I
AE
S
I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
Ar
t
if
icial
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
Vol.
14
,
No.
4
,
Augus
t
2025
,
pp.
2922
~
2934
I
S
S
N:
2252
-
8938
,
DO
I
:
10
.
11591/i
jai
.
v
14
.i
4
.
pp
29
22
-
29
34
2922
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
ai
.
iaes
c
or
e
.
c
om
E
st
imat
in
g b
r
oi
le
r
h
e
a
t
st
r
e
ss u
si
n
g
c
o
m
p
u
t
e
r
vi
si
on
an
d
m
ac
h
i
n
e
l
e
a
r
n
i
n
g
M
u
h
am
m
ad
I
q
b
al
Anggoro
Agun
g
1
,
E
k
o
M
u
r
s
it
o
B
u
d
i
2
,
M
irant
i
I
n
d
ar
M
an
d
as
ar
i
2
,
3
1
M
a
s
te
r
of
I
ns
tr
ume
nt
a
ti
on a
nd C
ont
r
ol
, F
a
c
ul
ty
of
I
ndus
tr
ia
l
T
e
c
hnol
ogy,
I
ns
ti
tu
t
T
e
knol
ogi
B
a
ndung,
B
a
ndung, I
ndone
s
ia
2
D
e
pa
r
tm
e
nt
of
E
ngi
ne
e
r
in
g P
hys
ic
s
, F
a
c
ul
ty
of
I
ndus
tr
ia
l
T
e
c
h
nol
ogy, I
ns
ti
tu
t
T
e
knol
ogi
B
a
ndung, B
a
ndung, I
ndone
s
ia
3
B
ui
ld
in
g P
hys
ic
s
R
e
s
e
a
r
c
h G
r
oup, F
a
c
ul
ty
of
I
ndus
tr
ia
l
T
e
c
hnol
ogy, I
ns
ti
tu
t
T
e
knol
ogi
B
a
ndu
ng, B
a
ndung, I
ndone
s
ia
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
J
ul
1
,
2024
R
e
vis
e
d
F
e
b
24
,
2025
Ac
c
e
pted
M
a
r
15
,
2025
T
o
o
p
t
i
m
i
ze
an
d
e
n
h
a
n
ce
t
h
e
eff
i
ci
e
n
cy
o
f
b
ro
i
l
er
c
h
i
ck
e
n
farmi
n
g
,
i
t
i
s
es
s
e
n
t
i
al
t
o
ma
i
n
t
ai
n
t
h
e
c
h
i
c
k
en
s
’
w
e
l
fare,
as
h
eat
s
t
re
s
s
ca
n
d
ecrea
s
e
g
ro
w
t
h
effi
ci
e
n
cy
.
T
h
e
t
emp
er
at
u
re
-
h
u
mi
d
i
t
y
i
n
d
e
x
(T
H
I)
i
s
a
k
e
y
i
n
d
i
ca
t
o
r
u
s
e
d
t
o
d
et
erm
i
n
e
i
f
ch
i
ck
e
n
s
are
e
x
p
er
i
en
c
i
n
g
h
e
at
s
t
res
s
.
Preci
s
i
o
n
l
i
v
es
t
o
c
k
farmi
n
g
(PL
F)
b
as
ed
o
n
co
m
p
u
t
er
v
i
s
i
o
n
i
s
o
n
e
met
h
o
d
t
h
at
can
as
s
i
s
t
farmers
i
n
co
n
t
i
n
u
o
u
s
l
y
a
n
d
au
t
o
ma
t
i
ca
l
l
y
m
o
n
i
t
o
ri
n
g
t
h
e
c
o
n
d
i
t
i
o
n
o
f
t
h
e
i
r
ch
i
c
k
en
s
.
T
h
i
s
res
earch
d
e
v
el
o
p
e
d
a
co
mp
u
t
er
v
i
s
i
o
n
-
b
as
e
d
PL
F
s
y
s
t
em
t
o
o
b
s
erv
e
ch
i
c
k
en
s
w
i
t
h
CP
7
0
7
s
t
rai
n
i
n
a
co
mmerci
a
l
farm
u
s
i
n
g
t
h
e
Mas
k
reg
i
o
n
-
b
as
e
d
co
n
v
o
l
u
t
i
o
n
al
n
e
u
ral
n
e
t
w
o
rk
(
Mas
k
R
-
CN
N
)
met
h
o
d
an
d
o
b
j
ect
t
rack
i
n
g
al
g
o
r
i
t
h
ms
t
o
an
a
l
y
ze
fea
t
u
re
s
s
u
ch
a
s
t
h
e
cl
u
s
t
er
i
n
d
e
x
,
u
n
re
s
t
i
n
d
ex
,
an
d
t
h
e
d
i
s
t
an
c
e
t
ra
v
el
e
d
b
y
b
r
o
i
l
ers
.
T
h
e
res
u
l
t
s
i
n
d
i
cat
e
d
t
h
at
al
l
feat
u
res
t
en
d
t
o
i
n
v
ers
el
y
co
rrel
a
t
e
w
i
t
h
t
h
e
T
H
I
v
a
l
u
e,
w
i
t
h
t
h
e
c
l
u
s
t
er
i
n
d
ex
s
h
o
w
i
n
g
t
h
e
mo
s
t
n
o
t
i
cea
b
l
e
t
en
d
en
c
y
.
A
d
d
i
t
i
o
n
a
l
l
y
,
i
t
w
as
f
o
u
n
d
t
h
a
t
ex
t
er
n
al
fact
o
rs
,
s
u
c
h
as
t
h
e
p
res
e
n
ce
o
f
farmers
aro
u
n
d
t
h
e
o
b
s
erv
a
t
i
o
n
area,
can
affect
t
h
e
ch
i
ck
e
n
s
'
b
eh
a
v
i
o
r,
al
t
h
o
u
g
h
t
h
e
cl
u
s
t
er
i
n
d
e
x
feat
u
re
i
s
rel
at
i
v
e
l
y
re
s
i
l
i
e
n
t
t
o
d
i
s
t
u
r
b
an
ce
s
i
f
t
h
e
o
p
erat
o
r
i
s
n
o
t
cap
t
u
re
d
b
y
t
h
e
camera.
It
w
as
c
o
n
c
l
u
d
ed
t
h
a
t
t
h
ere
i
s
a
re
l
at
i
o
n
s
h
i
p
b
et
w
een
t
h
e
feat
u
re
s
an
d
t
h
e
T
H
I
v
al
u
e;
h
o
w
e
v
er,
t
h
e
s
e
feat
u
res
are
n
o
t
y
e
t
s
u
ff
i
ci
e
n
t
t
o
d
i
s
t
i
n
g
u
i
s
h
t
h
e
co
n
d
i
t
i
o
n
o
f
ch
i
ck
e
n
s
u
n
d
er
h
i
g
h
an
d
l
o
w
T
H
I
co
n
d
i
t
i
o
n
s
i
n
r
eal
-
t
i
me.
K
e
y
w
o
r
d
s
:
B
r
oil
e
r
c
hicke
n
C
lus
ter
index
C
omput
e
r
vis
ion
I
ns
tanc
e
s
e
gmenta
ti
on
M
a
s
k
R
-
C
NN
P
r
e
c
is
ion
li
ve
s
tock
f
a
r
mi
ng
Unr
e
s
t
index
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
E
ko
M
ur
s
it
o
B
udi
De
pa
r
tm
e
nt
of
E
nginee
r
ing
P
hys
ics
,
F
a
c
ult
y
o
f
I
nd
us
tr
ial
T
e
c
hnology,
I
ns
ti
tut
T
e
knologi
B
a
ndung
St
.
Ga
ne
s
a
No.
10
,
L
e
ba
k
S
il
iwa
ngi,
C
oblong,
B
a
ndung,
J
a
wa
B
a
r
a
t
40132
,
I
ndone
s
ia
E
mail:
mu
r
s
it
o@it
b.
a
c
.
id
1.
I
NT
RODU
C
T
I
ON
T
he
c
ons
umpt
ion
of
poult
r
y
mea
t
a
s
a
pr
im
a
r
y
s
our
c
e
of
a
nim
a
l
pr
otein
globally
is
e
xpe
c
ted
to
incr
e
a
s
e
f
r
om
39.
4
%
in
2022
to
40
.
8%
by
2030
[
1]
.
