I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
E
lec
t
r
ical
an
d
Com
p
u
t
e
r
E
n
gin
e
e
r
in
g
(
I
JE
CE
)
Vol.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
,
pp.
569
~
579
I
S
S
N:
2088
-
8708
,
DO
I
:
10
.
11591/i
jec
e
.
v
15
i
1
.
pp
5
69
-
579
569
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
e
c
e
.
iaes
c
or
e
.
c
om
Hi
e
r
ar
c
h
ic
al
B
aye
si
an
o
p
t
imi
z
at
io
n
b
ase
d
c
o
n
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
f
or
c
h
e
st
X
-
r
ay
d
is
e
a
se
c
la
ssi
f
ic
at
io
n
B
h
ar
at
h
Ku
m
ar
Gowr
u
,
App
a
Rao
Gi
d
u
t
u
r
i,
A
n
u
r
ad
h
a
S
e
s
e
t
t
i
D
e
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ie
nc
e
a
nd E
ngi
ne
e
r
in
g
, F
a
c
ul
ty
of
E
ngi
ne
e
r
in
g, G
I
T
A
M
(
D
e
e
me
d t
o be
U
ni
ve
r
s
it
y)
, V
is
ha
k
a
pa
tn
a
m, I
ndi
a
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
un
3,
2024
R
e
vis
e
d
Aug
13,
2024
Ac
c
e
pted
Aug
20,
2024
Pn
eu
mo
n
i
a
i
s
an
i
n
fec
t
i
o
n
t
h
at
affec
t
s
t
h
e
l
u
n
g
s
,
ca
u
s
e
d
b
y
b
ac
t
er
i
a
o
r
v
i
r
u
s
e
s
i
n
h
al
e
d
t
h
r
o
u
g
h
t
h
e
ai
r,
l
ead
i
n
g
t
o
res
p
i
r
at
o
ry
p
r
o
b
l
ems
.
T
h
e
p
rev
i
o
u
s
res
earc
h
es
o
n
t
h
i
s
s
u
b
j
ec
t
h
av
e
l
i
m
i
t
a
t
i
o
n
s
o
f
h
i
g
h
d
i
me
n
s
i
o
n
a
l
feat
u
re
s
u
b
s
p
ace
an
d
o
v
erf
i
t
t
i
n
g
w
h
i
ch
m
i
n
i
mi
ze
t
h
e
c
l
as
s
i
fi
er
p
erfo
rma
n
ce.
In
t
h
i
s
res
earc
h
,
h
i
erarc
h
i
ca
l
Bay
es
i
an
o
p
t
i
mi
za
t
i
o
n
b
as
e
d
co
n
v
o
l
u
t
i
o
n
a
l
n
eu
ra
l
n
e
t
w
o
rk
(
H
BO
-
C
N
N
)
me
t
h
o
d
i
s
p
r
o
p
o
s
e
d
t
o
effe
ct
i
v
el
y
cl
a
s
s
i
fy
c
h
es
t
X
-
ray
d
i
s
ea
s
es
.
T
h
e
p
r
o
p
o
s
e
d
H
BO
al
g
o
r
i
t
h
m
o
p
t
i
mi
ze
s
h
y
p
er
p
arame
t
ers
o
f
CN
N
w
h
i
ch
mi
n
i
m
i
zes
t
h
e
o
v
erfi
t
t
i
n
g
i
s
s
u
e
an
d
en
h
a
n
ces
t
h
e
p
erfo
rma
n
ce
o
f
cl
a
s
s
i
fi
ca
t
i
o
n
.
T
h
e
h
y
b
r
i
d
Me
x
i
ca
n
a
x
o
l
o
t
l
o
p
t
i
m
i
zat
i
o
n
(MA
O
)
a
n
d
t
u
n
a
s
w
arm
o
p
t
i
m
i
zat
i
o
n
(T
SO
)
b
as
e
d
feat
u
re
s
el
ec
t
i
o
n
met
h
o
d
i
s
u
s
ed
f
o
r
s
el
ec
t
i
n
g
re
l
ev
a
n
t
feat
u
res
f
o
r
c
l
as
s
i
f
i
cat
i
o
n
t
h
at
mi
n
i
m
i
ze
s
t
h
e
h
i
g
h
d
i
me
n
s
i
o
n
al
feat
u
re
s
.
T
h
e
Res
N
e
t
5
0
met
h
o
d
i
s
u
s
e
d
fo
r
feat
u
re
ex
t
ract
i
o
n
t
o
ex
t
rac
t
h
i
erarc
h
i
cal
feat
u
res
fr
o
m
t
h
e
p
re
-
p
r
o
ces
s
ed
i
m
ag
es
t
o
d
i
fferen
t
i
a
t
e
t
h
e
cl
as
s
es
.
T
h
e
p
ro
p
o
s
ed
H
BO
-
C
N
N
t
ec
h
n
i
q
u
e
i
s
es
t
i
ma
t
ed
w
i
t
h
p
erfo
rma
n
ce
m
et
r
i
cs
o
f
acc
u
rac
y
,
p
rec
i
s
i
o
n
,
recal
l
,
an
d
F1
-
s
co
re.
T
h
e
p
ro
p
o
s
ed
met
h
o
d
at
t
ai
n
s
t
h
e
h
i
g
h
e
s
t
acc
u
racy
9
7
.
9
5
%
,
p
reci
s
i
o
n
9
2
.
0
0
%
,
recal
l
8
9
.
0
0
%
an
d
F
1
-
s
co
re
9
2
.
0
0
%
,
as
o
p
p
o
s
ed
t
o
t
h
e
co
n
v
e
n
t
i
o
n
a
l
met
h
o
d
s
,
d
eep
co
n
v
o
l
u
t
i
o
n
al
n
eu
ra
l
n
et
w
o
r
k
(D
C
N
N
).
K
e
y
w
o
r
d
s
:
C
onvolut
ional
ne
ur
a
l
ne
twor
k
Hie
r
a
r
c
hica
l
B
a
ye
s
ian
opti
mi
z
a
ti
on
M
e
xica
n
a
xolot
l
opti
mi
z
a
ti
on
P
ne
umoni
a
T
una
s
wa
r
m
opti
mi
z
a
ti
on
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
:
B
ha
r
a
th
Kuma
r
Gow
r
u
De
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ienc
e
a
nd
E
nginee
r
ing
,
F
a
c
ult
y
of
E
nginee
r
ing
,
GI
T
AM
(
De
e
med
to
b
e
Unive
r
s
it
y)
Vis
ha
ka
pa
tnam,
Andhr
a
P
r
a
de
s
h
,
I
ndia
E
mail:
bha
r
a
th
.
kumar
436@gmail
.
c
om
1.
I
NT
RODU
C
T
I
ON
Dur
ing
the
pa
nde
mi
c
,
C
OV
I
D
-
19
e
mer
ge
d
a
s
the
pr
im
a
r
y
global
e
mer
ge
nc
y,
im
pa
c
ti
ng
pe
ople
wor
ldwide.
T
he
vir
us
wa
s
t
r
a
ns
f
e
r
r
e
d
f
r
om
pe
r
s
on
to
pe
r
s
on
thr
ough
r
e
s
pir
a
tor
y
dr
oplets
a
nd
c
los
e
s
t
c
ontac
t
a
long
the
c
ontaminate
d
s
ur
f
a
c
e
[
1]
.
