TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.7, July 201
4, pp
. 5368 ~ 53
7
8
DOI: 10.115
9
1
/telkomni
ka.
v
12i7.528
1
5368
Re
cei
v
ed
De
cem
ber 4, 20
13; Re
vised
F
ebruary 21,
2014; Accept
ed March 1
1
, 2014
Characteristics Analysis and Detection Algorithm of
Mosquitoes
Jahan
gir Ala
m
S.M.*, Hu Guoqing, Ch
eng Ch
en
Dept. of Mecha
n
ical & El
ectric
al Eng
i
ne
eri
ng,
Xiam
en U
n
ive
r
sit
y
Room 2
28, Sci
ence Bu
ild
in
g, 361
00
5, Simin
g
Distric
t, Xiam
en, F
u
jia
n, Chi
na, telp/fa
x: +
86-59
2-21
86
39
3
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: jaha
ngir
_
u
i
ts@
y
ah
oo.com
A
b
st
r
a
ct
The systematic detection and elim
ination
of m
o
s
quit
oes
is a va
luable process, the
results of
w
h
ich coul
d be
imp
o
rtant in th
e fight aga
inst Malari
a.
In this study, imag
e process
i
ng is u
s
ed, all
o
w
i
ng t
h
e
researc
hers to
detect the mo
squito
es an
d their loc
a
tio
n
s. Mosquitoes' physical c
har
act
e
ristics, territorial
and
be
havi
o
ra
l
patterns w
e
re
also
an
aly
z
e
d
throug
h rec
o
g
n
itio
n tech
nol
o
g
y. It is foun
d
that mosq
uitoe
s
can be d
e
tected an
d differe
ntiated by the
i
r physical,
terr
itorial a
nd b
e
h
a
vior
al pattern
s through thes
e
meth
od
olo
g
i
e
s. In ad
diti
on to
mos
q
u
i
toes, fli
e
s a
nd
bees
ar
e als
o
i
n
cl
ude
d
in th
is study
a
nd w
e
re
an
aly
z
e
d
for their p
a
tterns, as w
e
ll as t
heir d
i
stin
guis
h
ing fe
atures. Si
z
e
, n
u
m
b
e
r of
obj
ects, probo
scis, body s
h
a
pe,
color, a
n
ten
n
a
e
, hin
d
l
egs,
a
nd sh
ap
e p
a
ra
meters
w
e
re
a
l
l factors co
nsid
ered f
o
r
mosq
uito d
e
tectio
n
w
i
th
imag
e proc
ess
i
ng. All th
ese
infor
m
ati
ons w
e
re us
ed
i
n
th
e Mosq
uito D
e
tection A
l
gor
i
t
hm. T
h
is stud
y
provi
des char
a
c
teristic descri
p
tions of al
l three
ins
e
cts and
statistical an
al
y
s
is of the data foun
d.
Ke
y
w
ords
: mosquito, pattern, m
o
squito
mo
d
e
l, detectio
n
, malari
a
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
Huma
n M
a
la
ria a
nd
so
m
e
othe
r di
se
ase
s
su
ch
a
s
yello
w fev
e
r a
nd
den
g
ue a
r
e
transmitted by mos
q
uitoes [1]. Partic
ularly, the
femal
e
mo
squito
es are
da
nge
ro
us ve
cto
r
s fo
r
infesting
hu
mans with
such
di
sea
s
e
s
. Malari
a tra
n
smi
ssi
on
ca
n be
expe
rie
n
ce
d in
ho
u
s
e
s
,
forest
s, farm
s and a
n
y other vecto
r
di
sea
s
e e
n
viro
nment. The
r
e
are some m
o
sq
uitoe
s
wh
ich
bite hu
man
s
ro
utinely. T
hese
routine
bites
with v
e
ctors carrying infe
ctiou
s
disea
s
e
s
af
fect
millions of p
eople pe
r year [2, 3].
Althoug
h there
are othe
rs m
o
sq
uitoe
s
wh
ich do not bi
te
human
s, they
are n
oneth
e
l
e
ss vecto
r
s f
o
r ani
mal
di
seases
su
ch
a
s
de
ngu
e, Zo
ono
sis, et
c. [4
,
5]. Every year a
bout o
n
e
million
pe
ople lo
se
th
eir live
s
du
e
to Mala
ria
cau
s
e
d
mai
n
ly b
y
Mosq
uitoe
s
. Among this figure, ab
out 85% are Chil
d
r
en un
der the
age of 5 years old [4]. Most
Malari
a ca
se
s that re
sult i
n
the loss of
life oc
cur in
developin
g
countrie
s
with
90% of malaria
death
s
o
c
curring i
n
Africa
. The e
c
o
n
o
m
ic imp
a
ct
o
f
malaria
de
a
t
hs d
ue to m
o
sq
uito bite
s in
developin
g
countrie
s
i
s
e
norm
o
u
s
.
Th
e
effect
s
in
cl
ude
l
o
w
life
expecta
ncy rates with a high
infant mortalit
y rate, all of
whi
c
h bri
ng redu
ct
ion in produ
ctivity and an increa
se
in governm
e
n
t’s
budg
et on he
alth at the expen
se
of othe
r so
cial am
en
ities.
