TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 13, No. 3, March 2
015,
pp. 561 ~ 56
7
DOI: 10.115
9
1
/telkomni
ka.
v
13i3.719
7
561
Re
cei
v
ed O
c
t
ober 2, 20
14;
Revi
se
d De
cem
ber 16, 20
14; Accepted
Jan
uary 5, 20
15
Modeling Singular Value Decomposition and K-Means
of Core Image in Clasification of Potential Nickel
Agung Praju
h
ana Putr
a*
1
, Agus Buo
n
o
2
, Bib Paruhum Silalahi
3
1,2
Departement
of Computer S
c
ienc
e, F
a
cult
y of Mathematic
s and Natur
a
l
Scienc
es,
Bogor Agr
i
cult
ural U
n
ivers
i
t
y
,
1668
0 Bog
o
r,
Indon
esi
a
, Ph/F
ax. +
62-2
51-6
284
48/6
229
61
3
Departem
ent of Mathematic,
F
a
cult
y
of
Mat
hematics a
nd
Natura
l Scienc
es,
Bogor Agr
i
cult
ural U
n
ivers
i
t
y
,
1668
0 Bog
o
r, Indon
esi
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: agun
gp
p_ma
il@
ya
ho
o.com
1
, pudesh
a
@
y
a
hoo.co.i
d
2
,
bib
paru
hum
1
@
yah
oo.com
3
A
b
st
r
a
ct
Explor
ation
is a mai
n
proces
s
in
th
e nicke
l mi
nin
g
activitie
s
. One of th
e
most
i
m
porta
nt
steps i
n
expl
oratio
n
is o
b
tain
soi
l
s
a
mp
les (c
ores) to
d
e
termin
e
th
e p
o
tentia
l
of nick
el i
n
th
e s
o
il.
L
abor
atory testi
n
g
is a w
a
y to
kno
w
how
muc
h
th
e nick
el c
onte
n
t
on th
e cor
e
.
T
h
is rese
arch
ai
ms to
utili
z
e
t
he c
o
re i
m
ag
e
of
the statistical c
haracter
i
stics
of
color
and te
xture, Bipl
ot a
nalysis
usi
ng
SVD, K-Means
and
ide
n
tificat
i
o
n
usin
g SVM met
hod w
i
th RBF
kerne
l
and
poly
n
o
m
i
a
l to deter
mi
ne the p
o
ten
t
ial of nicke
l.
Ke
y
w
ords
: SVD, K-Means, SVM, RBF
,
polyno
m
i
a
l
Copy
right
©
2015 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Indone
sia i
s
rich in natu
r
al
resource
s, o
ne of
whi
c
h i
s
a mine
ral.
One type of sedim
ent
whi
c
h h
a
s th
e potential
to
be u
s
e
d
a
s
mines is sedi
mentary late
rite. Sediment
widely
used
for
mining late
rite nickel ore a
nd / or iron
-ni
c
kel (Fe
-
Ni
) [1].
To obtain the
nickel ore
sh
ould be do
ne
by mi
ning activity exploration. Explorati
on is a
survey a
c
tivity of the invest
igatio
n an
d asse
ssme
nt area that
estimated
contain
s
valu
able
minerals. Th
ese
activities gene
rate in
formati
on
of the soil th
a
t
is usually
perfo
rmed
b
y
a
geolo
g
ist.
Accordi
ng to
Hazria [2
], the research using the data from
the
sat
e
llite fractal j
u
st take a
sampl
e
su
rfa
c
e, so it cann
ot be use
d
a
s
an i
ndicator t
o
obtain info
rmation on th
e percenta
g
e
o
f
nickel
conte
n
t
below the
surface. On
e
of the meth
o
d
s d
one
is th
e analy
s
is of
the nickel
co
ntent
of the beddi
ng to see t
he stratigra
p
h
ic. The
r
efo
r
e, this stud
y aimed to determin
e
the
stratig
r
ap
hic l
a
terite of nickel conte
n
t in the se
diment.
