TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.4, April 201
4, pp. 2683 ~ 2
6
8
9
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i4.4060
2683
Re
cei
v
ed Au
gust 6, 201
3; Re
vised O
c
to
ber 18, 20
13;
Accept
ed No
vem
ber 7, 20
13
Comparision of Several Preprocessing Algorithms
Based on Near Infrared Spectroscopic Measurement of
Glucose in Aqueous Glucose Solutions
Yan Zhang*,
Ya
w
e
n Deng
, Jin
w
e
i
Sun,
Chunling Ya
ng, Guoliang
Zhang, Dan
Liu
Schoo
l of Elect
r
ical En
gin
eeri
ng an
d Auto
ma
ton, Harbi
n
Instit
ute of T
e
chnolog
y
92, W
e
st Dazhi
Street, NanGang District, Har
b
in 1
5
0
001, Ch
ina
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: z
y
hit@
hit.ed
u
.cn
A
b
st
r
a
ct
Glucose
conc
entratio
n
me
a
s
ure
m
e
n
t is t
he
basis
of n
oni
nvasiv
e det
ection
of blo
o
d
g
l
ucos
e
conce
n
tratio
n. It is significant
in scie
n
ti
fic research. In this study, Near Infrared
Sp
ectrosc
o
p
y
(
NIRS
)
a
nd
regressi
on
ana
lysis metho
d
o
l
ogy w
e
re co
mbin
ed to
me
as
ure the g
l
uc
os
e conce
n
trat
io
n. T
he spectru
m
of
gluc
ose s
o
luti
o
n
s w
a
s obta
i
n
ed w
i
th the
F
o
urier T
r
a
n
sfor
me
d Infrare
d
Spectro
m
eter,
and th
en t
he
data
w
a
s used for
r
egress
i
on
an
al
ysis. In ad
ditio
n
, the
meth
od
of Partia
l L
e
a
s
t Squar
es
(
PL
S
)
wa
s u
s
ed to
achi
eve pr
inci
p
l
e co
mp
on
ents
and var
i
o
u
s spectral
prepr
oc
essin
g
metho
d
s
w
e
re discuss
ed. Duri
ng P
L
S
mo
de
lin
g, the
Savit
z
ky-Go
lay
coul
d i
m
pr
ove
the
Pre
d
i
c
ti
on
R
e
si
du
al
Erro
r Su
m
o
f
Sq
ua
res
(
PRESS
)
wi
th
in
6%. T
h
e
exp
e
r
iment res
u
lts
de
mo
nstr
ate t
hat NIRS
has
t
he p
o
tenti
a
l f
o
r the
m
eas
ur
ement of
gl
uc
ose
soluti
on.
Ke
y
w
ords
: n
e
a
r-infrare
d s
p
e
c
trum, p
a
rtial
l
east sq
uares,
s
pectral
pre
p
ro
cessin
g
, pre
d
ic
ted resi
du
al
er
ror
sum of sq
uare
s
Copy
right
©
2014 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
Glycuresi
s
is one kind of
global
illne
sses an
d it seri
ously ha
rm
s
th
e
he
althy of human
being
s. Seri
o
u
sly dia
betic
patients m
u
st
measur
e th
eir blo
od gl
u
c
o
s
e
conte
n
ts several tim
e
s
one d
a
y in th
e presently u
s
ed th
erapy. Instrum
ents
now
used fo
r the self-mo
n
i
toring
of blo
o
d
glucose are
almost all i
n
vasive
types that req
u
ire
a drop of
b
l
ood to b
e
withdra
w
n fro
m
a
fingertip o
r
ot
her me
asure
m
ent site on t
he body by
a
needl
e pun
ct
ure. Thi
s
re
q
u
ire
s
the dia
b
e
tic
patient to suffer pain
an
d
also i
n
volves a ri
sk
of
infection. M
o
re
freque
nt or
continuo
us
blo
o
d
glucose mo
ni
toring
is ne
ce
ssary fo
r
disti
n
ct bl
ood
glu
c
o
s
e co
ntrol, whi
c
h wo
uld more
effectiv
ely
reduce the ri
sk of compli
cations
from diabetes m
e
llitus. For thi
s
purpose, a n
oni
nvasive method
for blood gl
ucose mo
nitori
n
g
is highly de
sire
d.