This
study f
o
cu
se
s o
n
the dete
c
tion
of
mosq
uitoes th
rou
g
h
image p
r
o
c
e
ssi
ng
tec
hniques
in order to
des
troy them with
las
e
r tec
h
nologies
.
LASER
tec
h
nology is
a
preventative
measure that
coul
d be u
s
ed to avoid
mosq
uito bit
e
s the
r
eby
saving lives a
n
d
cutting do
wn
person
a
l and
national b
udg
ets on he
alth
care.
2. The Proposed Algo
rithm
2.1. Chara
c
terstic
s
and
Patte
rn An
aly
s
is for Mos
quitoes
w
i
th
Flies and Be
es
The frequ
en
cy of mo
squito
es va
rie
s
fro
m
200
Hz to
7
00Hz
and
the
avera
ge f
r
eq
uen
cy is
about 600
Hz [6]. General
ly, mosquitoe
s
hav
e one
prob
osci
s, two wing
s, an abdom
en wit
h
cha
r
a
c
teri
stics such a
s
whi
t
e and g
r
ay li
nes,
six le
g
s
(two front leg
s
, two mi
d le
gs a
nd two lo
ng
spe
c
ial
hind l
egs), a h
ead
and a th
roat.
The bo
dy of
a mo
squito i
s
esp
e
ci
ally lo
ng an
d thin.
The
length
of a
n
adult m
o
squit
o
is vari
ed
an
d it i
s
u
s
u
a
lly 16mm
o
r
l
o
n
ger (0.6
in
ch
) [7]. The
wei
ght
of a mo
sq
uito
is fo
und
to b
e
up
to 2.5m
g (0.0
4
g
r
ai
n
s
). All M
o
squi
toes
have
sle
nder bo
die
s
with
three segm
e
n
ts namely; (i) he
ad, (ii) thor
ax and
(iii) ab
dome
n
.
A mosquito
’s hea
d ha
s
a
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Cha
r
a
c
teri
stics Analysi
s
an
d Dete
ction Algor
ithm
of Mosq
uitoe
s
(Ja
hangi
r Alam
S.M.)
5369
spe
c
iali
zed
resp
on
se sen
s
or fo
r receiving inform
atio
n and fo
r fee
d
ing [8]. It has two eye
s
a
nd a
long pai
r of segmente
d
an
tennae. A Mosq
uitoe
s
'
an
tennae h
a
ve multiple pu
rp
ose
s
which can
be used a
s
a
sen
s
ing
org
an to dete
c
t other in
se
cts,
to s
m
ell, to touch, and to
tak
e
in mois
t
u
re
from the ai
r [9]. The co
mp
ound
eyes
of an ad
ult mo
squito
develo
p
in a
sepa
ra
te regio
n
of the
head. The
he
xagonal
patte
rn only
b
e
co
mes
visible
when
th
e ca
ra
pace
of
t
he st
age with sq
u
a
re
eyes i
s
molte
d
[10]. The
h
ead al
so
ha
s an elo
ngat
e
d
forward-pro
j
ecting
whi
c
h
is “sting
er-like”
prob
osci
s used for feedin
g
, and two sensory pulp
s
. Male mosq
uitoes h
a
ve longe
r
maxill
ary
pulp
s
whil
e th
e female
s ha
ve sho
r
ter o
n
e
s. Some fe
male mo
squit
oes
have elo
ngated p
r
o
b
o
sci
s.
A mosquito’
s
thorax is a l
o
com
o
tion
system an
d
is
con
n
e
c
ted to
the three pa
irs of leg
s
an
d a
pair of
wing
s.
Gene
rally, th
e traveling
ra
nge for a
mo
squito i
s
a
r
ou
nd 75
-10
0
mil
e
s. Mo
sq
uitoes
can fly for u
p
to four ho
urs
contin
uo
usly at
1k
m/
h to 2k
m/h (0.6-1mph) [11]. At night,
the
Anophel
es
m
o
sq
uito ca
n travel up to 1
2
km (7.5mile
s) [6]. The u
s
ual life spa
n
of a mosq
uito is
up to 3
0
d
a
ys o
r
mo
re. A
com
parative study of
t
h
e
t
h
ree
in
se
ct
s
su
ch a
s
M
o
s
s
quit
o
,
Fly
,
and
Bee [12-1
4
] has sho
w
n in Figure 1, Figure2, an
d Fi
g
u
re 3, that, the body sh
ape
of both flies and
a bee is fatter and thicker t
han a mo
squi
to and their le
gs are also compa
r
atively sho
r
ter.
(a)
(b)
(c
)
(d)
(e)
(f)
(g)
(h)
Figure 1. Images of Diffe
re
nt Types of Mosq
uitoe
s
(a)
(b)
(c
)
(d)
(e)
Figure 2. Images of Diffe
re
nt Types of Flies
(a)
(b)
(c
)
(d)
Figure 3. Images of Diffe
re
nt Types of Bees
The p
r
obo
sci
s
too is
so
me
what
sho
r
ter
and in
som
e
instan
ce
s the
r
e is
no p
r
ob
oscis
at
all. Both bees and flies h
a
ve spe
ckl
ed
hair on their bodies, unli
k
e mosq
uitoe
s
which la
ck this
type of chara
c
teri
stic. A compa
r
ative a
nalysi
s
of
mo
squito
es to b
ees a
n
d flies helps to d
e
tect
mosq
uitoe
s
a
nd differentia
te them fro
m
other in
se
cts sh
own
in T
a
ble 1, T
a
ble
2, and
Tabl
e
3.