T
he
major
ge
ne
r
a
l
s
ympt
oms
of
c
ough,
f
e
ve
r
,
a
nd
dys
pne
a
s
howe
d
f
or
2
-
14
da
ys
a
f
ter
the
vir
us
e
xpos
ur
e
[
2]
.
C
he
s
t
X
-
r
a
ys
(
C
XR
s
)
a
nd
c
omput
e
d
tom
ogr
a
phy
(
C
T
)
s
c
a
ns
a
r
e
uti
li
z
e
d
to
s
c
r
e
e
n
the
c
he
s
t
X
-
r
a
y
dis
e
a
s
e
s
,
a
nd
f
o
r
the
e
va
luation
o
f
d
is
e
a
s
e
pr
ogr
e
s
s
ion
in
the
a
dmi
tt
e
d
c
a
s
e
s
in
the
hos
pit
a
l
[
3]
–
[
5]
.
How
e
ve
r
,
the
e
f
f
e
c
ti
ve
s
e
ns
it
ivi
ty
de
tec
ti
on
of
thor
a
c
ic
a
bnor
malit
ies
by
uti
li
z
ing
C
T
s
c
a
ns
ha
s
numer
ous
c
ha
ll
e
nge
s
[
6]
.
T
he
C
T
s
c
a
nne
r
s
a
r
e
non
-
por
table
a
nd
ne
e
d
e
quipm
e
nt
s
a
nit
iz
ing
a
nd
im
a
ging
r
ooms
be
twe
e
n
pa
ti
e
nts
diagnos
e
s
[
7]
,
a
nd
a
ls
o,
their
dos
e
r
a
diation
is
gr
e
a
ter
than
the
X
-
r
a
ys
.
I
n
c
ontr
a
s
t,
por
table
unit
s
of
X
-
r
a
ys
a
r
e
major
ly
a
va
il
a
ble
with
a
f
e
a
s
ibl
e
a
c
c
e
s
s
in
major
hos
pit
a
l
s
[
8]
.
I
n
numer
ous
c
a
s
e
s
,
the
c
li
nica
l
s
it
ua
ti
on
of
the
pa
ti
e
nt
doe
s
not
a
ll
ow
f
or
C
T
s
c
a
ns
,
a
nd
ther
e
f
o
r
e
,
the
C
XR
is
a
n
e
f
f
e
c
ti
ve
c
hoice
f
or
pr
im
a
r
y
a
s
s
e
s
s
m
e
nt
[
9]
.
Di
s
e
a
s
e
s
li
ke
C
OV
I
D
-
19
a
nd
pne
umoni
a
a
r
e
de
te
c
ted
a
nd
c
las
s
if
ied
by
us
ing
c
he
s
t
X
-
r
a
y
im
a
ge
s
.
R
e
c
e
ntl
y,
a
wide
r
a
nge
o
f
r
e
s
e
a
r
c
he
s
f
or
the
a
utom
a
ti
c
de
t
e
c
ti
on
of
pne
umoni
a
in
c
he
s
t
X
-
r
a
y
im
a
ge
s
ha
ve
be
e
n
de
ve
loped
with
de
e
p
lea
r
ning
methods
[
10
]
,
[
11]
.
T
he
c
las
s
if
ica
ti
on
of
pne
umoni
a
gives
r
e
leva
nt
a
tt
e
nti
on
to
pne
umoni
a
pa
ti
e
nts
.
Nume
r
ous
r
e
s
e
a
r
c
he
r
s
ha
ve
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
569
-
579
570
de
ve
loped
a
utom
a
ti
c
methods
f
o
r
pne
umoni
a
dis
e
a
s
e
c
las
s
if
ica
ti
on
us
ing
c
he
s
t
X
-
r
a
y
im
a
ge
s
[
12]
.
T
he
de
e
p
lea
r
ning
tec
hnique
is
a
c
ompl
e
tely
a
utom
a
ti
c
f
e
a
tu
r
e
lea
r
ning
a
nd
e
xtr
a
c
ti
on
method
that
c
ons
umes
mor
e
ti
me
f
or
the
c
ompl
e
te
of
tr
a
ini
ng
[
13]
.
He
nc
e
,
s
uc
h
s
olut
ions
a
r
e
not
r
obus
t
due
to
the
maxi
mi
z
e
d
a
mount
of
da
tas
e
ts
.
De
e
p
lea
r
ning
methods
li
ke
c
onvolut
i
ona
l
ne
ur
a
l
ne
twor
ks
(
C
NN
)
ha
ve
ga
ined
a
tt
e
n
ti
on
f
or
pne
umoni
a
c
las
s
if
ica
ti
on
be
c
a
us
e
of
their
good
a
c
c
ur
a
c
y
a
nd
r
e
pr
e
s
e
ntation
o
f
f
e
a
tur
e
s
[
14
]
,
[
15]
.
How
e
ve
r
,
the
e
xis
ti
ng
r
e
s
e
a
r
c
he
s
ha
ve
dr
a
wba
c
ks
of
high
dim
e
ns
ional
f
e
a
tur
e
s
ubs
pa
c
e
a
nd
ove
r
f
it
ti
ng
is
s
ue
s
whic
h
mi
nim
ize
the
c
las
s
if
ier
’
s
pe
r
f
o
r
manc
e
.
I
n
thi
s
r
e
s
e
a
r
c
h,
a
n
opti
mi
z
a
ti
on
-
ba
s
e
d
f
e
a
tur
e
s
e
lec
ti
on
method
is
e
mpl
oye
d
to
s
e
lec
t
r
e
leva
nt
f
e
a
tur
e
s
whic
h
mi
n
im
ize
the
high
dim
e
ns
ional
f
e
a
tu
r
e
s
.
T
he
uti
li
z
a
ti
on
of
hype
r
pa
r
a
mete
r
tuni
ng
of
c
las
s
if
ier
pa
r
a
mete
r
s
e
nha
nc
e
s
the
c
las
s
if
ica
ti
on
pe
r
f
or
manc
e
a
nd
mi
ni
mi
z
e
s
the
ove
r
f
it
ti
ng
is
s
ue
by
f
indi
ng
the
be
s
t
c
ombi
na
ti
on
o
f
hype
r
pa
r
a
mete
r
s
.
Ha
gha
nif
a
r
e
t
al
.
[
16]
in
tr
oduc
e
d
a
method
to
de
t
e
c
t
im
a
ge
f
e
a
tur
e
s
o
f
pne
umoni
a
by
u
ti
li
z
ing
the
de
e
p
c
onvolut
ional
ne
ur
a
l
ne
twor
k
in
a
huge
da
ta
s
e
t.
I
n
the
int
r
oduc
e
d
method
,
s
e
ve
r
a
l
c
he
s
t
X
-
r
a
y
im
a
ge
s
f
r
om
numer
ous
s
our
c
e
s
we
r
e
ga
ther
e
d
a
nd
the
lar
ge
s
t
publi
c
ly
a
c
c
e
s
s
ibl
e
da
tas
e
t
wa
s
pr
e
pa
r
e
d.
