
pla
nt
ide
ntific
at
ion
base
d
o
n
te
xtures,
m
orp
ho
l
og
y,
a
nd
le
af
colo
rs
ha
ve
be
en
done
by
the
pr
evi
ous
resea
rch
e
rs.
T
he
res
earch
on
cl
assi
ficat
ion
base
d
on
c
olor
te
xtu
r
es
and
le
af
sh
a
pes
w
as
car
ried
out
by
the
resea
r
cher
s
[
6]
by
usi
ng
the
Proba
bili
sti
c
Neura
l
Netw
ork
with
th
e
su
pe
r
vised
t
rai
ning
an
d
th
e
F
eed
f
orwa
rd
st
ru
ct
ur
e.
Ba
ye
s
’
r
ule
of
the
K
ern
el
Fis
her
D
isc
ri
m
inant
A
na
ly
sis
was
us
e
d
to
cl
assify
a
nu
m
ber
of
le
a
f
cat
eg
or
ie
s.
Decisi
on
m
aking
was
ba
sed
on
the
r
es
ult
of
cal
culat
ing
th
e
distance
betwe
en
the
pro
bab
i
li
ty
den
sit
y
functi
on
of
the
char
act
e
risti
c
vecto
r
ba
sed
on
the
rou
ndne
ss
an
d
sle
nd
e
rn
e
ss
of
the
le
af
im
ages
[
6].
T
he
resea
rch
on
i
den
ti
fy
ing
pla
nts
ba
se
d
on
le
af
sh
a
pe
s
was
car
ried
out
by
the
Re
searc
he
r
s
[
7]
us
i
ng
m
ul
ti
la
ye
r
per
ce
ptr
on
ne
ural
net
w
ork
(MLP)
.
T
he
resea
rch
e
rs
use
d
per
ce
ptr
on
with
on
e
weig
ht
la
ye
r
w
hich
has
only
a
li
near
f
un
ct
ion
wit
h
the
i
nput
of
a
ppr
oxi
m
at
ely
6
sp
eci
es
out
of
19
7
le
aves
with
sim
il
ar
st
ru
ct
ur
es
s
uch
as
m
ang
o,
sap
ota,
guava
,
ne
e
m
,
and
cotto
n.
The
res
ult
showe
d
that
MLP
has
a
le
af
cl
assifi
cat
i
on
acc
ur
acy
va
lue
of
88.
20%
[7
]
.
I
den
ti
fi
kas
i
te
rh
ada
p
j
e
nis
ta
na
m
an
Ad
e
niu
m
dilakuk
a
n
[8]
m
eng
gu
nak
a
n
m
et
od
e
Lea
rn
i
ng
Ve
ct
or
Q
ua
ntiza
ti
on
.
T
his
m
et
ho
d
is
use
d
by
th
e
r
es
earche
rs
t
o
cl
assify
aden
i
um
plants
wh
e
re
eac
h
ou
tpu
t
will
re
pr
es
ent
a
cl
ass
or
cat
egory
of
Ad
e
niu
m
.
In
this
m
et
hod
the
re
m
a
y
be
m
ul
ti
ple
ou
t
puts
for
eac
h
cl
a
ss.
T
he
weig
ht
vecto
r
for
a
n
ou
t
pu
t
unit
is
us
ua
ll
y
a
ref
e
r
ence
to
the
cl
a
ss
in
wh
ic
h
the
un
it
is
locat
ed.
T
he
le
arn
in
g
m
eth
od
in
t
his
stu
dy
will
cl
assify
inp
ut
vecto
rs
,
cl
asses
an
d
s
pacin
g
be
twee
n
in
pu
t
vecto
rs.
If
tw
o
input
vecto
rs
ha
ve
ap
pro
xim
a
te
ly
equ
al
sp
ac
ing
,
t
hen
both
i
nput
vect
or
s
w
il
l
be
placed
into
the
sam
e
cl
ass
[8
]
.
A
nothe
r
rese
arch
in
plant
identific
at
io
n
ba
sed
on
le
af
ch
aracte
risti
cs
w
as
al
s
o
cond
ucted
by
t
he
re
searc
her
usi
ng
t
h
e
E
xtre
m
e
Learn
in
g
Ma
chine
(EL
M).
ELM
is
a
sing
le
-
la
ye
r
fe
edforwa
rd
neural
net
wor
k
or
usual
ly
abbre
viate
d
as
SLFNs.
T
here
are
m
any
ty
pes
of
fee
dfo
r
ward
a
rtific
ia
l
ne
ur
al
netw
orks.
T
he
le
arn
in
g
proc
ess
of
ELM
is
m
uch
slow
e
r
than
ex
pected
becau
se
al
l
par
am
et
ers
are
giv
e
n
m
anu
al
ly
and i
te
rati
ve
tu
ning
is require
d o
n ea
ch param
et
er [9].