Bees have e
c
on
omic a
nd indu
strial val
ue and are
al
so environme
n
tally friendly. Howeve
r, flies
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5368 – 53
78
5370
and mo
squito
es ca
n have a negative impact on the
e
n
vironm
ent and health and
can se
rve as a
cau
s
ative ag
ent of Malaria
.
Table 1. Ch
aracteri
stics of
Di
fferent Typ
e
s of Mo
squit
oes
Species
Head
Thora
x
Abdomen
Obj
e
c
t
Aedes aegypti
Pr
oboscis: Dar
k
;
Pulps: Tipped
w
i
th
silver
y
-
white;
Cl
y
peus: White
scales.
Wings Scales da
rk; Legs
color: Tarsal seg
m
ent 5 is
entirel
y
w
h
ite; W
h
ite basal
bands 2*6 for Hi
nd legs and 3
for mid & foreleg
s
.
Scutum: L
y
re
shaped, White
scales.
30
Aedes
albopictus (Stegom
yia
albopicta)
Pulps: Tipped
w
i
th
silver
y
-
white;
Cl
y
peus: Black.
Wings Scales da
rk; Legs
color: Tarsal seg
m
ent 5 is
entirel
y
w
h
ite; W
h
ite basal
bands 2*6 for Hi
nd legs and 3
for mid & foreleg
s
; Thora
x
:
Sides are man
y
silvery
-
white.
Scutum: One
middle silver
y
-
w
h
ite stripe
do
w
n
.
23
Aedes vexans
Pr
oboscis: Dar
k
,
Pulps: Dark.
Wings Scales da
rk; Legs
color: Tarsal seg
m
ent is
entirel
y
na
rro
w
w
h
ite; narro
w
White basal bands 2*6 for
Hind legs and 3 f
o
r mid &
forelegs.
Shape: V, Color:
Pale.
23
Cul
e
x
terri
tans
Pr
oboscis: Dar
k
,
Pulps: Dark
Wings Dark, narr
o
w
;
Legs
c
o
l
o
r: Dark
.
Col
o
r: Na
rro
w
APICAL bands.
15
Culex restuans
Pr
oboscis: Dar
k
;
Edge is light w
h
ite,
Pulps: Dark.
Wings Dark, narr
o
w
;
Legs
c
o
l
o
r: Dark
; s
o
m
e
whi
t
e
spots; Thora
x
: P
a
tches; pale
scales.
Dark &
w
h
ite
basal bands,
scutum: Copper
c
o
l
o
r; rarel
y
2
pale spots.
19
Deinocerites cancer
Pr
oboscis: Dar
k
;
Antenna longer
than prob
oscis,
Pulps: Dark.
Wings Dark, narr
o
w
;
Legs
c
o
l
o
r: Dark
.
Color: Copp
er
bro
w
n.
15
Ochlerotatus
baha
m
ensis
Pr
oboscis: Dar
k
,
Pulps: White
tipped.
Wings Dark, narr
o
w
;
Legs
color: Hind legs dark
w
i
th
White basal bands.
Sputum: lines of
golden &
w
h
ite
scales.
17
Ochlerotatus infir
m
atus
Pr
oboscis: Dar
k
;
Pulps: Dark.
Wings Dark, narr
o
w
;
Legs
c
o
l
o
r: Dark
.
Color: Da
rk scale
w
i
th basal;
23
Ochlerotatus tris
eriatus
Pr
oboscis: Dar
k
;
Pulps: Dark.
Wings Dark; narr
o
w
;
Legs
color: Dark; Tho
r
ax: patches
of silver
-
w
hite scales.
Scutum: Dark;
Legs are
w
h
ite of
edge.
30
Table 2. Ch
aracteri
stics of
Different Typ
e
s of Flie
s
Species
Head
Thora
x
Abdomen
Obj
e
c
t
Am
enia albo
m
a
culata
No pr
oboscis; Post
orbits y
e
llo
w
to
orange; met
a
llic
dark blue; bright
orange color fac
e
.
Dark blue-g
r
een
to bluish violet;
Head of male gol
den orang
e; Shin
y
w
h
ite spots patt
e
rn.
12 mm; shin
y
w
h
ite p
a
ttern;
Scutellum w
i
th
three pair of
marginal setae.
52
Am
enia leonina
No proboscis;
Head of male
golden Y
e
llo
w
.
Post orbits
y
e
llow
to orange.
Dark Blue-Green
to bluish violet.
12 mm;
Scutellum w
i
th 3
pairs of marginal
setae.
68
Dexia r
u
stica
No Pr
oboscis;
Color: Bro
w
n,
wh
i
t
e
.
Dark; Legs are
th
in and short;
Bro
w
n, da
rk.
B
r
ow
n
-
da
r
k
;
speckle.
108
Lucilia cupr
ina
No Proboscis; Red
ey
e
s
.
Silvery
he
ad.
8 mm; Green
metallic.
56
Neom
yia sp.
Black.
5 mm; Black.
