TELKOM
NIKA
, Vol.13, No
.3, Septembe
r 2015, pp. 1
021
~10
2
8
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v13i3.1786
1021
Re
cei
v
ed Ma
rch 2
4
, 2015;
Re
vised J
une
3, 2015; Accepted June 2
5
, 2015
Earth Image Classification Design Using Unmanned
Aerial Vehicle
Barlian Henr
y
r
anu Prasetio*
1
, Ahmad Afif Supianto
2
, Gembong Edhi Setiaw
a
n
1
,
Budi Da
rma Setia
w
a
n
2
, Imam Cholissodin
2
, Sabrians
y
a
h R Ak
bar
3
1
Computer S
y
s
t
em and Ro
boti
cs Lab, Progra
m
of
Information T
e
chnolo
g
y
and C
o
mput
er Scienc
e,
Univers
i
t
y
of Bra
w
ij
a
y
a, Mal
a
n
g
651
45, Indo
n
e
sia
2
Intellig
ent Co
mputin
g an
d Multime
d
ia
L
ab,
Program of Informatio
n
T
e
chnol
og
y a
nd Co
mputer Scie
nc
e,
Univers
i
t
y
of Bra
w
ij
a
y
a, Mal
a
n
g
651
45, Indo
n
e
sia
3
Computer N
e
tw
o
r
kin
g
La
b, Program of Infor
m
ation
T
e
chno
log
y
a
nd C
o
mp
uter Scienc
e, Univers
i
t
y
of
Bra
w
ij
a
y
a, Mal
ang 6
5
1
45, Ind
ones
ia
Corresp
on
din
g
author, e-mai
l
: barli
an@
ub.ac
.id,
afif.supia
n
to@u
b.ac.id, ge
mbon
g@u
b
.ac.id,
s.budi
darma
@
ub.ac.id, imam
cs@ub.
ac.i
d, sabri
an@
ub.ac.i
d
A
b
st
r
a
ct
T
he res
earch
ai
ms to
bu
il
d s
o
ftw
are that ca
n p
e
rfor
m th
e
classificati
on
o
f
earth
i
m
a
ge f
r
om UA
V
(Un
m
an
ne
d Ae
rial Ve
hic
l
e)
monitor
i
ng. T
he I
m
a
ge c
onverte
d into Y
U
V for
m
at th
en cl
assi
fied us
ing F
u
zzy
Supp
ort Vector
Mach
ine
(F
SV
M). UAVs w
ill
b
e
us
ed
for
monitoring as
fol
l
o
w
s: (1) the c
o
n
t
rol statio
n, w
h
i
c
h
used to se
nd o
r
receive
data,
and
disp
lay the
data in
gr
ap
hi
cal for
m
, (2) pa
yloa
d, camera
capture
d
i
m
a
g
e
s
and s
end to the control stat
ion, (3) comm
unication syst
em
using TCP/IP protocol, and (4)
UAV, using X650
qua
d c
opter
pr
oducts. T
h
e i
m
age
of th
e
mon
i
torin
g
carr
ied
out o
n
th
e
UA
V si
z
e
d
256
x
256
pix
e
ls
w
i
th 45
0
traini
ng data. It is 16x16 pix
e
l image d
a
ta. T
e
sts perform
ed to classify the i
m
a
ge into
3 classes, na
me
l
y
agric
ultura
l ar
e
a
, resid
enti
a
l
a
r
ea, a
nd w
a
ter
area.
T
h
e h
i
g
hest acc
u
racy
valu
e of
77.69
% obt
ain
ed
by
the
nu
mb
er of train
i
ng d
a
ta as mu
ch as 375.
Ke
y
w
ords
: image transfor
m
a
t
ion, e
m
be
d
d
e
d
system, FSVM, YUV format
Copy
right
©
2015 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Image cla
s
sification is the
pro
c
e
ss of groupin
g
image
pixels into a number of cl
asse
s,
so that ea
ch
class
can
descri
be a
n
entity wi
th certain
cha
r
a
c
teristics [1]. One a
r
ea of
appli
c
ation
s
develop
ed u
s
ing the
cl
assificati
o
n
tech
niqu
e is a satellite
image. Ob
ject
identificatio
n i
n
satellite im
age
be
com
e
s impo
rtant
given the
b
enefit
gain
e
d
is al
so so
gre
a
t. F
o
r
example, it can get inform
ation about th
e type and ex
tent of the area of food so
urces, buil
d
in
g or
resi
dential
area, and the
water
area
(su
c
h a
s
rive
rs, lakes, a
n
d
reservoi
rs).