Another re
se
arch i
s
con
d
u
c
ted to
dete
r
mine
the pot
ential of ni
ckel in
soil
by using XRF
(X-Ray Fluo
rescen
ce
). XRF is a
tool that use
s
sp
ect
r
ometry
meth
od to analyze
the content
of
particular mat
e
rial
elements. XRF
sp
ectrometry utilizes the X
-ray
s
em
itted by the materi
al that is
sub
s
e
que
ntly captu
r
e
d
by
the dete
c
tor to analy
z
e
th
e
conte
n
t of th
e elem
ent. So far, the
use
of
XRF in
stru
m
ent is to a
nal
yze metal
all
o
ys, copp
er,
aluminu
m
all
o
y, rocks, mi
neral
s, a
nd
crust.
Material
s
th
at
ca
n be anal
yzed are
in
the
fo
rm of
a massive solid
,
plate or po
wde
r
.
Elem
e
n
tal
analysi
s
pe
rforme
d both q
ualitatively and quantitativ
e
l
y. Quantitative analysi
s
is t
o
determi
ne the
numbe
r of ele
m
ents
contai
ned in the ma
terial [3].
Until n
o
w, th
e meth
od to
determi
ne th
e pote
n
tial of
nickel
in th
e
co
re
shoul
d
be d
one
throug
h labo
ratory testing
usin
g X-ray beam with a
relatively long
pro
c
e
ss that
is more than
10
hours
as
we
ll as the
co
nsid
era
b
le
cost. Based o
n
this, it is
necessa
ry to
investigate
the
cha
r
a
c
teri
stics of the m
odel extra
c
ti
on an
d
cl
a
ssifi
cation
m
e
thod
s ap
propriate
to the
cla
ssifi
cation
of nickel by
u
s
ing th
e ima
g
e
cla
s
sificatio
n
process ni
ckel
co
re
s that
can
be
done
more quickly.
Several studi
es ba
sed o
n
the identificati
on
of such imagery ha
s be
en develop
ed
for face
recognitio
n
with a 84 % a
c
cura
cy rate
[4
], and then developed reached 97.3
% [5]. In
this
resea
r
ch, the classification
of nickel in
soil sa
mpl
e
result
s of expl
oration
(core) base
d
on th
e
cha
r
a
c
teri
stics of
col
o
r an
d texture
wit
h
the
sin
gula
r
value
d
e
co
mpositio
n
(SVD)
and
K-M
ean
s
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 13, No. 3, March 2
015 : 561 – 5
6
7
562
as a
red
u
ctio
n of the ch
aracteri
stics, a
s
well a
s
the u
s
e of Sup
port
Vector M
a
ch
ine (SVM)
as a
method for
cl
assificatio
n
.
2. Rese
arch
Metho
d
2.1. Preproce
ssi
ng
This re
sea
r
ch
used 160
i
m
age data which are
divi
ded into
four cla
s
ses,
wh
ere
ea
ch
cla
ss
co
nsi
s
t
s
of 40 im
age
s. This i
m
age
come
s fr
o
m
the re
sult
s of the explo
r
atio
n perfo
rme
d
10
times at different place
s
in
2013. The ex
plorat
io
n re
su
lts image hav
e different di
mensi
o
n
s
, then
do croppi
ng
with dimen
s
io
ns of 120
0 x 120 pixel
s
.
Figure 1. Pre
p
ro
ce
ssi
ng
2.2.
Grouping
w
i
th K - Fold
Cross Validati
on
At this stage, the image of eac
h cl
ass will be divided or scramble
d into subg
ro
up
s, from
sub
g
ro
up k, take
n one
su
bgro
up to be
used a
s
dat
a validation test, and the
rest is u
s
e
d
for
training
d
a
ta,
the pro
c
e
s
s is rep
eated so that
all subgrou
ps
can b
e
used a
s
testing data. Fol
d
use
d
in this rese
arch con
s
ists of 4 su
bg
roup
s.
2.3. Feature
Extr
action
This re
se
arch
u
s
ed a ch
ara
c
teri
stic
e
x
tracti
on ba
sed
o
n
stati
s
ti
cal col
o
r and
texture.
Feature colo
r ca
n be o
b
tained th
rou
g
h
statisti
ca
l
cal
c
ulatio
ns
su
ch a
s
me
an (1
), stan
dard
deviation (2
), skewne
ss
(3) and ku
rto
s
is
(4) [6]. Fo
r ex
ample, ena
bl
e this feature can b
e
used t
o
identify the interest
s of orn
a
mental pla
n
ts [7
]. Calcul
a
t
ion imposed
on ea
ch com
pone
nt of R, G
and B.