At
las
t,
the
pa
r
a
digm
of
t
r
a
ns
f
e
r
lea
r
ning
,
the
C
he
XN
e
t
me
thod
wa
s
us
e
d
f
or
de
ve
lopi
ng
the
C
OV
I
D
-
C
X
Ne
t.
T
he
int
r
oduc
e
d
method
de
tec
ted
c
or
ona
vir
us
pne
um
onia
de
pe
nding
on
the
r
e
leva
nt
s
igni
f
ica
nt
f
e
a
tur
e
s
with
a
c
c
ur
a
te
loca
li
z
a
ti
on.
How
e
ve
r
,
the
int
r
oduc
e
d
method
c
ontaine
d
high
dim
e
ns
ional
f
e
a
tur
e
s
,
i
nc
ludi
ng
ir
r
e
leva
nt
or
r
e
dunda
nt
f
e
a
tur
e
s
whic
h
mi
nim
ize
d
t
he
c
las
s
if
ica
ti
on
pe
r
f
or
manc
e
.
C
houa
t
e
t
al.
[
17
]
s
ugge
s
ted
a
potential
de
e
p
tr
a
ns
f
e
r
lea
r
ning
to
de
ve
lop
the
c
las
s
if
ier
f
or
de
tec
ti
ng
C
OV
I
D
-
19
pa
ti
e
nts
by
uti
li
z
ing
C
T
s
c
a
ns
a
nd
C
XR
im
a
ge
s
.
T
he
a
ugmenta
ti
on
o
f
da
ta
wa
s
uti
li
z
e
d
to
maximi
z
e
the
t
r
a
ini
ng
da
ta
s
ize
to
s
olve
the
ove
r
f
it
ti
ng
is
s
ue
a
nd
im
pr
oving
ge
ne
r
a
li
z
a
ti
on
c
a
pa
bil
it
y
of
the
method.
T
he
s
ugge
s
ted
method
include
d
pr
e
-
tr
a
ined
de
e
p
ne
ur
a
l
ne
twor
ks
s
uc
h
a
s
R
e
s
Ne
t
50,
I
nc
e
pti
onV3,
VG
GN
e
t
-
19,
a
nd
Xc
e
pti
on
by
uti
li
z
ing
da
ta
a
ugmenta
ti
on
method.
T
he
s
ugge
s
ted
method
ha
d
pr
e
f
e
r
a
ble
ge
ne
r
a
li
z
a
ti
on
a
bil
it
y
a
nd
r
ob
us
tnes
s
.
How
e
ve
r
,
the
s
ugge
s
ted
method
f
a
c
e
d
ove
r
f
it
t
in
g
is
s
ue
s
due
to
it
ha
ving
many
laye
r
s
a
nd
pa
r
a
mete
r
s
.
Agr
wa
l
a
nd
C
houdha
r
y
[
18]
p
r
e
s
e
nted
de
e
p
C
NN
de
pe
nding
on
the
s
tr
uc
tu
r
e
to
de
tec
t
C
OV
I
D
-
19
by
uti
li
z
ing
c
he
s
t
r
a
diogr
a
phs
.
T
he
da
tas
e
ts
we
r
e
uti
l
ize
d
f
or
tr
a
ini
ng
a
nd
tes
ti
ng
the
method
on
va
r
iou
s
publi
c
r
e
pos
it
or
ies
.
T
he
pr
e
s
e
nted
method
ha
d
high
a
c
c
ur
a
c
y
a
nd
the
de
tec
ti
on
of
C
OV
I
D
-
19
wa
s
c
a
r
r
ie
d
out
in
c
ons
ult
a
ti
on
with
a
medic
a
l
c
li
nicia
n.
None
thele
s
s
,
the
pr
e
s
e
nted
method
ha
d
les
s
e
r
c
las
s
i
f
ica
ti
on
pe
r
f
or
manc
e
due
to
ove
r
f
i
tt
ing
be
c
a
us
e
the
tr
a
in
ing
da
ta
c
ontaine
d
many
ir
r
e
leva
nt
f
e
a
tur
e
s
,
a
nd
did
not
c
ons
ider
im
a
ge
r
e
s
izing
be
c
a
us
e
the
dif
f
e
r
e
nt
dim
e
ns
ions
we
r
e
dif
f
icult
to
be
ha
ndled
.
Agga
r
wa
l
e
t
al.
[
19]
de
ve
loped
the
tr
a
ns
f
e
r
lea
r
ning
method
with
a
c
ombi
na
ti
on
of
f
ine
-
tuned
pa
r
a
mete
r
s
to
c
las
s
if
y
the
c
he
s
t
X
-
r
a
y
im
a
ge
s
.
T
he
c
li
ppe
d
a
da
pti
ve
his
togr
a
m
e
qua
li
z
a
ti
on
(
C
L
AH
E
)
wa
s
us
e
d
to
e
nha
nc
e
the
c
ontr
a
s
t
of
im
a
ge
s
.
F
ur
ther
,
a
ugmenta
ti
on
of
da
ta
wa
s
pe
r
f
or
med
to
a
void
the
o
ve
r
f
it
ti
ng
is
s
ue
in
the
method.
T
he
de
ve
loped
tr
a
ns
f
e
r
lea
r
ni
ng
methods
s
uc
h
a
s
M
obil
e
Ne
tV2,
R
e
s
Ne
t50,
I
nc
e
pti
onV3,
NA
S
Ne
tM
obil
e
,
VG
G16,
Xc
e
pti
on,
I
nc
e
pti
onR
e
s
Ne
tV2,
a
nd
De
ns
e
Ne
t121
we
r
e
pe
r
f
o
r
med.
T
he
d
e
ve
loped
method
a
tt
a
ined
be
tt
e
r
pe
r
f
or
manc
e
in
dif
f
e
r
e
nt
c
l
a
s
s
e
s
of
a
s
mall
da
tas
e
t.
B
ut
it
did
not
c
hoos
e
the
r
e
leva
nt
f
e
a
tur
e
s
,
r
e
s
ult
ing
in
high
dim
e
ns
ional
is
s
ue
of
f
e
a
tur
e
s
a
nd
poor
c
las
s
if
ica
ti
on
pe
r
f
o
r
manc
e
.
M
ous
a
vi
e
t
al.
[
20
]
int
r
oduc
e
d
a
C
NN
-
long
s
hor
t
-
ter
m
memor
y
(
L
T
S
M
)
de
ve
loped
f
o
r
e
xtr
a
c
ti
ng
f
e
a
tur
e
s
f
r
om
the
r
a
w
da
ta.
T
o
make
it
much
mo
r
e
r
e
a
li
s
t
ic
a
nd
uti
li
z
e
the
int
r
oduc
e
d
model
in
the
pr
a
c
ti
c
a
l
f
ield,
white
Ga
us
s
ian
nois
e
wa
s
a
dde
d
to
the
r
a
w
im
a
ge
s
.
M
or
e
ove
r
,
the
int
r
oduc
e
d
method
wa
s
tes
ted
a
nd
e
xa
mi
ne
d
on
s
ix
da
tas
e
t
s
a
nd
two
a
ddit
ional
da
ta
s
e
ts
.
T
he
int
r
oduc
e
d
method
mi
nim
ize
d
the
medi
c
a
l
c
os
t.