T
he
r
e
f
o
r
e
,
maintaining
the
e
f
f
icie
nc
y
of
poult
r
y
gr
owth
is
c
r
uc
ial,
given
the
im
pa
c
t
of
he
a
t
s
tr
e
s
s
whic
h
c
a
n
hinder
the
g
r
owth
of
br
oil
e
r
c
hicke
ns
,
e
s
pe
c
ially
a
t
21
-
22
da
ys
of
a
ge
[
2]
,
[
3]
.
T
he
tempe
r
a
tur
e
-
humi
dit
y
index
(
T
HI
)
is
a
mea
s
ur
e
us
e
d
to
de
ter
mi
ne
the
he
a
t
s
tr
e
s
s
c
ondit
ion
in
b
r
oil
e
r
s
,
ba
s
e
d
on
tempe
r
a
tu
r
e
a
nd
humi
dit
y
va
lues
[
4]
,
[
5]
.
P
r
e
vious
s
tudi
e
s
ha
v
e
s
hown
that
the
opti
mal
T
HI
va
lue
f
o
r
br
oil
e
r
s
is
a
r
ound
21
°C
[
4]
,
[
5
]
.
F
ur
ther
mo
r
e
,
guidelines
f
or
b
r
oil
e
r
f
a
r
mi
ng
s
ugge
s
t
that
the
opti
mal
T
HI
va
lue
f
or
br
oil
e
r
s
a
ge
d
22
da
ys
is
a
r
ound
25
°C
[
6]
,
[
7
]
.
E
f
f
o
r
ts
to
opti
mi
z
e
p
r
oduc
ti
on
e
f
f
icie
nc
y
a
nd
a
ni
mal
we
lf
a
r
e
ha
ve
dr
iven
r
e
s
e
a
r
c
h
int
o
pr
e
c
is
ion
li
ve
s
tock
f
a
r
mi
ng
(
P
L
F
)
,
whic
h
e
na
bles
c
onti
nu
ous
a
nd
a
utom
a
ti
c
moni
tor
ing
us
ing
va
r
ious
s
e
ns
or
s
[
8]
.
I
mage
-
ba
s
e
d
a
ppr
oa
c
he
s
ha
ve
e
mer
ge
d
a
s
e
f
f
e
c
ti
ve
methods
to
s
tudy
poul
tr
y
be
ha
vior
unde
r
va
r
ying
e
nvir
onmenta
l
c
ondit
ions
[
9
]
.
T
he
s
e
methods
pr
ovide
non
-
invas
ive
ins
ight
s
int
o
poult
r
y
be
ha
vio
r
.
R
e
c
e
nt
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
E
s
ti
mating
br
oil
e
r
he
at
s
tr
e
s
s
us
ing
c
ompute
r
v
is
io
n
and
mac
hine
…
(
M
uhamm
ad
I
qba
l
A
nggor
o
A
gu
ng
)
2923
s
tudi
e
s
ha
ve
uti
li
z
e
d
c
omput
e
r
vis
ion
a
nd
ma
c
hine
lea
r
ning
tec
hnologi
e
s
to
obtain
be
ha
vior
a
l
f
e
a
tur
e
s
in
poult
r
y
a
c
ti
vit
ies
.
F
or
ins
tanc
e
,
J
oo
e
t
al.
[
10
]
us
e
d
M
a
s
k
r
e
gion
-
ba
s
e
d
c
onvolut
ional
ne
ur
a
l
ne
twor
k
(R
-
C
NN
)
a
nd
YO
L
Ov4
model
s
to
c
las
s
if
y
nine
c
hicke
n
pos
tur
e
s
a
nd
be
ha
vior
s
,
while
Guo
et
al.
[
11]
c
ompar
e
d
f
ive
c
onvolut
ional
ne
ur
a
l
ne
twor
k
(
C
N
N)
-
ba
s
e
d
mac
hine
lea
r
ning
models
to
de
tec
t
f
ou
r
dif
f
e
r
e
nt
c
hicke
n
be
ha
vior
s
.
M
or
e
ove
r
,
E
ij
k
e
t
al.
[
12]
e
va
luate
d
c
omput
a
ti
ona
l
r
e
s
our
c
e
e
f
f
icie
nc
y
in
de
tec
ti
ng
br
oil
e
r
s
by
c
ompar
ing
M
a
s
k
R
-
C
NN
a
nd
U
-
Ne
t
m
ode
ls
.
Additi
ona
ll
y,
M
a
s
s
a
r
i
e
t
al.
[
2]
e
va
luate
d
th
e
c
lus
ter
index
a
nd
unr
e
s
t
index
in
br
oil
e
r
s
unde
r
he
a
t
s
tr
e
s
s
by
s
e
gmenting
im
a
ge
s
ba
s
e
d
on
mor
phologi
c
a
l
ope
r
a
ti
ons
,
while
L
in
e
t
a
l.
[
13
]
e
xa
mi
ne
d
the
c
or
r
e
lation
be
twe
e
n
T
HI
va
lues
a
nd
c
hicke
n
m
ove
ment
de
tec
ted
us
ing
F
a
s
ter
R
-
C
NN
.
De
s
pit
e
s
e
ve
r
a
l
pr
e
vious
s
tudi
e
s
,
ther
e
h
a
s
les
s
r
e
s
e
a
r
c
h
o
n
c
omm
e
r
c
ial
pou
lt
r
y
f
a
r
ms
us
e
d
in
the
f
a
r
mi
ng
indus
tr
y
a
s
mos
t
s
tudi
e
s
ha
ve
be
e
n
c
onduc
ted
in
a
c
ont
r
oll
e
d
labor
a
to
r
y
e
nvir
onment
.
F
a
c
to
r
s
s
uc
h
a
s
na
tur
a
l
s
unli
ght
a
nd
lar
ge
r
incons
is
tent
number
s
of
c
hicke
ns
typi
c
a
l
in
c
omm
e
r
c
ial
f
a
r
ms
ha
ve
not
be
e
n
c
ons
ider
e
d
in
p
r
e
vious
r
e
s
e
a
r
c
h.
I
n
thi
s
r
e
s
e
a
r
c
h,
br
oil
e
r
wi
th
C
P
707
s
tr
a
in
is
obs
e
r
ve
d
on
a
c
o
mm
e
r
c
ial
f
a
r
m
to
e
xplor
e
a
nd
a
na
lyze
the
r
e
lation
be
twe
e
n
T
HI
va
lues
a
nd
br
oil
e
r
a
c
ti
vit
ies
r
e
pr
e
s
e
nted
a
s
th
e
c
lus
ter
index,
unr
e
s
t
index,
a
nd
a
ve
r
a
ge
br
oi
ler
move
ment
obtaine
d
us
ing
M
a
s
k
R
-
C
NN
model
f
or
ins
tanc
e
s
e
gmenta
ti
on
to
de
tec
t
ind
ivi
dua
l
c
hicke
ns
.
De
tec
ted
c
hicke
ns
a
r
e
t
r
a
c
ke
d
us
ing
a
n
objec
t
t
r
a
c
king
a
l
gor
it
hm
a
nd
the
r
e
s
ult
w
as
c
ompar
e
d
with
pr
e
vio
us
r
e
s
ult
s
to
tes
t
whe
ther
labor
a
tor
y
r
e
s
ult
s
c
a
n
be
a
ppli
e
d
in
c
omm
e
r
c
ial
f
a
r
ms
.
I
f
s
uc
c
e
s
s
f
ul,
thes
e
f
e
a
tur
e
s
will
be
us
e
d
to
de
ve
lop
a
s
ys
tem
f
or
de
tec
ti
ng
c
hicke
n
c
ondit
ions
unde
r
he
a
t
s
tr
e
s
s
.
2.
M
E
T
HO
D
I
n
thi
s
r
e
s
e
a
r
c
h,
the
c
omput
e
r
v
is
ion
-
ba
s
e
d
P
L
F
s
y
s
tem
a
r
c
hit
e
c
tur
e
s
hown
in
F
igur
e
1
wa
s
us
e
d
to
obtain
s
e
gments
of
e
a
c
h
c
hicke
n
to
be
tr
a
c
ke
d.
T
h
e
M
a
s
k
R
-
C
NN
ins
tan
c
e
s
e
gmenta
ti
on
model
wa
s
pr
opos
e
d
a
s
a
method
to
r
e
c
ognize
indi
vidual
c
hicke
ns
due
to
it
s
good
pe
r
f
o
r
manc
e
in
te
r
ms
of
inf
e
r
e
nc
e
s
p
e
e
d
a
nd
se
gmenta
ti
on
a
c
c
ur
a
c
y
pr
oduc
e
d
[
14]
.