72
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Cha
r
a
c
teri
stics Analysi
s
an
d Dete
ction Algor
ithm
of Mosq
uitoe
s
(Ja
hangi
r Alam
S.M.)
5371
Table 3. Ch
aracteri
stics of
Different Typ
e
s of Bee
s
Species
Head
Thora
x
Abdomen
Obj
e
c
t
Apis
andrenifor
m
is
Probocis: 2.80
mm; pulps: black
stripes on the
legs.
Tibia & dorsolateral: 8 or fe
wer;
Thora
x
: 70
-90 m
m
; Pigmentation:
Blacki
sh;
A circular bod
y
shape; Length:
47-54 um.
29
Apis florea
Proboscis: 3.27
mm.
Pigmentation: Y
e
llow
i
sh.
Index: 2.
86.
35
Honey bee (Apis
m
e
llifer
a)
Pr
oboscis: y
e
llow
band; Pulps:
Pomeranian
bro
w
n.
Stocky
bo
d
y
; B
r
o
w
n;
dark
coloration; Thora
x
:
Nigra
, heav
y
dark pigmentatio
n of the
w
i
ngs.
Blacki
sh, or
m
e
llifer
a
, rich
dark bro
w
n; color
:
yellow band.
31
2.2. Dete
ctio
n Process
Analy
s
is and
Proposed
Algorithm
Identifying th
e mo
sq
uitoes' phy
sical
sha
pe i
s
ve
ry im
portant
in th
e
pr
ocess of
d
e
tection.
If it is po
ssib
le to id
entify a mo
squito’
s
physi
cal
sha
pe, then
the
informatio
n
could
be fu
rth
e
r
pro
c
e
s
sed fo
r dete
c
tion a
nd po
sition tracking. As
a
result the Ro
botic visi
on
s
are the
n
abl
e
to
effective as the target an
d
destroy mo
squito
e
s
. The image processing u
s
ed al
so detecte
d the
mosq
uito’s p
o
sition
by te
rritori
al a
n
d
behavio
ra
l
p
a
ttern
analysis. Thi
s
im
pli
e
s th
at furth
e
r
resea
r
ch
coul
d be
do
ne
where
it is po
ssible
to
sho
w
ho
w roboti
c
vision
ca
n de
tect mo
squito
es
by using thi
s
method, so th
at the destru
c
tion of
mosqu
i
toes is mo
re
efficient and
su
ccessful. T
he
algorith
m
ic p
r
oce
s
se
s are
pre
s
ente
d
in the Figu
re 4.
Figure 4. Mosquito’s p
a
ttern detectio
n
al
gorithm (MPDA)
Start
Ima
g
e Ca
p
tu
re
Ima
g
e F
ilter
in
g
Re
g
ion Gro
w
in
g
Threshold
Ed
g
e Detection
Feature
Se
g
men
t
ation
Color Se
g
mentat
ion
Parts Se
g
mentati
on
Model Creation
Sha
p
e Se
g
ment
ation
Shape and C
o
rre
lation based
Anal
y
s
is
Statistical Ana
l
y
si
s
Mos
q
uito’s Pattern
End
Is It Mosquito?
No
Ye
s
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5368 – 53
78
5372
There are
many metho
d
s for dete
c
tion
of the
moving obje
c
t. The backgroun
d
subtractio
n method is ef
fective to detect for
slo
w
motion flying mosq
uito. The ba
ckgro
und
image
s pixel value and the
detected mo
ving image
s pixel are not the sam
e
[15]. The differen
c
e
betwe
en the
s
e two valu
es
can
be d
e
fine
d as th
e re
gio
n
of intere
st (ROI) fo
r mo
squito dete
ctio
n.
The segme
n
tation of ba
ckgrou
nd pixel
and the ta
rge
t
ed mosquito’
s
pixel value
is impo
rtant to
differentiate the modeli
ng
and up
dating
[15, 16].
The Adaptive
Surend
ra Algorithm (AS
A
) is
for ba
ckgroun
d esti
mation and t
he Inter-
Frame
Diffe
renci
ng Alg
o
ri
thm (IF
D
A) i
s
effe
ctive for flying
Mo
squito
dete
c
t
i
on. The
fra
m
e
differen
c
ing,
backg
rou
nd e
x
traction an
d updatin
g,
backgroun
d su
btractio
n and
motion dete
c
tion
[15] is import
ant for Mosq
uito detection
. The
template image is i
m
porta
n
t for post processi
ng
whi
c
h
ca
n b
e
an
alyze
d
f
o
r m
o
rpholo
g
i
cal filte
r
ing
to elimi
nate t
he tiny
noisy
re
gion. Im
a
g
e
filtering i
s
im
portant to
eli
m
inate the
no
ise of i
m
age
and
smo
o
thin
g the im
age t
o
get the
efficient
pixel value. T
he ba
ckgrou
nd segme
n
ta
tion woul
d
b
e
helpful by
usin
g thre
sh
o
l
ding an
d re
g
i
o
n
gro
w
ing
p
r
o
c
ess
whi
c
h i
s
helpful
for feature
segm
entation. T
h
e
body
pa
rts
of Mo
squito
are
compl
e
x thus flying Mosqu
i
to detection i
s
very diffi
cult. In this case
the model cre
a
tion and bo
dy
parts
analy
s
i
s
is imp
o
rta
n
t
to analyze
and dete
c
t m
o
sq
uitoe
s
. Th
e legs
of Mosquito
es a
r
e
no
t
the same a
s
other in
se
cts.