In
t
his
stu
dy
th
e
resea
rch
e
r
w
il
l
identify
pla
nt
ty
pes
by
the
ir
le
af
te
xt
ur
es
.
Leaf
featu
re
e
xtracti
on
is
done by cal
culat
ing
the a
rea
value,
pe
rim
e
t
er,
an
d
a
dd
it
io
nal f
eat
ures
of
the lea
f
im
ages
su
ch
a
s the ro
undnes
s
and
sli
m
ness
of
the
le
af
s
ha
pes.
T
he
res
ul
ts
of
the
e
xtra
ct
ion
will
then
be
sel
ect
ed
f
or
trai
ning
us
i
ng
t
he
backp
ropa
gation
ne
ur
al
net
w
ork.
T
he
trai
ni
ng
res
ult
(the
f
or
m
at
ion
of
t
he
trai
ning
set
)
will
be
the
cal
cula
ti
on
of
the
value
of
recogn
it
io
n
ac
cur
acy
wit
h
w
hich
the
featu
r
e
value
of
the
dataset
of
t
he
le
af
im
ages
is
t
hen
t
o
be
m
at
ched
.
2.
RESEA
R
CH MET
HO
D
2.1.
D
ata Sets
In
t
his
stu
dy,
t
he
treat
e
d
im
a
ges
a
re
the
le
af
dig
it
al
im
ages
obta
ined
f
ro
m
the U
CI
Ma
c
hi
ne
Le
ar
ning
Re
po
sit
ory
[10]
.
The
data
of
the
pla
nts’
le
af
i
m
ages
hav
e
the
siz
e
of
23
22x
4128
pix
el
s
with
the
jpg
-
form
atted
RGB
i
m
ages.
The
processe
d
im
a
ges
are
ta
ke
n
f
ro
m
the
le
af
dataset
co
m
pr
isi
ng
32
cl
asses
of
plants
as what
ca
n be
seen i
n
Fi
gure
1.
Figure
1. The
e
xam
ple o
f
t
he
l
eaf
dataset
[
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
Id
e
ntif
ic
ation
of
Plan
t
Types
by Leaf
Text
ur
e
s Ba
s
ed
on
th
e
Back
pr
op
agation Ne
ural
... (
T
au
fi
k
Hidayat
)
5391
Each
cl
ass
has
sp
eci
es
with
va
ryi
ng
am
ou
nt
s
of
data
per
c
la
ss.
The
total
nu
m
ber
of
im
a
ges
us
e
d
is
1605 im
ages o
f
plant leaves
. For
e
xam
ple, in
the G
e
ra
nium
class, Figu
re
1
has 15
var
ia
t
ion
s
of
leave
s as
wh
at
can
be
see
n
i
n Fi
gure
2
.
Figure
2. The
e
xam
ple o
f
t
he
l
eaf
dataset
[
10
]
2
.
2
.
Le
af Se
gm
ent
at
i
on
Leaf
im
age
segm
entat
ion
pr
ocess
is
done
to
se
par
at
e
t
he
foregr
ound
f
rom
the
bac
kground
of
eac
h
le
af.
T
he
le
af
im
age seg
m
entat
ion
process
c
onduct
ed by t
he
resea
rch
e
rs
is
don
e
th
rou
gh the
fo
ll
owin
g
st
ages:
a.
Exec
ute
the
bl
ue
c
ha
nn
el
e
xt
racti
on
of
eac
h
le
af
im
age.
The
blu
e
c
hannel
is
us
e
d
as
the
m
ai
n
co
l
or
because
the
blu
e
col
or
has
t
he
hi
gh
e
st
intensit
y
of
the
ot
he
r
tw
o
col
ors
–
re
d
an
d
gre
en
-
of
eac
h
im
age
within
t
he
sam
e ty
pe
of t
he
R
GB c
olo
r
s as
w
hat can
b
e
see
n i
n
Fi
gure
3
.
F
igure
3.
The
c
hannels
of the l
eaf im
ager
y
Her
e
is the m
at
hem
atical
f
or
m
ula for
f
i
nd
i
ng
the b
l
ue
c
ha
nnel
s [
11
]:
)
(
B
G
R
B
g
(1)
F
or
m
ula
(
1)
is
a
gr
ee
n
c
olor
cha
nn
el
wh
il
e
R,
G
a
nd
B
are
se
qu
e
ntial
ly
Re
d,
G
reen,
an
d
Bl
ue
.