∑∑
(1)
∑∑
(2)
∑∑
(3)
∑∑
(4)
While
the te
xture cha
r
a
c
teristi
c
s obtai
ned fr
om sta
t
istical Grey Level
Co-occurren
ce
Matrices (GLCM).
GLCM f
i
rs
t
prop
o
s
ed
[8] with
28
Feature to
explain th
e
spa
t
ial patterns [9].
GLCM u
s
e
s
textures in
se
con
d
-o
rd
er calcul
ation.
In the se
con
d
orde
r, the pairwise relatio
n
ship
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Modelin
g Singular Val
ue Decom
p
o
s
ition
and K-Mea
n
s
of Co
re… (Agung Praj
uh
ana Putra
)
563
betwe
en t
w
o
origin
al im
ag
e pixel
s
i
s
ta
ken
into
acco
unt [10]. To
g
e
t Featu
r
e
G
L
CM,
only
so
me
scale that
pro
posed by
Ha
ralick. Fo
r exa
m
ple [11] o
n
l
y
use
s
five
scale for
GL
CM
, in the form
of
angul
ar
se
co
nd mom
ent (ASM) (5
), Co
ntrast
(6
), inverse different
moment
(IDM
) (7
), ent
ropy
(8)
and correlatio
n (9).
∑∑
,
(5)
∑
∑
|
|
(6)
∑∑
,
(7)
∑∑
,
,
(8)
∑∑
,
′
′
′
′
(9)
Sample featu
r
e extra
c
tion
pre
s
ente
d
in Table 1.
Table 1. Unit
s for Mag
neti
c
Prop
ertie
s
Stat. Warna
Stat. Tekstur (GL
C
M)
0
o
45
o
90
o
135
o
mean_r:
117.0014
mean_g:
104.6245
mean_b:
86.0524
dev_r: 31.858
9
dev_g: 32.8633
dev_b: 33.0866
skew
_
r: 0.1130
skew
_g: 0.3
337
skew
_b: 0.6
607
cur_r: -0.
3372
cu
r
_
g
:
-
0
.
1
22
0
cur_b: 0.3154
asm: 1.5540e-0
0
4
Contrast: 478.
24
68
idm: 0.0936
entrop
y
: 9.123
7
correlation: 6.59
99e-
004
asm: 1.5270e-0
0
4
Contrast: 493.
18
46
idm: 0.0926
entrop
y
: 9.137
7
correlation: 6.51
91e-
004
asm: 3.6057e-0
0
4
Contrast: 83.9
0
5
8
idm: 0.1966
entrop
y
: 8.303
0
correlation: 7.88
58e-
004
asm: 1.5187e-0
0
4
Contrast: 499.
15
84
idm: 0.0918
entrop
y
: 9.141
1
correlation: 6.50
43e-
004
2.4.
SVD and Bip
l
ot Analy
s
is
Biplot was
de
rived from the
SVD deco
m
positio
n of the data matri
c
and the cha
r
a
c
teri
stic
extraction results will
be di
splaye
d in the form of
2 dimensi
on im
age Biplot visualization. From
the pi
cture,
we
ca
n
see
the relation
shi
p
bet
wee
n
th
e varia
b
le
s a
nd the
comp
arison
betwe
en
cla
s
ses that chara
c
te
rizes t
he potential o
f
nickel
in the
core. If the variabl
es
were
overlappi
ng or
contig
uou
s then these vari
able
s
have in common / linka
ge
s that can be red
u
ce
d into one ne
w
variable.
2.5. K-M
eans
K-Mean
s is u
s
ed to optimi
z
e the com
b
i
nation of
variable
s
that most represents
the core
of nickel
ch
a
r
acte
ri
stics, so that when
te
sted
with S
V
M model
s
will ge
ne
rate
value O
p
tim
u
m
accuracy.
Co
mbination val
ue (K
) ge
nerated from
th
e previo
us
st
age
s in the
biplot an
alysi
s
pha
se.
2.6.
Modeling Wi
th Suppor
t Vector M
achi
n
e
Modelin
g by usin
g Supp
ort Vector Ma
chine is
mult
i cla
ss;
S
V
M
h
a
s
sev
e
r
a
l m
e
t
hod
s in
comp
ari
ng o
b
ject
s, they a
r
e on
e-again
s
t-on
e an
d
o
ne-a
gain
s
t-all. In this re
se
arch will
be u
s
e
d
one-agai
nst
-
a
ll with RBF ke
rnel fun
c
tion
s and polynom
ial.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 13, No. 3, March 2
015 : 561 – 5
6
7
564
2.7.
