T
E
L
KO
M
NI
K
A
,
V
ol
.
14,
N
o.
3,
S
ept
em
ber
20
16,
pp.
95
6~
9
62
I
S
S
N
:
1
693
-
6
930
,
ac
c
r
edi
t
ed
A
b
y
D
IK
T
I,
D
e
c
r
e
e
N
o
:
58/
D
I
K
T
I
/
K
ep/
2013
D
O
I
:
10.
12928/
T
E
LK
O
M
N
I
K
A
.
v
1
4
i
3
.
3712
95
6
R
ec
ei
v
ed
M
ar
c
h
5
,
2
01
6
;
R
ev
i
s
ed
M
ay
20
,
20
1
6
;
A
c
c
e
pt
ed
J
u
ne 6
,
2
01
6
R
e
c
o
gni
ti
on of
O
d
or
C
h
ar
act
er
i
st
ics
b
ased
o
n
BP
N
eu
r
al
N
e
tw
o
r
k
W
u
L
e
i
*
1
,
F
a
n
g
J
i
a
n
d
o
n
g
2
,
Z
h
a
o
Y
u
d
o
n
g
3
1
,2
C
ol
l
ege
of
In
fo
r
m
a
ti
o
n
E
ng
i
n
eer
i
ng
,
I
n
ner
M
ongol
i
a
U
n
i
ve
r
si
t
y
of
T
ec
hn
ol
ogy
,
H
o
hhot
010
080
, P
.R
.C
h
i
n
a
3
I
nner
M
ongol
i
a
A
gr
i
c
ul
t
ur
e
a
nd
A
ni
m
al
H
u
s
b
andr
y
I
nf
or
m
a
t
i
on
C
ent
er
,
H
oh
hot
0100
11
,
P
.
R.
Ch
i
n
a
*
C
or
r
es
po
ndi
ng a
ut
hor
,
e
-
m
ai
l
:
w
l
m
ango@
16
3.
c
o
m
1
,
z
haoy
v
dong@
16
3.
c
om
3
A
b
st
r
act
T
hi
s
pap
er
i
nt
r
od
uc
e
s
t
he b
as
i
c
pr
i
nc
i
pl
e
and
c
a
l
c
ul
at
i
on
s
t
e
ps
of
B
P
ne
ur
al
net
w
or
k
a
l
gor
i
t
hm
f
o
r
r
ec
og
ni
t
i
on
of
o
dor
c
har
a
c
t
er
i
s
t
i
c
s
.
U
s
i
ng t
h
e P
E
N
3 el
e
c
t
r
o
ni
c
n
os
e
c
ol
l
ec
t
s
t
h
e v
o
l
at
i
l
e
c
om
pone
nt
s
o
f
m
i
l
k
.
S
i
m
ul
at
i
o
n of
B
P
ne
ur
al
net
w
or
k
al
g
or
i
t
hm
i
n M
A
T
LA
B
c
ond
i
t
i
o
n,
an
d c
om
p
ar
ed w
i
t
h
t
he par
t
i
al
l
ea
s
t
s
quar
es
i
dent
i
f
i
c
at
i
on a
l
gor
i
t
h
m
w
hi
c
h i
s
i
n t
h
e s
o
f
t
w
ar
e o
f
el
ec
t
r
on
i
c
n
os
e
.
I
t
i
s
v
er
i
f
i
e
d t
hat
t
h
e od
o
r
c
har
a
c
t
er
i
s
t
i
c
par
am
et
er
s
w
i
t
h no
nl
i
near
and
l
i
t
t
l
e
di
f
f
er
e
n
c
es
c
o
ul
d
be
be
t
t
er
i
de
nt
i
f
i
e
d by
B
P
neur
a
l
net
w
or
k
.
Ke
y
w
o
rd
s
:
B
P
n
eur
al
net
w
or
k
,
e
l
ec
t
r
o
ni
c
n
os
e
,
o
dor
,
r
ec
o
gn
i
t
i
on
.
C
o
p
y
r
i
g
h
t
©
20
16 U
n
i
ver
si
t
a
s A
h
mad
D
ah
l
an
.
A
l
l
r
i
g
h
t
s r
eser
ved
.
1
.
I
n
tr
o
d
u
c
ti
o
n
T
he s
ens
e of
s
m
el
l
i
s
an
ef
f
ec
t
i
v
e
w
a
y
t
o u
nder
s
t
a
n
d t
he
i
nf
or
m
at
i
on
of
t
he
e
x
t
er
nal
s
m
el
l
.
D
ue t
o t
he i
nf
l
uenc
e b
y
t
he od
or
c
om
ponent
c
o
m
pl
ex
i
t
y
and s
ubj
ec
t
i
v
i
t
y
of
t
he s
ens
e
or
gans
,
t
h
e s
m
el
l
s
t
i
l
l
c
o
n
'
t
i
s
ex
pr
es
s
ed b
y
s
pec
i
f
i
c
t
er
m
s
.
I
n r
ec
ent
y
e
ar
s
,
w
i
t
h t
he r
ap
i
d
dev
el
opm
ent
of
s
ens
or
t
ec
hno
l
og
y
a
nd c
om
put
er
t
ec
hno
l
og
y
,
t
he r
es
e
ar
c
h of
el
ec
t
r
on
i
c
nos
e
s
y
s
t
em
has
been pai
d m
or
e at
t
e
nt
i
on b
y
s
c
i
ent
i
s
t
s
and de
v
e
l
o
ped r
ap
i
d
l
y
.
E
l
ec
t
r
on
i
c
nos
e
s
y
s
t
em
c
ons
i
s
t
s
of
t
w
o p
ar
t
s
,
t
he s
e
m
i
c
onduc
t
or
g
as
s
ens
or
ar
r
a
y
an
d pat
t
er
n r
ec
ogni
t
i
o
n
s
y
s
t
em
.
P
at
t
er
n
r
ec
ogni
t
i
o
n i
s
one of
t
he
k
e
y
r
es
ear
c
h
s
of
el
ec
t
r
oni
c
nos
e
s
y
s
t
em
.
I
t
pl
a
y
s
an
i
m
por
t
ant
r
ol
e i
n t
h
e r
ea
l
i
z
a
t
i
on
of
t
he f
unc
t
i
on
of
el
ec
t
r
oni
c
n
os
e s
y
s
t
em
[1
-
3]
.
I
n t
he pr
oc
es
s
of
ac
t
ual
pr
oduc
t
i
on an
d l
i
f
e,
t
he e
l
ec
t
r
oni
c
nos
e c
an'
t
c
om
pl
et
el
y
t
ak
e
t
he p
l
ac
e of
t
he bi
ol
og
i
c
al
nos
e
[
4]
.