Thro
ugh
the
bl
ue
cha
nn
el
,
the
detai
ls
of
the
i
m
age
can
be
seen
cl
early
and
thoro
ughly
whil
e
the
us
e
of
t
he
red
c
hannel
wi
ll
on
ly
disp
la
y
i
m
age
restriction
s
,
and
th
r
ough
the
gree
n
channel
the
i
m
age
can
only
be
par
tl
y
seen
a
nd
there
is
a
lot
of
no
ise
.
I
n
fi
gure
3,
it
can
be
obse
rv
e
d
th
at
le
af
m
or
phol
og
y
ap
pea
rs
m
os
t
con
t
rasti
ng in
the
blu
e c
ha
nne
l com
par
ed
t
o
t
he red
and
gr
ee
n
c
hannels i
n
t
he
R
GB im
age
.
b.
Perfo
rm
the
bin
arizat
io
n
proc
ess
for
the
bl
ue
channel
le
af
i
m
ages.
The
in
pu
t
is
the
or
i
gin
al
i
m
age
and
th
e
ou
t
pu
t
is
the
i
m
age
resu
lt
ed
from
the
bin
ar
y
pr
oce
ss.
T
his
bin
arizat
io
n
c
an
be
perf
or
m
ed
us
i
ng
f
or
m
ula
2
[
12
]
:
T
y
x
f
T
y
x
f
y
x
g
)
,
(
1
)
,
(
0
)
,
(
(2)
The
obj
ect
e
xt
racti
on
f
ro
m
the
backg
rou
nd
is
to
sel
ect
t
he
th
res
ho
l
d
va
lue
T
(T
repr
esents
the
pi
xe
l
m
app
in
g
val
ue
)
that
sepa
rates
the
two
m
od
es
(0
a
nd
1). A
ft
erw
a
rds,
f
or
an
y
po
int
(
x,
y
)
t
hat
sat
isfie
s
f(
x,y)
>
T
is
cal
le
d
the
point
of
t
he
obj
ect
,
oth
e
rw
i
se
cal
le
d
the
ba
ckgr
ound
poi
nt
[
12
]
.
T
he
im
age
res
ulted
from
the b
i
nar
y
proc
ess can
b
e
see
n i
n
Fi
gure
4.
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5389
-
53
98
5392
Figure
4. The
le
af im
age b
ina
rizat
ion
proces
s
c.
Do
a
cl
os
in
g
operati
on
on
the
bin
ary
im
age
t
o
rem
ov
e
the
bl
ack
pix
el
s
in
side
the
le
af
obje
ct
by
enlargi
ng
the
oute
r
boun
dar
y
of
the
f
oregro
und
obj
ect
and
al
s
o
cl
ose
the
sm
all
hole
locat
ed
in
t
he
m
idd
le
of
t
he
obj
ect
with t
he
for
m
ula [
12
]
:
S
S
A
S
A
)
(
(3)
In
the
cl
os
in
g
op
e
rati
on,
the
researc
hers
us
e
disc
-
s
hap
e
d
st
ru
ct
ur
in
g
el
em
ents
to
ad
just
the
sh
a
pe
of
the lea
f
im
age.
The
r
es
ult o
f
t
he
cl
osi
ng
op
e
r
at
ion
ca
n be se
en
in
Fig
ure
5.
Figure
5. The
c
losin
g op
e
rati
on
of the leaf
b
i
nar
y i
m
age
2
.
3
.
Le
af Fe
ature Ex
tra
c
tio
n
The
pr
e
vious
r
esearch
only
use
d
the
te
xture
featur
e
a
s
the
on
e
im
age
featur
e
i
n
ide
ntifyi
ng
t
he
ty
pe
of
a
plant.
Th
e
stu
dy
of
t
he
featur
e
c
har
act
erist
ic
extracti
on
to
wards
t
he
Ethio
pia
c
off
ee
pla
nt
diseas
e
was
done
with
H
S
V
col
or
s
pace
wh
e
re
the
feat
ur
es
of
the
c
offee
le
aves
ha
d
diff
e
re
nt
colo
r
va
riat
ion
s
[
14]
.
The
pr
e
vious
resea
rch
di
d
not
use
the
sh
ape
fe
at
ur
e
w
hich
ca
n
visu
al
ly
sho
w
that
a
ver
y
diff
e
re
nt
plant
has
a
diff
e
re
nt
le
af
s
hap
e
from
oth
er
le
aves.
I
n
this
stu
dy,
the
resea
rc
hers
pr
opos
e
d
the
fea
ture
c
har
act
eri
sti
cs
in
identify
in
g
pla
nt
ty
pes
base
d
on
t
heir
le
a
ve
s
by
l
ooki
ng
at
the
ge
om
e
tric
sh
a
pe
of
t
he
le
af
obj
ect
-
the
rou
ndness
or
the
sle
nder
ness
-
by
cal
culat
ing
the
area
of
the
le
ave.
