In
d
onesia
n
Jo
u
rn
al
of
E
l
ect
rical En
gin
eerin
g an
d
Compu
t
er
S
cien
ce
V
ol
.
1
4
,
N
o.
1
,
A
pril
201
9
,
pp.
3
68
~
374
I
SS
N
:
25
02
-
4752
,
D
O
I
:
10
.
11591/ijeecs.
v1
4
.i
1
.
pp
368
-
374
368
Jo
u
rn
al
h
om
e
page
:
http:/
/
ia
es
core.
com/
journals
/index.
ph
p/ijeecs
Convol
utional
ne
ural net
wor
k vs bag of
feature
s for bamb
ar
a
ground
nut lea
f di
sease re
cogniti
on
Ha
f
i
z
atu
l
Ha
n
i
n
Ha
mzah
1
,
Nu
r
b
ai
t
y S
ab
ri
2
,
Z
ai
d
ah
Ibra
h
i
m
3
,
D
i
n
o I
sa
4
1,
3
Facult
y
of
Co
m
put
er and
Mathe
m
ati
ca
l
Sci
enc
es,
Un
i
ver
s
i
ti T
e
knol
ogi M
ARA
,
Shah
Al
a
m
,
Sel
a
ngor,
Ma
l
aysi
a
2
Facult
y
o
f C
o
mputer
and Mat
he
m
at
i
ca
l
Sc
i
enc
es,
Un
i
ver
s
i
ti T
ekn
ol
ogi M
ARA
,
C
ampus
Jas
i
n
,
Me
l
aka
,
Mal
ays
i
a
4
CON
N
E
CT Init
i
at
i
ve,
Crop
s f
or
t
he F
u
t
ure
,
Jal
an
Brog
a,
Se
m
enyi
h,
Se
l
angor
,
Mala
ysi
a
A
rt
i
cle In
f
o
A
B
S
T
R
A
C
T
A
r
t
i
c
l
e
h
i
story
:
R
ece
i
ve
d
S
ep
2
,
20
18
R
e
vi
se
d
N
ov
30
,
20
18
A
ccept
e
d
D
ec
12
,
2018
T
his
pape
r
i
nve
sti
gates
bambara
groundnut
l
eaf
disea
se
re
cog
ni
tion
us
i
ng
t
wo
popu
l
ar
t
ec
hni
ques
known
as
Convol
u
t
i
ona
l
Neura
l
Net
work
(CNN
)
and
Bag
of
Features
(BOF
)
w
i
t
h
Speede
d
-
up
Robust
Feature
(SU
RF
)
and
Support
Vec
t
or
Mac
hi
ne
(SV
M)
classi
fie
r.
L
ea
f
disea
se
re
cogniti
on
has
at
t
ra
cted
m
any
re
sea
rc
her
s be
ca
use
t
he
outc
om
e
i
s
use
f
ul
for
f
ar
m
ers.
One
of
t
he
cro
ps
t
hat
prov
i
de
h
i
g
h
i
ncome
f
or
f
armers
i
s
ba
m
bar
a
groundnut
bu
t
the
l
ea
ves
are
ea
si
l
y
i
n
f
ec
t
ed
wit
h
d
i
sea
ses
es
pec
i
all
y
a
f
t
er
t
h
e
ra
i
n.
T
h
is
could
aff
ec
t
t
he
cro
p
produc
t
i
vit
y.
T
hu
s,
au
t
oma
t
i
c
disea
se
re
c
ogni
tion
i
s
cru
c
i
al.
A
new
dataset
t
ha
t
cons
i
s
t
s
o
f
400
imag
es
of
t
he
i
n
f
ec
t
e
d
and
non
-
i
n
f
ec
t
ed
l
ea
ves
of
bambara
groundnut
has
bee
n
const
ruc
t
ed.
T
he
e
xper
i
m
ental
re
su
l
ts
i
nd
i
ca
t
e
t
hat
bo
t
h
o
f
t
h
ese
t
ec
hni
ques
produc
e
excel
l
ent
l
ea
f
d
i
sea
se
re
cognit
i
on
ac
cur
ac
y.
K
ey
w
ord
s
:
BoF
C
N
N
Leaf
disease
r
ecogn
i
ti
on
S
U
R
F
S
V
M
Copyri
ght ©
201
9
Institu
t
e of A
d
vanc
ed E
ngi
nee
ri
ng
and
Scienc
e.
A
ll r
i
ghts
reserv
ed
.
Corres
po
n
din
g Auth
or
:
H
afi
za
t
ul
H
anin H
am
zah
F
acul
ty
of
C
om
pu
t
er
an
d Ma
t
hem
at
i
ca
l
S
cien
ces,
U
niv
er
siti
Te
knol
ogi
M
A
R
A
,
S
hah
A
l
am
, S
e
l
an
gor
,
M
alay
s
i
a
Em
ail:
hani
nh
a
m
zah9
4@
gma
il.
com
1.
IN
T
R
ODU
C
T
ION
B
am
bara
gro
undn
ut
,
or
the
s
cient
i
f
i
c
nam
e
V
i
gna
sub
t
er
r
anea
(
L)
V
er
dc.
is
ori
gin
a
lly
plan
t
e
d
in
the
A
frican
c
onti
ne
nt
an
d
has
be
en
c
ultiv
a
t
e
d
in
t
r
opi
ca
l
A
fr
ic
a
f
or
cen
t
ur
i
e
s
[
1]
.
I
t
has
be
en
pl
a
nted
in
M
a
l
ays
i
a
du
e
to si
m
ilar
wea
t
her co
ndit
i
on
but
one
of
the ch
a
llenges
pl
a
ntin
g
it h
er
e
i
s tha
t
it
can
e
asi
l
y
be
infec
t
e
d
w
it
h
le
af
di
sea
ses
aft
er
hea
vy
r
ai
ns
[1
]
.