Stages o
f
Te
sting
The test dat
a used in th
e research p
o
sse
ss of 4 grou
ps, ea
ch
group
conta
i
ns 40
image
s, and
each of these
grou
ps
will b
e
use
d
as
te
st data alterna
t
ely, so that the wh
ole ima
ge
will be tested.
Model testin
g
will be used
the Support
Vector
Ma
chine mod
e
lin
g. Data teste
d
is the
data
redu
ctio
n that
re
sulte
d
from
trai
nin
g
data
to
yiel
d
the co
rrect
cla
ssifi
cation
of
re
du
ce
s
te
st
image
s.
The method used is
"on
e
- agai
nst
- all"
w
here the cl
assification is t
r
ain
e
d
by the e
n
tire
data to comp
are the a
c
curacy of each ke
rnel fun
c
tion
and feature
extraction.
2.8. Resul
t
An
aly
s
is
Traini
ng an
d
testing pro
c
e
ss u
s
e
d
the SVM mo
deling to yield the accura
cy or
su
ccessfuln
e
ss of identif
ying the cla
ssi
fication of po
tentia
l nickel and imag
e reco
gnition e
r
ror
rat
e
i
n
t
h
e
c
o
re
of
e
a
c
h
cla
ss.
A
c
cur
a
cy
is
calculat
ed b
a
sed
on
test
data
on
4 fold
valida
t
ion
pro
c
e
ss, in o
r
der to dete
r
m
i
ne the fault di
stributio
n by usin
g the co
n
f
usion mat
r
ix.
A
ccur
a
cy
x100%
3. Results a
nd Analy
s
is
Data co
re was
o
b
taine
d
from
the dat
a
of 10 tim
e
s expl
orat
io
n preprocessing by
cro
ppin
g
with
dimen
s
ion
of
1200x1
20 pi
xels, and
cl
assified b
a
sed
on lab
test
re
sults
data i
n
to 4
cla
s
ses with
40 ima
g
e
s
fo
r e
a
ch
class.
The fo
ur
cla
s
se
s a
r
e:
Cla
s
s 1:
Lo
w Pot
ential of Sm
o
o
th
Texture, Cl
ass 2: Lo
w Pot
ential of Rou
gh Textur
e, Cla
ss
3: Hig
h
Potential of
Smooth Text
ure,
Cla
ss 4: Hi
gh
Potential of Rou
gh Texture
Low potential
categ
o
ry
if the pe
rcentag
e of
≤
1.5 %
of nickel a
nd
high p
o
tential
if levels
≥
1.5 % of nickel. While th
e texture ca
n
be see
n
fr
o
m
the sha
pe,
if it is rocky then the cate
gory
is rou
gh texture and if it is not
then the category is
sm
ooth texture.
3.1.
Training and
Testing Data
Each
cla
ss i
m
age of K-F
o
ld method
will
be divided in
to subg
ro
up
s, from the sub
g
rou
p
k
will be taken a
subgroup
as
data
validation test, and
the
rest is used for trai
ning data. T
h
e
pro
c
e
s
s is
re
peated
so th
at all su
bg
ro
ups
ca
n be
use
d
a
s
test
data. Fold
used in thi
s
stu
d
y
con
s
i
s
ts of 4 sub
g
ro
up
s. The K-Fold di
stribution
of the testing an
d training d
a
ta is sh
own belo
w
.
Class Number
of
Data
1
Test Data
Training data
Training data
Training data
2 Training
data
Test Data
Training data
Training data
3
Training data
Training data
Test Data
Training data
4
Training data
Training data
Training data
Test Data
Table 2. Training an
d Test
ing Data
with K - Fold
Pattern
Training Data
Testing Data
1
11,12,13,14,
15,1
6
,17,18,19,2
0
,
21,22,23,24,
25,2
6
,27,28,29,3
0
,
31,32,33,34,
35,3
6
,37,38,39,4
0
1,2,3,4,5,6,7,
8,9,
10
2
1,2,3,4,5,6,7,
8,9,
10,21,22,23,
24,2
5
,
26,27,28,29,
30, 31,32,33,34,
35,
36,37,38,39,
40
11,12,13,14,
15,1
6
,17,18,19,2
0
3
1,2,3,4,5,6,7,
8,9,
10,
11,12,13,14,
15,1
6
,17,18,19,2
0
,
31,32,33,34,
35,3
6
,37,38,39,4
0
21,22,23,24,
25,2
6
,27,28,29,3
0
4
1,2,3,4,5,6,7,
8,9,
10,
11,12,13,14,
15,1
6
,17,18,19,2
0
,
21,22,23,24,
25,2
6
,27,28,29,3
0
31,32,33,34,
35,3
6
,37,38,39,4
0
3.2.