The seg
m
en
tation pr
o
c
e
s
s is u
s
eful fo
r mosquitoe
s
’
detection. T
h
e
sha
pe of th
e Mosquito
can
be
capt
ured
by mo
del setting. Espe
cially h
e
rein th
e rotated
recta
ngul
ar
model ha
s b
een create
d
to detect the
flying or moving Mosquito.
The abd
o
mi
nal
parts are
special stylist
such as white an
d black-gray segm
ented
color line.
3. Rese
arch
Metho
d
This stu
d
y consi
s
ts of two method
s: one is
an exte
rnal (b
ehavio
ral) an
alysi
s
and the
other meth
od
is an imag
e pro
c
e
ssi
ng a
nalysi
s
. The i
m
age p
r
o
c
e
s
sing p
a
rt wa
s done by Mat
l
ab
softwa
r
e. No
ise wa
s eli
m
inated
f
r
o
m
the
f
ilm by
a
n
o
ise reje
ction met
hod usi
ng
i
m
age
pro
c
e
ssi
ng.
(a)
(b)
(c
)
(d)
(e)
(f)
(g)
(h)
(i)
(j)
(k
)
(l)
(m)
(n)
(o)
(p)
Figure 5. Mosquitoe
s
pattern image
s (i)-(p)
from (a)-(h
) gray imag
es (of Figure 1)
The noi
se ex
traction a
nd rejectio
n pro
g
r
am was al
so
develope
d with Matlab. The cam
e
ra was
use
d
to
differentiate m
o
sq
uitoes,
bee
s
and flie
s. T
h
en the
ima
g
e
processin
g
data
whi
c
h
was
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Cha
r
a
c
teri
stics Analysi
s
an
d Dete
ction Algor
ithm
of Mosq
uitoe
s
(Ja
hangi
r Alam
S.M.)
5373
colle
cted on
the mosq
uito
es, flies and
bees was o
r
gani
ze
d and
analyzed through stati
s
tical
method
s
su
ch a
s
histog
ra
ms th
at p
r
od
uce
d
the
a
r
e
a
, mea
n
, me
dian, a
nd
sta
ndard
deviation
etc. After a
detailed
ch
aracte
ri
stic stu
d
y derived f
r
om the
film
s
images was collected,
a
recogni
za
ble
behavio
ral
an
d territorial
p
a
ttern e
m
er
g
e
d. The
in
se
cts’ lo
cation
s
and th
eir
pattern
cha
r
a
c
teri
stics ratios can
be fou
nd in
the hi
stog
ram
analy
s
is. T
h
e hi
stogram
wa
s a
pplied
to
analyze in
se
ct si
ze,
num
ber of o
b
je
cts, p
r
ob
os
ci
s, body
sh
ape,
col
o
r,
anten
n
ae
and
sha
p
es,
etc.
Whether at home or out of
home, one can gr
ab Mosquito(s) and put it/them inside the
15cm x13
c
m
x2cm Mosq
uito box. The images
a
r
e captu
r
ed by GigE 4900 a
nd Keyence XG
H20
0
M
cam
e
ra. Th
e exp
e
riment ha
s be
en a
nalyz
ed
by Hal
c
o
n
1
1
.
0 software.
Duri
ng th
e im
age
pro
c
e
ssi
ng, the media
n
filter is u
s
ed
to eliminate
the noisy
interfe
r
ence.
Features
whi
c
h
were in
clu
ded in th
e ta
ble
are ante
n
n
ae, probo
s
ci
s, hind l
e
g
s
, size an
d
body sha
pe
are dete
c
ted
by edge de
tection, st
ain
s
, pitch, edg
e width, edg
e angle, edg
e
positio
ns, sta
t
istical an
alysis and it
s cha
r
acte
ri
stics.
Fourie
r de
scri
p
t
ors, a
s
well as a diffe
renti
a
l
coeffici
ent a
nd di
stribute
d
pro
c
e
s
sing
were u
s
ed
in the patt
ern fo
r re
co
gnition [17].
The
mosq
uito’s
p
o
sition
al patt
e
rn
wa
s ob
se
rved an
d
the
Edge patte
rn
of the mo
sq
uito dete
c
ted
by
Matlab Edg
e
pro
c
e
ssi
ng
whi
c
h i
s
sho
w
n in
Figu
re
5, and Fi
gu
re 6. Th
e re
sults
sh
ow t
h
e
mosq
uito’s p
o
sition
al hi
stogra
m
. The
i
m
aging
p
r
ocess all
o
ws
re
sea
r
che
r
s to
determi
ne
wh
ere
the mosq
uito
is, and even
make predi
ctions ba
sed
on analyzi
ng
the pattern
s as to whe
r
e the
mosq
uito
will likely be
in t
he future a
s
i
n
sh
own in F
i
gure
5(i
)
-(p
)
. The hi
stog
ra
m analy
s
is
al
so
sho
w
e
d
that t
he mo
sq
uito’s territory i
s
smalle
r th
a
n
that of flies a
nd be
es. T
h
i
s
can
be
see
n
in
Figure 6.