By knowi
ng t
h
is
informatio
n t
hen th
e foo
d
se
cu
rity of certain
area
s
and
at certai
n times can
be o
b
tained.
One
alternative way of making
image
s that woul
d be cl
a
ssifie
d
is by using a UAV (Unma
nne
d Aeria
l
Vehicle
)
.
UAV (Unm
an
ned A
e
rial
Ve
hicle
)
or
dron
es
app
ear
d
u
e
to have sev
e
ral
advanta
g
e
s over
manned aircraft. Applications UAV
is wi
dely used to overcome hum
an limitations in their work,
for example,
to do the work that tend
s to r
outine
or dan
gerou
s, such a
s
the monitorin
g
of
ongoi
ng or t
hat can
not b
e
rea
c
h
ed b
y
human
s. By utilizing the Hu
man
UAV still req
u
ire
informatio
n di
rectly from th
ese pl
aces, e
i
ther
in the fo
rm of pictu
r
e
s
, video and
other data. T
h
e
result of information obtain
ed from the UAV c
an be
pro
c
e
s
sed int
o
useful data
for human
s such
as m
onitori
ng for land or territory
or al
so info
rmation on the monitori
ng of
the milit
ary operations
area
. With
a
d
vances in th
e devel
opme
n
t of an
in
te
grated
ele
c
tronic ci
rcuit (I
ntegrate
d
Circuit)
trigge
ring
ma
ss UAV
devel
opment
so
th
at today
the
UAV
can be obtaine
d
at a
n
affordable
price
[2] and be ine
x
pensive alte
rnative sol
u
tion in the field
of aviation.
UAV have be
come
sup
p
o
r
ting equipm
e
n
t in area
s monitorin
g
re
search such a
s
fore
st
fire monitori
n
g
[3], agricult
u
ral a
r
ea
s [4] and tran
sp
ort
traffic on the high
way [5].
Image from
UAV monit
o
ring i
s
sen
t
to the co
nt
rol statio
n, it is req
u
ires a
n
effici
ent
comm
uni
cati
on mechanism power.
It is because
UAVs have l
i
mited utilizat
ion power. T
h
e
probl
em i
s
h
o
w
to
sen
d
the
monitori
ng
d
a
ta carried
b
y
the UAV to
the groun
d st
ation comp
uter
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 3, September 20
15 : 1021 – 10
28
1022
and the
n
cla
ssify them
i
n
to thre
e a
r
eas
(resi
dent
ial, agri
c
ultu
ral and
waters)
usi
ng F
u
zzy
Suppo
rt Vector Ma
chin
e (FSVM).
The aim of rese
arch were
: (a) De
signi
n
g
a
UAV prototype monito
ring sy
stem that can
be ea
sily integrated
with th
e aircraft whe
r
e the
mo
nito
ring
system i
s
able to p
e
rf
orm lo
w-power
comm
uni
cati
on between t
he UAV flight control st
atio
n and sendi
n
g
earth ima
g
e
. (b) Desi
gni
ng
and buil
d
ing
station software to imple
m
ent the
YUV color m
ode
l and a FSVM to perform obje
c
t
cla
ssifi
cation
of ag
ricultural lan
d
in th
e re
su
lt
s of
UAV monito
ri
ng ima
ge. T
he results
of th
e
softwa
r
e impl
ementation fo
r furthe
r anal
ysis to det
e
r
mine the a
c
cura
cy level of the classifica
tion
p
r
oc
es
s
.
The b
enefits of this
re
se
arch a
r
e:
(a) t
he results
can
be
useful for m
onito
ring t
h
e
developm
ent
of an area
in need of UAVs as a to
ol for exploration. By utilizing lo
w po
wer
comm
uni
cati
on mechani
sm, is expected to be active UAV here lon
ger
becau
se po
wer
con
s
um
ption
is u
s
e
d
to
co
mmuni
cate te
nds to b
e
lo
w. (b) O
b
taine
d
info
rmation
agri
c
ultu
ral l
a
nd
obje
c
t class i
n
a UAV imagery. The re
sults of t
he classificatio
n
can be used to determin
e
how
the extents of the object. By sepa
rating
obje
c
ts
from
an image of f
a
rmla
nd, furt
her research
can
be devel
ope
d
and fo
cu
se
d
on the
ide
n
tification
of ag
ricultu
r
al l
and
obje
c
ts, b
e
in
g food
so
urce
area
s. Id
entification
of the
f
ood
so
urce
area
s
can
be u
s
e
d
to i
m
pleme
n
tatio
n
natio
nal fo
od
se
curit
y
inf
o
r
m
at
ion sy
st
e
m
.