T
he
s
e
gmenta
ti
on
r
e
s
ult
s
a
r
e
us
e
d
f
or
the
tr
a
c
ke
r
,
a
nd
f
e
a
tur
e
e
xtr
a
c
ti
on
will
be
c
a
r
r
ied
out
in
the
f
or
m
of
the
unr
e
s
t
index,
c
lus
ter
index,
a
nd
kinema
ti
c
f
e
a
tur
e
s
s
uc
h
a
s
the
a
ve
r
a
ge
dis
plac
e
ment
of
tr
a
c
ke
d
c
hicke
ns
.
F
inally
,
c
or
r
e
lation
a
na
lys
is
wa
s
c
a
r
r
ied
out
be
twe
e
n
the
f
e
a
tur
e
s
a
nd
the
mea
s
ur
e
d
T
HI
va
lue.
I
f
ther
e
is
a
c
or
r
e
lat
ion
be
twe
e
n
the
f
e
a
tur
e
s
a
nd
the
T
HI
va
lue,
a
s
upe
r
vis
e
d
lea
r
ning
-
ba
s
e
d
model
will
be
de
ve
loped
in
f
utur
e
r
e
s
e
a
r
c
h
to
model
c
hicke
n
b
e
ha
vior
unde
r
he
a
t
s
tr
e
s
s
c
ondit
ions
in
r
e
a
l
-
ti
me.
F
igur
e
1.
P
r
opos
e
d
s
ys
tem
a
r
c
hit
e
c
tur
e
f
or
br
oil
e
r
c
hicke
n
moni
tor
ing
2.
1.
E
xp
e
r
im
e
n
t
al
s
e
t
u
p
I
n
thi
s
s
tudy,
e
xpe
r
im
e
nts
we
r
e
c
onduc
ted
in
a
c
omm
e
r
c
ial
c
hicke
n
f
a
r
m
mea
s
ur
ing
120×
12×
2
m
loca
ted
in
S
uba
ng,
W
e
s
t
J
a
va
,
I
ndone
s
ia.
T
his
s
tu
dy
obs
e
r
ve
d
br
oil
e
r
s
with
s
tr
a
in
C
P
-
707
f
r
om
P
T
.
C
ha
r
oe
n
P
okpha
nd
I
ndone
s
ia
[
6]
.
B
r
oil
e
r
s
we
r
e
r
e
c
or
de
d
us
ing
a
n
I
P
C
a
mer
a
Unia
r
c
h
I
P
C
-
T
124
ins
talled
on
the
c
e
il
ing
of
the
f
a
r
m
with
a
r
e
s
olut
ion
a
nd
s
a
mpl
ing
r
a
ti
o
of
2560
×
1440@20
f
ps
.
T
he
I
P
C
a
mer
a
wa
s
p
os
it
ioned
a
s
s
hown
in
F
igur
e
s
2
(
a
)
a
nd
2
(
b)
,
c
a
ptur
ing
the
a
r
e
a
whe
r
e
c
hicke
ns
c
ons
is
tently
s
tay
in
f
r
a
me
d
ur
ing
the
br
ooding
pe
r
iod
,
including
e
quipm
e
nt
with
in
the
f
a
r
m.
T
he
I
P
c
a
mer
a
s
we
r
e
c
onne
c
ted
to
a
ne
two
r
k
video
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
292
2
-
2934
2924
r
e
c
or
de
r
(
NV
R
)
Da
hua
P
F
S
3010
-
8E
T
-
96
to
r
e
c
or
d
video
e
ve
r
y
da
y
f
r
om
06:00
to
20:00
(
loca
l
ti
me)
.
I
ndus
tr
ial
tempe
r
a
tu
r
e
a
nd
humi
di
ty
s
e
ns
or
s
(
S
M
7820B
)
we
r
e
c
onne
c
ted
to
a
mi
ni
-
PC
s
e
r
ve
r
to
r
e
c
or
d
da
ta
e
ve
r
y
s
e
c
ond.
T
he
s
e
r
ve
r
ti
me
a
nd
NV
R
we
r
e
th
e
n
s
ync
hr
onize
d
to
a
loca
l
ne
twor
k
ti
me
p
r
otoco
l
(
NT
P
)
s
e
r
ve
r
,
a
nd
a
ll
c
a
mer
a
a
nd
s
e
ns
or
r
e
c
or
ding
da
ta
wa
s
s
tor
e
d
on
the
s
e
r
ve
r
.
(
a
)
(
b)
F
igur
e
2.
E
xpe
r
im
e
nt
s
e
tup
o
f
:
(
a
)
c
a
mer
a
a
nd
s
e
r
ve
r
ins
tallation
a
nd
(
b)
loca
ti
on
of
c
a
mer
a
ins
tallatio
n
2.
2.
Dat
a
c
oll
e
c
t
ion
an
d
a
n
n
ot
at
ion
T
o
tr
a
in
the
ins
tanc
e
s
e
gmenta
ti
on
model
,
vi
de
o
da
ta
r
e
c
or
de
d
f
r
om
Augus
t
19,
2023
,
to
S
e
ptembe
r
23,
2023,
wa
s
us
e
d
a
s
the
da
ta
s
e
t.
A
s
a
mpl
e
of
50
im
a
ge
s
c
ontaining
br
oil
e
r
s
a
nd
3
kinds
of
c
oop
e
quipm
e
nt
wa
s
a
nnotate
d
a
s
s
hown
in
F
igu
r
e
3
us
ing
ope
n
-
s
our
c
e
labe
li
ng
s
of
twa
r
e
Any
L
a
be
li
ng.
F
r
om
th
e
s
e
s
a
mpl
e
d
im
a
ge
s
,
the
da
tas
e
t
wa
s
divi
de
d
int
o
tr
a
ini
ng,
va
li
da
ti
on
,
a
nd
tes
ti
ng
,
with
e
a
c
h
s
e
t
c
ontaining
da
ta
a
s
s
hown
in
T
a
ble
1.
T
he
tr
a
ini
ng
s
e
t
wa
s
us
e
d
to
tr
a
in
the
model
in
thi
s
r
e
s
e
a
r
c
h,
while
the
v
a
li
da
ti
on
s
e
t
wa
s
e
mpl
oye
d
dur
ing
the
t
r
a
ini
ng
p
r
oc
e
s
s
to
e
v
a
luate
model
pe
r
f
or
manc
e
a
nd
pr
e
ve
nt
ove
r
f
it
ti
ng
.
F
inally,
the
tes
ti
ng
s
e
t
wa
s
us
e
d
f
or
the
f
inal
e
va
luation
a
f
t
e
r
tr
a
ini
ng
to
a
s
s
e
s
s
the
pe
r
f
or
manc
e
of
the
t
r
a
ined
model.
F
igur
e
3.
Da
tas
e
t
a
nnotation
on
4
c
las
s
e
s
T
a
ble
1
.
Da
tas
e
t
s
pli
tt
ing
O
bj
e
c
t
T
r
a
in
in
g
(
30 I
mgs
(%)
)
V
a
li
da
ti
on
(
10 I
mgs
(%)
)
T
e
s
ti
ng
(
10 I
mgs
(%)
)
T
ot
a
l
(
50 I
mgs
(%)
)
f
e
e
dbowl
30
(
60)
10
(
20)
10
(
20)
50
(
100)
br
oi
le
r
3
,
010
(
59.54)
1
,
069
(
21.15)
976
(
19.31)
5
,
055
(
100)
dr
in
kl
in
e
220
(
62.14)
67
(
18.93)
67
(
18.93)
354
(
100)
f
e
e
dl
in
e
100
(
62.12)
31
(
19.25)
30
(
18.63)
161
(
100)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
E
s
ti
mating
br
oil
e
r
he
at
s
tr
e
s
s
us
ing
c
ompute
r
v
is
io
n
and
mac
hine
…
(
M
uhamm
ad
I
qba
l
A
nggor
o
A
gu
ng
)
2925
2
.
3
.
I
n
s
t
an
c
e
s
e
gm
e
n
t
at
io
n
an
d
m
u
lt
i
ob
j
e
c
t
t
r
a
c
k
in
g
T
he
s
e
gmenta
ti
on
p
r
oc
e
s
s
wa
s
c
onduc
ted
by
c
om
pa
r
ing
the
M
a
s
k
R
-
C
NN
model
[
14]
,
with
the
be
s
t
ba
c
kbone
then
us
e
d
a
s
the
tr
a
c
ke
r
f
or
mul
ti
-
objec
t
tr
a
c
king
us
ing
a
n
a
lgor
it
hm
ba
s
e
d
on
the
s
im
ple
on
li
ne
a
nd
r
e
a
lt
im
e
tr
a
c
king
(
S
OR
T
)
a
lgor
it
hm
[
15
]
.