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
(i)
(j)
(k
)
(l)
(m)
(n)
(o)
(p)
Figure 6. Statistics of Mosq
uitoes; (a
)-
(h
) histogram, a
nd (i)-(p) a
r
ea
histogram
From all the figure
s
of the territo
rial ed
ge
pa
tterns, it can be co
ncl
u
ded that a mosquito’
s
territori
al p
a
ttern
sh
ape i
s
narro
w an
d l
ong
whe
r
e th
e
pattern for flies i
s
alm
o
st
triang
ular or
‘V’
o
r
‘Z
’ or
‘C
ur
ve
’ or
‘
Λ
’ in
sh
ape
a
s
i
n
Figu
re
5(i
)
-(p
). Th
e p
a
tterns of Fli
e
s a
r
e
simil
a
r to
‘recta
ngul
ar’
or ‘a
rrow
hea
d’ sh
ape
as i
n
Figu
re 7
(
f)-(j). Fo
r Bee
s
,
it is ro
und
a
nd not ve
ry cl
ear.
These
re
sult
s a
r
e
sh
own
cle
a
rly in
Fi
gure
8
(
e)
-(h).
The
histo
g
ram of the
m
o
sq
uitoe
s
sh
ows
more
spikes i
n
con
s
iste
ncy
and
a n
a
rro
w
er wi
dth
th
an
flie
s a
n
d be
e
s
in
F
i
g
u
r
e 6
(
a)
-(
h)
, F
i
gu
r
e
7(k)-(o
)
, an
d
Figure 8
(
i)-(l
). It is rem
a
rkable that
th
e
histog
ram
reveals th
at the territory o
f
a
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5368 – 53
78
5374
mosq
uito is
much
sm
aller than that of
Flies a
nd B
ees i
n
Figu
re
6(i)-(p), Fig
u
r
e 7
(
p)-(t
), a
n
d
Figure 8(m
)
-(p).
(a)
(b)
(c)
(d
)
(e)
(f)
(g)
(
h
)
(i)
(j)
(k
)
(l)
(m)
(n)
(o)
(p)
(q)
(r)
(s
)
(t)
Figure 7. Pattern an
d Statistical Imag
es
of Flies;
(a
)-
(
e
) gr
ay
image
s
(fro
m Figu
r
e
2), (f)-
(j)
edge
s pattern
, (k)-(o
) hi
sto
g
ram,
an
d (p
)-(t) a
r
ea hi
sto
g
ram
(a)
(b)
(c
)
(d)
(e)
(f)
(g)
(h)
(i)
(j)
(k
)
(l
)
(m)
(n)
(o)
(p)
Figure 8. Pattern an
d Statistical Imag
es
of Bees
; (a
)-
(
d
) gr
ay
image
s (fro
m Figu
r
e
3), (e
)-
(h)
edge
s patter
n
, (i)-(l
) hist
ogr
am, and (m
)-
(
p
) ar
ea hi
stog
ram
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Cha
r
a
c
teri
stics Analysi
s
an
d Dete
ction Algor
ithm
of Mosq
uitoe
s
(Ja
hangi
r Alam
S.M.)
5375
4. Results a
nd Discu
ssi
on
From the
re
sults, it is evident that Mosq
ui
toes h
a
ve l
ong p
r
ob
osci
s than Flie
s a
nd Bee
s
.
The
b
ody sh
ape of
flies has
two ro
un
d
form
s,
a
n
d
they have
shorte
r a
nd th
icker l
e
g
s
th
an
Mosq
uitoe
s
. The abd
omin
al part of the Mosq
uito
is na
rro
w sl
e
nder a
nd White-G
r
ay stri
pe.
Ho
wever, th
e
flies h
a
ve n
o
White-G
r
ay
strip
e
on
th
e abd
ome
n
. The h
ead
of
the flies i
s
al
so
large
r
, a
s
i
s
t
he tho
r
ax. O
b
se
rvation
s
a
l
so
sho
w
th
at a flies colori
ng is a m
e
tall
ic blu
e
o
r
g
r
e
en
colo
r with b
r
o
w
n, da
rk g
r
ay
colo
ring
spre
ad thro
ugho
u
t. Its body length is le
ss th
an 12mm
whi
c
h
is small
e
r tha
n
a mosq
uito’
s
si
ze an
d bo
dy l
ength. Mathematically, it can be dete
r
mined that, the
body sh
ape o
f
fly is linear,
and the main
se
ctio
n
s
of the body are ful
l
y narro
w cyli
nders.
Bees are mo
re simila
r to flies than mo
squi
toe
s
. However, their bo
dy shape p
a
ttern is
linear in it
s
sp
heri
c
al
sh
ape
with
probo
sci
s
. Its
pr
o
b
o
s
cis i
s
sho
r
ter t
han th
at of m
o
sq
uitoe
s
’. T
he
body le
ngth i
s
n
o
rm
ally le
ss than
1
6
m
m
wherea
s t
he m
o
squito’
s
b
ody l
ength
is a
minimu
m of
16 mm. It wa
s al
so
observ
ed that the
b
ee ha
s ve
ry small yello
w
spi
k
e
s
on it
s
body. Tabl
e
4
summ
arie
s th
e physi
cal ch
ara
c
teri
stics
of Mosquito
e
s
, Flies a
nd Bees.