Color and Te
xture F
eatur
e Extrac
tion
K-Fold trainin
g
data
subg
roup
s extra
c
te
d ba
s
ed o
n
th
e col
o
r a
nd te
xture ch
aract
e
risti
cs.
Colo
r
extracti
on p
r
ovide
d
by cal
c
ul
atin
g the
mea
n
v
a
lue
of the
compon
ent
R
(re
d),
G
(gre
en),
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Modelin
g Singular Val
ue Decom
p
o
s
ition
and K-Mea
n
s
of Co
re… (Agung Praj
uh
ana Putra
)
565
and B
(blu
e).
Whil
e the te
xture extra
c
ti
on p
r
ovid
e
d
by GL
CM val
ue of th
e im
age. Th
e ove
r
all
value extracti
on dimen
s
io
n
of 160x32 is
sho
w
n in Ta
b
l
e 3.
Table 3. Col
o
r and Textu
r
e
Feature Extraction
1
2
3
.
..
..
.
3
2
1
9
6.6
5
7.6
2
7
1
3.8
12.2
8
.85
-
0.2
2
9
1.8
5
1.5
1
9
.
5
1
4.8
12.4
6
.96
-
0
.
44
3
7
5.9
4
2.4
2
1
1
3.3
12.6
1
0
.
1
-
0
.
14
4
1
08
9
4
56
.2
2
2
.7
22.4
2
2
.
3
-
0
.
38
.
8
8.7
7
7.6
4
6
.
7
2
5.7
30.4
3
2
-
0.2
.8
0
.
26
7
.
63
5
.
52
9
.
1
3
0
.
12
8
.
30
.
7
3
.
9
8.7
8
8
5
5
.
1
2
5.3
25.9
2
6
.
9
-
1
.
8
8.9
7
7.9
4
5
.
5
2
7.2
27.5
2
6
.
4
-
0
.
43
.
1
0
4
83
.1
4
4
.
2
25
.
2
26
2
4
.
8
0.
2
8
16
0
1
14
9
7
.8
58
.8
2
2
.4
2
2
20
.4
-
0
.18
3.3. Biplot
An
aly
s
is
From th
e d
a
ta matrix extraction
of 16
0x
32 pixel
s
with
singul
ar value d
e
co
mpositio
n
(SVD) te
chni
que can be d
e
scrib
ed in a
biplot, as sho
w
n in Figu
re
2.
Figure 2. Biplot relation
shi
p
of variable
with the obje
c
t.
Biplot analysi
s
is to
red
u
ce the varia
b
l
e
by
com
b
ini
ng varia
b
le
s that have rele
vance
or
nearly e
qual i
n
to one
ne
w
variable. T
h
e
biplot va
ri
abl
e re
du
ction a
nalysi
s
de
scri
bed in
Figu
re
3
and Tabl
e 4 b
e
low.
Table 4. Re
d
u
ction
with Biplot Analysis
No
Ne
w
Variable
Variable Reduction
1 V1
1
2 V2
2
3 R1
3,4,5,6,14,19,
24,
29
4 R2
16,21,26,31
5 R3
7,8,9,10,11,1
2
,1
7,22,27,32
6 R4
13,18,23,28
7 R5
15,20,25,30
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 13, No. 3, March 2
015 : 561 – 5
6
7
566
Figure 3. Vari
able re
du
ctio
ns with bi
plot analysi
s
Red
u
ctio
n variable
with bi
plot analysi
s
resu
lted in seven variabl
e
s
that rep
r
e
s
ent the
cha
r
a
c
teri
stics of the
nickel
co
re ima
ge.
The va
lue
of
7 variabl
e is then b
e
came
con
s
tant valu
es
in the K-Mea
n
s.
3.4.
