TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.6, Jun
e
201
4, pp. 4230 ~ 4
2
3
6
DOI: 10.115
9
1
/telkomni
ka.
v
12i6.486
0
4230
Re
cei
v
ed O
c
t
ober 2
0
, 201
3; Revi
se
d Decem
b
e
r
31, 2013; Accept
ed Ja
nua
ry 2
2
, 2014
The Recognition of Stored Grain Pest
s Based on The
Gabor Wavelet
and Sparse Representation
Hongliang F
u
1,2
, Jing Lu*
1
, Hua
w
ei Tao
1
, Beibei Zhang
1
1
Institute of Informatio
n
Scien
c
e & Engin
eer
i
ng, Hen
an U
n
i
v
ersit
y
of T
e
chnol
og
y
,
Z
hengz
ho
u 45
000
1, Chi
n
a
2
Ke
y
L
abor
ator
y of Grain i
n
for
m
ation pr
ocess
i
ng a
nd co
ntrol
of the MOE, H
ena
n Univ
ersit
y
of T
e
chnolo
g
y
,
Z
hengz
ho
u 45
000
1,
Chin
a
Addres: Bo
x 1
112
0, Lia
nhu
a
Street,
High-te
ch Z
one, Z
hen
gzho
u, Hen
a
n
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: lujin
ga
nn@
1
63.com
A
b
st
r
a
ct
In order to
i
m
p
r
ove the r
e
co
g
n
itio
n rate
and
accu
racy
of stored
grain pests classific
a
tion,
savin
g
classificati
on ti
me,
a n
e
w
recogn
ition
metho
d
bas
ed
on th
e Gabor
w
a
vel
e
t and
spars
e
repres
entati
on
i
s
prop
osed
in t
h
i
s
pap
er. In thi
s
pap
er, ni
ne t
y
pical
pes
ts
in
the store
d
gra
i
n ar
e re
gard
e
d
as th
e rese
a
r
ch
obj
ect, Gabor
ener
gy feat
ure
s
an
d
mor
pho
l
ogic
a
l fe
atur
es
are
extracted,
princ
i
p
a
l c
o
mpon
ent a
n
a
l
ysi
s is
used
to r
educti
on
di
me
nsi
on
a
nd s
parse
re
pr
esentati
o
n
is
u
s
ed to
ac
hiev
e
the cl
assificati
o
n
of stor
ed
gr
ai
n
pests. Si
mu
lati
on r
e
sults
sho
w
that, Gabor
ener
gy fe
atur
e
is
a
better c
h
o
o
se for
gr
ain
p
e
sts class
i
ficati
on
,
and the ov
erall perfor
m
anc
e
of Gabor
feat
ures and spars
e
repres
entation
is
better than the traditional
classificati
on methods.
Ke
y
w
ords
:
stored
grai
n p
e
sts, Gabor w
a
vel
e
t, sparse
r
epres
entati
o
n reco
gn
ition,
princi
pa
l co
mp
one
n
t
ana
lysis
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
Grain
sto
r
ag
e pe
sts i
s
an
importa
nt factor
of ca
usi
ng the lo
ss
of stored g
r
a
i
n, whi
c
h
seri
ou
sly affect the safety of food sto
r
a
ge. Co
nt
rol m
easure
s
m
u
st
be taken to redu
ce its im
p
a
ct
on the a
g
ri
cul
t
ural p
r
od
ucti
on ha
za
rd
s. Accu
rately
cl
assificatio
n
of
pest
s
is
one
of the mea
n
s for
the pest co
ntrol, whi
c
h is
based on vari
ous type
s of pest
s
trainin
g
sample
s to determi
ne wh
ich
categ
o
ry the
test sample
i
s
. Due to
the
st
rong
ope
rability and
e
a
sy to i
m
ple
m
ent, the im
age
recognitio
n
m
e
thod h
a
s b
e
en be
com
e
a
great d
e
velo
pment. In the
image
recog
n
ition metho
d
,
the pe
sts a
r
e cl
assified
usu
a
lly through th
e image p
r
e
p
roce
ssing, fe
ature extract
i
on,
cla
ssifi
cation and
re
co
gniti
on.