I
n
order
t
o
m
i
nim
i
ze
the
l
ea
f
disease
t
ha
t
induce
d
dam
age
dur
ing
t
he
growt
h
of
ba
m
bara
groun
dnut,
harve
st
a
nd
pos
t
-
harves
t
process
ing
,
as
w
ell
as
m
axim
i
ze
pro
du
c
ti
vity
a
nd
ens
ure
agric
ultura
l
s
usta
i
na
bi
l
it
y,
au
tom
ati
c
l
ea
f
disease
r
ecogn
i
ti
on
i
s
highly
i
m
por
t
a
nt
[
2].
The
ex
i
s
ti
ng
m
eth
od
fo
r
l
ea
f
pla
nt
disease
r
ecognit
i
on
i
s
si
m
ply
a
pply
ing
the
na
ke
d
eye
obser
va
tion
by
e
xp
er
t
s
[
3].
I
n
doing
so
,
a
l
ar
ge
t
eam
of
ex
pe
r
t
s
as
w
el
l
as
cont
i
nu
ous
m
onit
ori
ng
of
pl
a
nt
i
s
r
equired,
w
hi
c
h
i
nc
ur
co
st
s
fo
r
l
ar
ge
farm
s
[3]
.
P
l
ant
diseas
e
r
ecog
ni
t
io
n
by
visu
a
l
w
ay
i
s
m
ore
l
abo
r
i
ous
a
nd
t
i
m
e
consum
in
g
an
d
at
th
e
sam
e
t
i
m
e,
l
ess
accur
ate a
nd
can
be done
only
i
n l
i
m
ite
d ar
e
a
s
[4
]
.
I
n
order
to
a
da
pt
to
t
his
fas
t
ch
an
ging
e
nvi
r
on
m
e
nt,
ap
pro
priat
e
an
d
t
i
m
el
y
plan
t
le
af
disease
r
ecognit
i
on
i
s
cr
uci
a
l
.
H
owever,
m
os
t
pla
nt
l
eaf
diseases
gen
er
a
t
e
so
m
e
kin
d
of
m
anif
esta
t
io
n
in
the
visib
l
e
sp
ec
t
r
um,
so
t
he
na
ked
ey
e
exam
i
na
ti
on
of
a
t
r
ain
e
d
pro
fe
ssion
a
l
i
s
t
he
pr
i
m
e
t
echni
qu
e
adopt
e
d
in
pr
acti
c
e
fo
r
pla
nt
disea
se
r
ecognit
i
on
[5
]
.
A
n
au
t
om
ated
plan
t
l
eaf
disease
r
eco
gni
ti
on
sy
s
t
em
coul
d
be
of
gre
at
he
lp
fo
r
am
ate
urs
in
t
he
garde
ning
process
a
nd
als
o
t
r
ai
ned
prof
e
ssio
nals
as
a
ver
i
f
i
cat
i
on
system
i
n
diseas
e
di
a
gnost
i
cs
[6]
.
V
ar
i
ou
s
fe
atu
r
es
a
nd
c
la
ssifi
er
s
hav
e
been
inv
e
sti
gat
e
d
t
o
r
ec
ognize
pl
a
nt
diseases
automa
tica
l
ly
[
7
]
-
[
10
]
.
C
ol
our
feature
s
a
nd
B
ac
k
-
P
r
op
aga
tion
N
e
ural
N
e
tw
ork
(
B
P
N
N
)
hav
e
be
en
us
e
d
fo
r
cotton
a
nd
gro
undnut
disease
s
class
i
f
i
ca
tion
[
7
].
S
hap
e
an
d
co
lou
r
fe
at
ur
es
w
i
t
h
S
up
por
t
V
ect
or
M
a
chine
(
SVM)
classifier
hav
e
bee
n
util
i
ze
d
t
o
c
l
ass
i
f
y
r
i
ce
-
plan
t
disease
s
[
8
].
SVM
has
als
o
be
en
us
e
d
to
cl
assif
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndon
es
i
a
n J
El
ec
En
g &
C
om
p S
ci
I
SS
N
:
2502
-
4752
Con
v
olut
i
on
al
neura
l
ne
t
work
vs
ba
g o
f
f
ea
tu
res
f
or
bam
bar
a grou
ndnu
t
l
e
af
…
(
H
afi
z
atul
H
an
in
H
amza
h
)
369
cotton
l
ea
f
s
pot
disease
in
[
9
].
A
com
parative
s
tudy
has
bee
n
perfo
r
m
ed
am
ong
vari
ou
s
t
e
xtu
r
e
fe
atu
r
es
nam
ely
Loca
l
Bin
ar
y
P
at
t
er
n
(
LB
P
)
and
G
r
ay
Le
vel
Co
-
oc
curr
ence
M
a
tr
i
x
(
G
LCM)
and
c
l
ass
ifi
er
s
that
ar
e
P
r
ob
ab
i
lis
t
ic
N
eural
N
etw
ork
(
P
NN),
B
P
NN,
S
V
M
and
R
a
ndom
F
orest
(
R
F
)
t
o
cla
ss
i
f
y
diseases
i
n
grapes
and
the
r
esult
s
i
nd
i
cate
t
ha
t
G
LC
M
w
i
th
RF
achi
eve
t
he
bes
t
r
ecogn
i
ti
on
r
esult
s
[
1
1
].
Conv
olutio
nal
N
eural
N
etwork
(
C
N
N
)
i
s
getting
popul
ar
in
obje
ct
r
ecogn
itio
n
pro
blem
s
suc
h
as
l
ea
f
r
ecogn
i
ti
on
[
13
-
14]
,
frui
t
r
ecognit
i
on
[15
-
16]
,
chara
cter
r
ecog
ni
t
io
n
[17],
vehic
l
e
r
ecognit
i
on
[
18
]
a
nd
pa
l
m
oil
fresh
fru
it
bu
nc
h
r
i
pe
ness
grad
ing
r
ecog
niti
on
[19].
P
l
ant
di
sease
clas
sific
ation
based
on
C
N
N
pro
duce
out
s
t
a
nding
a
ccura
cy
r
esult
s
[
20]
.
B
oF,
on
e
of
t
he
m
any
m
ach
i
ne
l
ear
ning
t
ech
niques,
has
a
l
s
o
show
n
good
pe
r
fo
r
m
ance
in
obj
e
c
t
r
ecognit
i
on
[
21
-
22]
.