2. Rese
arch
Metho
d
The meth
od
s use
d
in a
re
sea
r
ch g
r
eatl
y
affe
ct the p
e
rform
a
n
c
e o
f
the system
to work
optimally. The method
s u
s
ed in the
rese
arch a
r
e:
(a) Mate
rial
Prepa
ration:
In the material
prep
aration carri
ed out
a
c
tivities
for co
llecting
literat
ure
revie
w
i
s
used
as the
ba
sis for th
e
impleme
n
tation of the re
search "Mo
n
itoring Syst
e
m
using
UAVs"
.
The coll
ecti
ons
wa
s obta
i
ned
throug
h a lite
r
ature revie
w
of previou
s
rese
arch
jo
urnal search
as sup
por
tin
g
literatu
r
e, bo
oks,
and al
so coll
ect the data
s
heet of Pi raspberry
equi
p
m
ent that will
be use
d
a
s
a ca
rgo
UAV and
UAV low
po
wer commu
nication devi
c
e
s
with the
con
t
rol station.
(b) Identif
ication: Identificat
ion
is to identify proble
m
s
with the obvious lim
itation
s
of obse
r
ving equipm
ent
used. Th
ese
observation
s we
re ma
de
to test ea
ch eleme
n
t of
the obje
c
t
of re
sea
r
ch. (c) Requi
rem
ent
Analysis: Aft
e
r all
the
dat
a obtai
ned
th
roug
h the
id
entification
p
hase, follo
we
d by a
sta
g
e
of
analysi
s
nee
ds. T
h
is sta
g
e
be
gin
s
to
analyze the
need
s
of bot
h ha
rd
wa
re
a
nd
softwa
r
e
i
s
in
confo
r
mity with the provisi
o
ns of whi
c
h
h
a
ve been d
e
signed in the i
dentificatio
n stage.
2.1. UAV
Design
UAV system
desig
n co
n
s
ist
s
of payload an
d gro
und statio
n. block dia
g
ra
m of the
sy
st
em
can b
e
see
n
in Fig
u
re 1.
Figure 1. UAV System Block
Diag
ram
Figure 1 sho
w
s th
e UAV i
s
eq
uippe
d
with a paylo
a
d
co
nsi
s
ting
of a ra
spb
e
rry pi, HD
USB cam
e
ra
, and XBee
as a
comm
unication p
r
o
t
ocol. The
g
r
oun
d statio
n
con
s
i
s
ts of
a
comp
uter e
q
u
ippe
d with X
B
ee wh
o se
rves as
a
re
ce
iver. At the ground
station,
image received
packet
s
are t
hen combin
e
d
into a whol
e file and displayed in the
images. Payload Appli
c
ati
o
n
flowchart is
shown in Figu
re 2.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Earth Im
age Cla
ssifi
cation
Desi
gn Usin
g Unm
anne
d Aerial Vehi
cle
(Barlian
Hen
r
yran
u Pra
s
et
io)
1023
Figure 2. Payload Appli
c
ati
on Flowch
art
The groun
d station system
flow
ch
art ca
n be se
en in
Figure 3.
Figure 3. The
control
syste
m
flowch
art
2.2.
Ground Stati
on Design
2.2.1. Data
Collec
t
i
o
n
Data obtai
ne
d on JPG fo
rmat. The dat
a that have b
een obtai
ned
are group
ed
into two
categ
o
rie
s
(training an
d testing). Th
e re
sea
r
ch used
450 traini
ng d
a
ta and ea
ch
cla
ss
con
s
isti
ng
of 150 data.