O
n
e of
t
he r
eas
o
ns
i
s
t
hat
m
os
t
o
f
t
he ex
i
s
t
i
ng e
l
ec
t
r
on
i
c
nos
e p
at
t
er
n r
ec
ogn
i
t
i
on
al
gor
i
t
hm
s
ar
e bas
ed
on s
t
at
i
s
t
i
c
al
m
odel
s
.
T
hi
s
t
ec
hn
i
q
ue i
s
bas
ed
on
t
he s
t
at
i
s
t
i
c
al
d
i
s
t
r
i
bu
t
i
o
n o
f
t
he obs
er
v
ed f
eat
ur
es
w
i
t
h s
i
gni
f
i
c
an
t
di
f
f
er
enc
es
[5
],
and m
os
t
o
f
t
he d
at
a
pr
oc
es
s
i
n
g i
s
l
i
nea
r
s
t
r
uc
t
ur
e,
s
uc
h as
par
t
i
al
l
eas
t
s
qu
ar
es
an
al
y
s
i
s
.
T
he
r
ef
or
e,
f
or
t
he
odor
da
t
a
w
i
t
h l
i
t
t
l
e d
i
f
f
e
r
enc
es
and n
onl
i
n
ear
s
t
r
uc
t
ur
e i
t
i
s
di
f
f
i
c
ul
t
t
o m
ak
e ac
c
ur
at
e
i
de
nt
i
f
i
c
at
i
on.
A
r
t
i
f
i
c
i
a
l
neur
a
l
net
w
or
k
i
s
one
of
t
he
t
y
p
i
c
al
m
et
hods
of
nonl
i
near
dat
a
pr
oc
es
s
i
n
g.
T
he
m
et
hod i
s
m
ai
nl
y
a k
i
nd o
f
al
gor
i
t
hm
t
o i
m
i
t
at
e t
he
hum
an br
ai
n
t
h
i
nk
i
ng e
ngi
neer
i
ng.
I
t
c
an
s
ol
v
e
t
h
e
r
ec
og
ni
t
i
o
n
pr
o
bl
em
t
hat
t
he
di
s
t
r
i
but
i
on
of
t
he
s
am
pl
e
i
s
unk
no
w
i
n
g
[6
-
8]
.
T
hi
s
pap
er
des
i
g
ns
a
m
odel
bas
ed
o
n
B
P
ne
ur
al
n
et
w
or
k
al
gor
i
t
h
m
t
o
i
dent
i
f
y
od
or
c
har
ac
t
e
r
i
s
t
i
c
s
.
A
c
as
e
s
t
ud
y
o
n t
he
odor
c
har
ac
t
er
i
s
t
i
c
s
of
m
i
l
k
w
i
t
h di
f
f
er
ent
qua
l
i
t
y
,
t
hr
ou
gh c
om
par
ed
w
i
t
h par
t
i
a
l
l
eas
t
s
quar
es
a
l
gor
i
t
hm
of
el
ec
t
r
on
i
c
nos
e s
of
t
w
ar
e,
i
t
i
s
v
er
i
f
i
ed t
hat
t
h
e odo
r
c
har
ac
t
er
i
s
t
i
c
par
am
et
er
s
w
i
t
h
n
on
l
i
n
ear
and l
i
t
t
l
e
di
f
f
er
enc
es
c
oul
d be
b
et
t
er
i
de
nt
i
f
i
e
d
b
y
B
P
n
eur
al
net
w
or
k
.
2.
BP
N
eu
r
al
N
e
tw
o
r
k
2.
1
.
D
e
fi
n
i
ti
o
n
a
n
d
S
t
ru
c
t
u
re
Mul
t
i
l
a
y
er
f
eed
f
or
w
ar
d
ar
t
i
f
i
c
i
al
ne
ur
al
ne
t
w
or
k
bas
ed
o
n
er
r
or
bac
k
pr
opagat
i
on
al
g
or
i
t
hm
i
s
c
al
l
ed
B
P
n
eu
r
al
n
et
w
or
k
or
B
P
n
eur
a
l
n
et
w
or
k
m
odel
.
B
P
ne
t
w
or
k
c
an
l
ear
n
an
d
s
t
or
e a l
ar
ge
num
ber
of
i
n
put
-
out
put
m
ode m
appi
n
g
r
el
at
i
ons
h
i
p,
an
d no
ne
ed
t
o r
ev
ea
l
a
nd
des
c
r
i
be
t
he
m
at
hem
at
i
c
al
equa
t
i
o
ns
of
t
he
m
appi
ng
r
el
at
i
o
ns
hi
p
.
I
t
s
l
ear
n
i
ng
r
ul
e
i
s
t
o
us
e
t
he
s
t
eepes
t
des
c
ent
m
et
hod,
b
y
bac
k
pr
opag
at
i
on
t
o c
o
ns
t
ant
l
y
adj
us
t
t
he
ne
t
w
or
k
w
ei
g
ht
s
a
nd
t
hr
es
hol
d,
w
hi
c
h m
ak
es
t
he s
um
o
f
net
w
or
k
'
s
er
r
or
s
quar
e
is
m
i
ni
m
u
m
.
T
he t
opo
l
o
g
y
s
t
r
uc
t
ur
e
of
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
K
A
I
S
S
N
:
1
693
-
6
930
R
ec
ogn
i
t
i
on of
O
dor
C
har
a
c
t
er
i
s
t
i
c
s
bas
ed on B
P
N
eur
a
l
N
et
w
or
k
(
W
u Lei
)
957
BP
ne
ur
al
net
w
or
k
m
odel
i
nc
l
udes
i
n
put
l
a
y
er
,
hi
dde
n
l
a
y
er
a
nd
out
p
ut
l
a
y
e
r
.
T
he
i
nt
er
c
on
nec
t
i
on b
et
w
e
en l
a
y
er
an
d l
a
y
er
,
eac
h
l
a
y
e
r
bet
w
een n
odes
i
s
not
c
onnec
t
ed.
T
he
num
ber
s
of
i
nput
l
a
y
er
n
o
des
us
ua
l
l
y
t
ak
e t
he
di
m
ens
i
ons
of
t
he
i
np
ut
v
ec
t
or
,
t
he n
um
ber
o
f
out
p
ut
l
a
y
er
n
odes
us
ua
l
l
y
t
ak
e t
he di
m
ens
i
ons
of
t
he out
p
ut
v
ec
t
or
,
t
he n
um
ber
of
hi
dden
l
a
y
er
nodes
t
her
e
i
s
no
def
i
n
i
t
e s
t
and
ar
d,
t
hr
oug
h r
ep
eat
e
d t
r
i
al
,
t
he
n ge
t
t
h
e f
i
na
l
r
es
ul
t
.