D
ue
t
o
prom
i
s
i
ng
r
esu
l
ts
pro
duced
by
BoF
an
d
C
N
N
,
t
hi
s
r
esear
ch
pla
ns
to
i
nve
stig
a
t
e
t
he
i
r
perfo
r
m
ances
in
r
ecog
ni
zing bam
bara
gro
undnut
l
eaf
disease.
2.
R
E
S
E
A
R
C
H METHOD
2.1.
C
onvo
l
u
t
i
ona
l
N
eu
ral
N
et
wo
r
k
(CNN)
The
ar
ch
itec
t
ur
e
of
C
N
N
i
s
st
r
uc
t
ure
d
as
a
ser
i
es
of
l
a
ye
r
s,
t
hat
co
nsis
ts
of
t
hree
l
a
ye
r
s
w
hi
c
h
ar
e
conv
olv
e
l
ayer
,
po
ol
i
ng
l
ayer
and
R
ec
tified
Linear
unit
(
Re
Lu)
[
16
]
.
C
on
volve
l
a
yer
ex
tr
acts
fea
t
ures
of
a
n
i
m
age
us
in
g
f
ilt
er
a
nd
i
m
age
pa
t
ch
tha
t
s
t
r
ides
ov
er
t
he
in
put
i
m
age.
R
e
Lu
l
ayer
r
ep
l
a
ces
al
l
neg
a
tive
pixe
l
values
i
n
the
f
eatu
r
e
m
a
p
w
ith
zer
o
w
hil
e
poolin
g
l
a
yer
allows
the
fea
tur
e
m
ap
t
o
be
do
w
n
-
sam
ple
d
af
t
er
R
eL
u
l
ayer
t
o
r
edu
ce
the
dim
ens
i
ona
l
it
y.
M
a
x
-
pool
i
ng
com
put
es
the
m
axi
m
um
lo
cal
of
fea
t
ure
m
ap.
N
eighbori
ng
pool
i
ng
t
a
kes
inpu
t
from
feature
m
aps
t
hat
ar
e
sh
i
ft
e
d
or
st
r
i
de
by
m
ore
t
ha
n
on
e
r
ow
s
or
columns.
F
i
gu
r
e
1 sho
w
s
t
he a
r
chitecture
of
a
C
N
N
[2
3].
F
i
gu
r
e
1.
T
he
ar
c
hi
te
c
t
ur
e
of
CN
N
[26]
2.2
.
B
ag
of
Featu
res
(
B
oF)
O
ne
m
ethod
that
r
eprese
nts
i
m
ages
as
or
derless
c
oll
ec
tions
of
l
oca
l
feature
s
i
s
ca
lled
B
a
g
of
F
eatu
r
es
(
B
oF)
[22].
I
n
t
hi
s
pr
oj
ec
t
,
S
peed
ed
up
Ro
bust
F
eatu
r
es
(
S
U
R
F)
has
been
us
e
d
in
BoF
beca
us
e
th
e
perfo
r
m
ance
of
this
fea
t
ure
i
s
excelle
nt
a
nd
only
r
equire
lo
w
com
put
at
iona
l
cos
t
[24].
I
t
i
s
a
de
scr
i
pt
or
t
ha
t
i
s
base
d
on
H
ess
i
an
m
atri
x
m
easure
s
a
nd
an
i
m
age
d
e
t
ect
or.
F
or
a
descri
pto
r
w
hi
c
h
us
es
only
64
di
m
e
nsions
l
ead
ing
t
o
quic
k
fea
t
ure
ex
t
r
actio
n,
and
it
als
o
us
es
a
2D
H
aa
r
w
avel
et
t
r
ans
f
orm
[2
4].
The
t
w
o
com
m
on
perspec
t
ives
f
or
t
he
BoF
i
m
ag
e
r
eprese
ntati
on
e
xpl
a
nation
w
hi
c
h
t
he
f
i
r
st
on
e
i
s
the
by
a
nalo
gy
fr
om
t
he
B
a
g
of
W
ords
r
epr
esen
t
at
ion
.
O
ne
r
eprese
nt
s
a
do
c
ume
nt
that
no
r
m
al
i
zes
his
t
ogr
am
of
w
ord
co
unt
s
w
ith
B
a
g
of
Wo
r
ds
,
[2
5].
Fi
gur
e
2
sh
ows
the
p
r
ocess
f
or
BoF
i
m
age
r
ep
r
esent
a
t
io
n
.
F
i
gu
r
e
2.
Process
fo
r
B
oF
i
m
age
r
eprese
nt
a
ti
on
[
22]
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
25
02
-
4752
I
ndon
es
i
a
n J
El
ec
En
g &
C
om
p S
ci,
V
ol
.
1
4
,
N
o.
1
,
A
pril
201
9
:
368
–
374
370
3.
R
E
S
U
L
T
S
A
N
D
A
N
A
L
YS
IS
3.1.
T
h
e
D
ataset
A
new
dat
ase
t
of
the
bam
ba
r
a
groundn
ut
le
af
i
m
ages
ha
s
been
c
onst
r
uc
t
ed
that
c
onsists
of
20
0
i
m
ages of
t
he
n
on
-
infec
t
e
d
l
e
aves
a
nd 2
00
im
ages
of l
ea
ve
s
w
ith
disease
s
.
They
w
er
e
ca
ptu
r
ed
from
a
f
ar
m
i
n
S
em
eny
i
h,
S
el
angor
usi
ng
a
m
o
bil
e
phone.
S
ome
sam
pl
e
i
m
ages
of
b
a
m
bara
groun
dnut
w
i
th
a
nd
w
i
thou
t
diseases
ar
e
i
l
lus
t
r
ated
in
F
i
gur
e
3 and
F
i
gur
e
4.
F
i
gu
r
e
3.
S
ome
sam
pl
e
i
m
ages
of
bam
bara
groundn
ut
w
itho
ut
l
ea
f d
i
seases
F
i
gu
r
e
4.