Traini
ng ima
ge data ca
ptured in 1
6
x1
6 pixels. Example trainin
g
data can be
seen in
Figure 4. Ea
ch 16x16
pixel
image d
a
ta i
s
p
r
o
c
esse
d
usin
g 3x3 pix
e
ls
wind
ow
a
nd shifted ev
er
y
singl
e hori
z
o
n
tally and vertically pixel. Each
windo
w is cal
c
ulate
d
the averag
e
for each col
o
r
cha
nnel. T
he
3x3 pixel
win
dow will
be
u
s
ed
to p
e
rfo
r
m testing
of t
e
st ima
g
e
s
. T
e
st ima
g
e
s
ta
ken
by 5 sample i
m
age
s with 2
56x256 pixel
s
. The
Example test data ca
n be se
en in
Figure 5.
a b
c
d
e
f g
h
i
j
k l
m
n
o
Figure 4. Examples of trai
n
i
ng data: a-e:
agri
c
ultu
re tra
i
ning data, f-j: resid
ential training d
a
ta,
and k-o: wate
r trainin
g
data
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 3, September 20
15 : 1021 – 10
28
1024
Figure 5. Examples of te
st image
s 256x
256 pixel
s
2.2.2. Algorithm
Design
Traini
ng
pha
se u
s
ing
wi
nd
ow-ba
s
ed
cl
a
ssifi
cation
techniqu
es. At t
h
is
pha
se
the
r
e
are
a
few ste
p
s, (i
)
the sel
e
ction
of training
obj
ects fo
r
ea
ch
cla
ss i
s
p
r
ed
etermin
ed, (ii) the co
nversi
on
of an RGB i
m
age to the
YUV format that will be u
s
ed as in
put to the next stage [6], (iii) feature
extraction of the input imag
e by performi
ng sig
nal ana
lysis for ea
ch
signal Y, U and V, and (i
v
)
the com
pon
e
n
ts cal
c
ul
atio
n of non-li
ne
ar cl
assifier
u
s
ing SVM [7], in this ca
se
usin
g a gau
ssian
function
RBF
as its ke
rne
l
func
tion trick. Traini
ng p
hase is inten
ded to get the value of the
para
m
eter w and
b
are
optimal, the
s
e pa
ram
e
ters
a
r
e
then
u
s
ed
to
con
s
t
r
uct
the
opti
m
al
hyperpl
ane a
s
a cla
s
sifier i
n
the testing
pha
se.
Testing phase will
process the image
data from
UAV
to cl
assify into three cl
asses. The
step
s are (i
) trackin
g
wi
nd
ow o
n
the testing image
s, (ii) imag
e co
n
v
ersio
n
into
YUV model
s,
(iii)
feature extraction by anal
yzing t
he
signal for every
signal Y, U
and V, (iii) image cl
assification
p
e
r
w
i
nd
ow
us
in
g F
SVM [8] w
i
th
pa
ra
mete
r
s
o
b
t
a
i
ned
from
the trai
ning
and
cl
assificatio
n
u
s
in
g
deci
s
io
n funct
i
on.
2.3. Algorithm
Implementa
tion
Stage algorit
hm impleme
n
tation is do
ne to build appli
c
ation
s
that will be use
d
to
determi
ne th
e out
come
of
the
cla
ssifi
cation of
th
e
prop
osed
me
thod. Based
on the
de
sig
n
of
algorith
m
s
de
scribe
d in th
e previou
s
section, th
e
g
eneral impl
e
m
entation
co
nsi
s
ts of t
w
o
main
module
s
, na
mely module
s
in the traini
ng st
ag
e and
the module
s
i
n
the testing
pha
se.
In the training modul
e there a
r
e se
veral pro
c
e
d
u
re
s that are neede
d to build
appli
c
ation
s
.
Procedu
re
s required to
bu
ild a trai
ning
module
con
s
i
s
ts
of a p
r
o
c
edure T
r
aini
n
g
Data,
F
eature
extra
c
tion
pro
c
ed
ures, and procedu
res
svmT
rain.
Trai
ning
Data p
r
o
c
ed
ure
aims
to regul
ate the process
of data retri
e
val and
g
r
o
uping the
m
into cla
s
ses
that have be
en
determi
ned.
Feature extra
c
tion procedu
re aim
s
to make the process of
extractio
n
of the feature
s
that transfo
rm RGB to YUV. The svmTrain pro
c
e
d
u
r
e
aims to make the training
for data traini
ng
pro
c
e
ss.
The
re
sults of th
ese
procedu
res
are
the
compon
ent p
a
r
amete
r
s to
be u
s
e
d
in
the
testing ph
ase
.
In testing
th
e mod
u
le
s a
r
e al
so
seve
ral p
r
o
c
e
dures
req
u
ired
to build
ap
plication
s
.