Variable Red
u
ction
w
i
th K –Mea
ns
The n
u
mbe
r
of biplot a
nal
ysis va
riabl
es becam
e a co
nstant (K)
i
n
t
he redu
ction
stage
by
usin
g the
K-Mean
s m
e
th
od. Thi
s
m
e
thod i
s
to
opt
imize th
e
co
mbination
th
at rep
r
e
s
e
n
ts the
nickel
co
re
o
n
the im
age
automati
c
all
y
. The
pai
r
wise
combi
n
ation of th
ese varia
b
le
s
wa
s
formed from the iteration p
r
oce
s
s
duri
ng
the training/te
sting mod
e
ls.
3.5.
Nickel Poten
t
ial Classific
a
tion Using SVM
In SVM mod
e
l buildin
g p
r
oce
s
s, the d
e
termin
ation
value of pa
rameter i
n
th
e ke
rnel
function i
s
ve
ry affected to
the output. T
he mo
re
o
p
timal the value
of para
m
eter then the b
e
tter
the re
sultin
g
model. T
he v
a
lue of
pa
ra
meter i
s
nee
ded to
produ
ce th
e p
r
e
c
isi
on o
n
a
mod
e
l. In
this research
, the ke
rnel
para
m
eters
o
b
tained
by
u
s
ing
gri
d
sea
r
ch
metho
d
within a
cert
ain
interval
a
nd with RBF ke
rnel
a
nd poly
nomial.
T
he cla
ssifi
cation
pro
c
e
s
s
u
s
in
g
SVM with one
again
s
t all m
e
thod fun
c
tio
n
s of the kernel, image
si
ze, and th
e redu
ction of di
fferent variab
les.
Some of them s
u
c
h
as
:
a)
Experiment
s
usin
g the
RB
F kern
el fun
c
tion pa
ram
e
ter
1, 5,
1
0
, 15, 2
0
, 25,
30,
35, 40, 45, 5
0
with a 0
-
90
% image re
d
u
ction
a
nd
with the co
nsta
n
t
K-Mean
s =
4, 5,
6, 7, 8, 9, 10.
b)
Experiment
s
usin
g a p
o
lynomial kernel f
uncti
o
n
pa
ra
meter d
= 1,
2, 3, 4, 5 with
a 0-
90 % ima
ge
redu
ction
an
d with
a com
b
ination
Mea
n
s= 4, 5,
6, 7, 8, 9, 10.
After
each p
r
o
c
e
s
s is
perfo
rme
d
,
re
sulting i
n
optimum
accura
cy in
pa
ra
meter
RBF v
a
lue
s
75, ima
g
e
red
u
ctio
n=
0
%
(120
0x120
) an
d
k-me
an
s vari
able
re
ductio
n
with
k=7.
The accuracy
of eac
h process is illustrat
ed in Figure 4.
Figure 4. SVM-RBF a
c
cu
racy re
sults u
s
ing K-Mea
n
s
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Modelin
g Singular Val
ue Decom
p
o
s
ition
and K-Mea
n
s
of Co
re… (Agung Praj
uh
ana Putra
)
567
4. Conclusio
n
Potential nickel classification in the image co
re is
usin
g SVD Biplot redu
ctio
n and K-
Mean
s and S
V
M as modeli
ng. Red
u
ctio
ns of the
ima
gery used in this re
se
arch
experim
ents
are
betwe
en 0% until
90%, wh
ile
the
SVM
model
usi
n
g
RBF
ke
rnel
functio
n
a
nd
kernel
polyn
o
m
ial
with the meth
od of one–
ag
ainst-all.
The re
se
arch
con
c
lud
ed a
s
follows:
1)
Biplot analysi
s
re
sult
s can
be used a
s
the value of K in K-Mean
s.
2)
The results
of the pote
n
tial nickel
cla
ssifi
cation
using SVM me
thod with
RBF ke
rn
e
l
function h
a
s
an accu
ra
cy rate of 71.875
% on K-Mean
s = 7.
3)
The results of
the potential
nickel
cla
ssifi
ca
tion
usi
ng
SVM with pol
ynomial kern
el functio
n
has a
n
accu
racy rate of 62
.5 % in the K-Mean
s = 7.
4)
The re
sult
s o
f
image cla
s
sificatio
n
of nick
el after
redu
ced va
ria
b
le with K-M
ean
s have
highe
r accu
ra
cy com
pare to before.