Ye
t,
the
int
r
oduc
e
d
method
did
not
r
e
s
ize
im
a
ge
s
,
a
nd
thes
e
im
a
ge
s
with
va
r
iant
a
s
p
e
c
t
r
a
ti
os
led
the
method
to
a
tt
a
ini
ng
a
m
ini
mi
z
e
d
pe
r
f
o
r
manc
e
.
F
r
om
the
ove
r
a
ll
a
na
lys
is
of
e
xis
ti
ng
methods
,
it
is
s
e
e
n
that
the
e
xis
ti
ng
methods
ha
ve
li
mi
tations
of
high
dim
e
ns
ional
f
e
a
tur
e
s
,
is
s
ue
s
of
ove
r
f
it
ti
ng,
a
nd
no
c
ons
ider
a
ti
on
o
f
im
a
ge
r
e
s
izing
whic
h
mi
nim
ize
d
the
c
las
s
if
ica
ti
on
pe
r
f
o
r
manc
e
.
T
he
pr
e
vious
r
e
s
e
a
r
c
he
s
ha
ve
li
m
it
a
ti
ons
of
high
dim
e
ns
ional
f
e
a
tur
e
s
ubs
pa
c
e
a
nd
ove
r
f
i
tt
ing
whic
h
a
ls
o
mi
nim
ize
d
the
c
las
s
if
ier
pe
r
f
o
r
manc
e
.
He
nc
e
,
the
pr
opos
e
d
s
tudy
include
s
im
a
ge
r
e
s
izing
f
or
a
djus
ti
ng
the
s
ize
of
the
im
a
ge
unif
or
ml
y
f
or
f
e
a
s
ibi
li
ty
in
ha
ndli
ng.
T
he
n,
the
hybr
id
op
ti
mi
z
a
ti
on
-
ba
s
e
d
f
e
a
tur
e
s
e
lec
ti
on
method
is
e
mpl
oye
d
to
s
e
lec
t
the
r
e
leva
nt
f
e
a
tur
e
s
that
mi
nim
ize
the
high
dim
e
ns
ionalit
y
o
f
f
e
a
tur
e
s
while
e
li
mi
na
ti
ng
the
ir
r
e
leva
nt
f
e
a
tu
r
e
s
,
a
f
ter
whic
h
the
ove
r
f
it
ti
ng
is
s
ue
is
tac
kled
thr
ough
h
ype
r
pa
r
a
mete
r
tuni
ng
of
C
NN
hype
r
pa
r
a
mete
r
s
us
ing
the
hier
a
r
c
hica
l
B
a
ye
s
ian
opti
mi
z
a
ti
on
(
HB
O
)
a
lgor
it
hm.
T
he
majo
r
c
ont
r
ibut
ions
of
the
r
e
s
e
a
r
c
h
a
r
e
given
a
s
be
low:
−
T
he
HB
O
-
ba
s
e
d
C
NN
is
pr
opos
e
d
to
c
la
s
s
if
y
the
c
he
s
t
X
-
r
a
y
dis
e
a
s
e
c
la
s
s
e
s
us
ing
c
he
s
t
X
-
r
a
y
im
a
ge
s
.
T
he
HB
O
opti
mi
z
e
s
the
pa
r
a
mete
r
s
of
the
C
NN
model
to
e
nha
nc
e
the
c
he
s
t
X
-
r
a
y
dis
e
a
s
e
c
las
s
if
i
c
a
ti
on
pe
r
f
or
manc
e
.
−
T
he
M
e
xica
n
a
xolot
l
opti
mi
z
a
ti
on
(
M
AO
)
a
nd
tu
na
s
wa
r
m
opti
mi
z
a
ti
on
(
T
S
O)
a
lgo
r
it
hms
a
r
e
c
om
bined
in
f
e
a
tur
e
s
e
lec
ti
on
pha
s
e
s
to
c
hoos
e
r
e
leva
nt
f
e
a
tur
e
s
f
r
om
the
whole
f
e
a
tur
e
s
ubs
e
t,
whic
h
e
f
f
e
c
ti
ve
ly
mi
nim
ize
the
dim
e
ns
ionalit
y
o
f
f
e
a
tur
e
s
a
nd
e
nha
nc
e
s
the
c
las
s
if
ica
ti
on
pe
r
f
or
manc
e
.
−
T
he
R
e
s
Ne
t
50
-
ba
s
e
d
f
e
a
tur
e
e
xtr
a
c
ti
on
method
is
e
mpl
oye
d
f
or
e
xtr
a
c
ti
ng
hier
a
r
c
hica
l
f
e
a
tur
e
s
f
r
o
m
the
pr
e
-
pr
oc
e
s
s
e
d
im
a
ge
s
whic
h
dif
f
e
r
e
nti
a
tes
f
e
a
tur
e
s
int
o
di
f
f
e
r
e
nt
c
las
s
e
s
f
or
c
las
s
if
ica
ti
on.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
Hie
r
ar
c
hical
B
ay
e
s
ian
opti
miz
ati
on
bas
e
d
c
onv
olut
ional
ne
ur
al
…
(
B
har
ath
K
umar
Gow
r
u
)
571
T
he
r
e
maining
s
e
c
ti
on
of
the
r
e
s
e
a
r
c
h
is
or
ga
nize
d
a
s
f
oll
ows
:
s
e
c
ti
on
2
pr
ovides
the
pr
opos
e
d
methodology
de
tails
,
while
s
e
c
ti
on
3
pr
ovides
the
r
e
s
ult
s
a
nd
di
s
c
us
s
ion.
At
las
t,
the
c
onc
lus
ion
of
thi
s
r
e
s
e
a
r
c
h
is
given
in
s
e
c
ti
on
4.
2.
P
ROP
OS
E
D
M
E
T
HO
D
T
his
r
e
s
e
a
r
c
h
pr
opos
e
s
a
n
HB
O
ba
s
e
d
C
N
N
method
to
c
las
s
if
y
c
he
s
t
X
-
r
a
y
dis
e
a
s
e
s
us
ing
c
he
s
t
X
-
r
a
y
da
tas
e
t
s
.
T
he
r
a
w
i
mage
s
a
r
e
pr
e
-
pr
oc
e
s
s
e
d
by
us
ing
im
a
ge
r
e
s
izing
a
nd
mi
n
-
max
nor
m
a
li
z
a
ti
on
tec
hniques
.
T
he
n,
the
h
ier
a
r
c
hica
l
f
e
a
tur
e
s
a
r
e
e
xt
r
a
c
ted
by
de
ployi
ng
the
R
e
s
Ne
t50
method,
a
nd
the
r
e
leva
nt
f
e
a
tur
e
s
a
r
e
c
hos
e
n
f
r
om
the
whole
f
e
a
tur
e
s
ubs
e
t
us
ing
the
hybr
id
opt
im
iza
ti
on
a
lgo
r
it
hm.
T
he
s
e
lec
ted
r
e
leva
nt
f
e
a
tur
e
s
a
r
e
c
las
s
if
ied
by
uti
li
z
ing
the
p
r
opos
e
d
HB
O
ba
s
e
d
C
NN
method.