F
i
gur
e 1
i
s
a
t
hr
ee l
a
y
er
B
P
neur
a
l
n
et
w
or
k
s
t
r
uc
t
ur
e
m
odel
.
F
i
gur
e 1.
T
he t
hr
e
e l
a
y
e
r
B
P
ne
ur
al
net
w
or
k
s
t
r
uc
t
ur
e m
odel
I
n F
i
gur
e
1,
12
n
,,
XX
X
a
r
e
t
he
i
n
put
v
al
ue
s
an
d
12
,,
m
YY
Y
ar
e
t
h
e pr
e
di
c
t
i
v
e
va
l
u
e
s
of
B
P
neur
a
l
net
w
o
r
k
,
ij
ω
and
jk
ω
ar
e B
P
ne
ur
al
n
et
w
or
k
w
ei
g
ht
s
.
A
s
c
an be
s
een
f
r
o
m
F
i
gur
e 1,
t
h
e B
P
ne
ur
al
ne
t
w
or
k
c
an be r
egar
de
d
as
a nonl
i
n
ear
f
unc
t
i
on
,
t
h
e net
w
or
k
i
nput
va
l
u
e
s
an
d t
h
e pr
ed
i
c
t
i
v
e
v
a
l
ue
s
r
es
pec
t
i
v
el
y
r
epr
es
ent
t
h
i
s
f
unc
t
i
on'
s
i
n
de
p
en
dent
v
ar
i
ab
l
es
and
dep
end
ent
v
ar
i
ab
l
es
.
W
h
en t
he i
np
ut
n
odes
ar
e
n
,
t
he
out
put
nod
es
ar
e
m
,
t
he
B
P
neur
al
net
w
or
k
i
s
ex
pr
es
s
ed t
he f
unc
t
i
on m
appi
n
g r
el
a
t
i
o
n
w
h
i
c
h i
s
f
r
o
m
n
i
nde
pen
dent
v
ar
i
ab
l
es
t
o
m
depe
nde
nt
v
ar
i
ab
l
es
[
6]
.
2.
2
. A
l
g
o
r
i
th
m
S
t
ep
s an
d
F
lo
w
C
h
a
rt
B
P
al
gor
i
t
hm
i
s
a k
i
nd of
s
u
per
v
i
s
ed
l
e
ar
ni
ng
al
g
or
i
t
hm
.
T
he m
ai
n i
d
ea
i
s
:
i
np
ut
l
e
ar
ni
n
g
s
a
m
pl
es
,
us
i
n
g t
he b
ac
k
-
pr
opa
gat
i
on a
l
g
or
i
t
hm
of
t
he
net
w
or
k
,
w
ei
ght
s
an
d dev
i
at
i
o
ns
ar
e
r
epeat
ed a
dj
us
t
m
ent
and t
r
ai
ni
ng,
m
ak
e t
he out
put
v
ec
t
or
s
and ex
pec
t
ed v
ec
t
o
r
as
c
l
os
e as
pos
s
i
bl
e.
W
hen t
he er
r
or
s
u
m
of
s
quar
es
of
t
he net
w
or
k
out
put
l
a
y
er
i
s
s
m
al
l
er
t
ha
n t
he
s
pec
i
f
i
ed er
r
or
,
t
r
ai
n
i
n
g i
s
c
om
pl
et
e,
s
a
v
e t
he net
w
or
k
w
ei
g
ht
s
and d
ev
i
at
i
o
ns
.
S
p
ec
i
f
i
c
s
t
eps
ar
e
a
s
f
o
llo
w
s
[
6]
:
T
he f
i
r
s
t
s
t
ep:
W
ei
ght
s
i
n
i
t
i
al
i
z
a
t
i
o
n.
A
c
c
or
d
i
ng
t
o t
h
e
s
y
s
t
em
i
nput
an
d out
put
s
e
quenc
e
(,
)
xy
det
er
m
i
ne t
he n
et
w
or
k
i
np
ut
l
a
y
er
'
s
nod
e num
ber
n
,
t
h
e hi
d
den
l
a
y
er
'
s
nod
e num
ber
l
,
t
he out
put
l
a
y
er
'
s
no
de n
um
ber
m
,
i
ni
t
i
al
i
z
e t
he
i
np
ut
l
a
y
er
,
h
i
d
den l
a
y
er
a
nd
out
p
ut
l
a
y
er
c
onnec
t
i
on
w
e
i
g
ht
s
bet
w
e
en
ne
ur
ons
ij
ω
and
jk
ω
,
and
i
ni
t
i
a
l
i
z
e
h
i
dd
en
l
a
y
e
r
'
s
t
hr
es
hol
d
a
,
out
p
ut
l
a
y
er
'
s
t
hr
es
ho
l
d
b
,
g
i
v
en
t
he
l
e
ar
ni
ng r
at
e a
nd t
h
e ex
c
i
t
at
i
o
n f
unc
t
i
o
n of
neur
ons
.
T
he s
ec
ond s
t
ep:
T
he hi
dd
en l
a
y
er
out
put
c
a
l
c
ul
a
t
i
o
n.
A
c
c
or
di
ng t
o t
he i
n
put
v
ar
i
abl
e
X
,
t
he c
o
nnec
t
i
on
w
e
i
g
ht
s
of
i
nput
l
a
y
er
an
d t
he
hi
dde
n
l
a
y
er
ij
ω
and
t
he
h
i
dd
en
l
a
y
er
'
s
t
hr
es
ho
l
d
va
l
u
e
a
,
c
al
c
u
l
at
e
t
he
hi
dd
en
l
a
y
er
'
s
out
put
H
.
(
1)
I
n t
he f
or
m
ul
a,
l
i
s
t
he
nu
m
ber
o
f
hi
dde
n no
des
;
f
i
s
t
he i
nc
ent
i
v
e f
unc
t
i
o
n of
t
he
hi
d
den
l
a
y
er
.
T
hi
s
f
unc
t
i
o
n has
m
an
y
k
i
nds
of
ex
p
r
es
s
i
on f
or
m
s
,
and t
h
i
s
p
aper
s
e
l
ec
t
ed
f
unc
t
i
on i
s
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1
6
9
3
-
6
930
T
E
L
KO
M
NI
K
A
V
o
l.
14
,
N
o
.
3,
S
ept
em
ber
2016
:
9
56
–
96
2
958
(
2)
T
he t
hi
r
d s
t
ep:
C
a
l
c
ul
a
t
e t
he ou
t
put
l
a
y
er
.