T
his
s
t
udy
uti
li
z
e
d
a
c
omput
e
r
r
unning
Ubuntu
22.
04
o
pe
r
a
ti
ng
s
ys
tem
with
P
ython
3.
11
.
9
a
nd
P
ytor
c
h
2.
3
.
0
wi
th
GPU
s
uppor
t
ins
talled.
T
he
r
e
s
e
a
r
c
h
wa
s
c
onduc
ted
on
a
n
P
C
ins
talled
with
P
r
oc
e
s
s
or
I
ntel
i7
-
13700K,
128
GB
s
ize
d
R
AM
,
a
nd
a
n
Nvidia
Ge
F
or
c
e
R
T
X
40
70
GPU
with
12
GB
of
memor
y
.
2
.
3
.
1.
M
as
k
r
e
gion
-
b
as
e
d
c
on
volu
t
ion
a
l
n
e
u
r
al
n
e
t
wor
k
M
a
s
k
R
-
C
NN
is
a
n
e
xtens
ion
of
the
F
a
s
ter
R
-
C
N
N
objec
t
de
tec
ti
on
a
lgo
r
it
hm
[
16]
,
whic
h
incor
por
a
tes
a
s
e
mantic
s
e
gmenta
ti
on
a
lgor
it
h
m
us
ing
a
f
ull
y
c
onvolut
ional
ne
twor
k
(
F
C
N)
[
17
]
on
the
r
e
gion
of
int
e
r
e
s
t
(
R
oI
)
laye
r
f
o
r
objec
t
s
e
gmenta
ti
on
[
14]
.
I
n
thi
s
s
tudy,
the
R
e
s
Ne
t50
ba
c
kbone
with
a
f
e
a
tur
e
pyr
a
mi
d
ne
twor
k
(
F
P
N)
[
18]
,
a
s
de
picte
d
i
n
F
igur
e
4,
wa
s
uti
l
ize
d.
Othe
r
than
that,
R
e
s
Ne
t
-
101
-
F
P
N
a
nd
R
e
s
Ne
Xt
-
101
-
F
P
N
ba
c
kbone
s
we
r
e
tr
a
ine
d
to
c
ompar
e
a
nd
e
va
luate
the
pe
r
f
or
manc
e
of
e
a
c
h
ba
c
kbone
model.
T
he
tr
a
ini
ng
p
r
oc
e
s
s
f
or
e
a
c
h
ba
c
kbone
w
a
s
c
onduc
ted
ove
r
3,
000
it
e
r
a
ti
ons
,
with
tr
a
ns
f
e
r
lea
r
ning
us
ing
model
that
ha
d
pr
e
vious
ly
be
e
n
p
r
e
tr
a
ined
o
n
the
M
S
C
OC
O
da
tas
e
t
we
r
e
pe
r
f
or
med
to
f
u
r
ther
tr
a
in
the
model
to
r
e
c
ognize
br
oil
e
r
s
us
ing
the
c
r
e
a
ted
da
tas
e
t.
I
nf
e
r
e
nc
e
us
ing
M
a
s
k
R
-
C
NN
pr
oduc
e
s
output
s
s
uc
h
a
s
objec
t
s
e
gmenta
ti
on
mas
ks
,
bounding
boxe
s
,
a
nd
objec
t
c
e
ntr
oid
c
oor
dinate
s
ba
s
e
d
on
the
s
e
gmenta
ti
on
r
e
s
ult
s
.
T
he
s
e
output
s
we
r
e
then
uti
li
z
e
d
a
s
the
t
r
a
c
ke
r
f
or
objec
t
t
r
a
c
king
us
ing
the
S
OR
T
-
ba
s
e
d
tr
a
c
king
a
lgor
it
hm.
F
igur
e
4.
M
a
s
k
R
-
C
NN
a
r
c
hit
e
c
tur
e
2
.
3
.
2.
S
i
m
p
le
on
li
n
e
an
d
r
e
alt
i
m
e
t
r
ac
k
in
g
S
OR
T
is
a
mul
ti
-
objec
t
tr
a
c
king
a
lgor
it
hm
that
ut
il
ize
s
the
Ka
lm
a
n
f
il
ter
[
19]
,
a
s
s
umi
ng
a
c
ons
tant
ve
locity
model
f
or
objec
t
mot
ion
[
15]
.
T
he
Ka
lm
a
n
f
il
ter
is
e
mpl
oye
d
to
e
s
ti
mate
c
ha
nge
s
in
objec
t
pos
it
ions
f
r
om
the
p
r
e
vious
f
r
a
me
a
nd
matc
h
them
with
de
tec
ti
ons
in
the
c
ur
r
e
nt
f
r
a
me.
Onc
e
matc
he
d,
thes
e
f
r
a
mes
a
r
e
us
e
d
to
upda
te
the
Ka
lm
a
n
f
il
ter
s
tate
[
15]
.
I
n
thi
s
s
tudy
,
the
tr
a
c
king
pr
oc
e
s
s
ba
s
e
d
on
th
e
S
OR
T
a
lgor
it
hm
wa
s
im
pleme
nted
us
ing
the
n
or
f
a
ir
f
r
a
m
e
wor
k
[
20]
,
a
s
il
lus
tr
a
ted
in
F
igur
e
5
.
F
igur
e
5.
S
OR
T
-
ba
s
e
d
tr
a
c
king
a
lgor
it
hm
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
292
2
-
2934
2926
I
n
thi
s
r
e
s
e
a
r
c
h,
the
c
e
ntr
oid
c
oor
dinate
s
f
r
om
the
s
e
gmenta
ti
on
r
e
s
ult
s
of
br
oil
e
r
c
las
s
we
r
e
us
e
d
a
s
f
e
a
tur
e
s
f
or
c
a
lcula
ti
ng
the
dis
tanc
e
be
twe
e
n
f
r
a
mes
.
T
he
c
los
e
s
t
objec
t
dis
tanc
e
s
mee
ti
ng
a
s
pe
c
if
ied
thr
e
s
hold
we
r
e
us
e
d
a
s
p
a
r
a
mete
r
s
to
matc
h
obje
c
ts
a
c
r
os
s
f
r
a
mes
.
T
he
tr
a
c
king
a
lgor
it
hm
output
s
objec
t
identit
ies
a
long
with
in
f
or
mation
of
s
e
gmenta
ti
on
mas
ks
,
bounding
boxe
s
,
a
nd
c
e
nt
r
oid
c
oo
r
dinat
e
s
f
or
e
a
c
h
tr
a
c
ke
d
f
r
a
me.
2
.
4
.
F
e
a
t
u
r
e
e
xt
r
ac
t
ion
T
o
model
br
oil
e
r
c
ondit
ions
dur
ing
he
a
t
s
tr
e
s
s
,
f
e
a
tur
e
e
xtr
a
c
ti
on
wa
s
pe
r
f
or
med
ba
s
e
d
on
the
tr
a
c
king
r
e
s
ult
s
.
T
his
s
tudy
f
oc
us
e
s
on
tes
ti
ng
a
nd
a
na
lyzing
thr
e
e
f
e
a
tur
e
s
,
na
mely
unr
e
s
t
index
[
21]
,
c
lus
ter
index
[
22]
,
a
nd
a
ve
r
a
ge
tr
a
ve
l
dis
tanc
e
[
13
]
.
T
he
f
ir
s
t
f
e
a
tur
e
,
unr
e
s
t
index
,
mea
s
ur
e
s
the
dif
f
e
r
e
nc
e
be
twe
e
n
s
e
ts
of
objec
ts
a
c
r
os
s
two
f
r
a
mes
withi
n
a
s
pe
c
if
ic
ti
me
int
e
r
va
l,
de
f
ined
by
(
1
)
.
U
nr
e
s
t
I
nd
e
x
(
i
,
i
−
1
)
=
.
dH
(
F
(
i
)
,
F
(
i
−
1
)
)
(
1)
W
he
r
e
dH
r
e
pr
e
s
e
nts
the
s
ymm
e
tr
ic
Ha
us
dor
f
f
dis
tanc
e
[
23]
be
twe
e
n
s
e
ts
of
c
e
ntr
oid
c
oor
dinate
s
of
ob
jec
ts
in
the
c
ur
r
e
nt
f
r
a
me
F
(
i
)
a
nd
the
pr
e
vious
f
r
a
me
F
(
i
−
1
)
.
is
the
pr
opor
ti
ona
li
ty
f
a
c
tor
of
the
im
a
ge
c
a
ptur
e
c
a
mer
a
,
de
f
ined
by
(
2)
.