Table 4. Pattern Cha
r
a
c
teristics of Mosq
uitoes, Flie
s and Bee
s
Bod
y
Shape
Leg
White
segment
Proboscis
(mm)
Bod
y
Shape
Length
(mm)
Recognition
Ratio
Mosquitoes
Narro
w
Slender
Long
yes
Long
(~6)
Λ
or V or
Z or
cur
v
e
>16
Higher
F
lies
Fat
C
y
linder
short
no
Short (~2
)
Linear or Pa
rt of
rectangle
5~12
Lo
w
e
r
Bees
Fat
C
y
linder
short
no
Shorter
(<2)
Linear or Pa
rt of
cir
c
le
<16
Lo
w
e
r
Most of the mosq
uitoe
s
h
a
ve angula
r
sha
ped bo
die
s
with an ab
domen p
a
rt, thorax &
head. Th
e le
gs a
r
e lo
ng, the ab
dome
n
is sl
end
er
a
n
d
narro
w
with dark
colo
r t
houg
h some
of
them have white circul
ar li
nes. Th
e hin
d
legs
are
bo
th light and d
a
rk i
n
col
o
r a
nd are thin. The
minimum l
e
n
g
th of mo
sq
u
i
toes i
s
a
bou
t 16mm
whi
c
h is ve
ry imp
o
rtant fo
r pat
tern
re
cogniti
on
desi
gn. The
s
e re
cog
n
ition
ratios
of Mo
squitoe
s
are hi
gher
wh
ere
a
s the re
cog
n
ition ratio
of Flies
and Bee
s
are lowe
r as
given in Tab
l
e 4.
The angle bet
wee
n
the head
and ab
dome
n
is
approximatel
y ±135
o
(ø)
while that of the reverse
sid
e
whi
c
h chan
ges d
u
rin
g
flying is ±2
35
o
(ø)
.
Hen
c
e th
e b
o
d
y sh
ape i
s
li
ke
a ‘V o
r
Λ
’
but if it is vie
w
ed
with th
e
addition
of th
e hind
leg
s
, t
hen
the patte
rn i
s
more li
ke
a ‘
Z
’ a
s
sho
w
n
i
n
Figu
re
9(a)-(c). If the
tho
r
ax
shap
e of
the mo
squito
is
more
cu
rved
then it woul
d
seem to
be a
fly or bee
a
s
indicate
d in
Figure 9(d), a
nd Figu
re 9
(
e
)
.
The p
r
opo
se
d demo m
o
d
e
l of mosquit
oes fo
r this
pattern
sha
p
e ha
s bee
n
mentione
d in
the
cro
s
s section
of the pattern
as in Figu
re
9(a
)
.
(a)
(b)
(c
)
(d)
(e)
Figure 9. Model and Patte
rn Shap
e of Mosq
uito with
Flies and Be
es; (a
) model
of
Mosq
uito, (b)
sha
pe of Mosquito, (c) sh
a
pe of
Mosq
uito with hind le
g, (d) shap
e of Fly,
and (e
) shap
e of Bee
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5368 – 53
78
5376
All of the feature
s
have b
e
en co
nsi
dere
d
to
detect th
e edge
s
of th
e Mosquito a
s
sh
own
in Figure 10.
The re
co
gni
tion ratio of
Mosq
uito
e
s
, bee
s and flie
s are simil
a
r as explai
ne
d in
Table 5. It is assume
d th
at the re
co
gn
ition ratio
i
s
t
he majo
r axi
s
len
g
th divid
e
d by the mi
nor
axis length in
pixels. The major axis le
ngth is
the total length of head, thorax, and abdo
mi
nal
part of th
e m
o
sq
uito. The
prob
osci
s
ca
n be
igno
re
d
for la
ck of
Robotic visio
n
becau
s
e it i
s
so
thin an
d n
o
t
clea
rly visibl
e
for i
m
agin
g
whe
n
the
Mo
squito
is flying. The
mino
r axis length
has
been
con
s
id
e
r
ed the l
ength
of abdom
en
or tho
r
ax pa
rt
as
width of
maximum pix
e
l value in mi
nor
axis. This ratio will en
sure that the Mosq
uito and its flying po
sition
s.
Table 5. Re
cognition
Ratio
of Mosquitoe
s
, Flies a
nd Bees
Length (L
pixel)
Width (W pixel)
R
L/W
=L / W
Mosquitoes
284.606
53.815
5.289
220.272
35.203
6.257
285.732
56.991
5.032
193.204
32.692
5.909
250.276
46.791
5.349
183.114
43.782
4.153
278.785
53.345
5.226
185.483
24.925
7.442
F
lies
588.596
280.781
2.096
497.653
232.982
2.136
489.230
248.110
1.972
455.539
186.843
2.438
489.230
248.110
1.972
Bees
113.054
34.363
3.290
204.662
60.754
3.369
554.116
180.360
3.072
204.662
60.754
3.369
If the major axis length is
L in pixel, the mi
nor axis le
ngth is W in
pixel the reco
gnition
ratio
can
be
obtain
ed from R
L/
W
=L/
W
[18].