A
c
c
or
di
ng
t
o t
he
hi
dde
n l
a
y
er
'
s
o
ut
put
,
c
onnec
t
i
on w
ei
ght
s
and t
hr
e
s
hol
d
,
c
al
c
u
l
at
e
t
he
B
P
n
eu
r
al
ne
t
w
or
k
'
s
pr
edi
c
t
i
v
e
out
put
.
(
3)
T
he f
our
t
h s
t
ep:
C
a
l
c
ul
a
t
e
t
he
er
r
or
.
A
c
c
or
di
n
g t
o
t
he
net
w
or
k
'
s
f
or
ec
as
t
out
put
and
ex
pec
t
ed out
put
,
t
he
net
w
o
r
k
‘
s
pr
edi
c
t
i
on
er
r
or
i
s
c
al
c
u
l
at
e
d.
(
4)
T
he
f
i
f
t
h
s
t
ep:
U
p
dat
e
t
he
w
ei
g
ht
.
A
c
c
or
di
n
g
t
o
t
he
n
et
w
or
k
'
s
pr
edi
c
t
i
on
er
r
or
,
u
pdat
e
t
he n
et
w
or
k
'
s
c
onnec
t
i
o
n
w
ei
g
ht
s
、
.
(
5)
(
6)
I
n t
h
e f
or
m
ul
a,
i
s
t
he
l
e
ar
ni
ng
r
at
e
.
T
he
s
i
x
t
h
s
t
ep:
U
pdat
e
t
he
t
hr
es
hol
d
.
A
c
c
or
di
n
g
t
o
ne
t
w
or
k
'
s
pr
edi
c
t
i
o
n
er
r
or
,
updat
e
net
w
or
k
node'
s
t
hr
es
ho
l
d
.
(
7)
(
8)
T
he s
ev
ent
h s
t
ep:
J
ud
ge
w
het
h
er
t
he
i
t
er
at
i
o
n of
t
he a
l
gor
i
t
hm
i
s
ov
er
.
I
f
not
ov
er
,
r
et
ur
n
t
o t
h
e s
ec
ond
s
t
ep.
T
he f
l
ow
c
har
t
of
B
P
ne
ur
al
net
w
or
k
al
gor
i
t
hm
i
s
s
ho
w
n i
n
F
i
g
ur
e 2.
F
i
gur
e 2
.
F
l
o
w
c
h
ar
t
of
B
P
al
g
or
i
t
hm
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
K
A
I
S
S
N
:
1
693
-
6
930
R
ec
ogn
i
t
i
on of
O
dor
C
har
a
c
t
er
i
s
t
i
c
s
bas
ed on B
P
N
eur
a
l
N
et
w
or
k
(
W
u Lei
)
959
3
.
O
d
o
r
S
ig
n
a
l
A
c
q
u
is
it
io
n
3.
1
.
E
x
p
e
r
i
m
e
n
t
E
q
u
i
p
m
e
n
t
T
hi
s
paper
us
e
s
t
h
e P
E
N
3 el
ec
t
r
oni
c
n
os
e s
y
s
t
em
t
o c
ol
l
ec
t
t
h
e od
or
da
t
a.
P
E
N
3
el
ec
t
r
on
i
c
nos
e
i
s
an an
al
y
t
i
c
a
l
i
ns
t
r
um
ent
w
h
i
c
h i
s
c
o
m
pos
e of
a s
et
of
c
o
m
p
os
i
t
e c
hem
i
c
al
s
ens
or
s
and
i
d
ent
i
f
i
c
at
i
o
n s
of
t
w
ar
e.
T
he har
d
w
ar
e s
t
r
u
c
t
ur
e of
t
he
s
y
s
t
em
i
nc
l
u
de
s
s
ens
or
ar
r
a
y
,
s
a
m
pl
i
n
g
and
c
l
e
ani
ng
c
h
anne
l
,
da
t
a
ac
qu
i
s
i
t
i
on
s
y
s
t
em
and
c
o
m
put
er
.
T
he
s
ens
or
ar
r
a
y
c
ons
i
s
t
s
of
10
m
et
al
ox
i
de s
ens
or
s
;
eac
h s
ens
or
s
en
s
i
t
i
v
e t
o t
he o
dor
m
ol
ec
ul
e
s
i
s
di
f
f
er
ent
.
W
h
en t
he o
d
or
c
ont
ac
t
ed
w
i
t
h
t
he s
ur
f
ac
e of
t
he
s
e
ns
or
,
t
he r
edox
r
e
ac
t
i
o
n o
c
c
ur
s
,
t
her
eb
y
af
f
ec
t
i
ng
t
he
c
i
r
c
ui
t
s
t
r
uc
t
ur
e
of
t
he
s
ens
or
.
A
f
t
er
s
am
p
l
e
s
t
an
d
f
or
a
c
er
t
ai
n
t
i
m
e,
t
he
t
o
p
em
pt
y
v
o
l
at
i
l
e
gas
t
hr
o
ugh
a bu
i
l
t
-
i
n pum
p,
i
t
i
s
a
ds
or
bed f
r
om
t
he ent
r
anc
e t
o t
h
e s
e
ns
or
c
hanne
l
,
t
hr
oug
h
t
h
e s
ens
or
ar
r
a
y
i
s
ex
c
l
ude
d
f
r
o
m
t
he
ex
por
t
s
.
F
i
gur
e
3 s
ho
w
s
t
he
r
es
p
ons
e
c
ur
v
e of
t
en s
ens
or
s
i
n t
he m
eas
ur
e
m
ent
pr
oc
es
s
.
D
at
a pr
oc
es
s
i
ng a
nd p
at
t
er
n r
ec
o
gn
i
t
i
o
n bas
ed
on t
he
r
at
i
o of
G
t
o G
0.
G
i
s
t
h
e r
es
i
s
t
anc
e v
a
l
ue of
t
he s
a
m
pl
e hea
ds
pac
e v
o
l
at
i
l
e
s
t
hr
ough t
he
s
ens
or
.
G
0 i
s
t
he r
es
i
s
t
anc
e v
al
u
e of
t
he
r
ef
er
enc
e ga
s
t
hr
ough
t
he
s
ens
or
[9
-
11]
.
F
i
gur
e 3
.
E
l
ec
t
r
o
ni
c
n
os
e r
e
s
pons
e s
i
g
na
l
3.
2
.