=
2
t
a
n
(
/
2
)
(
2)
T
he
va
lue
of
the
pr
opor
t
ionalit
y
f
a
c
tor
de
pe
nds
on
the
c
a
mer
a
he
ight
,
lens
a
pe
r
tur
e
a
ngle
,
a
nd
the
pixel
width
of
the
c
ha
r
ge
-
c
oupled
de
vice
(
C
C
D)
s
e
ns
or
.
T
he
ne
xt
f
e
a
tur
e
is
the
c
lus
ter
index
,
whic
h
r
e
pr
e
s
e
nts
how
de
ns
e
ly
objec
ts
a
r
e
c
lus
ter
e
d
a
t
a
gi
ve
n
ti
me,
de
f
ined
by
(
3)
.
Cl
u
s
te
r
I
nd
e
x
(
i
)
=
2
×
̅
×
√
ℎ
2
+
2
̅
×
̅
×
−
1
(
3)
W
he
r
e
̅
a
nd
̅
a
r
e
the
a
ve
r
a
ge
a
r
e
a
a
nd
pe
r
im
e
ter
o
f
de
tec
ted
objec
t
s
e
gments
,
ℎ
a
nd
a
r
e
the
he
ight
a
nd
width
of
the
im
a
ge
,
̅
is
the
a
ve
r
a
ge
dis
tanc
e
be
t
we
e
n
c
e
ntr
oids
of
s
e
gments
,
a
nd
is
the
number
of
de
tec
ted
s
e
gments
.
T
he
f
inal
f
e
a
tur
e
to
be
tes
ted
is
the
a
ve
r
a
ge
tr
a
ve
l
dis
tanc
e
,
c
a
lcula
ted
us
ing
(
4)
:
A
v
g
.
Dis
t
(
i
,
i
−
1
)
=
∑
(
j
(
i
)
,
j
(
i
−
1
)
)
j
=
1
(
4)
W
it
h
is
the
dis
tanc
e
f
unc
ti
on
be
twe
e
n
objec
t
s
e
gments
f
o
r
index
j
in
f
r
a
me
i
a
nd
the
p
r
e
vious
f
r
a
me
(
i
−
1
)
.
Va
r
ious
metr
ics
c
a
n
be
us
e
d
f
or
;
howe
ve
r
,
in
th
is
s
tudy,
E
uc
li
de
a
n
dis
tanc
e
wa
s
us
e
d
to
c
a
lcula
te
objec
t
dis
plac
e
ment
be
twe
e
n
f
r
a
mes
.
F
or
the
tes
ti
ng
p
r
oc
e
s
s
,
videos
r
e
c
or
de
d
on
S
e
pte
mber
2,
2023
,
f
r
om
06:00
to
20:00
wi
th
22
-
da
y
-
old
br
oil
e
r
s
[
2
]
we
r
e
s
e
lec
ted
.
T
his
pe
r
iod
wa
s
c
hos
e
n
to
obs
e
r
ve
a
nd
t
r
a
c
k
c
hicke
n
be
ha
vio
r
du
r
ing
h
e
a
t
s
tr
e
s
s
c
ondit
ions
.
T
r
a
c
king
wa
s
c
onduc
ted
with
a
s
a
mpl
ing
int
e
r
va
l
o
f
1
s
e
c
ond
to
c
a
ptur
e
s
igni
f
ica
nt
c
ha
nge
s
in
br
oil
e
r
be
ha
vior
.
F
oll
owing
f
e
a
tur
e
e
xtr
a
c
ti
on,
f
ur
ther
a
na
lys
is
wa
s
pe
r
f
or
med
to
de
ter
mi
ne
c
or
r
e
lations
be
twe
e
n
thes
e
f
e
a
tur
e
s
a
nd
T
HI
va
lues
.
2
.
5
.
T
e
m
p
e
r
at
u
r
e
-
h
u
m
id
i
t
y
in
d
e
x
c
alcul
at
io
n
T
HI
is
a
me
tr
i
c
d
e
ve
lo
pe
d
to
a
s
s
e
s
s
t
he
r
mal
c
on
dit
i
ons
in
li
ve
s
tock
.
F
o
r
b
r
oi
ler
c
h
icke
ns
,
T
H
I
is
typi
c
a
ll
y
c
a
lcula
ted
a
s
a
l
inea
r
c
omb
inat
ion
o
f
dr
y
bulb
te
mpe
r
a
tu
r
e
(
)
a
nd
we
t
bu
lb
te
mpe
r
a
tu
r
e
(
)
with
s
pe
c
if
i
c
we
igh
ts
that
de
pe
nd
on
t
he
type
o
f
l
ives
toc
k
be
ing
obs
e
r
ve
d
.
F
or
b
r
oi
ler
s
,
T
HI
is
de
f
ine
d
by
(
5
)
[
4
]
.
=
0
.
85
+
0
.
15
(
5)
I
n
thi
s
s
tudy,
the
we
t
bu
lb
tempe
r
a
tu
r
e
(
)
is
a
ppr
oxi
mate
d
us
ing
the
e
mpi
r
ica
l
e
qua
ti
on
given
by
(
6
)
[
2
4]
:
=
at
an
[
0
.
151977
(
%
+
8
.
313659
)
1
/
2
]
+
at
an
(
+
%
)
−
at
an
(
%
−
1
.
676331
)
+
0
.
00391838
(
%
)
3
2
at
an
(
0
.
023101
%
)
−
4
.
686035
(
6)
W
it
h
is
the
dr
y
bulb
tempe
r
a
tu
r
e
r
e
a
ding
f
r
om
th
e
tempe
r
a
tur
e
s
e
ns
or
,
a
nd
%
is
the
r
e
lative
humi
dit
y
r
e
a
ding
f
r
om
the
humi
d
it
y
s
e
ns
or
.
T
he
s
e
va
lues
a
r
e
us
e
d
to
c
a
lcula
te
the
we
t
bulb
tempe
r
a
tu
r
e
(
)
,
whi
c
h
is
then
us
e
d
in
(
5
)
to
de
ter
mi
ne
the
T
H
I
va
lue
in
th
e
s
tudy.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
E
s
ti
mating
br
oil
e
r
he
at
s
tr
e
s
s
us
ing
c
ompute
r
v
is
io
n
and
mac
hine
…
(
M
uhamm
ad
I
qba
l
A
nggor
o
A
gu
ng
)
2927
3.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
3.
1.
I
n
s
t
an
c
e
s
e
gm
e
n
t
at
io
n
u
s
in
g
M
as
k
R
-
CNN
T
he
tr
a
ini
ng
pr
oc
e
s
s
f
or
e
a
c
h
ba
c
kbone
of
the
M
a
s
k
R
-
C
NN
model
wa
s
c
onduc
ted
ove
r
3
,
000
it
e
r
a
ti
ons
.
M
ode
l
e
va
luation
wa
s
pe
r
f
or
med
us
ing
the
int
e
r
s
e
c
ti
on
ove
r
union
(
I
oU)
metr
ic,
de
f
in
e
d
a
s
the
r
a
ti
o
of
the
int
e
r
s
e
c
ti
on
to
the
un
ion
of
the
r
e
f
e
r
e
nc
e
s
e
gment
(
)
a
nd
the
p
r
e
dicte
d
s
e
gment
(
)
,
in
(
7)
.
(
,
)
=
∩
∪
(
7)
I
oU
r
a
nge
s
f
r
om
0
to
1
,
whe
r
e
a
va
lue
of
0
indi
c
a
tes
no
ove
r
lap
be
twe
e
n
the
s
e
gments
,
while
a
va
lue
of
1
indi
c
a
tes
pe
r
f
e
c
t
ove
r
lap.
A
pr
e
diction
is
c
ons
ider
e
d
a
tr
ue
pos
it
ive
(
T
P
)
i
f
it
mee
ts
thr
e
e
c
r
i
ter
ia:
it
ha
s
the
s
a
me
c
las
s
a
s
the
r
e
f
e
r
e
nc
e
,
th
e
pr
e
diction
pr
oba
bil
it
y
e
xc
e
e
ds
a
pr
e
de
f
ined
th
r
e
s
hold,
a
nd
the
I
o
U
va
lue
e
xc
e
e
ds
a
s
pe
c
if
ied
thr
e
s
hold.