In re
cent ye
ars,
som
e
co
mmonly u
s
e
d
feature extractio
n
meth
ods
and
cla
s
sificatio
n
algorith
m
s
ha
ve been
used
for the
store
d
grai
n pe
st
speci
e
s i
dentifi
c
ation. Z
hang
Hon
g
tao [1,
2]
use
d
ant col
ony optimiza
t
ion algorith
m
sele
ct
ed seven featu
r
es, then use
d
supp
ort vector
machi
ne to
id
entified ni
ne
kind
s
of g
r
ain
pe
sts,
a
nd
propo
sed
the
chara
c
te
risti
c
s of g
r
ain
in
se
cts
comp
re
ssion
method
ba
sed on
kernel
Fish
er
discriminant a
nal
ysis, effe
ctively red
u
ced
the
feature dim
e
nsion numbers while improv
ed the sepa
rability between classes. Yuan Jinli [3] used
extensio
n en
ginee
ring
ap
plicatio
ns to
the g
r
ain
pe
st
s
cla
ssifi
catio
n
, and
had
g
e
t better
re
su
lts.
Wang Keru [
4
] used artificial in
telligence and Internet technology
t
o
achieve
crop pests remote
image
re
cog
n
ition an
d di
agno
si
s. Zha
ng Hongm
ei
[5] used BP
neural net
wo
rk to
cla
s
sify and
identify pest
s
in sto
r
ed
gra
i
n. Lu Jun [6]
use
d
fu
zzy clusteri
ng
a
nal
ysis reali
z
ed
dynamic
fuzzy
clu
s
terin
g
an
alysis of the
store
d
g
r
ain
pest
s
. Ha
n Antai [7] used
comp
re
ssed
sen
s
ing to t
h
e
store
d
g
r
ain
pest
s
cl
assifi
cation,
re
cog
n
ition rate up
to 93%, whi
c
h b
r
o
ught n
e
w id
ea
s for
the
grain in
se
ct identificatio
n classificatio
n
.
For the cl
assification of grain pe
sts, pr
ede
ce
ssors often use
d
the geom
etri
c shap
es
feature
s
and
the colo
rs
ch
ara
c
teri
stic fe
ature
s
,
while
few studi
es t
he texture feature
s
, whi
c
h
is
not cond
uciv
e to the
tre
n
d
of
coal
esci
ng va
riety ch
ara
c
teri
stic fe
ature
s
. In
re
cent yea
r
s,
G
abor
energy feature has be
en applie
d in many fields
, but the method used for
stored g
r
ain p
e
st
s
classifi
cation
is still very
rare. In
this
paper, Gabor text
ure features of nine typical pest
s in t
h
e
store
d
g
r
ain
has bee
n ext
r
acte
d, then
combi
ned
wit
h
the 1
5
-dim
ensi
onal
morpholo
g
y featu
r
es,
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
The Re
co
gnit
i
on of Stored
Grain Pe
sts
Based o
n
Th
e Gabo
r Wavelet and… (Honglia
ng Fu
)
4231
prin
cipal
co
mpone
nt ana
lysis is
used
to dimen
s
io
n red
u
ctio
n. In the view
of cla
ssifi
cation,
spa
r
se re
pre
s
entation which is ba
sed o
n
the emer
gi
ng
theory of co
mpre
ssed
se
nsin
g is u
s
ed
for
the grain pe
sts cla
ssification. Comp
re
ssed sen
s
ing i
s
a new si
gn
al acqui
sition
, encodin
g
a
nd
decodin
g
the
o
ry, which m
a
ke
spa
r
se
repre
s
e
n
tation
to a
ne
w re
sea
r
ch
boom
. Wri
ght
J
et al.
prop
osed a
robu
st face
re
cog
n
ition via
spa
r
se
re
pre
s
entatio
n [8], which mainl
y
based
on i
dea
that hum
an v
i
sual
sy
stem
has t
he
ch
aracteri
stics of
image
sp
arse
re
pre
s
e
n
tation [9], an
d th
en
use
d
sig
nal sparse recon
s
truction in
re
dund
ant
dicti
onary to achi
eve face re
cognition [10
-
12].
Tanaya
Guh
a
et al. used
spa
r
se reco
gnition in
hu
man a
c
tion
reco
gnition,
which
re
con
s
truct
feature
dictio
nary
suitable
for recogniti
on an
d di
scu
ss fe
asi
b
ility of OMP (o
rth
ogon
al mat
c
h
i
ng
pursuit algo
ri
thm) in re
co
gnition [13]; Chi Cai
et al
. used sp
arse recognitio
n
in weed se
eds
recognitio
n
[1
4], howeve
r
, it didn’t con
s
i
der feat
u
r
e of
wee
d
se
ed,
so result wa
s poor; Xue
M
e
i
et al. u
s
ed
sp
arse
re
cog
n
ition vehi
cle
tra
cki
ng [1
5]. In
this p
ape
r,
sp
arse
rep
r
e
s
e
n
tation i
s
u
s
e
d
for the grai
n pest
s
cla
s
sification, re
co
g
n
ition rate a
c
hieve 94% ab
ove.