Nea
r
inf
r
ared
sp
ectrosco
p
y
(NIRS
)
h
a
s
bee
n kno
w
n
to have th
e
potential to
reali
z
e
noninva
s
ive
blood
glu
c
o
s
e mo
nitorin
g
, and
the
r
e
h
a
ve be
en
ma
ny trial
s
fo
r
monitori
ng
bl
ood
glucose cont
ents u
s
ing
NIRS over
these years. In the stoich
iom
e
tri
c
analysi
s
field, the
resea
r
chers
have take
n up the relate
d resea
r
ch
. Gary W.Sma
ll team from America [1] and
Kasem
s
um
ra
n S team from Japa
n [2] and other
re
se
arche
r
s
strive
to be the first one to dig de
ep
in
th
e
appli
c
at
ion area of n
ear-infra
re
d
spectrum. The
Chin
ese re
s
earche
r
s like
Xu Ke-Xin
in
the
Tianjin
University [3],
Hua
ng L
a
n
in Sh
angh
ai [4]
al
so
have
som
e
relate
studi
es i
n
develop
ing
instru
ment
s b
a
se
d on
nea
r infrared
spe
c
troscopy.
Ho
wever, th
ere
are
som
e
p
r
o
b
lems whi
c
h
are
difficult to de
sovle
and thi
s
imp
ede
the
devel
op
men
t
of non
-inva
s
ive me
asurement of
blo
o
d
glucose.
In the p
r
e
s
e
n
t study, n
e
a
r inf
r
ared
spectra
of different glu
c
o
s
e
con
c
entrations
were
colle
cted
with
a FT
-IR
sp
ectromete
r
. The
n
seve
ral
spe
c
tral
preproce
ssi
ng m
e
thod
s a
nd th
e PL
S
algorith
m
were used to
an
alyse the
dat
a. Finally
, the
experim
ental
re
sults
we
re
evaluated
wi
th
the Predi
ction
Resi
dual Error Sum of Square
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 4, April 2014: 2683 – 2
689
2684
2. Theor
y
2.1. Absorp
ti
on Char
ac
te
ri
stics of
Ch
emical Bond
A molecul
a
r bond vibrat
ion ab
sorsb
near infrare
d
light and it is the principl
e of
appli
c
ation
o
f
NIRS. In t
h
is
study, t
he mo
st
n
o
table
point i
s
a
freq
uen
cy-do
uble
d
a
nd
combi
nation
-
t
one bo
nd vib
r
ation of met
h
yl in gl
uco
s
e mole
cule
s.
As a re
sult, the absorpti
o
n
pea
k of wate
r sho
u
ld be
avoided. Thi
s
abso
r
pti
on b
and is initiall
y selected o
v
er the 4200
to
4800 cm
-1
spectral ran
g
e
with sam
p
les mai
n
tai
ned at roo
m
temperatu
r
e. The different
preprocessing algo
rithms were eval
uated by judging
the ability to determine glucose
con
c
e
n
tration
s
from a set of predictio
n spectra.
2.2. Lambert-Be
er La
w
The continu
ous wave spectrosco
py
wa
s
u
s
ed
to reali
z
e
glucose con
c
entratio
n
measurement
, Lamb
e
rt-Be
e
r
Law is
th
e ma
cro b
a
se of the
ab
orption p
r
o
c
e
s
s. The
Lamb
e
r
t-
Beer La
w [5] coul
d be defi
ned a
s
:
1
()
n
ii
i
Ac
L
(1)
Whe
r
e
A
is
the vector
o
f
abso
r
ba
nce
,
n
is the n
u
mbe
r
of variou
s solute
s being
observed,
α
is
th
e
c
oeffic
i
ent related to the
s
p
ecific
wa
velength
λ
an
d
L
is the opti
c
al di
stan
ce.
Acco
rdi
ng to
the La
mbe
r
t-Beer
La
w, gl
uc
os
e co
nc
en
tr
a
t
io
n ca
n
b
e
ac
qu
ir
ed
o
n
c
e
th
e
spe
c
tru
m
is
obtaine
d. The absorb
a
n
c
e of near in
f
r
ared spe
c
tro
s
copy is po
si
tively related to
glucose co
ncentration, whi
c
h is
the b
a
si
s of the mea
s
urem
ent of gluco
s
e
con
c
e
n
tration.