M
ode
l
pe
r
f
or
ma
nc
e
wa
s
e
va
luate
d
us
ing
the
C
OC
O
e
va
luation
method,
a
s
s
e
s
s
ing
a
ve
r
a
ge
pr
e
c
is
ion
(
AP)
va
lues
a
t
I
oU
t
hr
e
s
holds
r
a
nging
f
r
om
50%
to
95
%
in
inc
r
e
ments
of
5
%
.
I
n
thi
s
s
tudy,
A
P
e
va
luation
wa
s
c
onduc
ted
a
t
I
oU
thr
e
s
holds
of
50%
,
75%
,
a
nd
the
mea
n
AP
ove
r
t
he
r
a
nge
50
-
95%
mea
n
a
ve
r
a
ge
pr
e
c
is
ion
(
mAP
)
s
hown
in
T
a
ble
2.
T
a
ble
2.
E
va
luation
r
e
s
ult
f
r
om
t
r
a
ini
ng
M
a
s
k
R
-
C
NN
No
B
a
c
kbone
B
box mAP
B
box AP
50
B
box AP
75
S
e
gm m
A
P
S
e
gm AP
50
S
e
gm AP
75
1
R
e
s
N
e
t
-
50
-
FPN
74.8
83.5
81.6
75.2
83.5
81.5
2
R
e
s
N
e
t
-
101
-
FPN
70.6
81.6
79.7
73.2
81.6
80.2
3
R
e
s
N
e
X
t
-
101
-
FPN
71.6
80.0
76.8
71.3
80.0
77.6
B
a
s
e
d
on
T
a
ble
2,
f
or
the
tr
a
ini
ng
pr
oc
e
s
s
with
3,
000
it
e
r
a
ti
ons
,
the
be
s
t
r
e
s
ult
s
we
r
e
a
c
hieve
d
with
the
R
e
s
Ne
t
-
50
-
F
P
N
ba
c
kbon
e
,
s
howing
a
va
lue
of
2
-
3
higher
AP
c
ompar
e
d
to
the
ne
xt
be
s
t
ba
c
kb
one
.
T
his
pe
r
f
or
manc
e
im
pr
ove
ment
wa
s
obs
e
r
ve
d
both
in
bounding
box
e
va
luation
a
nd
s
e
gmenta
ti
on
r
e
s
ult
s
.
T
he
c
ompar
i
s
on
of
AP
e
va
luation
a
c
r
os
s
I
oU
thr
e
s
hold
va
r
iations
f
or
e
a
c
h
c
las
s
c
a
tegor
y
is
il
lus
tr
a
ted
in
F
igur
e
6.
B
a
s
e
d
on
F
igur
e
6
,
it
is
e
vident
that
AP
va
lues
f
o
r
e
a
c
h
ba
c
kbone
de
c
r
e
a
s
e
a
r
ound
the
I
oU
r
e
gion
of
~85%
.
Ge
ne
r
a
ll
y,
the
AP
va
lues
f
or
s
e
gmenta
ti
on
in
the
br
oil
e
r
c
las
s
a
r
e
be
tt
e
r
than
the
AP
va
lues
f
or
b
ounding
boxe
s
,
while
f
or
other
c
las
s
e
s
,
the
AP
va
lues
f
or
bounding
boxe
s
a
r
e
the
s
a
me
o
r
be
tt
e
r
than
thos
e
f
or
s
e
gmenta
ti
on.
T
his
indi
c
a
tes
that
the
s
e
gmenta
ti
o
n
r
e
s
ult
s
,
e
s
pe
c
ially
f
o
r
the
br
oil
e
r
c
las
s
,
a
r
e
mor
e
s
uit
a
ble
f
or
us
e
a
s
f
e
a
tur
e
s
in
the
objec
t
tr
a
c
king
p
r
oc
e
s
s
.
T
he
model
with
R
e
s
Ne
t
-
50
-
F
P
N
a
s
the
ba
c
kbo
ne
a
c
hieve
d
the
be
s
t
r
e
s
ult
s
,
pa
r
ti
c
ular
ly
f
o
r
the
b
r
oil
e
r
c
las
s
,
a
s
s
hown
in
F
igur
e
6(
a
)
.
At
a
n
I
oU
t
hr
e
s
hold
of
0.
70,
the
AP
va
lues
did
not
s
how
a
s
igni
f
ica
nt
de
c
li
ne
.
I
n
c
ontr
a
s
t,
models
with
R
e
s
Ne
t
-
101
-
F
P
N
a
nd
R
e
s
Ne
Xt
-
101
-
F
P
N
e
xhibi
t
a
no
table
de
c
li
ne
in
AP
va
lues
a
t
the
s
a
me
I
oU
th
r
e
s
hold,
a
s
il
lus
tr
a
ted
in
F
igur
e
s
6(
b
)
a
nd
6
(
c
)
.
Among
the
f
ou
r
c
las
s
e
s
tr
a
ined
with
va
r
ious
ba
c
kbone
s
,
the
dr
inkl
ine
c
las
s
pe
r
f
or
me
d
the
wo
r
s
t,
a
s
indi
c
a
ted
by
it
s
AP
va
lues
,
whi
c
h
we
r
e
c
ons
ider
a
bly
lowe
r
than
thos
e
of
the
othe
r
c
las
s
e
s
.
Additi
ona
ll
y,
the
f
indi
ngs
f
r
o
m
the
vis
ua
li
z
a
ti
on
r
e
s
ult
s
of
the
s
e
gmenta
ti
on
models
s
how
that
the
R
e
s
Ne
t
-
50
-
F
P
N
ba
c
kbone
(
F
igu
r
e
6(
a
)
)
p
r
oduc
e
s
be
tt
e
r
r
e
s
ult
s
with
f
e
we
r
e
r
r
o
r
s
c
ompar
e
d
to
other
ba
c
kbone
s
.
Othe
r
ba
c
kbone
s
(
F
igu
r
e
s
6(
b
)
a
nd
6(
c
)
)
e
xhibi
t
s
igni
f
ica
nt
e
r
r
or
s
.
T
he
s
e
include
in
c
or
r
e
c
tl
y
de
tec
ti
ng
pa
r
ts
of
the
ope
r
a
tor
c
a
ptur
e
d
on
c
a
mer
a
.
W
hich
a
r
e
les
s
pr
onounc
e
d
with
the
R
e
s
Ne
t
-
50
-
F
P
N
ba
c
kbone
.
3.
2.
F
e
a
t
u
r
e
c
om
p
ar
is
on
wi
t
h
T
HI
Af
ter
the
s
e
gmenta
ti
on
pr
oc
e
s
s
wa
s
s
uc
c
e
s
s
f
ull
y
c
onduc
ted,
f
e
a
tur
e
va
lues
we
r
e
c
a
lcula
ted
us
ing
(
1)
to
(
4
)
,
a
nd
the
T
HI
va
lues
we
r
e
c
omput
e
d
us
i
ng
(
5)
a
nd
(
6)
.
F
igu
r
e
7(
a
)
s
hows
the
T
H
I
g
r
a
ph
in
blue,
while
the
ba
r
gr
a
ph
r
e
pr
e
s
e
nts
the
his
togr
a
m
of
nor
malize
d
a
ve
r
a
ge
va
lues
f
or
e
a
c
h
f
e
a
tur
e
to
f
a
c
il
it
a
te
vis
ua
li
z
a
ti
on
a
nd
obs
e
r
va
ti
on,
with
a
95
%
c
on
f
idenc
e
int
e
r
va
l
(
C
I
)
.
I
t
c
a
n
be
obs
e
r
ve
d
that
be
twe
e
n
a
ppr
oxim
a
tely
9:30
a
nd
14
:30,
ther
e
we
r
e
f
luctua
t
ions
in
T
HI
va
lues
due
to
the
a
utom
a
ti
c
c
ontr
ol
s
ys
tem
in
the
f
a
r
m
,
whic
h
a
c
ti
va
ted
the
c
ooli
ng
pump
to
c
oo
l
the
incoming
a
ir
a
s
the
T
HI
ins
ide
the
f
a
r
m
r
os
e
t
oo
high,
r
e
a
c
hing
up
to
30
°C
.
B
a
s
e
d
on
the
obs
e
r
va
ti
on
of
the
f
e
a
tur
e
s
,
the
c
lus
ter
index
ge
ne
r
a
ll
y
e
xhibi
ted
a
n
inver
s
e
r
e
lations
hip
with
the
T
HI
va
lues
,
e
s
pe
c
ially
a
t
the
be
ginni
ng
whe
n
the
T
HI
va
lues
we
r
e
ve
r
y
low
,
c
ons
is
tent
with
the
f
indi
ngs
of
P
e
r
e
ir
a
’
s
[
22]
r
e
s
e
a
r
c
h.