2. Image Preproces
sing
Image
pre
-
p
r
oce
s
sing
can
improve th
e
imag
e d
a
ta
and
highli
ght
the im
age
f
eature
s
whi
c
h involve
d
in sub
s
eq
u
ent wo
rk. Th
e pre-p
r
o
c
e
s
sing
of ri
ce
weevil are
a
s
follows: grayscale
pro
c
e
ssi
ng,
Figure 1
(
b);
5×5 Ga
uss filterin
g,
Figure 1
(
c);
Otsu
algo
rithm ba
ckground
Segmentatio
n is u
s
ed, Fi
gure
1(d
)
; ra
dius of
3
structural elem
ents di
sc
op
ening o
p
e
r
ati
on is
use
d
to extract the la
rg
e
s
t co
nne
cted
com
pone
n
t, Figure 1
(
e);
interse
c
ted
with the o
r
igi
nal
image, Figu
re
1(f).
(a) Original image
(b) G
r
ay
scale
images
(c) Gauss filtering
(d) Ba
ckgrou
nd Segme
n
ta
tion
(e) Split po
stpro
c
e
ssi
ng
(f) Inters
ec
ted with the original
image
Figure 1. The
Pre-p
r
o
c
e
ssi
ng of Rice Weevil
3. Extrac
tion
and Compre
ssion of Ga
b
o
r Energ
y
Fe
ature
3.1. Extrac
tion of Gab
o
r Energ
y
Feature
2D
Gab
o
r
wavelets
can
descri
be th
e
feeling
s
of
neuron
s bi
ol
ogical visio
n
pro
b
lem
s
better, its ai
rspa
ce
and f
r
eque
ncy d
o
m
ain
cha
r
a
c
teristi
c
s can
be adj
uste
d
according to
the
need
s of visi
on. The diffe
rent frequ
en
cy scale
s
an
d texture o
r
ient
ation informat
ion of imag
e
can
be extra
c
ted
throu
gh
2D
Gabo
r
wavel
e
ts [16], a
s
well a
s
th
e
chara
c
te
risti
c
f
o
r
cla
ssifi
cati
on
pest
s
. 2D Ga
bor filter fun
c
tion ca
n be ex
pre
s
sed a
s
a function of th
e form:
2
2
2
2
,
,
2
,
2
2
2
,
exp
exp
exp
z
k
j
z
k
v
u
v
u
v
u
v
u
k
z
(
1)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 6, June 20
14: 4230 – 4
236
4232
Whe
r
e
u
is the
nucl
ear direct
ion of
Gab
o
r,
v
is
nu
c
l
e
a
r
sc
ale
,
z
is the
im
age
co
ordi
nat
es
of the
given p
o
sitio
n
.
,
uv
k
controls
the wi
dth of
the
Gau
ssi
an
wind
ow functio
n
, wavelength
an
d
direction of t
he oscillations.
is the radi
us of the Ga
ussian fu
n
c
ti
on, whi
c
h pro
v
ides 2
D
gab
or
wavelet size. In
natural
im
a
ges,
2
,
v
u
k
is to co
mpen
sate fo
r attenuation o
f
the energy spectrum
determi
ned
b
y
the fre
que
n
c
y,
2
2
2
,
2
exp
z
k
v
u
is Ga
uss
envelop
e fun
c
tion
;
z
k
j
v
u
,
exp
is pla
ne
wav
e
of co
mplex
values,
real
part i
s
z
k
v
u
,
cos
, imaginary p
a
rt i
s
z
k
v
u
,
sin
.
2
exp
2
is
DC
c
o
mponent.
Let image
(,
)
f
xy
siz
e
is
M
N
(M is the
number of pi
xels of the x-axis and N is the
numbe
r of pixels of the y-a
x
is), t
hen the
2D Ga
bor tra
n
sform functi
on is:
,,
(,
)
(
,
)
uv
uv
st
Gx
y
f
x
s
y
t
(
2)
Whe
r
e
s
is the length of the filter module,
t
is the width
of the filter module,
x
is the length of
the image,
y
is the
width
of the imag
e
.