T
h
e
Mo
squitoe
s
,
Flie
s, an
d
Bees minim
u
m
recognitio
n
ra
tio are
4.15
3, 1.972,
and
3
.
072, re
sp
ecti
vely and th
e
maximum
re
cognition
ratio
s
are
7.442,
2.
438, a
nd
3.3
69, resp
ectiv
e
ly. Theref
ore it can
be
defined
that
the mo
squito
’s
recognitio
n
ra
tio is relativel
y
higher. After com
p
a
r
ing t
he length a
n
d
width, it ca
n be sig
n
ify that
the length i
s
bigge
r and t
he width i
s
smaller of
the
Mosq
uito. In this phe
nom
e
non it can
be
pro
c
ee
ded to
differentiate a
nd dete
c
t the Mosq
uito in F
i
gure 1
0
.
(a)
(b)
(c
)
(d)
Figure 10. Mosq
uito Dete
ction (a) o
r
igin
al image (b) create mo
del
with edg
es (c) sha
pe mo
de
l
based ed
ge
s detectio
n
(d
) correl
ation ba
sed mo
del de
tection
The shap
e a
nd co
rrelatio
n based a
nal
ysis can
b
e
confirme
d by the expe
rime
ntal tests
and it ca
n be
signified to
detect a
nd track po
sition
of the Mosqu
i
to as p
r
e
s
en
ted in Figu
re
11.
The stati
s
tica
l method can
determin
e
d
e
tection in
fo
rmation, part
s
, segme
n
tatio
n
, position
s
a
n
d
the moving
a
ngle of th
e m
o
sq
uito. The
all dete
r
mine
d inform
ation
woul
d b
e
he
lpful to de
stroy
the Mos
quito
by LASER
s
y
s
t
em.
The c
o
ntour
informatio
n of images
c
a
n be prov
ided
by Halcon
softwa
r
e. Du
ring the det
ection, the backg
ro
u
nd
of the image has be
en
subtra
cted
by
thresholdi
ng
and
regio
n
g
r
owi
ng m
e
tho
d
so th
at
it can be
dete
c
ted a
s
qui
ckly as p
o
ssibl
e
as
sho
w
n in Fig
u
re 11.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Cha
r
a
c
teri
stics Analysi
s
an
d Dete
ction Algor
ithm
of Mosq
uitoe
s
(Ja
hangi
r Alam
S.M.)
5377
(a)
(b)
(c
)
(d)
(e)
(f)
Figure 11. Backgro
und Se
gmentation fo
r Mosquito De
tection; (a
) filtering & mod
e
l cre
a
tion, (b
)
pattern dete
c
t
i
on, (c) thre
sh
old witho
u
t model, (d
) regi
on gro
w
in
g, (e) thre
sh
old o
f
original
image, and
(f) regi
on growi
ng image afte
r thre
shol
d ap
plied
After setting up the mode
l, the background
s
ubtraction method i
s
applie
d. Next, the
threshold ima
ge and the region g
r
owi
n
g image ru
n
pro
c
e
ss i
s
ad
opted to dete
ct the mosqu
i
to
one by o
ne in
the re
al flying sp
ace. Before d
e
tect
io
n, it is ne
ce
ssa
r
y to pro
c
e
s
s
the optimization
of statistical a
nalysi
s
. The
grab
bing im
a
ge ca
n
be p
r
oce
s
sed a
nd
the Mosquito
can b
e
dete
cted
.
The Mo
sq
uito are flying from on
e po
sit
i
on to a
nothe
r po
sition th
e
n
the rotated
re
ctang
ular
are
sea
r
ching the
pattern of the model in Fi
gure 1
2
.
(a)
(b)
(c
)
(d)
Figure 12. De
tected Mo
squ
i
to after Mode
l Setting; (a) model setting
, and (b)-(d) d
e
tected in
diffetent posit
ions d
u
rin
g
flying time
The wh
ole d
e
tection p
r
o
c
essed h
a
s b
een an
alyze
d
by Halcon
Software. About 20
Mosq
uitoe
s
were grabb
e
d
and put in
side the M
o
s
quito box for experime
n
ta
l exerci
se. In
this
process
the applied algorit
h
m
is
effectiv
e and the
statistical
re
sults
have shown stability.
Duri
n
g
detectio
n
p
r
o
c
e
ss, mi
nimu
m scor
e, max
i
mum sco
r
e,
detectio
n
tim
e
and
an
gula
r movem
ent
has
been
ob
se
rved a
s
give
n
Table
6. The
time to det
ect the
Mo
squito is very
impo
rtant. Each
Mosq
uito det
ection p
r
o
c
e
s
s minimum ti
me is 1.95 m
s
and the m
a
ximum extended time is 4
.
87
ms that i
s
eff
e
ctive an
d
stable
fo
r dete
c
tion of
Mo
sq
uito. It can
b
e
consi
d
red
as th
e si
gnifi
cant
results to de
sign a Mosquit
o
destroy con
t
roller [19].
Table 6. Statistical
Re
sults for Mosq
uito Dete
ction
Range
Score
Time for each
M
o
squito detection (ms)
Angle (
ø)
Minimum 60%
1.95
-1.21
o
Maximum
100%
4.87
248.04
o
Evaluation Warning : The document was created with Spire.PDF for Python.