A
si
m
ple
way
to
c
al
culat
e
the
area
of
a
le
ave
ob
j
ect
is
by
co
unti
ng
the
nu
m
ber
of
pix
el
s
on
the
obj
ect
.
The
le
a
f
feat
ur
e
e
xtra
ct
ion
is
base
d
on
th
e
m
easur
em
ent usi
ng the
obj
ect
geo
m
et
ry appr
oach
w
hich
inc
lud
es:
a.
The
a
rea
value
wh
ic
h
is t
he num
ber
o
f pixels
per
ta
ini
ng in
t
he
seg
m
ented
i
m
age r
e
g
io
n of t
he
le
af
.
b.
The
pe
rim
et
er
is
a
ci
rcu
m
fer
ence
that
ex
presses
the
le
ngth
of
the
s
urroundin
g
e
dg
e
of
the
le
af
im
age
obj
ect
as
what
can
be
see
n
i
n Fi
gure
6
.
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
Id
e
ntif
ic
ation
of
Plan
t
Types
by Leaf
Text
ur
e
s Ba
s
ed
on
th
e
Back
pr
op
agation Ne
ural
... (
T
au
fi
k
Hidayat
)
5393
Figure
6. The
leaf im
age p
eri
m
et
er
The
ed
ge
of
th
e
le
af
obj
ect
is
pr
oce
ssed
us
i
ng
a
chai
n
co
de
,
so
that
the
pe
rim
e
te
r
can
be
cal
culat
ed
us
in
g
t
he
f
orm
ula [8]
:
odd
even
X
X
P
e
r
i
me
t
e
r
(4)
wh
e
re
:
even
X
=
Eve
n
Co
de
odd
X
=
Odd
C
ode
Figure
7
is
an
exam
ple
of
us
i
ng
a
chai
n
co
de
in
cal
culat
ing
the
per
im
et
er
of
the
le
af
im
a
ge.
T
he
area
of the leaf
ob
j
e
ct
is cal
culat
ed
b
y c
ountin
g
th
e num
ber
of
pix
el
s
on the lea
f
obj
ect
.
Figure
7. The
c
hain
c
ode in
de
te
rm
ining
t
he
l
eaf im
age p
eri
m
et
er
c.
The
le
af
im
age
al
te
rn
at
ive
fea
ture
w
hich
is
use
d
t
o
determ
i
ne
the
ed
ge
va
r
ia
ti
on
of
t
he
obj
ect
on
t
he
le
af
i
m
age is b
ase
d on
:
1)
The
shape
r
oundnes
s.
It
is
a
com
par
iso
n
betwee
n
the
obj
ect
a
rea
a
nd
the
pe
rim
et
er
s
qu
a
re
cal
culat
e
d
by u
si
ng the
fo
rm
ula o
f rou
ndness
:
A
P
e
r
i
m
e
t
e
r
R
at
i
o
4
(5)
Wh
e
re
[13]
:
2
r
A
(6)
The
area and the p
erim
et
er v
al
ues
wh
ic
h
ar
e the p
rope
rtie
s o
f
this ci
rcle can b
e cal
culat
ed
on the leaf
i
m
age
reg
io
ns
wh
ic
h
are
e
xtr
act
ed
as
the
ba
sic
fo
rm
of
the
r
oundne
ss
s
iz
e
[
13]
.
T
he
R
rati
o
f
or
a
c
ircl
e
is
wh
ic
h
t
he
m
ini
m
u
m
value
for
each
re
gion
is
.
The
rati
o
R
value
will
pro
duce
val
ues
ra
ngin
g
from
0
to
1,
wh
e
re
the
valu
e
0
is
assum
ed
that
the
le
af
i
m
age
obj
ect
is
ci
rcu
la
r
as
wh
at
ca
n
be
s
een
in
fi
gure
8
of
the
com
po
ne
nt of t
he
ci
rcle
obj
ect
.
The use
of the
le
ng
th
and t
he width
f
eat
u
res
of a leaf
obje
ct
b
ase
d o
n
the
ra
ti
o
can
sho
w
t
hat a leaf
sh
a
pe wit
h
r
ou
nd or sle
nd
e
r
s
pecifica
ti
on ca
n be
disti
nguis
hed to
facil
it
ate
the ide
ntific
a
ti
on
process
of
the
plant ty
pe base
d on it
s leaf
fe
at
ur
es
4
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5389
-
53
98
5394
.Fig
ur
e
8. T
he rep
rese
ntati
on
of the s
ha
pe ro
undness
of a le
af im
age
2)
Shape
Slende
rness.
It
is
the
com
par
ison
for
m
between
the
m
ajo
r
a
xis
le
ngth
a
nd
th
e
m
i
nor
axis
le
ngt
h
as
w
hat
ca
n be
seen i
n
F
ig
ure
9.