S
ome sam
pl
e
i
m
ages
of
bam
bara
groundn
ut
w
ith
l
eaf
diseases
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndon
es
i
a
n J
El
ec
En
g &
C
om
p S
ci
I
SS
N
:
2502
-
4752
Con
v
olut
i
on
al
neura
l
ne
t
work
vs
ba
g o
f
f
ea
tu
res
f
or
bam
bar
a grou
ndnu
t
l
e
af
…
(
H
afi
z
atul
H
an
in
H
amza
h
)
371
3.2
.
CNN
A
st
ack
of
C
N
N
con
s
i
s
t
of
c
onvolve
l
a
yer,
poolin
g
l
ayer
and
R
e
Lu
l
a
ye
r
w
hi
l
e
ad
diti
onal
st
ac
k
of
l
ayers
can
be
add
e
d
to
i
m
pro
ve
the
perf
ormance.
C
N
N
t
ak
es
color
i
m
age
s
and
the
fea
t
ur
es
ar
e
aut
oma
t
i
ca
ll
y
ext
r
ac
t
ed
by
the
c
onvo
l
ve
la
yers.
The
si
z
e
of
filter
s
i
n
t
he
c
onvolve
la
yer
an
d
the
va
l
ue
of
st
r
i
de
i
n
t
he
poolin
g
l
a
yer
r
eprese
nt
the
nu
m
ber
of
c
olu
m
ns
t
o
be
s
kippe
d
f
or
the
slidi
ng
w
i
nd
ow
t
hro
ugh
t
he
i
m
age
.
These
va
l
ue
s
c
an
be
c
hange
d
as
the
y
ca
n
a
f
fect
the
r
es
ult
of
the
r
eco
gnit
i
on
perf
orm
ance.
B
esi
des
tha
t
,
the
values
of
e
pochs
r
eprese
nt
t
he
num
ber
of
i
t
er
at
io
n
f
or
t
he
t
r
ai
ning
process
a
nd
i
ni
t
ia
l
l
ear
ni
ng
r
at
e
t
ha
t
r
eprese
nt
the
va
l
ue
of
the
w
e
i
gh
t
to
be
a
djuste
d
du
r
i
ng
th
e
t
r
ai
ning
proc
ess,
can
be
c
ha
ng
e
d
to
view
the
i
r
eff
ec
t
to
the
r
ecogn
i
ti
on
r
a
t
e
.
The
i
m
age
s
i
ze
r
equired
f
or
bas
i
c
C
N
N
i
s
22
4
x
224
pix
e
l
s.
Exp
er
im
ent
al
r
esult
s
w
er
e
c
onducte
d
on
the
c
omb
i
na
tion
of
t
hese
values
a
nd
the
r
e
sult
s
ar
e
s
ho
w
n
in
Ta
ble
1.
The
f
i
r
st
column
r
epres
ents
the
s
i
ze
of
t
he
filter
an
d
the
num
ber
of
f
i
lter
s
in
t
he
c
onvo
l
ve
l
a
yer.
By
r
efer
r
i
ng
to
Tab
l
e
1,
w
e
can
see
tha
t
a
10
0%
accur
a
cy
i
s
ac
hi
e
ve
d
w
i
th
[
5,
20
]
in
the
f
i
r
st
c
onvo
l
ve
l
a
ye
r
and
[
3,3
2]
in
the
seco
nd
co
nvol
ve lay
er
,
and
F
i
gur
e
5 sho
w
s
the
r
esult
s
of
th
i
s
t
r
ai
ning
an
d vali
datio
n proc
esses.
F
i
gu
r
e
5.
The resu
l
t
of
C
N
N
w
i
th
[5
,
20]
i
n t
he fi
r
s
t
conv
olv
e
l
ayer
an
d [3,
32
]
i
n t
he se
co
nd
co
nvol
ve la
yer
Tab
l
e
1.
Exp
er
i
m
ent
a
l
R
e
sul
ts
on
P
ar
a
m
eter
Tuni
ng
fo
r
B
a
si
c
C
N
N
No
of
Stac
k
of
L
ay
e
r
s
C
o
n
v
o
lv
e
L
ay
e
r
Po
o
lin
g
la
y
e
r
a
n
d
Str
id
e
Acc
u
r
acy
(
%)
Tota
l
T
i
m
e/s
1
[
3
,16
]
3
7
8
.82
3
0
sec
[
5
,20
]
3
8
3
.59
2
8
sec
2
[
3
,16
]
[
3
,3
2
]
3
9
1
.79
2
7
sec
[
5
,
2
0
][
3
,3
2
]
2
1
0
0
.00
2
7
sec
3
[
5
,
2
0
][
3
,3
2
]
[
3
,
3
2
]
2
7
5
.90
4
1
sec
[
5
,
2
0
][
3
,3
2
]
[
3
,16]
2
7
4
.87
3
4
sec
By
l
oo
king
a
t
Tab
l
e
1,
w
e
can
see
tha
t
a
s
the
nu
m
ber
of
l
a
yers
i
ncre
ases,
t
he
accu
r
acy
i
s
al
so
i
ncrea
se
d.
But
w
hen
t
he
num
ber
of
l
a
yers
i
s
m
ore
t
ha
n
3,
the
acc
uracy
be
gin
s
to
dr
op
.
T
his
m
eans
t
ha
t
t
w
o
st
ac
ks
of
l
a
yers
plu
s
1
clas
sifi
ca
tion
l
ayer
pro
du
ce
t
he
be
st
accur
ac
y
f
or
bam
bara
groun
dnut
l
ea
d
disease
r
ecognit
i
on.
3.3
.
B
ag
of
Featu
res
(
B
oF)
The
si
z
e
of
a
n
i
m
age
us
e
d
f
or
B
oF
i
s
227
x
227
pi
xels
an
d
t
he
acc
uracy
pr
oduced
i
s
10
0%
.
F
i
gu
r
e
6
sh
ows
the
r
esu
l
t
of
visu
a
l
w
or
ds
occ
urr
ence
pro
du
ce
d
by
BoF
fo
r
ou
r
da
t
ase
t
.