E
x
p
e
r
i
m
e
n
ta
l
M
at
er
i
a
l
s
Labor
at
or
y
t
em
per
at
ur
e
i
s
25
degr
ees
C
el
s
i
us
,
t
he
s
a
m
e
qual
i
t
y
of
m
i
l
k
w
er
e
s
t
or
ed
f
or
one da
y
,
t
w
o d
a
y
s
,
t
hr
e
e d
a
y
s
,
f
our
da
y
s
,
f
i
v
e da
y
s
.
T
ak
e t
he s
a
m
e qual
i
t
y
of
f
r
e
s
h
m
i
l
k
and
f
i
ve
di
f
f
er
ent
da
y
s
of
m
i
l
k
i
nt
o t
h
e t
es
t
t
u
be.
3.
3
.
E
x
p
e
r
i
m
e
n
ta
l
M
e
th
o
d
s
T
he
l
abor
at
or
y
t
em
per
at
ur
e
of
25
degr
ees
C
el
s
i
us
,
F
r
es
h
m
i
l
k
and
f
i
ve
k
i
nds
of
s
t
or
ed
w
i
t
h
di
f
f
er
ent
num
ber
of
da
y
s
of
m
i
l
k
eac
h
t
ak
e
out
1
0
m
l
put
i
n
t
he
t
es
t
t
ub
e.
E
ac
h
t
y
p
e
of
m
i
l
k
s
a
m
pl
es
w
as
pr
ep
ar
ed
50
r
epl
i
c
at
e s
am
pl
es
.
A
f
t
er
s
t
and
i
ng f
or
10m
i
n,
t
he
t
op e
m
p
t
y
v
ol
a
t
i
l
e
gas
t
hr
ou
gh a
bu
i
l
t
-
i
n p
um
p
,
i
t
i
s
a
ds
or
be
d f
r
o
m
t
he ent
r
anc
e t
o t
h
e s
ens
or
c
ha
nne
l
,
t
hr
o
ugh t
he
s
ens
or
ar
r
a
y
i
s
ex
c
l
ud
ed
f
r
om
t
he
ex
por
t
s
.
T
he
det
ec
t
i
on
t
i
m
e
i
s
150s
,
s
ens
or
c
l
ean
i
ng
t
i
m
e
i
s
60s
.
A
na
l
y
s
i
s
of
t
h
e s
i
gna
l
i
n a
s
t
ead
y
s
t
at
e
,
t
h
e s
i
gna
l
of
t
he
80s
i
s
us
ed
as
t
he
t
i
m
e poi
nt
of
t
he
el
ec
t
r
on
i
c
n
os
e an
al
y
s
i
s
.
T
abl
e
1
s
h
o
w
s
t
he
r
el
at
i
v
e c
ond
uc
t
i
v
i
t
y
v
a
l
ues
of
t
he s
i
x
gr
ou
ps
of
s
a
m
pl
es
i
n
t
en s
e
ns
or
s
.
D
ue t
o
t
he
l
i
m
i
t
at
i
on
of
t
hi
s
paper
l
eng
t
h,
her
e ar
e
on
l
y
t
he
r
es
u
l
t
s
of
t
hr
ee d
u
p
l
i
c
at
e s
am
pl
es
.
4
.
R
e
s
u
l
t
A
n
al
ysi
s
4.
1
/
P
a
r
ti
a
l
L
e
a
s
t
S
q
u
a
r
e
s
A
n
a
l
y
s
i
s
o
f
E
l
e
c
t
r
o
n
i
c
N
o
s
e
S
o
ftw
a
r
e
P
ar
t
i
al
l
eas
t
s
quar
es
(
PL
S)
r
egr
es
s
i
on
i
s
a
m
ul
t
i
v
ar
i
at
e
s
t
at
i
s
t
i
c
al
da
t
a
ana
l
y
s
i
s
m
et
ho
d
,
w
hi
c
h b
y
m
i
ni
m
i
z
i
ng t
he
er
r
or
s
u
m
of
s
quar
es
of
a s
et
of
dat
a t
o f
i
nd t
h
e
bes
t
m
at
c
hi
ng
f
unc
t
i
on.
I
t
us
es
t
he m
os
t
s
i
m
pl
e
m
et
hod t
o obt
a
i
n s
o
m
e
unk
now
n t
r
ut
h v
al
ue,
a
nd t
he s
um
o
f
er
r
or
s
quar
e
i
s
m
i
ni
m
u
m
[
6]
.
P
L
S
i
s
on
e of
t
h
e c
om
m
o
nl
y
us
e
d p
at
t
er
n
r
ec
ogn
i
t
i
o
n a
l
gor
i
t
hm
s
f
o
r
el
ec
t
r
on
i
c
n
os
e.
F
i
gur
e
4
i
s
t
he p
ar
t
i
al
l
e
as
t
s
quar
es
ana
l
y
s
i
s
m
i
l
k
'
s
qual
i
t
y
i
n t
he s
of
t
w
ar
e
of
el
ec
t
r
on
i
c
nos
e
.
C
an b
e s
e
en f
r
o
m
t
he f
i
gur
e t
he m
i
l
k
of
al
l
k
i
nds
of
qual
i
t
y
‘
s
P
L
S
v
al
ue,
t
hr
oug
h
t
he
c
om
par
i
s
on
of
t
hei
r
s
i
z
e
c
an
be
j
ud
ged
t
he
t
es
t
s
am
pl
e’
s
qua
l
i
t
y
.
R
and
om
l
y
s
el
ec
t
ed
eac
h
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1
6
9
3
-
6
930
T
E
L
KO
M
NI
K
A
V
o
l.
14
,
N
o
.
3,
S
ept
em
ber
2016
:
9
56
–
96
2
960
s
a
m
pl
es
of
10%
,
20%
and 40%
f
or
t
he t
e
m
pl
at
e,
a
nd t
he r
es
t
f
or
t
he t
es
t
s
a
m
pl
es
,
t
he r
es
ul
t
s
of
t
he ex
p
er
i
m
ent
as
s
ho
w
n i
n
T
abl
e 2.
T
abl
e 1
.
T
he e
l
ec
t
r
i
c
a
l
c
on
duc
t
i
v
i
t
y
of
m
i
l
k
w
i
t
h
di
f
f
er
ent
s
t
or
ag
e
da
y
s
i
n 8
0s
R
es
i
s
t
anc
e
Q
ual
i
t
y
R(
1
)
R(
2
)
R(
3
)
R(
4
)
R(
5
)
R(
6
)
R(
7
)
R(
8
)
R(
9
)
R
(
10)
F
re
s
h
S
a
m
pl
e1
1.
096
1.
095
1.
034
0.
887
1.
016
1.
427
1.
628
1.
039
1.
749
1.
410
S
a
m
pl
e2
1.
079
1.
151
1.
029
0.
836
1.
011
1.
191
1.