Acco
rdin
g
to the results of the 2D
Gabo
r
wavel
e
t
transfo
rm, en
ergy inform
ation are
cal
c
ul
ated acco
rdin
g to formula
(
3):
,,
(,
)
(
,
)
uv
uv
xy
Ex
y
G
x
y
(
3)
The direct u
s
e of ene
rgy
information
is lik
ely to cause errors,
so u
s
ually th
e mean
energy inform
ation are u
s
e
d
as texture f
eature
s
[17]:
,
1
(,
)
(
,
)
uv
xy
uv
E
x
y
MN
(
4)
In
this pap
er,
filter con
s
i
s
ting
of 40 gab
or wa
velet filt
er(five
scale
s
eight
dire
cti
ons) i
s
use
d
to tran
sform the grai
n pest
s
pictu
r
e. Figu
re 2 i
s
a Gab
o
r
wavelet transf
o
rm. Com
put th
e
mean en
ergy informatio
n after tran
sform,
then get
a total of 40 features a
s
texture
feature
s
.
Acco
rdi
ng to
the extra
c
tion method i
n
the
literatu
r
e
[2] \, area, perim
eter, el
ongatio
n,
stand
ard
pr
o
duct, comple
xity, duty cycle, ci
rcul
a
r
ity, equivalent
radiu
s
, 1
-
7
orde
r mo
m
ent
invariant
s are
extracted, g
e
t 15 morph
o
l
ogical featur
es in total, with 40 Gabo
r feature
s
toget
her
as featu
r
e vector.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
The Re
co
gnit
i
on of Stored
Grain Pe
sts
Based o
n
Th
e Gabo
r
Wavelet and… (Honglia
ng Fu
)
4233
(a) After p
r
e-
pro
c
e
ssi
ng
(b) Ene
r
gy
im
age after G
a
b
o
r tran
sfo
r
m
Figure 2
.
Gab
o
r Wavelet Transfo
r
m
3.2. Featur
e Compre
ssio
n
Princi
pal
com
pone
nt analy
s
is (P
CA) [1
8
]
is a
linea
r transfo
rmatio
n
of multiple va
riable
s
to elect a l
e
ss i
m
po
rtant
variable
an
d as
many
as p
o
ssibl
e
to refle
c
t the
origin
al vari
able
informatio
n. The n
o
ise a
nd the
data
redu
nda
nc
y
can
be
rem
o
ved by PCA,
as
well
as
the
redu
ction
dim
ensi
onality of
the o
r
igin
al complex d
a
ta.
In this
pape
r,
55-dimen
s
io
nal (40
D
G
a
b
o
r
feature
s
and
15D mo
rph
o
logy features) fe
at
ure
vector a
r
e redu
ced dim
ensi
onality a
n
d
optimizatio
n by PCA. The ma
in process is as follo
ws:
1)
Cal
c
ulate the
data mean
x
an
d covari
an
ce
S.
2) Let
12
,,
,
n
px
x
x
x
x
x
,
n
=
55,
calculate
the eig
envalue
s
and
eigenve
c
tors
of S through
T
pp
n
S
1
,
and eig
e
nval
ues
in desce
nding o
r
de
r.
3)
Select th
e first m l
a
rgest
eigenvalu
e
s
corr
e
s
p
ondin
g
eig
enve
c
tors a
s
ba
sis ve
ctors,
transfo
rme
d
in the minimu
m mean squa
re erro
r co
ndi
tions:
12
,,
,
pca
p
c
a
pca
p
cam
VV
V
V
(5)
X
V
Y
pc
a
(6)
Y is prin
cipa
l compo
nent
s, X is cha
r
acteri
stic va
ri
able
s
,
pca
V
is
the firs
t m larges
t
eigenvalu
e
s correspon
ding eigenve
c
tors.