Figure
9. The
m
ajo
r
ax
is a
nd
m
ino
r
a
xis le
ngth
of a
le
af
im
age
The
us
e
of
the
le
ng
th
a
nd
the
width
featu
res
of
a
le
af
ob
j
ec
t
based
on
the
rati
o
can
s
how
that
a
le
af
sh
a
pe
with
rou
nd
or
sle
nder
sp
eci
ficat
io
n
c
an
be
disti
ngui
sh
e
d
to
facil
it
at
e
the
id
entifi
c
at
ion
process
of
the
plant ty
pe base
d on it
s leaf
f
e
at
ur
es.
2
.
4
.
Le
af Im
age Tr
aining
b
y
th
e
B
ackp
r
opa
gatio
n Neur
al N
e
twor
k
The
trai
ning
proces
s
perfor
m
ed
to
i
den
ti
f
y
plants
base
d
on
thei
r
le
af
i
m
age
te
xture
s
re
qu
i
res
a
trai
ning
set
of
par
am
et
ers
of
the
le
af
im
age
char
act
e
risti
cs
su
c
h
as
the
fe
at
ur
e
of
t
he
ar
ea
value
,
pe
ri
m
et
er,
and
al
te
r
native
featur
es
of
th
e
le
af
i
m
age
con
sist
in
g
of
fe
at
ur
es
of
rou
ndne
ss
an
d
sle
nder
ness
.
Both
of
these
featur
e
s ar
e the
input to th
e n
e
ur
al
n
et
wor
k.
As
a tria
l, it
is
us
e
d
as m
any a
s 1
60
5
im
ages
of
plant leaves
ta
ke
n
from
32
cl
asses
of
plants.
T
he
featur
e
value
is
then
us
e
d
as
the
input
to
th
e
trai
nin
g
proc
ess
us
in
g
Le
ve
nb
e
r
g
-
Ma
rquardt as
wh
at
ca
n be
se
en
in
Fig
ure
10.
Figure
10.
T
he
sch
em
e
o
f
t
he t
rainin
g proces
s
w
it
h ne
ur
al
ne
twork
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
Id
e
ntif
ic
ation
of
Plan
t
Types
by Leaf
Text
ur
e
s Ba
s
ed
on
th
e
Back
pr
op
agation Ne
ural
... (
T
au
fi
k
Hidayat
)
5395
2
.
5
.
Tr
ainin
g T
arg
e
t
The
init
ia
l
sta
ge
of
the
t
rain
ing
process
is
to
est
ablish
t
he
ta
r
get
m
a
trix
of
32
plant
cl
asses.
F
or
exam
ple,
if
the
re
are
10
cl
ass
es
of
pla
nts,
a
t
arg
et
m
at
rix
is
form
ed
with
a
10x10
m
at
rix
order
a
s
w
hat
can
be
seen i
n
the
form
at
ion
of the
targ
et
m
at
rix
i
n Fi
gure
11.
Figure
11. T
he
il
lustrati
on of
the esta
blis
hme
nt
of
plant cla
ss tar
geted
m
atr
ic
es
2
.
6
.
Le
af Im
age Tr
aining
P
arameter
The
resea
rch
e
r
s
us
e
d
the
bac
kpr
opagati
on
m
et
hod
with
t
wo
hidde
n
la
ye
rs
us
in
g
t
he
e
xtra
ct
ion
of
the
le
af
featu
re
extracti
on
of
featur
e
c
ha
r
act
erist
ic
s.
The
le
af
im
age
recog
niti
on
pro
cess
us
i
ng
t
he
backp
ropa
gation
Ne
ur
al
Net
work
is
do
ne
by
determ
ining
so
m
e
par
am
et
e
rs
as
w
hat
can
be
see
n
i
n
fi
gure
12
of n
e
ur
al
netw
ork
sc
hem
e.
Figure
12. T
he
n
e
ur
al
netw
ork
sc
hem
e
2
.
7
.
Tr
ainin
g Accur
acy
Accuracy
with the
ep
och
is
th
e
rati
o
betwee
n
the
ou
t
pu
t
te
sti
ng
a
nd
the r
es
ulted
outp
ut
w
hich
is
the
n
div
ide
d by th
e
total
trainin
g r
esult as
wh
at
is
g
i
ve
n
i
n
th
e f
ol
lowing
ps
e
udoc
od
e
:
[m
,n
]
=
find (o
utput == t
ar
get
);
accuracy
=
s
um
(m
)/tota
l_i
m
ag
es*
100
Exam
ples o
f
th
e outp
ut traini
ng il
lustrati
ons
and the
outp
ut
resu
lt
s ca
n be s
een in Fig
ure
13.