This
re
sea
r
ch ca
n be d
e
v
eloped
with
fitted
the selectio
n of
method
s for
obtainin
g
optimal value
param
eter in
SVM kernel
comp
are
with other c
l
ass
i
fic
a
tion methods
such as
K-NN
method.
Ackn
o
w
l
e
dg
ements
I woul
d li
ke t
o
express
my sp
eci
a
l tha
n
k
s
of g
r
atitud
e supe
rviso
r
commi
ssion
who
gave
me golde
n g
u
idan
ce until
I am able to compl
e
te this re
se
arch,
as well a
s
the Dire
cto
r
a
t
e
Gene
ral
of E
ducation i
n
Higher Edu
c
ati
on
(DIKTI
), who re
spe
c
tive
ly
have cont
ri
buted
fu
nd
s
to
study and
su
gge
stion
s
for me.
Referen
ces
[1]
Simanj
untak. D
e
termin
a
tion of
Nickel Co
ntent
on Sedim
ent s
t
ratigrap
h
y
late
rite. 1994.
[2]
Hazria. L
a
terite
nickel i
n
sedi
ments. 2007.
[3]
Rudi S
u
r
y
a
d
i.
Determin
a
tio
n
of Nickel
Cont
ent on S
edim
e
nt stratigrap
h
y
laterite i
n
Ke
n
dari. T
hesis.
Surab
a
y
a: Uni
v
ersit
y
H
a
l
uol
e
o
; 2011.
[4]
Yang J
i
a
ng a
nd Z
h
a
ng
Dav
i
d. A Ne
w
Ap
proac
h
to Ap
p
eara
n
ce-Bas
ed
F
a
ce Re
pres
entatio
n a
n
d
Reco
gniti
on.
IEEE Transactio
n
on Pattern A
nalysis
and Ma
chin
e on Intel
l
i
genc
e.
200
4; 26(1): 1-9.
[5]
Le T
hai Hoa
n
g
,
Bui Len. F
a
ce
Recog
n
iti
on B
a
sed o
n
SVM and 2
D
PCA
. In
ternatio
nal J
o
u
r
nal of Si
gna
l
Procesi
ng, Ima
ge an
d Pattern
Recog
n
itio
n P
r
ocesi
ng.
20
11
; 4(3): 85-93.
[6]
Martinez, W
L
and Marti
nez, AR. Computati
ona
l
Statistics Han
dbo
ok
w
i
th
Matlab. CRC
Press LL
C
F
l
orida. 2
0
0
2
.
[7]
Abdu
l Kad
i
r an
d AdhiS
u
sa
nto.
T
heor
y
and A
pplic
atio
n of Image Proc
essin
g
. Andi Yo
g
y
ak
arta. 2013.
[8]
Haralic RM, K
Shanmugam,
ItshakDin
st
ein. Ima
g
e
te
xture
cl
assific
a
tion
. IEEE
T
r
ansacti
ons
o
n
Systems, Man
and Cy
bern
e
tic
s
.
1973; 3(6).
[9]
Kulkar
ni, AD. Artificial Ne
ura
l
Net
w
o
r
k for Image U
nderst
and
ing. Va
n N
o
strand R
e
in
h
o
ld, Ne
w
Y
o
rk.
199
4.
[10]
Hall-B
a
yer, M. HIS Co
-repre
s
entatio
n of cir
c
ular a
nd n
on-
circ
ular v
a
ria
b
l
e
s usin
g harm
onic a
nal
ys
i
s
param
eter.
Ca
nad
ian
Jo
urna
l
of Re
mote S
e
nsin
g
. 20
07;
3
3
(5): 4
16-4
21.
(in this
cas
e
V
o
l.33, Issu
es 4,
and p
a
g
e
41
6-421).
[11]
Ne
w
s
am
S, K
a
mmath
C.
C
o
mpari
n
g
Sh
a
pe
and
T
e
xtur
e F
i
tures
for
Pattern R
e
co
g
n
itio
n i
n
th
e
Simulati
on
Dat
a
. On the IS&
T
/SPIE '
S
Annual S
y
m
pos
iu
m on E
l
ectron
i
c
Imagin
g
. Sa
n Jose,
USA.
200
5.
Evaluation Warning : The document was created with Spire.PDF for Python.