Procedu
re
s required to bu
ild a testing module
co
n
s
i
s
ts of a Data
Testing p
r
o
c
edure, Featu
r
e
extraction
proce
dures,
sv
mCla
ssify p
r
oce
dures
, FS
VM pro
c
e
dures, an
d a
c
cu
racy
pro
c
e
d
u
r
es.
Data te
sting
pro
c
ed
ure ai
ms to regul
ate the proc
ess of data colle
ction, tra
cki
n
g
win
dow. F
e
ature
extraction
p
r
o
c
ed
ure
aim
s
t
o
ma
ke
the
p
r
ocess of feat
ure
extra
c
tion
, namely th
e t
r
an
sform
a
tion
of RGB to YUV on each
wi
ndo
w re
sults
of tracking.
SvmClassify
pro
c
ed
ure ai
ming to
cla
s
sify
the wi
n
dow into th
ree
stage
s a
c
cordi
ng
hyperpl
ane fo
rmed, na
mely (1) the hyp
e
rplane b
e
twe
e
n
cla
ss a
g
ri
cultural lan
d
to
resi
dential, (2)
the hyperpl
a
ne between
cla
ss a
g
ri
cult
ural lan
d
wit
h
water, an
d
(3) the hyp
e
rpla
ne bet
ween
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Earth Im
age Cla
ssifi
cation
Desi
gn Usin
g Unm
anne
d Aerial Vehi
cle
(Barlian
Hen
r
yran
u Pra
s
et
io)
1025
cla
ss
settlem
ents with
wat
e
r. FSVM pro
c
ed
ure ai
ms t
o
determi
ne the wind
ow int
o
the cla
s
sro
o
m
[9]. Accura
cy pro
c
ed
ure ai
ms to cal
c
ul
at
e the accuracy of the test result
s.
3. Results a
nd Analy
s
is
Testing fo
cu
sed on UAV sy
stem commu
nicatio
n
and i
m
age cl
assifi
cation the
n
a
nalyzin
g
the results of
tests such as
maximum accu
racy an
d o
p
timal amou
n
t
of test data.
3.1. UAV
Sy
stem
UAV transmit images u
s
in
g the XBee protocols.
Data sent in a text and sent 10 times.
UAV system t
e
sting fo
cu
se
d on XBee da
ta commu
nications a
nd Se
rial data rece
ption.
3.1.1.
Xbee Da
ta Communicati
on
Testing
wirel
e
ss d
a
ta
co
mmuni
cation
is
don
e by
con
n
e
c
ting t
he XBee
mo
dule
with
seri
al port on
a compute
r
via USB to Serial c
onvert
e
r. This com
puter is a
s
signed to tran
smit
seri
al data
which
will be t
r
an
smitted u
s
ing ra
di
o wa
ves by the XBee modul
e. The tran
smit
ted
data will the
n
be captu
r
e
d
by another XBee modul
e is also
con
necte
d to the compute
r
. The
received data will then be displayed
on the
co
mputer. The test is
perform
e
d usi
ng X-CTU
prog
ram
issu
ed by
Digi Int
e
rnatio
nal. T
h
is p
r
o
g
ra
m i
s
u
s
ed
to con
f
igure
and
pe
rform te
sting
for
XBee module
.
The block di
agra
m
of wire
less dat
a co
mmuni
cation
s test is sho
w
n in Figure 6.
Figure 6. The
block diag
ra
m of wi
rele
ss data com
m
un
ication
s
test
In this t
e
st,
the
sen
d
e
r
is i
n
the
l
e
ft hand
of
the figu
re
6
will
se
nd t
he
words
"abcd
e
fghij
k
l
m
nopq
rstuvwxyz1234
567
8
90" twice by
writing th
e text into the X-CT
U p
r
og
ram
twice. The p
aper
will the
n
be tran
smi
tted wi
rele
ssl
y
using XBee module
s
. Tran
smitted
data
captu
r
ed
by
the XBee re
ceiver an
d d
i
splaye
d
into
the X-CTU pro
g
ra
m on
the re
cipi
en
t's
comp
uter.
3.1.2. Serial
Data
Rec
e
ption
Data
s
e
nt in
a text file that is
s
ent
to be
br
ok
e
n
u
p
in
to
se
ve
r
a
l
parts
.
In this
tes
t
the
origin
al text file si
ze of 39
00 bytes
will
be split
into 10 files
with
each si
ze i
s
390 byte
s. Thus
sen
d
ing th
e
whol
e file is
done i
n
10 ti
mes. Th
e
breakdo
wn of t
he text file can be
se
en i
n
Figure 7.