S
peed
e
d
-
U
p
R
ob
us
t
F
eatu
r
e
(
SURF
) a
nd
Su
pp
ort V
ecto
r
M
ac
hin
e
(
S
VM
)
i
s
bein
g uti
l
i
ze
d i
n t
he BoF
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
25
02
-
4752
I
ndon
es
i
a
n J
El
ec
En
g &
C
om
p S
ci,
V
ol
.
1
4
,
N
o.
1
,
A
pril
201
9
:
368
–
374
372
F
i
gu
r
e
6.
V
i
sua
l
w
ord
occu
r
r
ences
r
esult
Tab
l
e
2
show
s
an
ov
er
vi
ew
of
the
acc
uracy
perfo
r
m
ance
of
C
N
N
c
ompar
ed
t
o
BoF
base
d
on
our
bam
bara
groun
dnut
l
ea
f
dat
a
s
et.
B
y
lo
oking
at
Tab
l
e
2,
w
e
can
see
t
ha
t
BoF
i
s
better
t
ha
n
bas
i
c
C
N
N
but
it
t
oo
k
a
longer
tim
e
to
ac
hieve
this
r
es
ul
t
.
Th
i
s
i
s
beca
us
e
e
xt
r
ac
t
in
g
of
th
e
S
U
R
F
feature
s
i
s
lo
ng
er
c
om
pared
t
o t
he t
i
m
e
t
o ext
r
ac
t
t
he l
ow
-
le
vel
an
d m
i
dd
le
l
evel
fea
tur
es
by
t
he
C
N
N
.
Tab
l
e
2.
The
Per
fo
r
m
ance
O
vervi
ew
fo
r
B
as
i
c
C
N
N
and
BoF
fo
r
B
am
bara
D
ataset
M
o
d
e
l
B
a
sic
C
NN
B
o
F
Va
l
i
d
a
t
i
o
n
a
c
c
u
r
a
c
y
100
100
E
la
p
se
d
Tim
e
(s
)
27
31
4.
C
ONC
L
U
S
IO
N
I
n
th
is
pa
pe
r
,
a
c
om
pa
r
is
on
be
tw
e
e
n
CN
N
a
nd
B
oF
w
a
s
pe
r
f
or
m
e
d
w
it
h
r
e
s
pe
c
t
to
a
c
c
ur
a
c
y
a
nd
e
la
ps
e
d
t
im
e
.
Th
e
e
xp
e
r
im
e
nt
r
e
s
ul
ts
s
ho
w
th
a
t
Bo
F
a
c
hi
e
ve
d
th
e
s
a
m
e
a
c
cu
r
a
c
y
r
a
te
a
s
CN
N
w
hi
c
h
is
1
00
%
.
H
ow
e
ve
r
,
Bo
F
r
e
qu
ir
e
s
a
hi
gh
e
r
e
la
ps
e
d
t
im
e
du
e
to
th
e
la
r
ge
nu
m
be
r
of
S
U
RF
f
ea
tu
r
e
s
r
e
qu
ir
e
d
to
b
e
e
xt
r
a
c
te
d
.
A
lt
h
ou
gh
t
he
n
um
be
r
of
la
ye
r
s
a
f
f
e
c
ts
t
he
a
c
c
ur
a
c
y
pe
r
f
or
m
a
nc
e
,
th
e
c
om
pl
e
x
it
y
o
f
t
he
CN
N
a
r
c
hi
te
c
tu
r
e
do
e
s
no
t
gu
a
r
a
n
te
e
a
be
tt
e
r
r
e
s
ul
t
.
T
he
e
xp
e
r
im
en
ta
l
r
e
s
ul
ts
in
t
hi
s
r
e
s
e
a
r
c
h
in
di
c
a
te
t
ha
t
tw
o
s
ta
c
ks
of
la
ye
r
s
pr
od
u
c
e
be
tt
e
r
a
c
c
ur
a
c
y
c
om
pa
r
e
d
to
th
r
e
e
s
ta
c
ks
of
la
ye
r
s
.
T
he
us
e
of
CN
N
is
r
e
c
om
m
e
nd
e
d
f
or
le
a
f
di
s
e
a
s
e
r
e
c
og
n
it
io
n
if
th
e
pr
o
c
e
s
s
in
g
t
im
e
is
no
t
a
n
is
s
ue
.
F
or
f
ut
ur
e
w
or
k
,
m
or
e
de
e
p
l
e
a
r
ni
ng
m
o
de
l
s
a
nd
pu
bl
ic
ly
a
va
i
la
bl
e
da
ta
s
e
ts
w
il
l
be
te
s
te
d.
A
C
KNOWL
ED
GE
M
E
N
T
S
The
a
uthor
s
w
oul
d
li
ke
t
o
than
k
F
acu
lt
y
of
Com
put
e
r
and
M
a
the
m
ati
ca
l
S
cienc
es,
U
ni
vers
i
ti
Tek
nologi
M
A
R
A
,
S
hah
A
l
a
m
, Selangor,
fo
r
sp
onso
r
i
ng
this
r
esear
ch.
R
E
F
E
R
E
N
C
E
S
[1]
Azm
an,
R.
,
Maye
s,
S.,
&
L
u,
C.
Nut
r
iti
ona
l
P
rofi
l
e
O
f
Bamb
ara
Groundnut
And
It
s
Po
t
ent
i
al
For
Food
Prod
uct
Deve
l
opm
ent
In
Mal
aysi
a.
Mod
el
Int
ern
ati
onal
Underut
il
i
sed
Le
gum
e
Research
And
Bree
di
ng
Journal
.
,
24(4),
.
(2016) 429.
Htt
p
:
//D
x
.
Do
i
.
Org/
1
0.
4314/
Acs
j.V
2
4i
4.
9)
.
[2]
Fang,
Y.,
&
Ramasam
y,
R
.
Cu
rre
nt
And
Pros
p
ec
t
i
ve
Met
hods
For
Pl
an
t
Disea
se
Det
ec
t
i
on.
B
i
osensors,
5(3),
.
Htt
p:/
/
Dx
.
Doi.
O
rg/
10.