After the prin
cipal comp
on
ent transfo
rm
we
get 55 princip
a
l com
p
onent
s
, and intercept
the
first
1
0
prin
cipal co
m
pone
nts
fo
r analysi
s
of
cumulative co
ntribution
rat
e
,
the
re
sult
s
is
sho
w
n i
n
Fig
u
re 3. T
he first 1, 2, 3 p
r
in
cipal
com
pon
ents
cumul
a
tive contri
butio
n rate of o
r
yzae
rea
c
h
86.5%. Studies in Li
terature [19]
have sho
w
n t
hat the
re
cog
n
ition rate is
better
whe
n
t
he
cumul
a
tive
contributio
n
ra
te more th
an
85%
of
the
prin
cipal
com
pone
nt, there
f
ore, the
pa
p
e
r
sele
cted 1, 2,
3 princi
pal compon
ents a
s
the gr
ain in
se
ct
cla
ssif
i
c
a
t
i
on t
e
st
f
e
at
ure v
e
ct
o
r
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 6, June 20
14: 4230 – 4
236
4234
Figure 3. Cu
mulative Cont
ribution
Rate
4. Classifica
tion
In re
ce
nt yea
r
s,
Ca
ndè
s
a
nd
Don
oho
et
al e
s
tabli
s
h
e
d
Comp
re
ssi
ve Sen
s
ing
(CS) [2
0-
22]. CS is a full use of sig
nal spa
r
sity or com
p
re
ssibi
lity of the new sign
al acq
u
isition, en
co
ding
and
de
codi
ng
theo
ry. The
theory
sugge
sts that
wh
en the
si
gnal
is spa
r
se or
co
mpre
ssible,
t
h
e
sign
al
can
b
e
ap
proximated a
c
cu
ratel
y
re
con
s
tru
c
t
thro
ugh
coll
ect a
small
amount
of
si
gnal
proje
c
tion. T
he p
r
op
ose o
f
CS ma
ke
sparse
rep
r
e
s
entation to
a
new hei
ght. The g
r
ain
pe
sts
cla
ssifi
cation
based on
spa
r
se
rep
r
e
s
ent
ation [23] model is a
s
follo
ws:
There a
r
e t di
stinct
obje
c
t
cla
s
ses, fe
ature
s
of
ea
ch
cla
s
s co
mpo
s
e the training
sam
p
le
matrix
1,
,
i
mn
i
A
Ri
k
,
,
wh
er
e
,1
,
2
,
[,
,
,
]
i
i
ii
i
n
A
vv
v
. Different types of trai
nin
g
sam
p
le
matrix com
p
ose
a compl
e
te sam
p
le
matrix
A
12
,
t
1,
1
1
,
2
,
[,
,
]
[
,
,
,
]
k
tn
A
AA
v
v
v
. Literature
[24] states: fo
r any ne
w te
st sample f
r
o
m
the sa
me
cla
s
s
mn
yR
ca
n
be approa
che
d
by
linear
spa
c
e
con
s
tituted by t
r
aini
ng
sam
p
le
i
A
. Therefore
cla
ssifi
cation
p
r
oblem
can
b
e
tra
n
sfo
r
me
d
into solving th
e followin
g
eq
uation:
yA
x
(7)
Whe
n
mn
,
(
7
)
become
s
und
erdete
r
min
e
d
equation, wh
ich can be
sol
v
ed by (8):
1
arg
m
i
n
.
x
xs
t
y
A
x
(8)
Coeffici
ents o
f
solve is
,1
,
2
,
[
0
,
,
0,
,
,
,
,
0,
,
0
]
i
Tn
ii
i
n
x
R
only the i category
coeffici
ent is
not zero, then
the test sam
p
le belo
nging
to class i , so as to achi
eve the purp
o
se
s
of classification.
5. Experimental Re
sults
and An
aly
s
is
The im
ag
s a
bout
cadell
e
, Grai
n Bo
rer, Alphitobiu
s
diape
rin
u
s
pan
zer,
Ory
z
aephil
u
s
suri
nam
en
sis, Cryptole
ste
s
tu
rci
c
u
s
,
Callosobruc
hu
s chin
en
sis, Rice weevil, Long
valley stolen,
Triboli
u
m castaneum a
r
e
selecte
d
in thi
s
pa
per, th
e
r
e are
9 pe
sts in all [25]. Construct traini
ng
sampl
e
s matrix throug
h ext
r
actin
g
featu
r
es
of
13
5 p
e
st image
s
(ea
c
h pe
st h
a
s 15
pictu
r
e
s
) an
d
con
s
tru
c
t te
st sam
p
le
s ma
trix throug
h e
x
tracting fe
ature
s
of 4
5
p
e
st ima
ge
(e
ach
pe
st ha
s 5
picture). Fou
r
kind
s of experime
n
tal pro
g
ram a
r
e:
1)
Only 15 morp
hologi
cal feat
ure
s
are used
.
2)
Only 40 texture features a
r
e use
d
.