Figure 7. The result
s of sol
v
ing
a text file
that will be sent
3.1.3.
Applica
t
ion Image Da
ta
Image data
obtaine
d fro
m
image ca
pture p
r
o
c
e
s
s usi
ng the step
s de
scrib
ed in the
previou
s
test.
Images that have been o
b
tained a
r
e then processe
d to obt
ain fragment
s of data
from the ima
ge pixel data
fragment
s. In each frag
m
ent re
sults fraction
al imag
e data is in
se
rted
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 3, September 20
15 : 1021 – 10
28
1026
attributes to
assist in th
e pro
c
e
s
s o
f
forming an
image o
n
the monito
r
appli
c
ation a
n
d
che
c
ksums t
o
verify the
data. Fragm
ents
of th
e
pixel data
is then
se
nt o
ne by
one
t
o
the
monitori
ng a
pplication. The dat
a re
ce
ived by the
appli
c
ation a
nd then re
assembl
ed into
an
image that is
displ
a
yed by the monito
ring
application.
3.2.
Image Class
ification
Testing
ph
ase is
usi
ng t
w
o sce
nari
o
s.
There
a
r
e te
st the effect of
paramete
r
a
c
cura
cy
value and te
st the effect of
training d
a
ta numbe
r on a
c
curacy level.
The first sce
nario
wa
s te
sted usi
ng 1
5
0
image
s
dat
a wh
ere
ea
ch cla
s
s con
s
i
s
ts of 5
0
image d
a
ta. Then
σ
pa
ramete
rs
det
ermin
ed in t
h
is te
st ran
g
ing fro
m
0.
7 to 1.5. T
he
determi
nation
of these pa
ramete
rs i
s
the paramete
r
value
σ
achieve the highe
st level of
accuracy in t
he value
of 0.7 [
10].Thu
s, the testing
para
m
eters
σ
startin
g
fro
m
0.7. The t
e
st
results in this sce
nari
o
ca
n
be seen in T
able 1.
Accuracy value in bold are the highe
st accu
racy
score fo
r ea
ch test image
s.
Table 1. Valu
e accuracy wi
th 150 trainin
g
data
Parameter
σ
Images Accurac
y
(%)
1 2
3
4
5
6
7
8
9
10
0.7 62.97
59.73
42.41
48.69
47.00
45.70
42.05
35.20
47.00
47.23
0.8 67.18
65.75
42.27
48.44
46.95
45.60
42.84
35.31
48.38
47.54
0.9 70.55
70.06
42.18
47.96
46.89
45.43
43.40
35.25
50.06
48.60
1.0 72.65
72.99
42.16
47.31
46.80
45.32
44.28
35.69
51.59
49.08
1.1 73.54
74.74
42.25
46.50
46.68
45.23
43.86
35.61
52.23
48.93
1.2 74.19
75.70
42.33
45.85
46.64
45.25
42.28
34.78
51.47
48.17
1.3
74.60
76.25
42.38
45.55
46.66
45.27
40.55
33.38
49.74
47.02
1.4 74.59
76.66
42.42
45.42
46.66
45.28
39.61
31.90
46.92
45.62
1.5 74.44
76.68
42.40
45.40
46.66
45.33
39.34
31.11
44.15
44.55
From
Tabl
e 1
,
it can
be
se
en that fo
r
e
a
ch
of the
te
st ima
g
e
s
, th
e maximum
accuracy
obtaine
d with
different p
a
rameter val
u
e
s
. The 1
st
i
m
age cla
s
sification with 1
50
trai
ning d
a
ta
and pa
ram
e
ter
=1.3 sho
w
ed o
n
Figu
re 8.
Figure 8. The
1
st
image cla
ssifi
cation
with 150 traini
ng
data and pa
rameter
=1.3
: (a) Image
Testing; (b) I
m
age cl
assifi
cation result
The
se
con
d
scen
ario
is int
ende
d to
det
ermin
e
the
ef
fect of trainin
g
data
num
b
e
r
on th
e
accuracy. T
h
e test
s
con
d
u
cted
for t
r
ai
ning d
a
ta
with the n
u
mb
e
r
rangin
g
fro
m
150,
225,
300,
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Earth Im
age Cla
ssifi
cation
Desi
gn Usin
g Unm
anne
d Aerial Vehi
cle
(Barlian
Hen
r
yran
u Pra
s
et
io)
1027
375, a
n
d
45
0 ima
ge
dat
a with
differe
nt pa
ramete
r value
s
as acco
rdi
ng to
the
re
sults t
e
st
accuracy rate
previou
s
ly [11]. The level of accu
racy f
o
r ea
ch traini
ng data n
u
m
ber i
s
sh
own
in
Table 2.