Figure
1
3
. T
he
il
lustrati
on exam
ple
o
f
the
outp
ut traini
ng
a
nd
the
outp
ut result
of
ger
a
ni
um
leaf i
m
ager
y
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5389
-
53
98
5396
3.
RESU
LT
S
A
ND AN
ALYSIS
3.1.
Ex
peri
ment
al Im
ag
e
Segm
entat
ion
done
to
t
he
pla
nt
cl
ass
is
base
d
on
the
le
a
f
fe
at
ur
e
te
xt
ur
e
a
s
wh
at
ca
n
be
s
ee
n
in
ta
ble
1.
As
can
be
s
een
in
Ta
ble
1,
the
overall
im
ages
ha
ve
s
ucc
essfu
ll
y
been
s
egm
ented
well
so
that
the
e
dges
of
the
le
af
i
m
ages
are
well
s
egm
ented
as
wh
at
can
be
see
n
in
Figure
13.
Cl
a
ssific
at
ion
is
done
with
re
fer
e
nce
t
o
the
sp
eci
es
of
each
plant
ty
pe
as
wh
at
can
be
seen
in
Ta
bl
e
3.
As
can
be
seen
in
Table
2,
the
re
is
an
error
in
the
res
ult
of
cl
assifi
cat
ion
.
T
he
or
i
gin
al
im
age
of
a
Cr
oton
is
know
n
as
the
Ly
chee
ty
pe
le
af
im
age.
This
i
s
du
e
t
o
the
la
c
k
of
s
pecies
in
the
pla
nt
cl
ass
dataset
f
or
a
Croton
s
o
t
ha
t
the
si
m
il
arity
of
e
dges
between
a
Croton a
nd a
L
yc
hee r
es
ults i
n
s
uc
h
cl
assifi
c
at
ion
e
rror
s
as
wh
at
ca
n be
se
en
in
Fig
ure
14.
Table
1.
T
he
L
eaf I
m
age S
e
gm
entat
ion
Table
2.
E
rro
r
in
the Res
ult
of Cl
assifi
cat
ion
Plan
t Class
Bin
ary
I
m
ag
e
Seg
m
en
tatio
n
Resu
lt
Gera
n
iu
m
Ash
an
ti
Blo
o
d
Bitter
Orang
e
Ch
o
co
late
Tr
ee
Egg
Plan
t
Ficu
s
Cro
to
n
Ly
ch
ee
Pap
ay
a
Sweet
Po
tato
Plan
t Class
Origin
al
I
m
ag
e
Back
p
rop
ag
atio
n
Neu
ral
N
etwo
rk
1
1
2
2
3
3
4
2
5
5
6
6
7
8
8
8
9
9
10
10
Figure
13. T
he
i
m
age seg
m
entat
ion
pr
ocess of
a
bitt
er orange
leaf
Figure
14. T
he
ex
am
ple
of th
e erro
r
in
classi
fyi
ng a
cro
t
on leaf
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
Id
e
ntif
ic
ation
of
Plan
t
Types
by Leaf
Text
ur
e
s Ba
s
ed
on
th
e
Back
pr
op
agation Ne
ural
... (
T
au
fi
k
Hidayat
)
5397
In
Ta
ble
2,
the
re
is
al
so
a
n
er
ror
in
ide
ntifyi
ng
t
he
ty
pe
of
Cho
c
olate
Tre
e
le
af
that
is
identifie
d
as
Ash
a
nti
Bl
ood
le
af.
T
his
is
due
to
the
sim
ilar
val
ue
of
the
sh
a
pe
rou
ndne
ss
of
bo
t
h
s
pe
ci
es
as
w
hat
can
be
seen i
n
Fi
gure
15.
Table
3.
T
he
Num
ber
of P
la
nt Cla
ssific
at
ion
s
Clas
sif
icatio
n
Nu
m
b
e
r
Plan
t
Clas
sif
icatio
n
1
Gera
n
iu
m
2
Ash
an
ti Blo
o
d
3
Bitter O
rang
e
4
Ch
o
co
late
Tr
ee
5
Egg
Plan
t
6
Ficu
s
7
Cro
to
n
8
Ly
ch
ee
9
Pap
ay
a
10
Sweet Potato
Figure
15. T
he
ex
am
ple o
f
th
e erro
r
in
classi
fyi
ng a
cho
c
olate
tree l
eaf
Fr
om
the
over
al
l
resu
lt
s
of
t
he
cl
assifi
cat
io
n
trai
ni
ng
tria
ls
on
32
plant
c
la
sses
with
16
0
5
pla
nt
le
af
i
m
ages
there
w
ere
cl
ass
ific
at
ion
e
rro
rs
in
id
entify
ing
48
spe
ci
es
ou
t
of
1557
s
pecies
wh
i
ch
we
re
s
ucces
sfu
ll
y
identifie
d wit
h
the r
es
ulted
ac
cur
acy
of 97%
cal
culat
ed by :
%
1
0
0
D
P
e
r
i
m
e
t
e
r
A
c
c
u
r
a
c
y
(7)
Wh
e
re
P
is
a
com
par
ison
be
tween
plant
cl
ass
sp
eci
es
tha
t
are
correct
ly
cl
assifi
ed
by
the
total
nu
m
ber
of
sp
ec
ie
s.