505
0.
955
1.
064
1.
369
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
S
am
p
l
e
50
1.
089
1.
163
1.
035
0.
866
1.
016
1.
258
1.
507
0.
980
1.
598
1.
450
O
ne
day
S
a
m
pl
e1
1.
097
1.
044
1.
038
0.
934
1.
017
1.
557
1.
618
1.
058
1.
605
1.
460
S
a
m
pl
e2
1.
037
1.
090
1.
030
0.
862
1.
012
1.
185
1.
538
0.
932
1.
524
1.
373
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
S
am
p
l
e
50
1.
079
1.
112
1.
035
0.
870
1.
017
1.
182
1.
499
0.
925
1.
486
1.
438
T
wo
day
s
S
a
m
pl
e1
1.
143
2.
222
1.
054
1.
017
1.
026
2.
155
3.
220
1.
254
3.
275
1.
636
S
a
m
pl
e2
1.
080
1.
098
1.
031
0.
885
1.
014
1.
244
1.
523
0.
961
1.
551
1.
407
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
S
am
p
l
e
50
1.
090
1.
134
1.
038
0.
891
1.
019
1.
280
1.
505
0.
974
1.
561
1.
483
T
hr
ee
day
s
S
a
m
pl
e1
1.
089
1.
366
1.
050
1.
082
1.
028
2.
318
1.
763
1.
275
1.
651
1.
653
S
a
m
pl
e2
1.
090
1.
402
1.
051
1.
176
1.
028
2.
476
1.
788
1.
344
1.
714
1.
675
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
S
am
p
l
e
50
1.
090
1.
406
1.
052
1.
165
1.
026
2.
331
1.
727
1.
305
1.
699
1.
709
F
our
day
s
S
a
m
pl
e1
1.
111
1.
293
1.
049
0.
977
1.
026
1.
795
1.
756
1.
168
1.
810
1.
610
S
a
m
pl
e2
1.
110
1.
348
1.
048
0.
931
1.
023
1.
678
2.
013
1.
092
2.
036
1.
607
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
S
am
p
l
e
50
1.
091
1.
153
1.
041
0.
869
1.
023
1.
311
1.
563
0.
970
1.
551
1.
509
F
i
ve
day
s
S
a
m
pl
e1
1.
094
1.
081
1.
037
0.
915
1.
019
1.
480
1.
630
1.
039
1.
627
1.
460
S
a
m
pl
e2
1.
082
1.
120
1.
034
0.
883
1.
016
1.
297
1.
561
0.
979
1.
545
1.
434
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
S
am
p
l
e
50
1.
089
1.
134
1.
039
0.
893
1.
020
1.
313
1.
482
0.
979
1.
498
1.
500
T
abl
e 2.
E
x
per
i
m
ent
al
r
es
ul
t
s
T
he nu
m
eber
of
t
e
m
pl
at
e
s
a
m
pl
e
20%
of
t
he
t
o
t
al
s
a
m
pl
e
40%
of
t
he
t
o
t
al
s
a
m
pl
e
80%
of
t
he
t
o
t
al
s
a
m
pl
e
T
he nu
m
ber
of
t
r
ai
ni
ng
s
a
m
pl
e
10
20
40
T
he nu
m
ber
of
t
es
t
s
am
p
l
e
40
30
10
P
LS
a
c
c
ur
ac
y
r
at
e
（
%
）
21%
34%
53%
B
P
ac
c
ur
ac
y
r
at
e
（
%
）
35%
46%
90%
F
i
gur
e 4.
P
ar
t
i
al
l
e
as
t
s
quar
es
ana
l
y
s
i
s
4.
2
.
B
P
N
eu
r
al
N
et
w
o
r
k
A
n
al
y
si
s
T
he s
t
r
uc
t
ur
e of
B
P
ne
ur
a
l
n
et
w
or
k
i
s
det
er
m
i
ned a
c
c
or
di
ng t
o t
he c
h
ar
ac
t
er
i
s
t
i
c
s
of
s
y
s
t
em
i
nput
and
o
ut
p
ut
dat
a.
T
he
el
ec
t
r
oni
c
n
os
e
has
10
s
ens
or
s
t
o
m
ea
s
ur
e
t
he
od
or
c
har
ac
t
er
i
s
t
i
c
p
ar
am
et
er
,
s
o t
he
od
or
i
n
put
s
i
gn
al
ha
s
10 d
i
m
ens
i
ons
.
T
her
e ar
e s
i
x
k
i
nds
of
di
f
f
er
ent
qua
l
i
t
y
of
m
i
l
k
,
s
o
t
he
ou
t
put
s
i
g
na
l
h
as
6
d
i
m
ens
i
ons
.
A
c
c
or
d
i
ng
t
o
t
h
e
t
r
i
al
t
o
det
er
m
i
ne,
t
h
e
n
um
ber
of
n
odes
i
n
t
h
e
h
i
dd
en
l
a
y
er
i
s
9
bes
t
.
S
o
t
he
s
t
r
uc
t
ur
e
of
t
he
B
P
neur
a
l
net
w
or
k
i
s
10
-
9
-
6
.
S
e
l
ec
t
ed
eac
h
s
am
pl
es
of
10%
,
20
%
and
40%
f
or
t
he
t
em
pl
at
e,
an
d
t
he
r
es
t
f
or
t
he t
es
t
s
am
pl
es
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
K
A
I
S
S
N
:
1
693
-
6
930
R
ec
ogn
i
t
i
on of
O
dor
C
har
a
c
t
er
i
s
t
i
c
s
bas
ed on B
P
N
eur
a
l
N
et
w
or
k
(
W
u Lei
)
961
F
i
gur
e
5
i
s
t
he
m
ean
s
quar
e
er
r
or
an
d
t
h
e
pr
e
di
c
t
i
ve
odor
s
i
g
na
l
of
t
he
t
es
t
s
am
pl
e
f
or
f
r
es
h m
i
l
k
.
M
ean s
quar
e er
r
or
i
s
t
he
m
ean s
quar
ed di
f
f
er
enc
e bet
w
een t
h
e out
p
ut
and t
he t
ar
get
t
hat
i
s
t
he
e
v
al
u
e.
T
he bes
t
v
er
i
f
i
c
at
i
on p
er
f
or
m
anc
e
m
or
e c
l
os
e t
o 0 i
nd
i
c
at
es
t
h
a
t
t
he t
r
ai
n
i
n
g
m
odel
i
s
bet
t
er
.