3)
Morp
holo
g
ica
l
features a
n
d
texture
features
set up 55
D integrated feature
s
.
1
2
3
4
5
6
7
8
9
10
0
10
20
30
40
50
60
70
80
90
10
0
11
0
P
r
i
n
ci
p
a
l
co
m
p
o
n
e
n
t
s
C
u
m
u
l
a
t
i
v
e
c
o
n
t
r
i
b
u
ti
o
n
ra
te
%
42.
3
78
.1
86
.5
92
.1
94.
6
97
.3
96
.2
99
.3
98
.7
97
.9
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
The Re
co
gnit
i
on of Stored
Grain Pe
sts
Based o
n
Th
e Gabo
r Wavelet and… (Honglia
ng Fu
)
4235
4)
Integrated fe
ature
s
thro
ug
h prin
cipal
co
mpone
nt anal
ysis an
d processing, then t
a
ke
the first three
prin
cipal
com
pone
nts.
Table 1. Perf
orma
nce Co
mpari
s
o
n
of Different Cl
assificatio
n
Sch
e
mes
Program 1
Program 2
Program 3
Program 4
Success rate of classif
i
cation
72.45%
89.12%
90.33%
94.03%
Time-consuming
52.35001
78.51286
80.84916
43.06575
Table 1
sh
o
w
s th
e success rate
and
time-con
su
ming of fou
r
different
cla
ssifi
cation
prog
ram
s
. From the time
con
s
um
ptio
n point of
vi
ew, with th
e
increa
sing
n
u
mbe
r
of fea
t
ure
vectors, pro
g
r
am 1 program 2 prog
ra
m 3 has
mo
re and mo
re
time consu
m
ption, while
the
con
s
um
ption
of prog
ram 4 is almo
st half of t
he progra
m
3, which fu
lly demonstra
t
ed throug
h the
PCA dimen
s
i
on re
du
ction
can
red
u
ce redun
dant a
n
d
save time.
From th
e cla
s
sificatio
n
succe
s
s
rate p
o
int of
view, progra
m
4 ha
s the
highe
st su
c
c
e
s
s
r
a
t
e
.
The
cla
ssif
i
cat
i
on
su
cc
es
s
rat
e
of
prog
ram
2 is
large m
o
re than 20% of progra
m
1,
this fully proves
Gabo
r wavel
e
t energy feature
excee
d
morp
hology feature for the
grai
n inse
ct cla
ssification.
As a wh
ole, classificatio
n
reco
gnition rate
based on
sparse rep
r
esent
ation in thi
s
pap
er
is high than
the neare
s
t
neighbo
r cl
assificatio
n
in literature [1], the extension e
ngine
e
r
ing
method in lite
r
ature [3], and the BP neural
netwo
rk m
e
thod in litera
t
ure [5].
6. Conclusio
n
This pa
per
re
sea
r
ch the re
cog
n
ition of st
ored grain p
e
sts b
a
sed o
n
Gabo
r wav
e
let and
spa
r
se repre
s
entatio
n, Ga
bor
ene
rgy fe
ature
s
of t
he
typical nin
e
ki
nds
of sto
r
ed
grain
pe
sts
a
r
e
extracted, p
r
inci
pal com
pone
nt anal
ysis is u
s
e
d
to dimen
s
ion redu
cti
on and sp
arse
rep
r
e
s
entatio
n is use
d
to the cla
ssifi
cati
on of st
ored
grain p
e
st
s. Experime
n
tal result
s sho
w
that,
Gabo
r wavel
e
t energy feature ex
cee
d
morp
hology
f
eature fo
r the grain in
se
ct cla
s
sificati
on.
Cla
ssifi
cation
based o
n
sparse re
presentat
ion exceed BP neural netwo
rk a
nd the nearest
neigh
bor
cla
s
sificatio
n
. A combinatio
n of
both ma
ke g
r
ain in
se
ct cl
assificatio
n
rate rise to 94
%,
the overall cl
assificatio
n
system
pe
rformance i
s
en
h
anced. Thi
s
p
aper also h
a
s som
e
thing to
be
perfe
cted:
su
ch
as: m
u
ch
more featu
r
es to
be
i
n
tegrate
d
, different
algo
rithms i
n
spa
r
se
rep
r
e
s
entatio
n to be comp
ared.