T
he
r
esearche
r
a
djust
ed
the
c
onne
ct
ion
weig
ht
duri
ng
t
he
trai
nin
g
proce
ss
of
a
num
ber
of
da
ta
set
s
ta
ken
f
ro
m
the
resu
lt
of
the
e
xtracti
on
of
th
e
le
af
featur
e
c
har
act
erist
ic
s
to
m
ini
m
iz
e
the
err
or
val
ue.
I
n
this
stud
y,
the
c
hai
n
r
ule
was
im
ple
m
ented
to
cal
culat
e
the
infl
ue
nce
of
eac
h
w
ei
gh
t
on
the
e
rror
functi
on
in
order
to m
ini
m
iz
e the leaf im
age id
entifi
cat
ion
e
rror.
4.
CONCL
US
I
O
N
The
segm
entat
ion
of
1605
le
af
i
m
ages
has
bee
n
su
c
cessf
ully
do
ne
us
ing
the
m
orp
ho
l
og
ic
a
l
op
e
rati
ons.
T
he
extracti
on
of
the
te
xtu
re
c
ha
racteri
sti
cs
ha
s
al
so
bee
n
suc
cessf
ully
do
ne
based
on
the
area
value,
the
pe
rim
et
er
and
su
c
h
a
dd
it
io
nal
fe
at
ur
es
of
the
le
af
im
ages
as
th
e
rou
ndness
an
d
the
sli
m
ness
of
t
he
le
af
sh
ape
.
Th
e
trai
nin
g
of
th
e
extracti
on
of
le
af
i
m
age
char
act
erist
ic
s
w
as
su
ccess
fu
ll
y
per
f
orm
ed
us
ing
th
e
backp
ropa
gation
neural
netw
ork.
Ba
sed
on
the
ov
e
rall
resu
lt
s
of
the
cl
assifi
cat
ion
te
sti
ng
tria
l
on
32
cl
asses
of
plants
with
1605
pla
nt
le
af
i
m
ages
there
was
a
cl
assifi
c
at
ion
er
r
or
of
48
s
pecies
out
of
15
57
sp
eci
es
wh
i
c
h
wer
e
su
cce
ssf
ul
ly
iden
ti
fied,
r
esulti
ng in
a
n
a
ccur
acy
of 97
%.
ACKN
OWLE
DGE
MENTS
We
are
in
deb
t
ed
to
t
he
e
xp
e
rts
w
ho
ha
ve
con
t
rib
uted
t
owar
ds
de
velo
pm
ent
of
t
he
te
m
pla
te
.Th
e
auth
or
s
wo
uld
li
ke
to ac
know
le
dg
e t
o Guna
da
rm
a U
niv
er
sit
y.
REFERE
NCE
S
[1]
Pers
oon
and
W
ee
rd,
“
Biodi
v
ersity
and
Natur
al
R
esourc
e
Man
agem
ent
in
Insular
Southea
st
As
ia
”
,
Island
Studies
Journal,
Vol
.
1
,
No.
1,
pp.
81
-
10
8
,
2006
.
[2]
Shankar
,
Ahm
ad,
Pasrichaa
an
d
Sastr
y
,
“
Bior
educ
t
ion
of
Chloroa
ura
te
Ions
by
Gera
nium
Le
ave
s
and
It
s
Endoph
y
ti
c
Fun
gus Yie
lds Gol
d
Nanopa
rticle
s
of
Diffe
r
ent
Shap
e
s
”
,
Journal
of
M
ate
rials Ch
emist
ry
,
Iss
ue
7
,
2003
.
[3]
Melly
Br
Bang
un,
Yeni
Herdi
y
en
i,
Elis
Nina
Herl
i
y
a
na
,
“
Morphological
Fe
at
ure
Ext
r
action
of
Jabon’s
L
e
af
Seedl
ing
Pa
thog
en
using
Micros
copi
c
Im
age
”
,
TEL
KOMNIKA
(Tele
communic
a
t
ion,
Computing
,
El
e
ct
roni
c
and
Control)
,
Vol.
14
,
No.1
,
Pp.
254
-
261,
2016
.
[4]
Le
m
baga
Ilmu
Penget
ahu
an
In
donesia
,
“
Biore
s
ourc
es
for
Gree
n
Ec
onom
y
De
vel
opm
ent
(in
Baha
sa)”
,
Jaka
r
t
a
:
LIPI,
2013
.