I
t
c
an
b
e
s
een
f
r
o
m
t
he
f
i
gur
e
t
ha
t
t
he
t
r
ai
ni
ng
m
odel
of
B
P
ne
ur
al
n
et
w
or
k
i
s
r
eas
ona
bl
e
;
t
he bes
t
v
er
i
f
i
c
at
i
on p
er
f
or
m
anc
e i
s
0.
00
1119
7 i
n t
h
e f
i
f
t
y
-
se
c
on
d t
r
ai
ni
ng.
I
n t
he
pi
c
t
ur
e of
pr
ed
i
c
t
i
v
e od
or
s
i
gna
l
,
t
he d
ot
t
e
d l
i
ne
i
n
t
he f
i
gur
e r
epr
es
e
nt
s
t
he
t
ar
get
out
put
c
at
egor
y
;
t
he s
o
l
i
d l
i
n
e s
ho
w
s
t
he
ac
t
ua
l
o
ut
pu
t
c
at
e
go
r
y
.
F
r
om
t
he f
i
gur
e c
a
n b
e s
een t
he
ac
t
ua
l
out
p
ut
a
nd t
he t
ar
get
out
pu
t
i
s
bas
i
c
al
l
y
co
n
si
st
e
n
t
.
(
a)
Mean
s
quar
e
er
r
or
(
b)
P
r
ed
i
c
t
i
v
e o
dor
s
i
gna
l
F
i
gur
e 5.
B
P
neur
al
net
w
or
k
anal
y
s
i
s
T
he i
dent
i
f
i
c
at
i
on r
es
u
l
t
s
f
or
al
l
t
es
t
s
am
pl
es
ar
e s
how
n
i
n
T
abl
e 2.
A
s
c
an be s
een f
r
o
m
t
he dat
a
i
n
T
abl
e
2
,
t
he m
or
e num
ber
of
t
r
ai
ni
ng
s
a
m
pl
es
is
t
he hi
gher
r
ec
og
n
i
t
i
on r
at
e
of
t
he
al
g
or
i
t
hm
.
U
nder
t
h
e
c
ond
i
t
i
on
of
t
he
s
am
e
t
r
ai
n
i
ng
s
a
m
pl
e
num
ber
,
t
he
r
ec
ogni
t
i
on
r
at
e
of
B
P
neur
a
l
net
w
or
k
al
gor
i
t
hm
i
s
hi
gh
er
.
T
o
s
u
m
up,
f
or
t
he
i
dent
i
f
i
c
at
i
on
of
odor
c
har
ac
t
er
i
s
t
i
c
s
w
i
t
h
non
l
i
n
ear
d
at
a s
t
r
uc
t
ur
e
a
n
d
l
i
t
t
l
e
d
i
f
f
er
enc
e,
t
he r
ec
o
gni
t
i
o
n ac
c
ur
ac
y
of
B
P
ne
ur
al
net
w
or
k
i
s
hi
g
her
t
h
an t
he
par
t
i
al
l
e
as
t
s
quar
e a
l
g
or
i
t
hm
.
5.
C
o
n
c
l
u
s
i
o
n
U
s
i
ng t
he m
et
hod
of
t
hi
s
paper
,
B
P
n
eur
a
l
n
et
w
or
k
al
g
or
i
t
hm
i
s
us
ed t
o
id
e
n
t
if
y
t
h
e
c
har
ac
t
er
i
s
t
i
c
of
odor
.
T
he
r
ec
ogni
t
i
o
n
ac
c
ur
ac
y
c
an
r
eac
h
9
0%
.
C
om
par
e
w
i
t
h
t
he
p
ar
t
i
a
l
l
eas
t
s
quar
es
a
nal
y
s
i
s
w
h
i
c
h i
s
i
n
t
he
e
l
ec
t
r
on
i
c
n
os
e
s
of
t
w
ar
e
,
B
P
ne
ur
al
ne
t
w
or
k
al
gor
i
t
h
m
r
ec
ogni
t
i
o
n ac
c
ur
ac
y
i
s
m
u
c
h hi
g
her
t
ha
n t
h
e
p
ar
t
i
a
l
l
eas
t
s
quar
es
al
gor
i
t
hm
.
A
n
d t
he
r
ec
ogni
t
i
o
n ac
c
ur
ac
y
w
i
l
l
i
nc
r
eas
e
w
i
t
h
t
he
i
nc
r
eas
e of
t
he
num
ber
of
t
r
ai
n
i
ng s
am
pl
es
.
T
her
ef
or
e,
f
or
t
he i
dent
i
f
i
c
at
i
o
n
of
odor
c
har
ac
t
er
i
s
t
i
c
s
w
i
t
h n
on
l
i
n
ear
dat
a s
t
r
uc
t
ur
e and l
i
t
t
l
e
di
f
f
er
enc
e,
neur
a
l
n
et
w
or
k
c
an m
or
e ac
c
ur
at
el
y
an
d ef
f
ec
t
i
v
el
y
i
de
nt
i
f
y
t
he
odor
c
har
ac
t
er
i
s
t
i
c
s
.
A
c
k
n
o
w
l
e
d
g
m
e
n
ts
T
hi
s
w
or
k
i
s
s
uppor
t
ed
b
y
N
at
ur
a
l
S
c
i
enc
e F
o
und
at
i
o
n of
I
nner
Mo
ngo
l
i
a
A
u
t
on
om
ou
s
R
egi
on (
N
o.
2
014
MS
061
9)
,
and I
n
ner
Mon
go
l
i
a S
c
i
enc
e
and T
ec
hnol
og
y
P
r
oj
ec
t
(
N
o.
2012
030
4)
.
It'
s
r
es
ear
c
h
f
i
ndi
n
gs
t
hat
ap
pl
i
c
at
i
on
s
y
s
t
em
of
i
nt
el
l
i
ge
nt
d
e
c
i
s
i
on m
ak
i
ng
t
ec
hno
l
og
y
,
b
y
i
n
dus
t
r
y
i
n
n
ov
a
t
i
o
n t
a
l
e
nt
s
t
eam
of
I
nn
er
Mon
go
l
i
a
A
ut
on
om
ous
R
egi
on.
R
ef
er
en
ces
[1
]
L
iu
H
o
ng
-
xi
u
,
L
uo
De
-
han.
O
dour
R
e
c
ogn
i
t
i
on A
l
gor
i
t
hm
s
f
or
M
ac
hi
ne O
l
f
ac
t
i
on S
y
s
t
em
.
C
h
i
n
es
e
j
our
na
l
of
s
e
ns
or
s
a
nd a
c
t
u
at
o
r
s
.
20
06
;
19(
6)
:
251
8
-
25
22.
[2
]
W
ang
W
ei
-
we
i
,
Z
ha
ng
S
ai
-
n
an.