3390
/
Bi
o
s
5030537
(2016)
537
-
561.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndon
es
i
a
n J
El
ec
En
g &
C
om
p S
ci
I
SS
N
:
2502
-
4752
Con
v
olut
i
on
al
neura
l
ne
t
work
vs
ba
g o
f
f
ea
tu
res
f
or
bam
bar
a grou
ndnu
t
l
e
af
…
(
H
afi
z
atul
H
an
in
H
amza
h
)
373
[3]
Si
ngh
,
V.,
&
Mi
sra,
A.
Det
e
cti
on
O
f
P
l
ant
L
ea
f
Disea
ses
Us
i
ng
I
m
age
Segm
entat
i
on
An
d
Soft
Com
pu
ti
ng
T
ec
hni
ques.
In
f
o
rm
at
i
on
Proce
ssi
ng
In A
gr
i
cul
t
ur
e,
4(1H
t
t
p
://Dx.Doi
.
Org/
10
.
101
6/
J
.
Inpa
.
2016.
10
.
005
(2017),41
-
49
[4]
Shet
ye,
H
.
,
Ran
e,
T
.
,
Pawar,
T
.
,
&
Dandwat
e,
P.
An
Anal
ys
i
s
Of
Met
hodo
l
o
gi
es
F
or
L
ea
f
Di
sea
se
Det
ec
ti
on
T
ec
hni
ques.
In
t
e
rna
t
i
onal
Journal
O
f
Resea
rc
h
In
Advent
T
ec
hnol
ogy,
Vo
l
.
4
,
No.
2
,
Februa
ry
2016,
Vol.
4(No
.
2),
53
-
58.
H
ttp://
Dx.Do
i
.
Org
/E
-
Iss
n
:.
(2
016).
2321
-
9637
[5]
Ibra
hi
m
,
Z.,
Sa
bri
,
N.,
&
Mangshor,
N
.
N.
A
L
ea
f
Rec
ogni
tion
u
si
ng
T
e
x
t
ure
Features
for
Herba
l
P
l
ant
Ide
nt
if
i
ca
tion, IJEE
C
S, vol.
9
,
.
(
2018).
no.
1,
15
2
-
156.
[6]
Sl
adojev
i
c,
S.,
Ars
enovic,
M.,
Anderl
a,
A.
,
Cul
i
brk
,
D.,
&
St
ef
anovic,
D
.
.
Dee
p
Neur
al
Net
works
Based
Rec
ogni
t
i
on
O
f
Pl
ant
Disea
ses
By
L
ea
f
Im
age
Cl
ass
if
i
ca
tion.
C
om
p
ut
a
t
i
onal
In
te
l
ligence
And
Neurosci
enc
e,
201
6.
(2016)
[7]
Pil
li
,
S.
K.,
Nal
l
a
t
hamb
i
,
B.
G
eor
ge,
S.
J.
and
Diwanji,
V
.
,
“
eAGRO
B
OT
-
A
Robo
t
f
or
Ea
rl
y
Crop
D
isea
se
Det
ec
t
i
on
u
si
ng
Im
age
Proce
ss
i
n
g”,
IEEE
2
nd
Inte
rna
t
i
onal
Con
f
e
re
nce
on
E
l
ec
t
roni
cs
and
Comm
uni
ca
t
i
on
Syst
em
(ICE
CS
),
2015.
[8]
Yao,
Q
.
et
.
al
.
“Applica
t
i
on
o
f
s
upport
vec
t
or
ma
chine
f
or
detec
t
i
ng
ri
ce
d
i
sea
ses
us
i
ng
shape
and
colour
t
extur
e
f
ea
t
ure
s”,
In
t
ern
ati
onal
Con
f
ere
nce
on
E
ngi
nee
ri
n
g
and Com
pu
t
at
ion,
2009,
pp.
79
-
83.
[9]
Pat
il
,
S
.
P.
and
Z
ambre,
R
.
S.
“
Cl
ass
if
i
ca
tion
o
f
Cott
on
L
ea
f
Sp
ot
D
i
sea
se
U
si
n
g
Support
Vec
t
o
r
Mac
hi
ne”
,
Jou
rna
l
of
E
ng
i
nee
ri
ng
Resea
rc
h
and Appl
i
ca
tions, Vol.
4,
I
ss
ue 5
,
Version 1,
2010
,
pp
.
1
0
-
20.
[10]
Fuji
t
a
,
E
.
,
Kawa
saki
,
Y
.
,
Uga,
H.,
Kagi
wada
,
S
.
and
Iya
t
om
i,
H.,
“Ba
si
c
i
nvestigati
on
o
n
a
robust
a
nd
pra
cti
ca
l
pl
ant
di
agnostic s
y
st
e
m
”,
15
th
IEEE In
t
ern
ati
onal
Con
fe
re
nc
e on
Mac
hi
ne L
ea
rni
ng
and
Appli
ca
t
i
on
s, 20
16,
pp.
989
-
992.
[11]
Sandi
ka,
B.
,
Avil
,
S
.
,
Sanat,
S
.
and
Sri
ni
vasu
,
P
.
,
“
Random
Fores
t
base
d
Cl
assifi
c
ati
on
o
f
Disea
ses
i
n
Grape
s
f
ro
m
Im
age
s
Captured
i
n
Uncont
ro
l
l
e
d
E
nvi
ron
m
ents”
,
IE
EE
Int
ern
at
i
onal
Con
f
ere
nc
e
on
Si
gnal
Pro
ce
ss
i
ng,
2016,
p
p.
1775
-
1780.
[12]
Sasano,
S.,
Han,
X.
and
Chen,
Y.
,
“Food
Rec
ogni
tion
by
Com
b
i
n
ed
Bags
of
Col
o
r
Features
and
T
exture
Features”
,
IE
EE
9
th
In
t
ern
ati
ona
l
C
ongre
ss
on
Im
age
and
S
igna
l
Proce
ssing,
Bi
o
Medi
ca
l
E
n
gi
nee
ri
ng
and
In
f
or
m
ati
cs
(CIS
P
-
BME
I),
2016,
p
p.
815
-
819.