3. Experiments
3.1. Instrum
e
nts
Nea
r-i
nfra
red
sp
ect
r
o
s
c
opi
c d
a
ta
we
re
m
easured wit
h
a
JAS
C
O FT-IR spe
c
trometer.
Sample
s we
re co
ntained i
n
1mm qu
art
z
col
o
rim
e
tric
dish.The
ele
c
troni
c b
a
lan
c
e Sa
rtoriu
s
BS
224S wa
s u
s
ed to quantify the gluco
s
e.
3.2. Reage
n
ts
For the NI
R data, reage
nt-grad
e
crystall
ine
glucose was di
ssolved in deioni
zed
water to
configure 12 sampl
e
s of
di
fferent
glucose concentrati
on. Electroni
c balance has
the range ability
of 220
g a
n
d
the me
asure
m
ent a
c
cura
cy is ±0.1mg
. A 500
mL v
o
lumetri
c
fla
s
k
wa
s
used f
o
r
mother liqu
o
r
and the 1
00mL on
es
for sam
p
le
s.
Durin
g
dilut
i
on, measuri
ng cylinde
r
with
accuracy of ±0.2mL wa
s u
s
ed. Ta
ble 1 sho
w
s the co
nce
n
tration of
sample
s.
T
able 1. Numb
er of Samples
and C
o
rresp
on
din
g
Conc
entra
tion
n
u
m
b
e
r
o
f
samp
l
es
1 2 3 4 5
6 7
8
9
10
11
12
co
n
cen
tr
atio
n
()
mg/dL
100
400
420
550
350
250
270
30
450
150
80
300
3.3. Procedu
r
es
Firstly, spectrum
of the em
pty
infrasil glass cell
s
and ce
lls filled with
sam
p
les
were
colle
cted
se
perately. Se
con
d
ly, spe
c
trum data
were imp
o
rted
to the com
puter an
d were
cal
c
ulate
d
wit
h
the Matlab
softwa
r
e. Th
e
band b
e
twe
e
n
4200
~4
800
cm
-1
we
re
sel
e
cted a
nd u
s
ed
to gene
rate the targ
et array. Finally, gluco
s
e a
b
sorption ability can b
e
cal
c
u
l
ated by usin
g
averag
e valu
e from
hom
o
l
ogou
s
data.
For exampl
e, K
i
is th
e
averag
e
spe
c
trum
of the
n
th
sampl
e
,
K
0i
is the average
spe
c
tru
m
of the
n
th empty glass cell, then the corre
s
p
ondin
g
gluco
s
e
absorptio
n ca
n be define
d
as:
0
ln
/
ii
i
A
KK
(2)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Com
pari
s
ion
of Several Prepro
c
e
s
sing
Algorit
hm
s Base
d on Near Infrared
… (Y
an Zhan
g)
2685
4. Spectral Data Prepr
o
ce
ssiing
Nea
r-i
nfra
red
spect
r
um reflect the informat
ion abo
ut the chemical
compo
s
ition
and the
con
c
e
n
tration
of substan
c
es. It also ca
n be affe
cted
by material visco
sity, particle den
sity and
stray light, etc. Therefore, the
eliminatio
n of these factors
coul
d im
prove the pe
rforman
c
e of the
measurement
. The original
spe
c
trum
col
l
ected
wi
th the JASCO FT
-IR spe
c
tro
m
e
t
er is sh
own in
Figure 1. Several p
r
ep
ro
ce
ssi
ng metho
d
o
logie
s
were i
n
trodu
ce
d to analyse the spectrum.
Figure 1. Orig
inal Spect
r
um
4.1. Standar
d
ization
Con
s
id
erin
g
the obviou
s
spe
c
tru
m
differen
c
e
amo
n
g
the differe
nt wavele
ngt
hs, the
nonlin
earity o
f
the dete
c
tor
coul
d
cau
s
e
different m
e
a
s
ureme
n
t error. Stand
ardi
zation
is
used
to
degrade its
effects on th
e model. Autoscalin
g is
o
ne kin
d
of standa
rdi
z
ation
.