Additi
ona
ll
y,
t
he
C
NN
pa
r
a
mete
r
s
a
r
e
opti
mi
z
e
d
by
e
mpl
oying
the
pr
o
pos
e
d
HB
O
a
lgor
it
hm.
F
igur
e
1
r
e
pr
e
s
e
nts
the
pr
oc
e
s
s
e
s
invol
ve
d
in
the
pr
opos
e
d
method.
F
igur
e
1.
P
r
oc
e
s
s
of
pr
opos
e
d
methodology
2.
1.
Dat
as
e
t
T
he
da
tas
e
t
us
e
d
in
thi
s
r
e
s
e
a
r
c
h
is
C
XR
s
[
21]
im
a
ge
da
tas
e
t
whic
h
is
publi
c
ly
a
c
c
e
s
s
ibl
e
.
T
he
r
e
a
r
e
a
tot
a
l
of
576
im
a
ge
s
in
the
C
OV
I
D
-
19
c
las
s
,
4
,
2
73
im
a
ge
s
in
the
pne
umoni
a
c
las
s
,
a
nd
1
,
583
im
a
ge
s
in
the
nor
mal
c
las
s
.
T
he
da
tas
e
t
is
divi
de
d
int
o
tr
a
ini
ng
a
nd
tes
ti
ng
s
e
t
in
the
r
a
ti
o
o
f
80:20
a
nd
the
de
s
c
r
ipt
ion
of
the
da
tas
e
t
is
dis
playe
d
in
T
a
ble
1
,
whi
le
the
s
a
mpl
e
im
a
ge
s
a
r
e
given
in
F
igur
e
2.
T
a
ble
1.
Da
tas
e
t
de
s
c
r
ipt
ion
C
la
s
s
e
s
T
ot
a
l
im
a
ge
s
T
r
a
in
in
g
T
e
s
ti
ng
C
O
V
I
D
-
19
576
460
116
P
ne
umoni
a
4
,
273
3
,
418
855
N
or
ma
l
1
,
583
1
,
266
317
F
igur
e
2.
S
a
mpl
e
im
a
ge
s
of
the
c
he
s
t
X
-
r
a
y
da
tas
e
t
2.
2.
P
r
e
-
p
r
oc
e
s
s
in
g
T
he
c
he
s
t
X
-
r
a
y
im
a
ge
s
f
r
om
the
da
tas
e
t
a
r
e
p
r
e
-
pr
oc
e
s
s
e
d
by
r
e
s
izing
a
nd
nor
maliza
ti
on
of
im
a
ge
s
us
ing
mi
n
-
max
nor
maliza
ti
on
[
22]
.
T
he
de
tailed
e
xplana
ti
on
of
thes
e
pr
e
-
pr
oc
e
s
s
ing
tec
hniques
is
e
xplaine
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
569
-
579
572
he
r
e
.
T
he
thr
e
e
c
las
s
e
s
of
da
tas
e
ts
na
mely,
C
O
VI
D
-
19,
pne
umoni
a
,
a
nd
no
r
mal
ha
ve
dif
f
e
r
e
nt
s
ize
s
of
256×
256
to
1024×
1024
pixels
.
T
he
im
a
ge
s
a
r
e
r
e
s
ize
d
to
a
de
f
ined
s
ize
of
224×
224
to
e
ns
ur
e
the
unif
or
m
s
ize
of
input
da
ta
f
or
f
e
a
tur
e
e
xt
r
a
c
ti
on
with
th
e
R
e
s
Ne
t
50
method.
R
e
s
izing
of
im
a
ge
s
mi
ni
mi
z
e
s
the
c
omput
a
ti
ona
l
c
ompl
e
xit
y.
T
he
no
r
maliza
ti
on
of
im
a
ge
s
is
a
s
igni
f
ica
nt
pha
s
e
f
or
pr
e
s
e
r
ving
n
umer
ica
l
s
tabili
ty.
T
he
nor
maliza
ti
on
e
ns
ur
e
s
quick
lea
r
ning
with
a
s
table
g
r
a
dient
map
in
the
im
a
ge
s
pa
c
e
.
T
h
e
r
e
s
ize
d
im
a
ge
s
a
r
e
nor
malize
d
to
a
pa
r
ti
c
ular
r
a
nge
[
0
,
1]
thr
ough
mi
n
-
max
nor
maliza
ti
on
whic
h
is
s
igni
f
i
c
a
nt
f
or
s
tanda
r
dizing
the
input
da
ta.
Af
ter
the
pr
e
-
pr
oc
e
s
s
ing,
the
im
a
ge
s
a
r
e
given
a
s
input
to
the
f
e
a
tur
e
e
xtr
a
c
ti
on
pha
s
e
s
to
e
xtr
a
c
t
the
pivot
a
l
f
e
a
tur
e
s
.
2.
3.
F
e
a
t
u
r
e
e
xt
r
ac
t
ion
T
he
inf
or
mative
f
e
a
tur
e
s
a
r
e
e
xt
r
a
c
ted
f
r
om
the
p
r
e
-
pr
oc
e
s
s
e
d
im
a
ge
s
by
uti
li
z
ing
the
C
NN
ba
s
e
d
pr
e
-
tr
a
ined
model
(
i
.
e
.
,
R
e
s
Ne
t
50)
[
23]
.
T
he
R
e
s
Ne
t
50
-
ba
s
e
d
f
e
a
tur
e
e
xtr
a
c
ti
on
invol
ve
s
the
us
a
ge
of
c
onvolut
ional
laye
r
s
of
the
ne
two
r
k
f
or
c
a
ptur
in
g
hier
a
r
c
hica
l
f
e
a
tur
e
s
f
r
om
the
c
he
s
t
X
-
r
a
y
im
a
ge
s
.
T
he
mea
ningf
ul
f
e
a
tur
e
s
a
r
e
e
xtr
a
c
ted
f
r
om
the
pr
e
-
pr
oc
e
s
s
e
d
im
a
ge
s
by
c
a
ptur
ing
dif
f
e
r
e
nt
pa
tt
e
r
ns
a
nd
s
tr
uc
tur
e
s
r
e
leva
nt
to
the
c
he
s
t
X
-
r
a
y
dis
e
a
s
e
.
F
u
r
ther
mor
e
,
the
R
e
s
Ne
t
50
model
include
s
s
kip
c
o
nne
c
ti
on
whic
h
tac
kles
the
is
s
ue
of
va
nis
hing
gr
a
dient
a
n
d
he
lps
in
lea
r
ning
inf
or
mat
ive
f
e
a
tur
e
s
.
T
he
R
e
s
Ne
t
50
model
pr
oc
e
s
s
e
s
the
input
i
mage
by
their
laye
r
s
a
nd
e
xtr
a
c
ts
the
low
-
leve
l
f
e
a
tur
e
s
o
f
e
dge
s
a
nd
te
xtur
e
s
in
the
ini
ti
a
l
laye
r
s
a
nd
high
-
leve
l
s
e
mantic
f
e
a
tur
e
s
,
li
ke
the
pa
r
ts
of
objec
ts
in
the
de
e
p
laye
r
s
.
F
inally,
a
tot
a
l
of
2048
f
e
a
tur
e
s
a
r
e
e
xt
r
a
c
ted
by
us
ing
the
global
pooli
ng
laye
r
o
f
the
R
e
s
Ne
t
50
model.