[13]
Z
aidah
Ibra
hi
m,
Nurbai
t
y
Sabri
and
Di
no
Isa,
"
Mul
ti
-
m
axpool
ing
Convol
ution
al
Neura
l
Net
w
ork
f
or
Medi
c
i
n
al
Herb
L
ea
f
Rec
ogni
tion",
Proce
edings
o
f
t
he
6
t
h
IIAE
Int
ern
at
i
o
nal
Conf
ere
nce
on
Int
elli
gent
S
yst
em
s
and
Im
ag
e
Proce
ss
i
ng,
Sh
imane,
Japa
n,
Sept
ember 2018 ,
p
p.
327
-
331.
[14]
Nurbai
t
y
Sabri,
Z
al
i
l
ah
Abd
Aziz
,
Z
aidah
Ibra
hi
m
,
Muhammad
Akm
al
Rasydan
bi
n
Mohd
Rosni
and
Abdul
Haf
iz
bi
n
Abd
Ghapul
,
"
Com
par
i
ng
Co
nvol
utional
Neur
al
Net
work
Mod
els
f
or
L
ea
f
Recogni
tion",
Int
ern
ati
onal
Journal
o
f
E
ngi
nee
ri
ng
T
ec
hnol
ogy,
7
(3
.
15
) (2018)
p
p.
141
-
144.
[15]
Nur
Azi
da
Muham
m
ad,
A
m
el
i
n
a
Ab
Nasi
r,
Z
ai
dah
Ibra
hi
m
and
Nurbai
t
y
Sabri,
"
E
valuat
i
on
of
CNN
,
A
l
exne
t
and
Googl
eNe
t
f
or
F
rui
t
Rec
ogni
tion
"
,
Indone
si
an
Jo
urna
l
of
Elec
t
ri
c
al
E
ng
i
nee
ri
ng
and
Com
pu
t
er
Sci
enc
e
(IJE
ESC),
Vol
.
12,
No
.
2
,
Novem
ber
201
8,
pp.
468
-
475
.
[16]
Hou,
L.,
W
u,
Q
.
,
Sun
,
Q.,
Yang
,
H.,
&
L
i,
P
.
F
rui
t
re
cogni
t
i
on
base
d
on
convol
ut
i
on
neur
al
net
work.
2016
12th
Int
ern
ati
onal
Co
nf
ere
nce
on
Nat
ura
l
Com
pu
t
at
i
o
n,
Fuzzy
Sy
st
e
m
s
and
Knowledg
e
Di
scove
ry,
ICNC
-
FS
K
D
20
16,
18
-
22.
[17]
Muhaa
f
i
dz
Md
Saufi
,
Mohd
A
f
iq
Z
amanhur
i
,
N
ora
si
ah
Mohammad
and
Z
a
i
dah
Ibra
hi
m
"
Dee
p
L
ea
rni
ng
f
or
Ro
m
an
Handwrit
t
en
Ch
ara
cter
R
ec
ogni
t
i
on"
,
Int
ern
ati
on
al
Journal
of
E
le
ctri
ca
l
E
ngi
nee
ri
ng
and
Com
put
er
Sci
enc
e,
Vol.
12,
No
.
2
,
Novem
ber
2018,
pp.
4
55
-
460.
[18]
Raja
Durrat
un
Safi
yah
,
Z
aid
A
bdul
Rahim,
Sy
ams
ul
Sya
fi
q
,
Z
aidah
Ibra
hi
m
and
Nurbai
t
y
S
abr
i
"
Perf
orm
anc
e
E
valuation
f
or
Vi
s
i
on
-
Based
V
ehicl
e
Cl
ass
ifica
t
i
on
Using
Conv
ol
utional
Neura
l
Net
work"
,
In
t
ern
ati
onal
Journal
o
f
E
ngi
nee
ri
ng
& Te
chnology,
7
(3
.
15),
(2018),
pp.
86
-
40.
[19]
Z
aid
ah
Ibra
hi
m,
Nurbai
t
y
Sabri
and
Di
no
Isa,
“Palm
Oil
Fresh
Fruit
Bunch
Ri
p
ene
ss
Gradi
ng
Rec
ogni
t
i
on
U
sing
Convol
utional
N
eur
al
Net
work”,
Journal
o
f
T
elec
om
m
un
i
ca
tion,
E
l
ec
t
ron
i
c
&
Co
m
put
er
E
ng
i
nee
ri
ng,
Vo
l
9,
No
.
3
-
2,
2018,
pp.
109
-
113.
[20]
Gupt
a,
D.,
Ja
i
n,
K.,
Ja
i
n,
A.,
&
Anal
y
t
i
cs
Vi
dh
ya
Cont
ent
T
ea
m
.
Archi
t
ec
t
ure
of
Convo
l
u
t
i
ona
l
Neura
l
Net
wor
ks
(CNN
s)
de
m
ys
t
ified.
(2017
,
June
29).
[21]
Schm
i
d
,
Cordel
i
a.
"
Bag
-
of
-
f
ea
t
ure
s
f
or
ca
t
egor
y
class
ifica
tion."
E
NS
/
I
NRIA
Vi
sua
l
Rec
ogni
tion
and
Mac
hi
n
e
L
ea
rni
ng
Su
mm
er S
chool L
ec
t
ur
e 25
-
29
Ju
l
y
(20
11).
[22]
O'H
ara
,
St
ephe
n,
and
Bruce
A.
Drape
r.
"
Int
rodu
cti
on
t
o
t
he
bag
of
f
ea
t
ure
s
par
a
di
gm
f
or
i
m
age
class
ifica
t
i
on
an
d
re
t
ri
eva
l
.
"
arXi
v
pre
pri
nt
arX
i
v:
1
101.
3354
(2011)
.
[23]
Muham
m
ad,
A.,
Nasi
r
A
.
,
Ibra
him,
Z
,
Sabri
,
N.
,
E
valuation
o
f
C
NN
,
A
l
exne
t
and
Googl
eNe
t
f
or
Fruit
Rec
ogni
ti
on.
Indone
si
an Jo
ur
nal
o
f
E
l
ec
t
r
i
ca
l
E
ngi
nee
ri
ng
and Com
pu
t
er S
c
i
enc
e 2018,
pp.
468
~475
[24]
Ahm
ad,
K
.