Centerin
g a
nd
norm
a
lization
are two
step
s of this process.
Duri
ng
ce
nte
r
ing, the
ab
sorba
n
ce
spe
c
trum
data
a
t
the same
wavele
ngth
p
o
ints
but
from different
sample
s sub
t
ract the average value, in
dicate
d as ex
pre
ssi
on (3):
XX
(3)
And the norm
a
lizatio
n re
sul
t
X
z
can be e
x
presse
d as:
1
1
1
1
zi
m
ij
j
XX
x
m
(4)
i
X
is the d
a
ta on
the
i
th wavel
ength,
σ
i
s
th
e stan
dard d
e
viation of th
e
i
th wavelength,
m
is
the numbe
r o
f
sample
s,
ij
x
is the absorptio
n of the
j
th sa
mple at the
i
th wavele
ngth
points. After
the stan
da
rdi
z
ed
processi
ng, the
spe
c
t
r
um
can
be
acq
u
ire
d
. Fig
u
re
2 sho
w
s
the sp
ect
r
a a
fter
stand
ardi
zati
on .
Figure 2. Spectra after Sta
ndardization
4.2. Sa
v
i
tzk
y
-Gola
y
The
appli
c
ati
on of
Savitzky-Golay i
s
ba
sed
on
the
a
s
sumption
th
at the
noi
se
containe
d
in the spe
c
trum is
white
noise, whi
c
h
can
be
d
e
g
r
ade
d by cal
c
ulatin
g the
spe
c
tral
data
of
4
100
4200
4300
440
0
4500
46
00
4700
4800
4900
1.
5
2
2.
5
3
cm
-
1
Ab
so
r
p
t
i
o
n
4100
4200
4300
44
00
4500
4600
4700
4800
4900
-2
-1
0
1
2
cm
-
1
Ab
s
o
r
p
ti
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 4, April 2014: 2683 – 2
689
2686
adja
c
ent wavelength p
o
int
s
. Du
ring thi
s
pro
c
e
ssi
ng, the data in th
e mobile
win
dow i
s
sm
oot
hed
by different p
o
lynomial se
quen
ce
s. Ho
wever, the
wi
dth of the mo
ving wind
ow
must be
ca
re
fully
cho
o
sed, oth
e
rwi
s
e
so
me
useful i
n
form
ation c
ould
b
e
lost a
nd th
e pro
c
e
s
sing
method
can
not
obtain the ide
a
l re
sults. Fig
u
re 3
sho
w
s the sp
ect
r
um
pro
c
e
s
sed u
s
ing win
d
o
w
width of 10 an
d
100 wavelen
g
th points in t
u
rn.
Figure 3. a) Spectrum processed u
s
in
g
Savitzky-G
ol
ay with Wind
ow of 10
Wavele
ngth p
o
ints.
Figure 3. b) Spectrum processed u
s
in
g
Savitzky-G
ol
ay with Wind
ow of 100
Wavele
ngth p
o
ints
4.3. Direct O
r
thogo
nal Signal Corr
ection (DOS
C)
The above
spectral pre
p
roce
ssing me
thods
carry out the data processin
g
without
c
o
mputing dens
i
ty matrix, only relating to s
p
ec
tr
a dat
a. Duri
ng
DO
SC metho
d
, the spe
c
tral a
r
ray
vary co
rre
sp
o
ndingly an
d is orthog
onal to
the con
c
e
n
tration [6]. Afte
r the multivari
a
te calib
ratio
n
,
the model p
r
e
s
ent mo
re ro
bust and p
r
e
d
iction a
b
ility [7].
Figure 4. Spectra Proces
se
d by DOSC
Method
If Y prese
n
t the con
c
entration an
d X prese
n
t the sp
ectral
array
,
DOSC can b
e
divided
into the following four ste
p
s.
(1) Projec
t Y on X,
T1
T
ˆ
((
)
)
X
PY
X
X
Y
Y
(2)
1
ˆ
ˆ
((
)
)
Z
XY
Y
X
, the step ensure that Z
is orthog
onal
to Y and
ˆ
Y
.