F
igur
e
3
r
e
pr
e
s
e
nts
the
pr
oc
e
s
s
of
f
e
a
tur
e
e
xtr
a
c
ti
on
by
the
R
e
s
Ne
t
50
method.
F
igur
e
3.
P
r
oc
e
s
s
of
HB
O
ba
s
e
d
C
NN
2.
4.
F
e
a
t
u
r
e
s
e
lec
t
ion
T
h
e
e
x
t
r
a
c
te
d
20
48
f
e
a
tu
r
e
s
a
r
e
gi
ve
n
a
s
in
pu
t
to
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
b
y
s
e
l
e
c
t
i
ng
t
he
r
e
l
e
va
n
t
f
e
a
t
u
r
e
s
f
r
o
m
t
he
f
e
a
tu
r
e
s
ubs
e
t
.
I
n
th
is
r
e
s
e
a
r
c
h
,
o
pt
im
iz
a
t
ion
-
ba
s
e
d
f
e
a
tu
r
e
s
e
le
c
t
io
n
m
e
t
ho
ds
a
r
e
us
e
d
t
o
s
e
lec
t
r
e
le
va
n
t
f
e
a
t
ur
e
s
f
r
om
the
f
e
a
t
u
r
e
s
u
bs
e
t
.
T
he
o
pt
i
mi
z
a
t
i
on
a
l
go
r
it
hm
s
e
a
r
c
h
e
s
t
he
be
s
t
f
e
a
tu
r
e
s
f
r
o
m
w
ho
le
f
e
a
t
ur
e
s
ubs
e
t
w
hi
c
h
e
nh
a
n
c
e
s
t
he
c
he
s
t
X
-
r
a
y
c
l
a
s
s
i
f
ica
t
ion
pe
r
f
o
r
ma
nc
e
.
T
he
M
A
O
a
n
d
T
S
O
a
lg
or
i
th
ms
a
r
e
c
om
b
ine
d
t
o
c
ho
os
e
t
he
r
e
lev
a
n
t
f
e
a
tu
r
e
s
f
o
r
c
las
s
i
f
i
c
a
ti
on
.
T
h
is
s
t
e
p
i
nv
o
lves
th
e
c
o
mb
in
in
g
o
f
t
he
e
xp
lo
r
a
t
io
n
a
n
d
e
xp
lo
i
ta
ti
on
s
t
r
a
te
gi
e
s
o
f
M
AO
a
n
d
T
S
O
a
l
go
r
i
thm
s
f
o
r
f
e
a
t
u
r
e
s
e
l
e
c
ti
on
t
o
e
f
f
i
c
i
e
n
tl
y
s
e
a
r
c
h
t
he
be
s
t
f
e
a
t
u
r
e
s
ubs
e
t
a
nd
f
i
nd
th
e
be
s
t
o
pt
im
a
l
f
e
a
t
u
r
e
s
.
T
he
M
A
O
a
l
go
r
i
th
m
e
x
pl
o
r
e
s
t
he
f
e
a
t
u
r
e
s
u
bs
p
a
c
e
a
nd
s
e
a
r
c
h
e
s
v
a
r
i
ous
f
e
a
tu
r
e
s
by
a
d
j
us
t
in
g
the
f
e
a
tu
r
e
s
ba
s
e
d
on
t
he
e
x
p
lo
r
a
ti
on
s
t
r
a
te
gy
o
f
a
xo
lo
t
l
b
e
h
a
v
io
r
.
T
he
n
,
t
he
T
S
O
a
lg
o
r
i
th
m
e
xp
l
oi
ts
t
he
f
e
a
t
u
r
e
s
u
bs
pa
c
e
s
f
ou
nd
by
the
M
AO
a
l
go
r
it
hm
a
n
d
r
e
f
i
ne
s
to
th
e
be
s
t
f
e
a
t
ur
e
s
.
2.
4.
1.
M
e
xican
axolo
t
l
op
t
im
iza
t
ion
T
he
M
e
xica
n
a
xolot
l
opt
im
iza
ti
on
(
M
AO
)
a
lgor
it
hm
is
a
na
tu
r
e
-
ins
pir
e
d
a
lgor
it
hm
that
m
im
ics
the
li
f
e
of
a
xolot
l
a
nd
it
s
population
is
d
ivi
de
d
int
o
male
a
nd
f
e
male
.
T
he
M
AO
a
lgor
it
h
m
ha
s
f
ou
r
it
e
r
a
ti
on
pha
s
e
s
of
tr
a
ns
it
ioni
ng
f
r
om
lar
va
e
to
the
a
dult
s
tage
,
r
e
pr
oduc
ti
on
a
nd
a
s
s
or
tm
e
nt,
inj
u
r
y,
a
nd
r
e
s
tor
a
ti
on
[
24]
.
I
nit
ially,
the
population
is
ini
ti
a
li
z
e
d
r
a
ndoml
y,
a
nd
ne
xt,
e
ve
r
y
indi
vidual
is
a
ll
oc
a
ted
to
a
male
or
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
Hie
r
ar
c
hical
B
ay
e
s
ian
opti
miz
ati
on
bas
e
d
c
onv
olut
ional
ne
ur
al
…
(
B
har
ath
K
umar
Gow
r
u
)
573
f
e
male
be
c
a
us
e
of
a
xolot
ls
de
ve
loped
by
it
s
s
e
x,
whe
r
e
in
2
s
ubpopulations
a
r
e
a
tt
a
ined.
T
he
male
ind
ivi
dua
ls
tr
a
ns
mi
t
wa
ter
f
r
om
a
dult
lar
va
e
thr
ough
a
djus
ti
ng
it
s
body
pa
r
ts
'
c
olor
towa
r
ds
male
.
T
he
e
f
f
e
c
ti
v
e
a
da
pted
indi
viduals
ha
ve
s
upe
r
ior
c
a
mouf
lage
a
nd
other
in
divi
dua
ls
c
ha
nge
their
c
olor
.
B
y
that
pos
s
ibi
li
ty,
th
e
a
xolot
l
is
c
hos
e
n
to
c
a
mouf
lage
towa
r
ds
the
s
upe
r
ior
.
C
ons
ider
ing
the
a
s
a
s
upe
r
io
r
ly
a
da
pted
male
,
r
e
pr
e
s
e
nts
the
tr
a
ns
it
ion
pa
r
a
mete
r
[
0
,
1]
f
or
male
a
xolot
l,
f
ur
ther
modi
f
ying
their
body
pa
r
ts
,
whe
r
e
i
n
the
numer
ica
l
e
xpr
e
s
s
ion
is
mathe
matica
ll
y
e
xpr
e
s
s
e
d
a
s
(
1)
.
T
he
f
e
male
a
xolot
ls
modi
f
y
the
bodies
f
r
o
m
lar
va
e
to
a
dult
s
towa
r
ds
the
f
e
male
s
a
long
a
s
upe
r
ior
a
da
ptation,
whos
e
numer
ica
l
e
xp
r
e
s
s
ion
is
f
or
mul
a
ted
in
(
2)
.