,
Kha
n,
R.
,
Ah
m
ad,
N.
and
Khan,
J.
“
E
va
l
uati
on
o
f
SIF
T
and
SU
RF
usi
ng
Bag
o
f
W
ords
Mode
l
o
n
a
Ver
y
L
arg
e Dat
a
set
,
”
Si
ndh
Un
i
v
.
Res.
Jour
.
(
Sci
.
Ser.) Vol.45
(3)
492
-
495
(2013).
[25]
S.
H.
L
ee
,
C.
S.
Chan,
P.
W
ilk
in,
and
P.
Remagni
no,
“De
ep
-
pla
nt
:
P
l
ant
i
den
t
ifica
t
i
on
w
i
t
h
co
nvol
utional
neur
al
netw
orks,
” I
EEE
Int
ern
at
i
onal
Co
nf
ere
nce
on
Im
age
Proce
ss
i
ng
(I
C
IP),
2015
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
25
02
-
4752
I
ndon
es
i
a
n J
El
ec
En
g &
C
om
p S
ci,
V
ol
.
1
4
,
N
o.
1
,
A
pril
201
9
:
368
–
374
374
B
IO
GRA
PHI
E
S
OF
A
U
T
HORS
Hafi
za
t
ul
Hani
n
Ham
za
h
i
s
a
Ma
st
er’
s
s
t
udent
at
t
he
Facult
y
o
f
C
om
put
er
and
Ma
t
hematica
l
Sc
i
enc
es,
Uni
ver
s
i
ti T
ekno
l
ogi
MA
RA
,
Sha
h
Al
am
,
Sel
ango
r, Mal
ays
i
a, whe
re
she
i
s c
ont
i
nu
ing
her
edu
ca
t
i
on
i
n
t
he
f
i
e
l
d
o
f
Co
mputer
Sci
enc
e
i
n
W
eb
T
ec
hnol
ogy.
Her
are
a
of
i
n
t
ere
st
s
are
i
m
ag
e
proc
essi
ng,
im
age
re
cognit
i
on
,
and
m
ult
i
m
ed
i
a com
put
i
ng
.
Z
aidah
Ibra
hi
m
i
s
an
Associate
Prof
essor
a
t
the
Facult
y
o
f
C
om
put
er
and
M
athem
ati
ca
l
Sci
e
nce
s,
Uni
ver
s
i
ti
T
ekno
l
ogi
M
ARA
,
Sha
h
Al
am,
Sel
ango
r,
Mal
aysi
a.
She
has
bee
n
t
eachi
ng
cour
ses
r
elat
ed
t
o
Artif
i
cia
l
Int
e
l
li
genc
e
f
or
m
ore
t
han
t
en
yea
rs.
She
i
s
ac
t
i
ve
l
y
involved
i
n
re
sea
rc
h
and
publ
i
ca
t
i
ons
under
Di
gital
I
m
age
,
Audi
o
an
d
Speec
h
T
ec
hn
ol
ogy
(DIA
ST
)
re
sea
rc
h
i
nt
ere
st
group
t
hat
i
nc
lude
s
comput
er v
i
s
i
on
and i
nt
e
l
ligent s
yst
em
s.
Nurbai
t
y
Sabri
i
s
a
l
ec
t
ure
r
at
t
he
Facult
y
o
f
Com
put
er
and
Mat
hematica
l
Sci
enc
es,
Uni
ve
rsit
i
T
eknolog
i
M
AR
A,
Jasi
n
,
Mel
aka
.
She
t
ea
che
s
progra
m
m
i
ng
l
angua
ge
and
i
m
ag
e
proc
essi
ng.
Sh
e
i
s
a
m
ember
of
D
i
g
i
t
a
l
Im
age
,
Aud
i
o
and
Speec
h
T
ec
hnol
ogy
(DlAS
T
)
re
sea
rc
h
group
and
cur
r
entl
y
par
t
i
cipating
i
n
var
i
ous
re
sea
rc
hes
re
l
ated
t
o
ima
ge
proc
essi
ng.
R
ec
entl
y,
she
has
publ
ished
pape
rs
and
co
-
authored
i
n
i
nt
ern
ati
onal
conf
ere
nce
s
and
j
ourna
l
s.
Her
r
ese
ar
ch
i
nt
ere
st
s
i
ncl
ude
i
m
age
proc
e
ss
i
ng
,
comput
er v
i
s
i
on
and patt
ern
re
cogniti
on
.
Prof.
D
i
no
Isa
is
a
prof
essor
o
f
Int
elli
gent
Syst
ems
at
Un
i
ver
sit
y
o
f
Nottingham
Mal
ays
i
a
Campus,
Sem
enyih,
Sel
an
gor,
si
nce
2018,
and
has
publ
i
sh
ed
m
ore
t
han
100
pape
rs
i
n
hi
s
fi
eld
o
f
re
sea
rc
h.
He
has
bee
n
appoin
t
ed
as
Di
re
ctor,
CON
N
E
CT
In
itiati
ve,
Crops
f
or
t
he
Fut
ure
,
i
nvo
lving
i
n
agr
i
cult
u
re
-
re
l
ated
proj
ec
t
s
usi
ng
m
ac
hi
ne
le
arn
i
ng
t
o
i
n
f
er
t
he
l
eve
l
of
pro
fi
ts
ava
i
l
able
t
o
fa
rm
ers
when
under
-
ut
il
i
ze
d
cro
ps
ar
e
grown
and
i
t
s
der
i
vati
ves
are
proc
essed
on
a
c
om
m
erc
i
al
sca
l
e.
He
i
s
cur
re
ntly
t
he
chief
consultant
t
o
Tiger
So
l
utio
ns
Sdn.
Bhd
.
,
an
i
nt
egr
ated
oil
a
nd
gas
t
ec
hnol
ogy
company
i
nt
e
re
st
ed
i
n
u
si
ng
m
ac
hi
n
e l
ea
rni
ng
t
o
pre
di
ct f
a
i
l
ure
i
n
o
il
and gas
p
i
pel
i
nes.
Evaluation Warning : The document was created with Spire.PDF for Python.