(3)
Z is proce
s
sed by P
r
in
cipal Compo
n
ent Analysi
s
(PCA) a
nd the
score matrix
t
can
be
acq
u
ire
d
. Th
e weig
ht vector
1
wX
t
, then we ca
n cal
c
ul
ate score ve
ctor
s
t
X
w
,and the
loadin
g
vecto
r
T
T
s
s
s
s
Xt
t
p
t
.
(4)
T
DO
S
C
s
s
XX
t
p
In the above
step
s, there
are two mai
n
co
mp
one
nts were u
s
ed
d
u
ring P
C
A. After the
DOSC, mo
st
information
about the
sa
mple charac
t
e
risti
cs i
s
lo
st and the sp
ectru
m
abo
ut the
con
c
e
n
tration
s
ca
n be arra
nged. The
sp
ectra p
r
o
c
e
ssed by DOSC
method is
sh
own in Fig
u
re
4.
41
00
42
00
43
00
44
00
45
00
46
00
47
00
48
00
2
2.
5
3
cm
-
1
Ab
s
o
r
p
t
i
o
n
a)
W
=
10
410
0
4
200
43
00
4
400
4
500
46
00
47
00
480
0
49
00
1.
5
2
2.
5
3
cm
-
1
Ab
s
o
r
p
t
i
o
n
b)
W
=
10
0
41
00
4
200
4300
440
0
45
00
4
600
4700
4800
4
900
0
1
2
3
4
cm
-
1
Ab
s
o
r
p
t
i
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Com
pari
s
ion
of Several Prepro
c
e
s
sing
Algorit
hm
s Base
d on Near Infrared
… (Y
an Zhan
g)
2687
5. NIRS and
PLS
5.1. PLS Theor
y
PLS is a method combini
n
g factor a
nalysis
a
nd re
gre
ssi
on an
alysi
s
. It has two steps.
Step 1: Facto
r
analysi
s
. De
comp
ose X and Y.
T
X
VP
E
(5)
T
YU
Q
F
(6)
W
h
er
e
s
u
p
e
rs
cr
ip
t T me
ans
tr
an
sp
os
e
d
ma
tr
ix,
V
is th
e sco
r
e
matri
x
of
X
,
U
i
s
th
e sco
r
e
matri
x
of
Y
,
P
is loading m
a
trix of
X
,
Q
is lo
ading m
a
trix of
X
,
E
and
F
are the error matrix. It is
importa
nt to
note that
T
i
s
ortho
gon
al to
P
. And
i
t
refl
ects the info
rmation of
sp
e
c
trum
matrix
X
whe
n
it wa
s
c
onv
eyed by v
e
ctors
i
p
.The remainin
g info
rmation i
s
co
nsid
ere
d
to b
e
incl
ude
d i
n
the error mat
r
ix [8-9]. In a s
i
milar way,
Y
is
de
comp
osed.
Step 2: Regression a
nalysis.
B
is the co
rrel
a
tion coefficient matrix.
UV
B
(7)
T1
T
()
BV
V
V
U
(8)
The predi
cted
value of
unknown co
ncen
tration
Y
p
can
be defined a
s
:
PP
YT
B
Q
(9)
Whe
r
e
P
T
is got from the sp
ectrum of unkno
wn sample
s a
nd the loadi
n
g
matrix
P
.
5.2. Dete
rmination th
e Number of PL
S Compone
nts
Duri
ng the
PLS modeli
ng,
the num
ber of com
pon
e
n
ts is
an im
portant
elem
ent. At
pre
s
ent, the
most commo
n method to
determi
ne th
e numb
e
r of
PLS comp
on
ents i
s
Predi
ction
Re
sidu
al Error Su
m of S
quares (P
RE
SS). The
cro
s
s validatio
n
method
i
s
u
s
ed
to a
naly
s
e
PRESS [10].
Negl
ectin
g
th
e
i
th
wavele
n
g
th poi
nts every time,
buil
d
PLS m
odel
with
h
com
p
onent
s
by using th
e
rest d
a
ta. Th
en plug th
e
i
th wavele
ngth
points into
regre
s
sion
eq
uation an
d g
e
t
()
()
ˆ
ij
xh
. The forecast
ing error
squ
a
re sum of
x
i
can b
e
define
d
as:
2
()
1
()
ˆ
()
ji
j
n
i
ij
PRESS
h
x
x
h
(10)
Whe
r
e
j
=
1
, 2, ……,
p
.