T
he
u
nr
e
s
t
index
a
ls
o
s
howe
d
a
n
inver
s
e
tr
e
nd,
though
it
wa
s
les
s
pr
onounc
e
d
c
ompar
e
d
to
the
c
lus
ter
index
.
T
his
dif
f
e
r
e
nc
e
is
a
tt
r
ibut
e
d
to
the
les
s
s
igni
f
ica
nt
tempe
r
a
tur
e
va
r
iation
withi
n
the
f
a
r
m,
a
li
gning
with
the
f
indi
n
gs
of
Va
ll
e
e
t
al
.
[
21]
,
who
noted
that
the
unr
e
s
t
i
nde
x
f
or
br
oil
e
r
s
s
howe
d
s
igni
f
ica
nt
c
ha
nge
s
a
t
t
e
mper
a
tur
e
s
a
r
ound
35
°C
,
c
or
r
e
s
ponding
to
T
HI
v
a
lues
of
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
292
2
-
2934
2928
a
ppr
oxim
a
tely
33
to
34
°C
.
T
he
a
ve
r
a
ge
dis
plac
e
ment
index
wa
s
pr
im
a
r
il
y
noti
c
e
a
ble
be
twe
e
n
0
8:00
a
nd
16:00,
with
a
dis
tr
ibut
ion
s
im
il
a
r
to
the
f
indi
ngs
o
f
L
in
e
t
al
.
[
13
]
.
(
a
)
(
b)
(
c
)
F
igur
e
6.
C
ompar
is
on
of
ba
c
kbone
s
(
a
)
R
e
s
Ne
t
-
50
-
F
P
N,
(
b)
R
e
s
Ne
t
-
101
-
F
P
N,
a
nd
(
c
)
R
e
s
Ne
Xt
-
101
-
FPN
A
de
e
pe
r
a
na
lys
is
of
the
index
da
ta
s
a
mpl
e
d
pe
r
s
e
c
ond,
a
s
s
hown
in
F
igu
r
e
7
(
b)
,
r
e
ve
a
led
s
e
ve
r
a
l
ins
tanc
e
s
whe
r
e
f
e
a
tur
e
va
lues
s
igni
f
ica
ntl
y
incr
e
a
s
e
d.
F
ur
ther
a
na
lys
is
a
nd
obs
e
r
va
ti
on
of
the
video
r
e
c
or
dings
indi
c
a
ted
that
thes
e
pe
a
ks
c
or
r
e
s
ponde
d
to
ins
tanc
e
s
whe
r
e
a
n
ope
r
a
tor
or
f
a
r
mer
e
nter
e
d
the
f
a
r
m
a
nd
wa
s
c
a
ptur
e
d
on
c
a
mer
a
,
a
s
il
lus
tr
a
ted
in
F
igu
r
e
8.
Othe
r
pe
a
ks
in
the
a
ve
r
a
ge
dis
plac
e
ment
f
e
a
t
ur
e
we
r
e
due
to
s
udde
n
a
nd
s
im
ult
a
ne
ous
moveme
nts
of
the
c
hicke
ns
,
tr
igger
e
d
by
the
pr
e
s
e
nc
e
of
a
f
a
r
me
r
or
ope
r
a
tor
not
c
a
ptur
e
d
by
the
c
a
mer
a
.
T
his
s
ugge
s
ts
that
the
c
lus
ter
index
a
nd
unr
e
s
t
index
f
e
a
tu
r
e
s
a
r
e
mor
e
r
e
s
il
ient
to
e
xter
na
l
dis
tur
ba
nc
e
s
c
a
us
e
d
by
f
a
r
mer
s
o
r
ope
r
a
tor
s
in
the
da
ta
c
oll
e
c
ti
on
a
r
e
a
c
ompar
e
d
to
the
a
ve
r
a
ge
dis
plac
e
ment.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
E
s
ti
mating
br
oil
e
r
he
at
s
tr
e
s
s
us
ing
c
ompute
r
v
is
io
n
and
mac
hine
…
(
M
uhamm
ad
I
qba
l
A
nggor
o
A
gu
ng
)
2929
(
a
)
(
b)
F
igur
e
7.
T
H
I
c
a
lcula
ti
on
(
a
)
c
umul
a
ti
ve
(
b)
s
a
mpl
e
d
e
ve
r
y
s
e
c
ond
F
igur
e
8.
E
xa
mpl
e
of
ope
r
a
tor
c
a
ptur
e
d
on
c
a
mer
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
292
2
-
2934
2930
T
he
c
lus
ter
index
is
mor
e
r
e
s
is
tant
to
dis
tur
ba
nc
e
s
be
c
a
u
s
e
it
s
c
a
lcula
ti
on
is
not
ba
s
e
d
on
two
dif
f
e
r
e
nt
f
r
a
me
c
ondit
ions
.
T
he
c
lus
ter
index
va
lue
only
incr
e
a
s
e
s
if
the
f
a
r
mer
or
ope
r
a
tor
is
c
a
ptur
e
d
by
the
c
a
mer
a
,
c
a
us
ing
many
c
hicke
ns
to
move
a
wa
y
a
nd
n
ot
be
c
a
ptur
e
d.
A
r
e
duc
e
d
number
of
c
hick
e
ns
(
)
r
e
s
ult
s
in
a
mi
n
im
um
va
lue
in
the
de
nomi
na
to
r
of
(
3)
,
lea
ding
to
a
n
incr
e
a
s
e
in
the
c
lus
ter
index
va
l
ue
.
I
f
the
f
a
r
mer
or
ope
r
a
tor
is
not
c
a
ptur
e
d
by
the
c
a
mer
a
,
the
c
lus
ter
index
va
lue
is
not
s
igni
f
ica
ntl
y
a
f
f
e
c
te
d
e
ve
n
if
the
c
hicke
ns
move,
be
c
a
us
e
they
a
r
e
s
ti
ll
de
tec
ted
a
nd
c
a
ptur
e
d
by
the
c
a
mer
a
.
I
n
c
ontr
a
s
t,
both
the
unr
e
s
t
index
a
nd
a
ve
r
a
ge
d
is
plac
e
ment
f
e
a
tur
e
s
de
pe
nd
on
the
c
ompar
is
on
be
twe
e
n
the
two
f
r
a
mes
.
W
he
ther
the
ope
r
a
to
r
is
c
a
ptur
e
d
or
not,
if
the
ope
r
a
tor
o
r
f
a
r
me
r
is
ne
a
r
the
c
hi
c
ke
ns
a
nd
c
a
us
e
s
incr
e
a
s
e
d
moveme
nt,
both
in
de
x
va
lues
will
r
is
e
.
W
he
n
c
ompar
ing
the
unr
e
s
t
i
nde
x
a
nd
a
ve
r
a
ge
dis
plac
e
ment,
the
unr
e
s
t
index
is
mo
r
e
r
e
s
is
tant
to
dis
tur
ba
nc
e
s
be
c
a
us
e
it
us
e
s
the
Ha
us
dor
f
f
dis
tanc
e
,
whic
h
only
c
ompar
e
s
the
s
pr
e
a
d
of
c
e
ntr
a
l
point
s
be
twe
e
n
two
f
r
a
mes
without
c
ons
id
e
r
ing
the
identit
y
of
e
a
c
h
objec
t.
F
or
e
xa
mpl
e
,
i
f
two
obje
c
ts
s
witch
pos
it
ions
be
twe
e
n
two
f
r
a
mes
,
the
unr
e
s
t
index
c
a
lcula
ti
on,
whic
h
doe
s
not
a
c
c
ount
f
or
the
identi
ty
of
e
a
c
h
objec
t,
r
e
s
ult
s
in
a
mi
nim
a
l
Ha
us
dor
f
f
dis
tanc
e
e
va
luation
due
to
the
ne
a
r
ly
identica
l
pos
it
ions
.
I
n
c
ontr
a
s
t,
the
a
ve
r
a
ge
dis
plac
e
ment
c
a
lcula
ti
on
tr
a
c
ks
the
moveme
nt
of
objec
ts
by
maintaining
the
s
a
me
i
de
nti
ty;
thus
,
whe
n
pos
it
ions
a
r
e
s
witche
d,
ther
e
is
s
ti
ll
dis
plac
e
ment,
r
e
s
ult
ing
in
a
n
on
-
z
e
r
o
dis
plac
e
ment
va
lue.