←
+
(
,
−
)
∗
(
1)
←
+
(
,
−
)
∗
(
2)
W
he
r
e
,
the
r
e
pr
e
s
e
nts
the
be
s
t
f
e
male
a
nd
r
e
pr
e
s
e
nts
the
pr
e
s
e
nt
f
e
male
a
xolot
l.
How
e
ve
r
,
by
t
he
inver
s
e
pos
s
ibi
li
ty
of
t
r
a
ns
it
ion,
indi
viduals
a
r
e
un
s
uc
c
e
s
s
f
ul
in
c
a
mouf
laging
thems
e
lves
towa
r
d
s
upe
r
ior
it
y,
a
nd
ha
ve
their
c
olor
s
c
hos
e
n.
I
f
the
r
a
ndom
numb
e
r
∈
[
0
,
1
]
is
les
s
e
r
than
the
inver
s
e
tr
a
ns
it
ion
pos
s
ibi
li
ty,
the
r
e
s
pe
c
ti
ve
indi
vidual
is
c
hos
e
n.
T
o
mi
nim
ize
t
he
is
s
ue
,
the
male
a
xolot
l
with
a
n
opti
mal
va
lue
is
c
hos
e
n.
T
he
numer
ica
l
e
xpr
e
s
s
ion
f
or
inver
s
e
tr
a
n
s
it
ion
pos
s
ibi
li
ty
is
given
a
s
(
3)
.
W
he
r
e
,
r
e
pr
e
s
e
nts
the
f
e
male
a
xolot
l
with
opti
mi
z
a
ti
on
va
lue
o
f
a
nd
t
he
numer
ica
l
f
or
mul
a
ted
is
given
in
(
4
)
,
a
nd
the
wor
s
t
indi
viduals
ha
ve
higher
c
ha
nc
e
s
of
r
a
ndom
t
r
a
ns
it
ions
.
T
he
s
e
indi
viduals
tr
a
ns
it
their
ℎ
body
pa
r
ts
r
a
ndoml
y
a
nd
thei
r
numer
ica
l
e
xpr
e
s
s
ion
is
given
i
n
(
5)
a
nd
(
6
)
.
=
∑
(
3)
=
∑
(
4)
←
+
(
−
)
∗
(
5)
←
+
(
−
)
∗
(
6)
F
r
om
(
5)
a
nd
(
6
)
,
the
∈
[
0
,
1
]
r
e
pr
e
s
e
nts
the
r
a
ndom
number
s
e
lec
ted
f
o
r
e
ve
r
y
ℎ
body
pa
r
t.
T
he
indi
viduals
with
the
r
a
ndom
tr
a
ns
mi
s
s
ion
a
r
e
c
ho
s
e
n
by
the
opti
mi
z
a
ti
on
f
unc
ti
on
va
lue.
B
y
movi
n
g
a
c
r
os
s
wa
ter
,
a
xolot
ls
s
uf
f
e
r
a
c
c
idents
a
nd
c
a
n
be
hur
t.
T
his
pr
oc
e
dur
e
is
r
e
pr
e
s
e
nted
a
s
inj
ur
y
in
the
r
e
s
tor
a
ti
on
s
tage
.
F
or
e
ve
r
y
a
xolot
l
in
population,
whe
ther
the
pos
s
ibi
li
ty
of
da
mage
(
)
is
c
ompl
e
ted,
a
xolot
l
los
e
s
a
c
e
r
tain
pa
r
t
o
f
their
body
pa
r
t.
I
n
thi
s
pr
oc
e
s
s
,
the
r
e
ge
ne
r
a
ti
on
pos
s
ibi
li
ty
(
)
is
uti
l
ize
d
pe
r
bit
.
2.
4.
2.
T
u
n
a
s
war
m
op
t
i
m
izat
ion
T
he
tuna
s
wa
r
m
opt
im
iza
ti
on
(
T
S
O)
a
lgor
it
hm
is
a
na
tur
e
-
ins
pir
e
d
a
lgor
it
hm
that
a
dopts
the
p
r
oc
e
s
s
of
s
pir
a
l
f
or
a
ging
s
tr
a
tegy,
a
ggr
e
ga
tes
to
the
s
pir
a
l
s
ha
pe
s
a
nd
de
ter
mi
ne
s
the
pr
e
y
in
s
ha
ll
ow
wa
ter
r
e
gions
[
25]
.
T
he
de
tailed
e
xplana
ti
on
of
population
ini
ti
a
li
z
a
ti
on,
pa
r
a
boli
c
f
or
a
ging
a
nd
s
pir
a
l
f
or
a
ging
s
tr
a
tegie
s
a
r
e
given
be
low.
T
he
r
e
a
r
e
NP
tunas
in
the
tun
a
s
wa
r
m
a
nd
a
t
the
s
wa
r
m
in
it
ializa
ti
on
s
tage
,
t
he
T
S
O
a
lgor
it
hm
r
a
ndoml
y
ge
ne
r
a
tes
the
ini
ti
a
l
s
wa
r
ms
i
n
the
s
e
a
r
c
h
s
pa
c
e
.
T
he
numer
ica
l
e
xpr
e
s
s
ion
to
ini
ti
a
li
z
e
tuna
indi
viduals
is
given
in
(
7
)
.
=
∙
(
−
)
+
(
7)
W
he
r
e
,
r
e
pr
e
s
e
nts
the
ℎ
tuna
,
a
nd
s
igni
f
y
the
uppe
r
a
nd
lowe
r
r
a
nge
bounda
r
ies
in
tuna
e
xplor
a
ti
on,
while
de
notes
a
r
a
ndom
va
r
iable
with
unif
or
m
dis
tr
ibu
ti
on
f
r
om
0
-
1.
P
a
r
ti
c
ular
ly
,
e
v
e
r
y
indi
vidual
r
e
pr
e
s
e
nts
the
c
a
ndidate
s
olut
ion
of
T
S
O.
E
ve
r
y
indi
v
idual
tuna
ha
s
a
gr
oup
o
f
dim
e
n
s
ion
number
s
.
a.
P
a
r
a
boli
c
f
or
a
ging
s
tr
a
tegy
I
n
the
pr
e
da
ti
on
pr
oc
e
s
s
,
e
ve
r
y
tuna
f
oll
ows
th
e
pa
s
t
indi
viduals
a
nd
a
ll
tuna
s
wa
r
ms
f
or
m
a
pa
r
a
bola
f
or
s
ur
r
ound
ing
the
pr
e
y
.
Additi
ona
ll
y,
the
tuna
s
wa
r
m
ut
il
ize
s
a
s
pir
a
l
f
or
a
ging
s
tr
a
tegy,
c
ons
ider
ing
the
pos
s
ibi
li
ty
of
the
tuna
s
wa
r
m,
s
e
l
e
c
ti
ng
the
s
tr
a
tegy
a
nd
a
numer
ica
l
e
xpr
e
s
s
ion
is
given
a
s
(
8)
a
nd
(
9
)
.
T
he
r
e
p
r
e
s
e
nts
the
ℎ
it
e
r
a
ti
on
that
is
pr
e
s
e
ntl
y
p
r
oc
e
s
s
e
d,
a
nd
r
e
pr
e
s
e
nts
the
highes
t
number
of
it
e
r
a
ti
ons
that
e
xis
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
569
-
579
574
+
1
=
{
+
∙
(
−
)
+
∙
2
∙
(
−
)
,
<
0
.