The fore
ca
sti
ng error
squ
a
r
e su
m of X= (
x
1
,
x
2
,,
……
x
p
)
T
ca
n be
defined a
s
:
1
()
()
p
j
j
P
RES
S
h
P
RES
S
h
(11)
At the sam
e
time, we
build
the PLS mo
del with
h
co
mpone
nts
by usin
g all d
a
ta
.
,
pi
j
x
is
the redi
cted
value of
the
i
th
wavel
engt
h
poi
nt.
The
forecastin
g e
rro
r sq
uare sum
of
x
i
ca
n
be
defined a
s
:
2
1
,
()
(
)
jp
i
j
i
j
n
i
SS
h
x
x
(12)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 4, April 2014: 2683 – 2
689
2688
The fore
ca
sti
ng error
squ
a
r
e su
m of
x
p
can b
e
define
d
as:
1
()
()
j
p
j
SS
h
S
S
h
(13)
When the
minimum PRESS(
h
) i
s
a
c
hieve
d
, the
app
rop
r
iate
num
ber of
the PLS
comp
one
nts
can b
e
determined. It is defined a
s
:
2
1(
)
/
(
)
h
Q
P
RESS
h
SS
h
(14)
If the condit
i
on that
2
0.0975
h
Q
ca
n be achieved
by using h
comp
onent
s, the
comp
uting p
r
oce
s
s stop
ed
.
6. Results
This pap
er di
scusse
d
sev
e
ral
metho
d
s for
sp
ectral
prep
ro
ce
ssin
g. Tabl
e 2
sh
ows the
experim
ent result
s. From t
he table
s
, it can be o
b
taine
d
that Savitzky-Golay dimi
nish
ed the
ro
ot
mean squa
re
resi
dual a
nd
maximum rel
a
tive erro
r.
Table 2. Para
meters of Th
e First Pre
d
iction Set
Processin
g
Met
hod
Num
b
er o
f
Co
m
pone
nts
r
Maximu
m Rel
a
ti
v
e
Error
RMSE
(mg/
dL
)
not
hin
g
4 0.0539
11.85875
A
u
t
o
scali
n
g
5 0.0671
12.64839
DOSC
3 0.0725
17.77251
S-G
4 0.0524
11.59880
In this study, The DOS
C
and auto
s
cali
ng
method
s
did not exhibi
t good perfo
rmance.
This may
be
denp
ende
nt on
the ch
ara
c
teri
stics
of
t
he dat
a. Savitzky-Golay
showed the
b
e
st
result in our
experim
ents
whe
n
the wid
t
h of
moving wind
ow was
15 wavele
ngt
h points an
d the
numbe
r of PLS comp
one
nts wa
s 4. T
he re
sult
s pr
ese
n
ted in th
e table dem
o
n
strate
d that the
maximum rel
a
tive erro
r is l
e
ss than 8% and the maxi
mum RMSE i
s
less than 1
8
mg/dL.
7. Conclusio
n
In this
study,
several p
r
ep
roce
ssing
met
hod
s an
d PL
S were
comb
ined to m
e
a
s
ure th
e
glucose
con
c
entration.
We
discu
s
sed
several
pr
eproce
s
sing
me
thods in
adv
ance. By u
s
i
n
g
autoscali
ng,
DOSC, Savit
zky
-Golay, th
e
PRESS
is
calcul
ated. Th
e maximum
relative erro
r
wa
s
confin
ed to 8
%
. The experiment re
sult
s co
uld
dem
onstrate that the appi
l
c
atio
n of NIRS a
n
d
PLS has the
potential for the mea
s
u
r
em
ent and an
alysis of glu
c
o
s
e
solution.
Ackn
o
w
l
e
dg
ements
The a
u
tho
r
s
are
grateful f
o
r the
supp
ort from the
National S
c
ien
c
e
Foun
dation
of Chi
n
a
(No.
6120
10
17, 613
780
4
6
), China P
o
stdo
ctor
al Scien
c
e
F
o
u
ndation (No.
2013
M531
0
27),
Heilo
ngjian
g
Postdo
ctoral Fund (No.