Table 2. List
of accura
cy level of trainin
g
data
Image
Data
Accuracy
w
i
th
nu
mber of tr
aining data (%
)
150 225
300
375
450
1
st
74.60
75.74
71.95
75.08
71.06
2
nd
76.68
77.28
76.17
77.69
74.94
3
td
42.42
42.17
41.96
42.67
43.69
4
th
48.69
51.01
52.93
52.43
48.89
5
th
47.00
47.44
47.85
48.07
47.76
6
th
45.70
45.62
45.42
46.23
46.95
7
th
44.28
49.69
51.08
49.29
40.38
8
th
35.69
41.77
43.27
40.39
37.29
9
th
52.23
63.10
63.85
60.15
46.15
10
th
49.08
55.51
56.73
53.73
47.76
Average
51.43
54.93
55.12
54.57
50.49
Maximum
76.47
77.28
76.17
77.69
74.94
Table
2
sho
w
s that the
cha
nge in
the
am
ount
of trainin
g
data
used
h
a
ve an i
n
flue
nce
on
the a
c
cura
cy
of the
re
sult
s, but fo
r
all
the
test
data
used
ca
nnot
be
determin
ed exa
c
tly at
a
certai
n am
ou
nt of data to be ge
nerated
value ma
x
i
m
u
m ac
cu
ra
cy
.
I
t
was
sho
w
n
f
r
om t
he r
e
s
u
lt
s
of the acqui
si
tion value of the maximu
m
accuracy for
each test ima
ge data.
3.3. Analy
s
is
Table 1 sho
w
ed, for the
1
st
image, the maximu
m accuracy value obtain
ed from
para
m
eters d
e
termin
ation
= 1.3
a
nd t
he 2
nd
i
m
age,
it is obtai
ned
with
the
pa
rameter value
s
= 1.5,
re
spe
c
tively 74.60 %
and
76.68 %
.
The 3
rd
ima
ge, it is o
b
tai
ned from d
e
termin
ation of
the
para
m
eter va
lues
= 1.4
with an
accu
racy 42.4
2
%. The 4th,
5
th
, and 6
th
im
ag
e, the maxim
u
m
accuracy valu
es
obtained
with the pa
ra
meter value
s
= 0.7 respe
c
tively 48.69
%, 47.00 %,
and
45.70 %. The
7
th
, 8
th
image, and 1
0
th
image
obtaine
d the maximu
m accu
ra
cy p
a
ram
e
ter val
u
e
s
with
= 1.0
respe
c
tively 44.28 %, 35.69 %, and 4
9
.08 %. The
9
th
image o
b
tained from
the
determi
nation
of paramete
r
values
= 1.
1 with a
n
a
c
cura
cy 52.23
%. These
re
sults indi
cate t
hat
the next determinatio
n of the pa
ramete
rs
cann
ot use a singl
e value,
but different for ea
ch test
image
s acco
rding to the maximum accu
racy value.
From Ta
ble 2
,
it can be se
en that 5 of t
he 10 test im
age
s, the highest a
c
cura
cy occu
rs
whe
n
numb
e
r of training data is 300. So it is recom
m
ende
d by 300 trainin
g
da
ta. And base
d
on
the
average accuracy re
sults,
it
app
ears that the
acq
u
isition val
ue
of the hig
h
e
s
t accu
ra
cy wh
en
the traini
ng d
a
ta is 3
00. M
a
ximum a
c
cu
racy
rate of
al
l test imag
es
wa
s 77.6
9
% i
n
the 2
nd
ima
ge
with numb
e
r
of training dat
a for 375.
4. Conclusio
n
The d
a
ta
se
nt by the
UAV aircraft
wirele
ssly throu
gh XBee
ca
n
be
received
by the
monitori
ng
station applications. Y
U
V m
odel
transformation method an
d Fuzzy Support Vector
has
be
en su
cce
ssfully perf
o
rme
d
UAV
i
m
age
cla
s
sification.
parameter
value
cha
nge
s and
the
amount
of training
data
give effect
to the
chan
g
e
the
a
c
cura
cy value,
bu
t the maxim
u
m
para
m
eter a
c
curacy value
is different. M
a
ximum accu
racy value o
b
t
ained in the
2nd imag
e wi
th
77.69% a
c
cu
racy
with d
e
termin
ation of
the pa
ramet
e
rs o
c
cura
cy
obtaine
d with
=
1.5 an
d
the
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 3, September 20
15 : 1021 – 10
28
1028
amount
of tra
i
ning d
a
ta a
s
375. T
he
hig
hest
accu
ra
cy level of 10
of test d
a
ta i
s
5
5
.12%
wh
en
training d
a
ta of 300.