Ackn
o
w
l
e
dg
ements
Suppo
rted by Henan
sci
en
ce and te
chn
o
logy i
nnovati
on outstan
din
g
youth fund proje
c
ts
(104
100
510
0
08). In
novation Spe
c
ial
P
r
og
ram
of
G
r
aduate
Edu
c
ation in
Hen
an
University of
Tech
nolo
g
y (2012Y
JCX
5
6
)
. The Nation
al Natu
ral Science Fun
d
No: 612
013
8
9
.
The Natio
nal
863 pla
n
proj
ect No
:
201
2
AA10160
8. Proje
c
t of Education
De
p
a
rt
ment of Hen
an Provin
ce
No:
14A550
001.
Referen
ces
[1]
Z
hang
H
ongta
o
, Mao
H
anp
i
ng, H
a
n
Lvh
u
a
. Base
d
on
Kerne
l
F
i
sh
er
Discrimi
nant
A
nal
ysis
i
n
sect
F
eature Com
p
r
e
ssio
n
metho
d
.
Jiangs
u Univ
e
r
sity Journa
l.
2012; 33(
1): 16-
20.
[2]
Z
hang
Hon
g
ta
o, Mao Ha
np
in
g, Qui Da
o
y
in.
F
eatur
e e
x
tra
c
tion for the st
ored-
grai
n ins
e
ct detectio
n
s
y
stem b
a
sed
on ima
ge reco
gniti
on tech
nol
og
y.
T
r
ansacti
ons of the CSA
E
. 2009; 25(
2): 126-1
30.
[3]
Yuan
Jin
li, Ji
Hai
y
a
n
, Gu
o
Zhitao.
Rese
arch
on e
x
te
nsio
n
e
n
g
i
ne
e
r
ing in
stored
grai
n pest
s
classification.
Agricult
ural En
gin
eeri
ng Jo
ur
nal.
20
04; 20(
5
)
: 170-17
2.
[4]
W
ang Keru.
Rese
arch of
base
d
on Ima
ge Rec
o
g
n
itio
n dia
g
n
o
sis o
f
crop pests.
Acade
my o
f
Agricult
ural Sci
ences
. Bei
jin
g, Chin
ese. 2
005.
[5]
Z
hang
Ho
ngm
ei, F
an Y
anfe
n
g
. Rese
arch
of
base
d
o
n
dig
i
tal ima
ge
proc
essin
g
tech
nol
og
y of
grai
n
pests classification.
Hen
an U
n
iversity of T
e
ch
nol
ogy Jo
urna
l
. 2005; 2
6
(1): 1
9
-22.
[6]
Lu Jun, T
an Z
uoju
n
, Che
n
Jianj
un. An
al
ysi
s of granar
y pests d
y
nami
c
fuzz
y
cluster
i
ng.
Ch
ines
e
Cere
als an
d Oils Journ
a
l.
20
0
9
; 24(9): 11
6-1
18.
[7]
Han A
n
tai, Gu
o
Xia
o
h
ua, L
a
o
Z
hon
g, et a
l
. Class
ificatio
n
of agric
ultur
a
l
pests b
a
sed
o
n
compr
e
sse
d
sensi
ng the
o
r
y
.
Transactions
of the CSAE.
2011; 6: 20
3-20
7.
[8]
W
r
ight JA, Ganesh A, et al. Rob
u
st fa
ce re
cogn
ition vi
a s
parse re
prese
n
t
ation.
IEEE trans. on pattern
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 6, June 20
14: 4230 – 4
236
4236
Analys
is and M
a
chi
ne Intel
lig
e
n
ce (PAMI)
. 2009; 31(2): 2
10-
227.
[9]
B Olshaus
en,
D F
i
el
d. le
arni
n
g
a D
i
ctio
nar
y
of Sh
a
pe-C
o
m
pon
ents i
n
Vis
ual
Corte
x
: C
o
mpariso
n
w
i
t
h
Neur
ons, Hum
ans an
d Mach
i
nes PhD d
i
sser
t
ation, MIT
.
2006.
[10]
D Do
no
ho. F
o
r Most Lar
ge
Und
e
rderterm
i
ned
S
y
stems
of Li
near
Equ
a
tions
the M
i
n
i
mal
l1-n
or
m
Soluti
on Is Also the Spars
e
st Soluti
on.
Co
mm. Pure a
nd A
ppli
ed Math.
2
006; 59(
6): 797
-829.
[11]
E Cand
es, J Romberg, T
T
ao. Stable
Signa
l Rec
o
ver
y
from I
n
comp
lete an
d Inaccurat
e
Measur
ements
.