L
B
H-Z1
209
3),
the Fund
ame
n
tal Re
se
arch Fund
s fo
r the
Central Unive
r
sitie
s
(No. HI
T.NS
RIF.20
1
3010, No. HIT.NSRIF.20
1
146).
Referen
ces
[1]
Mark AA, Gary
WS
.
Determ
inati
on
of Ph
ys
iolo
gic
a
l
Leve
l
s of Glucos
e i
n
al
l Aq
ue
ous
Matrix
w
i
t
h
Digita
l
l
y
Filter
e
d
Fourier T
r
ansform Near-Infrared Sp
ectra.
Analytic
al Ch
e
m
istry.
199
0; 6
2
(14): 14
57
–
146
4
.
[2]
Bai G.
T
he Measurem
ent an
d
Anal
ysis of Gl
ucose
C
once
n
t
r
ation C
han
ges
in Bloo
d
w
i
th
Near-i
nfrare
d
Spectrosco
p
y
.
W
uhan: HUST
.
2008.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Com
pari
s
ion
of Several Prepro
c
e
s
sing
Algorit
hm
s Base
d on Near Infrared
… (Y
an Zhan
g)
2689
[3]
Li QB, W
a
n
g
Q, Xu
K
X
, W
a
ng B.
Ne
ar-Infrared
Sp
ectros
copic
Assa
y
of
Princ
i
pa
l Mi
lk
Constitu
ents.
F
ood Scie
nce.
200
2, 23(6): 12
1–1
24.
[4]
Hua
ng L, D
i
ng
HS
,
W
ang GZ
.
T
he Prelimi
nar
y Stud
y
on
Noni
nv
as
ive
D
e
tection Usin
g NIR
Diffusio
n
Reflecta
nce S
pectrum for Monitor
i
ng Bl
oo
d Glucose.
Sh
ang
hai Institut
e of Metall
urg
y
, Acad
em
y
o
f
Scienc
es of Ch
ina.
Materials
Physics and Chem
istry.
2
0
00.
[5]
Yuan
HF
, Lu
W
Z
. Near i
n
frared s
pectr
al a
nal
ys
is te
chno
log
y
is r
api
dl
y marc
hi
ng i
n
to th
e
petroch
emica
l
field.
Oil R
e
fini
n
g
and C
h
e
m
ic
a
l
Industry
. 199
8; 29(9): 47
–50
.
[6]
Joha
n AW
, S deJo
ng, Smil
d
e
AK.
Direct o
r
thogo
nal s
i
g
n
a
l correcti
on.
Che
m
om. Inte
l
l
. Lab. Syst
.
200
1; 56: 13–
2
5
.
[7]
Liu
GJ
,
Dai D
M
,
Gao HT
.
T
he al
gorit
hms of orthog
ona
l sign
al co
rrecti
o
n
and its ap
plic
a
t
ion in sp
ectr
a
process
i
ng
.
J. Shan
do
ng Un
i
v
. Arehit. Eng.,
2005; 2
0
(2): 8
5
.
[8]
Gao J, Z
e
n
g
XP, Z
h
ang
X, Ch
en
X, Z
h
eng
DT
. Cavity V
e
rte
x
Reg
ener
ation
thro
ugh
Opti
m
a
l
Energ
y
M
ode
l
for Restorati
o
n of W
o
rn P
a
rts.
T
E
LKOMNIKA Indon
esia
n Jour
na
l
of Electrica
l
Engi
neer
in
g.
2013; 11(
5): 249
0–2
50
1.
[9]
Z
hang
XS, W
ang M
H
, Ma
J. Sparse R
e
prese
n
tati
o
n
for Detecti
on
ofMicrocalc
i
fic
a
tion
Clusters.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
ng.
2013, 1
0
(3): 5
45-5
50.
[10]
Rajkum
ar P, W
ang N, EIma
sr
y
G, Rag
hav
an G
SV, Garie
p
y
Y. Stud
ies
on b
ana
na frui
t qualit
y a
n
d
maturit
y
sta
ges
using h
y
p
e
rs
p
e
ctral ima
g
in
g.
Journ
a
l of F
o
o
d
Engi
ne
erin
g
. 201
2; 108(
1): 194–
20
0.
Evaluation Warning : The document was created with Spire.PDF for Python.