T
o
a
na
lyze
f
ur
the
r
whe
ther
thes
e
thr
e
e
f
e
a
t
ur
e
s
c
a
n
c
las
s
if
y
the
c
ondit
ion
of
c
hicke
ns
unde
r
he
a
t
s
tr
e
s
s
,
a
n
e
xplo
r
a
tor
y
da
ta
a
na
lys
is
(
E
DA
)
o
f
the
f
e
a
tu
r
e
s
wa
s
c
onduc
ted
by
plot
ti
ng
mea
n
o
f
e
a
c
h
f
e
a
tur
e
s
on
e
a
c
h
T
HI
r
a
nge
,
c
a
lc
ulating
P
e
a
r
s
on
c
or
r
e
lation
ma
tr
ix
[
25]
to
de
ter
mi
ne
li
ne
a
r
c
or
r
e
lation
be
twe
e
n
the
va
r
iab
les
,
a
nd
las
tl
y
plot
ti
ng
a
ll
the
f
e
a
tur
e
s
on
3D
s
c
a
tt
e
r
plot
,
with
T
HI
va
lues
a
s
the
he
a
tm
a
p
of
the
da
ta
po
int
.
F
igur
e
9
s
hows
mea
n
of
e
a
c
h
f
e
a
tur
e
on
dif
f
e
r
e
nt
T
HI
r
a
nge
s
,
ha
ving
ne
ga
ti
ve
tr
e
nd
on
c
lus
ter
index
a
nd
unr
e
s
t
index
whic
h
a
li
gn
with
the
r
e
s
ult
s
f
r
om
pr
e
vious
r
e
s
e
a
r
c
h
[
2]
,
[
21]
,
[
22]
,
while
dis
plac
e
ment
s
how
s
ne
ga
ti
ve
tr
e
nd
in
low
tempe
r
a
tur
e
s
a
nd
pos
i
ti
ve
tr
e
nd
on
mi
ddle
a
nd
higher
tempe
r
a
tur
e
.
T
his
ha
ppe
n
s
be
c
a
us
e
ther
e
w
as
dis
tur
ba
nc
e
f
r
om
the
ope
r
a
tor
s
whic
h
ma
d
e
the
c
hicke
ns
move.
He
nc
e
f
ur
ther
da
ta
pr
oc
e
s
s
ing
s
uc
h
a
s
da
ta
c
lea
ning
or
f
il
ter
ing
we
r
e
ne
e
de
d
to
pr
oc
e
s
s
the
da
ta
f
or
f
u
r
ther
us
e
.
F
igur
e
9.
Nor
malize
d
f
e
a
tur
e
s
mea
n
on
d
if
f
e
r
e
nt
T
HI
r
a
nge
T
he
P
e
a
r
s
on
c
or
r
e
lation
matr
ix
in
F
igur
e
10
s
how
s
that
c
lus
ter
index
ha
ving
a
ne
ga
ti
ve
c
or
r
e
lation
with
va
lue
of
-
0.
5
with
T
H
I
while
the
other
two
ha
ve
a
low
c
or
r
e
lation
with
T
HI
.
T
his
mea
ns
that
c
lus
ter
index
ha
s
s
ome
li
ne
a
r
c
or
r
e
lation
inver
s
e
ly
to
s
ome
e
xtend
with
T
HI
,
whic
h
c
or
r
e
s
pond
with
the
r
e
s
ult
s
f
r
om
F
igur
e
7.
Unr
e
s
t
index
a
ls
o
ha
s
a
ne
ga
ti
ve
c
or
r
e
lation
with
low
va
lue,
indi
c
a
ti
ng
that
thes
e
f
e
a
tur
e
s
a
r
e
ha
v
e
a
n
inver
s
e
c
or
r
e
lation
,
bu
t
not
li
ne
a
r
ly
c
or
r
e
late
d
with
T
H
I
.
S
im
il
a
r
r
e
s
ult
s
s
how
on
the
dis
p
lac
e
ment
f
e
a
tur
e
that
ha
ve
ve
r
y
low
c
or
r
e
lation
with
T
HI
,
be
c
a
u
s
e
of
the
s
a
me
dis
tur
ba
nc
e
by
the
ope
r
a
tor
s
a
r
ound
the
obs
e
r
va
ti
on
a
r
e
a
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
E
s
ti
mating
br
oil
e
r
he
at
s
tr
e
s
s
us
ing
c
ompute
r
v
is
io
n
and
mac
hine
…
(
M
uhamm
ad
I
qba
l
A
nggor
o
A
gu
ng
)
2931
F
igur
e
10.
P
e
a
r
s
on
c
or
r
e
lation
matr
ix
of
T
H
I
a
nd
e
a
c
h
f
e
a
tur
e
F
igur
e
s
11(
a
)
a
nd
11
(
b
)
s
how
that
c
ondit
ions
wit
h
high
(
r
e
d)
a
nd
low
(
blue)
T
H
I
va
lues
a
r
e
quit
e
ove
r
lapping,
making
it
dif
f
icult
to
dis
ti
nguis
h
the
c
hicke
n's
c
ondit
ion
unde
r
high
a
nd
low
T
HI
.
How
e
ve
r
,
the
dir
e
c
ti
on
of
c
lus
ter
index
a
xis
s
tands
out
a
s
it
c
a
n
be
tt
e
r
dif
f
e
r
e
nti
a
te
b
e
twe
e
n
high
a
nd
low
T
HI
c
ondit
ions
.
C
ompar
e
d
to
the
othe
r
two
a
xis
,
with
be
tt
e
r
vis
ibi
li
ty
of
the
s
e
pa
r
a
ti
on
be
twe
e
n
r
e
d
a
nd
blue
c
olor
.
(
a
)
(
b)
F
igur
e
11.
S
c
a
tt
e
r
p
lot
o
f
3
f
e
a
tur
e
s
on
(
a
)
3D
S
c
a
t
ter
P
lot
a
nd
(
b
)
2D
S
li
c
e
of
the
3D
plot
B
a
s
e
d
on
the
f
ind
ings
,
a
ll
th
r
e
e
f
e
a
tur
e
s
a
r
e
r
e
late
d
to
the
c
hicke
ns
'
c
ondit
ions
unde
r
h
igh
a
nd
low
T
HI
,
a
s
obs
e
r
ve
d
in
the
c
umul
a
ti
ve
e
va
luation
r
e
s
ult
s
in
F
igur
e
7
(
a
)
.
How
e
ve
r
,
a
mor
e
in
-
de
pth
a
na
lys
is
r
e
ve
a
ls
that
it
's
not
ye
t
c
lea
r
i
f
thes
e
f
e
a
tur
e
s
a
lon
e
a
r
e
s
uf
f
icie
nt
f
o
r
r
e
a
l
-
ti
me
c
hicke
n
c
ondit
ion
r
e
c
ognit
ion.
F
r
om
F
igur
e
s
7,
9
,
a
nd
11,
the
c
lus
ter
index
f
e
a
tur
e
a
ppe
a
r
s
to
ha
ve
a
be
tt
e
r
c
a
p
a
bil
it
y
to
dis
ti
nguis
h
c
hicke
n
c
ondit
ions
unde
r
high
a
nd
low
T
HI
c
ompa
r
e
d
to
th
e
other
two
f
e
a
tur
e
s
.
T
h
e
da
ta
s
e
g
me
nt
a
t
i
on
p
r
o
c
e
s
s
i
n
th
is
s
t
ud
y
,
by
u
s
in
g
t
he
M
a
s
k
R
-
C
NN
me
t
ho
d
ma
y
in
f
lue
nc
e
f
e
a
t
u
r
e
e
xt
r
a
c
t
i
on
.
P
r
e
v
i
ous
s
tu
di
e
s
us
e
d
ma
the
ma
t
ica
l
s
e
gm
e
n
ta
t
io
n
me
t
ho
ds
l
ik
e
f
i
lt
e
r
i
ng
,
ma
the
ma
t
ica
l
m
or
p
ho
lo
gy
o
pe
r
a
t
io
ns
,
a
nd
b
in
a
r
i
z
a
t
i
on
f
o
r
f
e
a
t
u
r
e
e
x
t
r
a
c
ti
on
o
f
t
he
c
l
us
te
r
in
de
x
a
nd
u
n
r
e
s
t
i
n
de
x
[
2
]
,
[
21
]
,
[
2
2]
.
T
he
s
e
methods
a
ll
ow
f
or
c
ons
is
tent
de
tec
ti
on
o
f
c
hicke
ns
a
s
long
a
s
they
ha
ve
a
dis
ti
nc
t
c
olo
r
c
ont
r
a
s
t
f
r
om
Evaluation Warning : The document was created with Spire.PDF for Python.