[5]
Yan
Qing,
Lian
g
Dong,Z
hang
J
ingj
ing,
“
Rese
ar
ch
of
Plant
-
L
eaves
Cla
ss
ifi
c
at
io
n
Algorit
hm
base
d
on
Supervise
d
LLE
”
,
TEL
KOM
NIKA
(
Tele
com
municat
ion
,
Co
mputing,
El
e
ct
ro
nic
and
Control
)
,
Vol.
11,
No.
6,
pp.
3265
-
327
0,
2013.
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5389
-
53
98
5398
[6]
Kadir
,
Nugroho
,
Sus
ant
o
and
Sa
ntosa,
“
Perform
anc
e
Im
prove
m
ent
of
L
ea
f
Ide
nt
ifi
c
at
ion
S
y
stem
Us
ing
Princi
pa
l
Com
ponent
Ana
l
y
sis
”
,
In
te
rnatio
nal
Journal
of
A
dvanc
ed
Sc
ie
nc
e
and
Techno
logy
,
Vol
.
44
,
2012.
[7]
Kadir
,
Nugroho
,
Sus
ant
o
and
Santosa,
“
Le
a
f
Cla
ss
ifica
t
ion
Us
ing
Shape,
Color,
and
T
ext
ure
Feat
ur
es
”
,
Inte
rnational
Jo
urnal
of
Comput
er
Tr
ends
and
Technol
og
y
,
2011
.
[8]
Rest
y
W
ula
nn
in
grum
,
Bagus
Fa
dze
ri
e
Robb
y
,
“
Le
arn
ing
Vec
tor
Quanti
zation
I
m
age
for
Ide
nti
f
ic
a
ti
on
Adenium
”
,
Indone
sian J
our
nal
of
Elec
tric
al
Engi
ne
ering
and
Computer
Sc
ie
n
ce
(
IJE
ECS
),
Vol.
4
,
No
.
2
,
pp
.
3
83
-
389,
2016
.
[9]
Chuan
-
Min
Zha
i
,
Ji
-
Xiang
Du,
“
Appl
y
ing
ext
re
m
e
le
arn
ing
m
achine
to
pl
ant
spe
ci
es
ide
n
ti
fi
catio
n
”
,
Int
ernati
ona
l
Confe
renc
e
on
I
nformation
and
Aut
omation
(
ICIA)
,
2008.
[10]
Le
af
Dat
ase
t
,
ht
t
ps://
ar
chi
ve
.
i
cs.
u
ci
.
edu/
m
l/
da
ta
se
t
s/le
af
,
Ac
ce
s Da
t
e
:
April
5,
2017
[11]
Sus
et
ia
ningtias
D.T
,
Made
nda
,
Rahay
u
D.A
,
R
odia
h,
“
Ret
in
al
Microa
neur
s
y
m
Dete
c
ti
on
using
Maximally
St
able
Ext
ern
al
Regi
on
Algorit
hm
”
,
Inte
rnational
Jour
nal
on
Adv
anced
Sci
ence
Engi
n
ee
ring
Informati
on
Technol
ogy
,
Vol.
6
,
No.5
,
IS
SN
:
2088
-
5334,
2016.
[12]
Gonza
lez
and
W
oods,
“
Digit
al
Im
age
Proce
ss
ing
”
,
Th
ird
Editi
on,
Pear
son
Prenti
c
e
Hall.
ISBN
0
-
13
-
168728
-
x,
200
8
[13]
Savel
i
ev
Pete
r
,
“
Mea
suring
obje
ct
s
,
Com
pute
r
Vision
&
Math
cont
ai
ns:
m
at
h
emati
cs
cour
ses
,
cove
rs:
image
ana
l
y
sis
and
d
ata
an
aly
sis
,
provi
des:
image
an
alys
is soft
ware
art
i
c
le
”
,
Ava
il
ab
le From
:
htt
p
:/
/
inpe
r
c.
com/
,
2011
.
[14]
Abrham
De
basu
Mengistu,
Seffi
Gebe
y
ehu
Meng
istu,
Dagna
ch
ew
Mele
sew,
“
An
Autom
at
ic
Coffe
e
Plant
Disea
ses
Ide
nti
f
ic
a
ti
on
Us
ing
H
y
br
id
Appr
oac
hes
of
Im
age
Proce
ss
ing
and
Dec
ision
Tre
e
”
,
Indone
sian
Jour
nal
of
Elec
tri
cal
Engi
ne
ering
and
Computer
Sc
ie
n
ce
(
IJEECS)
,
Vol.
9
,
No
.
3
,
pp
.
8
06
-
811,
2018
.
Evaluation Warning : The document was created with Spire.PDF for Python.