Comm. Pure
and Ap
pli
ed M
a
th
. 200
6; 59(8
)
: 1207-1
2
2
3
.
[12]
E Can
des, T
T
ao. Near-Optimal S
i
gn
al
R
e
cove
r
y
form
Ran
dom Pr
oje
c
tions: Un
ivers
a
l Enc
o
d
i
n
g
Strategies.
IEEE Trans. Inform
ation Theory.
2006; 5
2
(12):
540
6-54
25.
[13]
T
anya Guh
a
, Rab
ab K W
a
rd
. Learn
i
ng S
p
a
r
se Repr
ese
n
tations for
Hum
an Actio
n
Rec
ogn
ition
. IEEE.
T
r
ans. on patte
rn ana
lysis an
d
mach
in
e intel
l
i
genc
e
. 201
1:1
57-1
588.
[14]
Cai
Chi,
Z
han
g
Ming, Z
hu J
u
n
p
in
g. W
eed
se
eds c
l
assific
a
ti
on
base
d
on c
o
mpress
ed s
e
nsin
g.
Scienc
e
chin
a press.
20
10; 40: 16
0-17
2.
[15]
Xu
e M
e
i, H
a
ib
i
n
Li
ng.
Rob
u
st
Visu
al T
r
ackin
g
a
nd V
e
h
i
cle
Classific
a
tio
n
v
i
a S
parse
Re
p
r
esentati
on.
IEEE Trans on Pattern Analys
is Machine.
201
1; 33(11): 2
259
-227
2.
[16]
Z
hang Gan
g
, Ma Z
ongmin.
One kind of u
s
ing Ga
b
o
r
w
a
vel
e
t texture
feature e
x
tract
i
on meth
od.
Chin
ese Jo
urn
a
l of Ima
ge a
n
d
Graphics.
20
10; 15(2): 2
47-
254.
[17]
Z
hang G
ang,
Ma Z
ongm
in.
An ap
pro
a
ch
of usin
g g
a
b
o
r
w
a
ve
lets for
textur
e featur
e e
x
tractio
n
.
Journ
a
l of Ima
ge an
d Graph
i
cs.
2010; 15(
2)
: 247-25
4.
[18]
Hua
ng Qua
n
l
o
ng. F
a
ce Rec
ogn
ition B
a
se
d on PCA.
Xi'
an Un
iversity
of Electron
ic Scienc
e an
d
T
e
chno
logy.
2
012.
[19]
Xi
an
Ho
ng. B
a
sed
o
n
e
l
ect
r
onic
tong
ue
Lon
gj
in
g te
a ta
ste
qu
al
i
t
y
te
sti
n
g re
sea
r
ch
. Zh
e
j
i
a
ng
Univers
i
t
y
. 2
0
1
0
: 50-69.
[20]
DL Do
noh
o. Compress
ed se
nsin
g.
IEEE Tr
ans. on Infor
m
ation Theory
. 2
006; 52(
4): 128
9-13
06.
[21]
E Can
des, J
R
o
mber
g, T
e
rence
T
ao. Rob
u
s
t
uncerta
int
y
pr
incip
l
es: E
x
act sign
al r
e
co
nstruction fr
o
m
hig
h
l
y
inc
o
mpl
e
te freque
nc
y i
n
formatio
n
.
IEEETrans. on Inform
ation
Th
eory
. 2006; 52(
2): 227-2
54.
[22]
Jian
hua
Z
hou,
Si
w
a
ng Z
h
o
u
, Qiang
F
a
n
.
Mat
hematics
Appro
a
ch
es
in C
o
mpress
e
d
Sens
in
g.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
ng.
2013; 1
1
(9): 5
435-
544
0.
[23]
Xinsh
eng
Z
h
a
ng, Mi
ng
hu W
ang.
S
pars
e
R
epres
entati
on f
o
r D
e
tectio
n of
Microc
alcific
a
tion
Cl
uster
s
.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2012; 1
0
(3): 5
45-5
50.
[24]
K Le
e, J
Ho,
D
Krie
gman.
La
mbertai
n
R
e
fle
c
tion
an
d
Lin
e
a
r Su
bsp
a
ces.
IEEE Trans.Pattern Analysis
and Mac
h
in
e Intelli
ge
nce
. 20
06; 28(3).
Evaluation Warning : The document was created with Spire.PDF for Python.