Referen
ces
[1]
Richar
ds Jo
hn
A. Remote S
e
nsin
g Di
gita
l Image A
n
a
l
ysis
: An Introducti
on. Verl
ag B
e
rl
in H
e
id
el
berg
:
Sprin
ger. 20
12
.
[2]
Mahmo
od S. U
n
man
ned A
e
ria
l
Vehic
l
e C
o
m
m
unic
a
tions
. M
a
ster T
hesis. Sw
e
d
en: Bleck
i
n
g
Institute of
T
e
chnolog
y; 2
007.
[3]
Merino
L, Caballero F, Mart
inez RJ, Maza I, Ollero A.
Autom
a
tic Forest
Fi
re Monitor
i
ng a
n
d
Measur
e
m
ent
usin
g U
n
man
n
ed A
e
ria
l
V
ehi
cles
. VI Intern
a
t
iona
l C
onfer
e
n
ce
on F
o
r
e
st
F
i
re Res
earc
h
DX Vi
eg
as. 20
10.
[4]
Her
w
itz
SR, J
ohns
on
LF
, D
una
ga
n SE,
H
i
ggi
ns
RG, et
al. Ima
g
i
ng f
r
om a
n
U
unm
ann
ed
Aeri
al
Vehic
l
e: Agric
u
ltural Surv
eil
l
a
n
ce a
nd D
e
cisi
on Su
pport.
Co
mp
uters a
nd El
ectronics i
n
Ag
riculture
44
.
200
4; 49(6
1
).
[5]
Coifman B, Mc
Cord M, M
i
shalani RG, Is
w
a
lt M, Ji Y.
R
o
a
d
w
ay T
r
affic Monitori
ng fro
m
a
n
U
n
man
n
e
d
Aerial V
ehic
l
e
. IEE Proc. Intell.
T
r
ansp. S
y
st. 1. 2006; 1
53.
[6]
Rub
e
rt C, F
onseca L, Vel
h
o
L.
Lear
nin
g
Based S
uper-
R
esol
utio
n Usi
ng YUV Mo
de
l for Re
mo
t
e
Sensi
ng Imag
e
s
. Conferenc
e. Procee
din
g
s of
WT
DCGPI. - [s.l.] : WT
DCGPI. 2005.
[7]
Abe, Shi
ngo,
Abe S, T
a
ku
ya I.
F
u
zz
y S
u
pport Vector
Machi
ne for
Multiclass Pr
o
b
le
ms
. Brug
es
,
Belg
ia: Euro
pe
an S
y
mpos
ium
on Artificial N
e
ural N
e
t
w
ork. 2
002.
[8] T
a
kuy
a
I.
F
u
zzy Support Vect
or Machin
e for
Multiclass Pro
b
le
ms
. Euro
pe
an S
y
m
pos
ium
on Artificial
Neur
al Net
w
o
r
ks. Bruges, Bel
g
ia. 20
02.
[9]
Yuy
o
ng C, Zhiy
uan Z.
Remot
e
Sensi
ng Ima
ge Class
ificati
on Base
d on the HSI T
r
ansformatio
n
an
d
F
u
zz
y
S
upp
or
t Vector Ma
chin
e
. IEEE International
Confer
enc
e on Future Computer and
Commun
i
cati
o
n
. 2009: 6
32-6
35.
[10]
Sutrisno, Afif
S, Imam C. Impleme
n
tasi
Me
tode
W
a
ter
s
hed
Da
n Mor
f
olog
i U
n
tuk S
egme
n
tasi P
a
d
a
Citra Satel
i
t Area Un
iversitas
Bra
w
ij
a
y
a. 20
1
2
: 1-8.
[11]
Gonzal
ez RC, E W
oods R. Di
gital Imag
e Pro
c
essin
g
. 2nd E
d
itio
n. Prentice
Hall. 20
02.
Evaluation Warning : The document was created with Spire.PDF for Python.