E
l
ec
t
r
oni
c
n
os
e
s
y
s
t
e
m
b
as
e
d on P
C
A
an
d B
P
hy
br
i
d neur
al
net
w
or
k
al
gor
i
t
h
m
.
T
r
an
s
d
uc
er
and
M
i
c
r
os
y
s
t
em
T
e
c
hn
ol
o
gi
e
s
.
2
013
;
33(
4)
:
90
-
9
2.
0
10
20
30
40
50
10
-4
10
-3
10
-2
10
-1
10
0
B
est
V
al
i
d
at
i
o
n
P
er
f
o
r
m
an
ce i
s 0.
0011197 at
ep
o
c
h
52
M
ean
S
q
u
ar
ed
E
r
r
o
r
(
m
se)
58 E
p
o
ch
s
T
r
a
i
n
V
al
i
dat
i
on
T
e
s
t
B
es
t
1
1.
5
2
2.
5
3
3.
5
4
4.
5
5
5.
5
6
-0
.
2
0
0.
2
0.
4
0.
6
0.
8
1
1.
2
T
em
pl
at
e o
u
t
pu
t
A
c
t
ua
l
o
ut
p
ut
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1
6
9
3
-
6
930
T
E
L
KO
M
NI
K
A
V
o
l.
14
,
N
o
.
3,
S
ept
em
ber
2016
:
9
56
–
96
2
962
[3
]
D
e H
an,
Y
aw
en S
hao.
D
i
s
c
r
i
m
i
n
at
i
on o
f
C
hi
n
es
e H
e
r
bal
M
edi
c
i
ne by
M
ac
hi
n
e
O
l
f
ac
t
i
on
.
T
EL
KO
M
N
I
KA
T
el
ec
om
m
uni
c
at
i
on
C
om
put
i
ng E
l
ec
t
r
oni
c
s
a
nd C
ont
r
ol
.
2013
;
11(
2)
:
63
0
-
6
36.
[4
]
S
hi
Z
hi
bi
a
o,
H
uan
g
S
he
ngqu
a
n,
F
an
X
uebi
ng,
L
i
Y
ang.
S
t
ud
y
on
E
l
ec
t
r
oni
c
N
os
e
B
a
s
ed
o
n
B
i
ol
ogi
c
O
l
f
ac
t
i
on.
C
h
i
na
M
ec
hani
c
al
E
ngi
ne
er
i
n
g
.
2
007
;
18(
2
3)
:
28
10
-
2812
.
[5
]
Q
I
N
S
h
u
-
ji,
X
u
C
hun
-
hua,
W
A
N
G
Z
han
-
s
ha
n.
E
f
f
e
c
t
of
A
r
t
i
f
i
c
i
al
N
eu
r
al
N
et
w
or
k
on P
er
f
or
m
an
c
e
o
f
E
l
ec
t
r
oni
c
N
os
e.
J
our
n
al
of
T
o
ngj
i
U
ni
v
er
s
i
t
y
.
200
5
;
3
3
(6
):
804
-
808.
[6
]
G
ao T
i
an,
W
a
n
g Q
i
.
T
ar
ge
t
R
ec
og
ni
t
i
on A
l
gor
i
t
hm
B
as
ed o
n B
P
N
et
w
or
k
s
a
nd I
nv
ar
i
an
t
M
om
ent
s
.
T
EL
KO
M
N
I
KA
T
el
ec
om
m
uni
c
at
i
on
C
om
put
i
ng E
l
ec
t
r
oni
c
s
a
nd C
ont
r
ol
.
2013
;
11(
2)
:
96
9
-
9
74.
[7
]
H
ong
Xu
e
-
z
hen,
W
ei
Z
he
n
-
bo
,
H
ai
Z
he
n
g,
W
ang
J
un
.
A
p
pl
i
c
at
i
on
of
E
l
ec
t
r
on
i
c
N
o
s
e
and N
eur
a
l
N
et
w
or
k
i
n B
e
ef
F
r
es
h
ne
s
s
D
et
e
c
t
io
n
.
M
oder
n F
ood S
c
i
e
nc
e
and T
e
c
h
nol
o
gy
.
201
4;
3
0
(4
):
280
-
28
5.
[8
]
H
anaa T
,
EI
-
m
ad
any
,
F
at
en
H
F
ahm
y
.
S
pa
c
e
c
r
af
t
pow
er
s
y
s
t
em
c
ont
r
o
l
l
er
bas
ed o
n neur
a
l
net
w
or
k
.
A
c
t
a A
s
t
r
on
aut
i
c
a
.
201
1
;
69
(
7
-
8
):
65
0
-
657
.
[9
]
G
ao
Li
-
pi
n
g,
W
an
g
J
un,
C
ui
S
hao
-
qi
ng,
et
al
.
E
v
al
u
at
i
o
n
of
f
r
e
s
h
j
u
i
c
e
of
s
t
r
aw
ber
r
i
es
at
d
i
f
f
er
e
n
t
degr
ee
s
of
r
i
pen
es
s
us
i
ng
e
l
ec
t
r
o
ni
c
nos
e
and
e
l
e
c
t
r
on
i
c
t
ongu
e.
J
o
ur
na
l
of
Z
h
ej
i
a
ng
U
ni
v
er
s
i
t
y
.
2012
;
38(
6)
:
7
15
-
724
.
[1
0
]
Z
hou
H
ong
bi
a
o
,
Z
han
g
Y
u
lin
,
D
i
ng
Y
ou
w
ei
,
L
iu
J
ia
jia
.
A
p
p
l
i
c
a
t
i
o
n of
adapt
i
v
e pr
oba
bi
l
i
s
t
i
c
neur
al
net
w
or
k
i
n C
h
i
ne
s
e
l
i
q
uor
E
-
N
os
e.
C
A
A
I
T
r
an
s
a
c
t
i
o
ns
on
I
n
t
el
l
i
gent
S
y
s
t
em
s
.
2013
;
8
(2
):
17
7
-
182
.
[1
1
]
O
l
unl
oy
o
V
O
,
I
bi
da
po
T
A,
D
i
n
ri
f
o
R
R.
N
eur
al
net
w
or
k
-
b
as
e
d
el
ec
t
r
oni
c
nos
e
f
or
c
o
c
oa
b
ea
ns
qua
l
i
t
y
as
s
es
s
m
e
nt
.
A
gr
i
c
ul
t
ur
al
E
n
gi
neer
i
n
g I
n
t
er
na
t
i
on
al
:
C
I
G
R
J
o
ur
nal
.
201
2;
1
3
(4
):
1
-
1
7.
Evaluation Warning : The document was created with Spire.PDF for Python.