5
∙
2
∙
,
≥
0
.
5
(
8)
=
(
1
−
)
(
⁄
)
(
9
)
b.
S
pir
a
l
f
o
r
a
ging
s
tr
a
tegy
T
he
ne
xt
e
f
f
e
c
ti
ve
c
oope
r
a
ti
ve
f
or
a
ging
s
tr
a
tegy
is
known
a
s
the
s
pir
a
l
f
o
r
a
ging
s
tr
a
tegy
.
W
he
n
c
ha
s
ing
pr
e
y,
many
tunas
c
a
nnot
s
e
lec
t
the
c
or
r
e
c
t
dir
e
c
ti
on
,
but
a
s
mall
a
mount
of
tuna
guides
the
s
wa
r
m
to
s
wim
in
the
c
or
r
e
c
t
dir
e
c
ti
on
.
T
he
e
f
f
e
c
ti
ve
ind
iv
iduals
a
r
e
una
ble
to
c
a
us
e
the
s
wa
r
m
to
c
a
ptur
e
t
he
pr
e
y
e
f
f
icie
ntl
y.
T
he
tuna
c
hos
e
r
a
ndom
indi
v
iduals
i
n
a
s
wa
r
m
to
f
oll
ow.
T
he
numer
ica
l
e
xpr
e
s
s
ion
f
or
s
pir
a
l
f
or
a
ging
s
tr
a
tegy
is
g
iven
a
s
(
10)
.
+
1
=
{
1
∙
(
+
∙
|
−
|
+
2
∙
1
∙
(
+
∙
|
−
|
+
2
∙
−
1
)
,
<
1
∙
(
+
∙
|
−
|
+
2
∙
)
,
≥
1
∙
(
+
∙
|
−
|
+
2
∙
−
1
)
(
10)
T
he
+
1
r
e
pr
e
s
e
nts
the
ℎ
tuna
in
+
1
it
e
r
a
ti
on
.
T
he
pr
e
s
e
nts
a
n
opti
mal
ind
ivi
dua
l
a
nd
de
notes
the
r
a
ndoml
y
c
hos
e
n
tuna
s
wa
r
m.
1
de
notes
the
we
ight
c
oe
f
f
icie
nt
f
or
c
ontr
oll
ing
the
s
wimm
ing
of
tuna
indi
viduals
f
or
the
opti
mum
indi
vidual
.
T
h
e
2
de
notes
the
we
ight
c
oe
f
f
icie
nt
f
or
c
ontr
oll
ing
tuna
indi
viduals
a
nd
de
notes
the
dis
tan
c
e
pa
r
a
mete
r
that
mana
ge
s
the
dis
tanc
e
be
twe
e
n
tuna
indi
vidu
a
ls
a
nd
opti
mum
indi
viduals
.
F
inally,
the
1228
r
e
leva
nt
f
e
a
tur
e
s
a
r
e
s
e
lec
ted
f
r
om
the
whole
f
e
a
tur
e
s
ubs
pa
c
e
by
us
ing
hybr
id
M
OA
a
nd
T
S
O
a
lgo
r
it
hms
.
2.
5.
Clas
s
if
icat
ion
u
s
in
g
HB
O
-
b
as
e
d
CN
N
T
he
s
e
lec
ted
r
e
leva
nt
f
e
a
tur
e
s
a
r
e
given
a
s
input
to
the
C
NN
model
f
or
c
las
s
if
ying
the
d
if
f
e
r
e
nt
c
las
s
e
s
of
C
OV
I
D
-
19,
pne
umoni
a
,
a
nd
nor
mal
.
T
he
C
NN
a
r
c
hit
e
c
tur
e
include
s
f
ive
c
onvolut
i
ons
,
f
ive
pooli
ng,
two
f
ul
ly
c
onne
c
ted
a
nd
one
dr
opou
t
lay
e
r
.
T
he
C
NN
model
ha
s
f
e
we
r
pa
r
a
mete
r
s
a
s
c
om
pa
r
e
d
to
c
e
r
tain
c
onve
nti
ona
l
f
e
e
df
or
wa
r
d
ne
twor
ks
whic
h
r
e
s
ult
in
a
f
e
a
s
ibl
e
tr
a
ini
ng
p
r
oc
e
s
s
.
He
r
e
,
the
r
e
c
ti
f
ied
li
ne
a
r
unit
(
R
e
L
U)
is
us
e
d
a
s
a
n
a
c
ti
va
ti
on
f
unc
ti
o
n
a
nd
the
numer
ica
l
e
xpr
e
s
s
ion
is
given
a
s
(
11)
.
W
he
r
e
,
r
e
pr
e
s
e
nts
the
ne
ur
on
input
.
T
he
r
e
s
ult
of
thi
s
a
c
ti
v
a
ti
on
f
unc
ti
on
is
0
whe
n
the
va
lue
of
input
is
les
s
than
0
or
e
ls
e
,
a
f
ter
whic
h
it
r
e
tur
ns
a
non
-
ne
ga
ti
ve
input
va
lue.
T
he
f
e
a
tur
e
map
of
ℎ
c
onvolut
ional
laye
r
th
r
ough
uti
li
z
ing
the
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
is
given
a
s
(
12)
.
I
n
(
12
)
,
the
a
nd
de
notes
the
we
ight
s
a
nd
bias
e
s
of
ℎ
laye
r
.
T
he
M
a
x
-
pooli
ng
is
a
non
-
li
ne
a
r
down
-
s
a
mpl
ing
method,
s
e
pa
r
a
ti
ng
the
c
onvolved
da
ta
int
o
×
dis
joi
nt
pa
r
ts
.
T
his
laye
r
is
ne
xt
to
the
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on
a
nd
is
uti
li
z
e
d
f
or
e
xe
c
uti
ng
the
la
s
t
f
e
a
tur
e
ve
c
tor
.
T
he
output
laye
r
of
a
nd
the
numer
i
c
a
l
e
xpr
e
s
s
ion
is
given
in
(
13
)
.
(
)
=
m
a
x
{
0
,
}
(
11)
ℎ
=
(
(
)
+
)
(
12)
ℎ
=
(
(
)
)
(
13)
I
n
(
13
)
,
ℎ
r
e
pr
e
s
e
nts
the
ℎ
f
e
a
tur
e
map
o
f
s
ize
in
a
gi
ve
n
c
onvolut
ional
laye
r
with
pixel
c
oor
d
inate
s
.
T
h
e
dr
opout
r
e
gular
iza
ti
on
method
is
im
pleme
nted
to
a
void
the
ove
r
-
f
it
ti
ng
is
s
ue
.
At
e
ve
r
y
c
onvolut
io
na
l
laye
r
block,
the
pe
r
c
e
ntage
of
node
is
dr
oppe
d
to
a
tt
a
in
a
mi
nim
ize
d
pr
oc
e
s
s
of
the
ne
twor
k
.
T
he
hype
r
pa
r
a
mete
r
s
of
ba
tch
s
ize
,
opt
im
ize
r
,
los
s
,
lea
r
ning
r
a
te,
e
poc
h,
a
nd
a
c
ti
va
ti
on
f
unc
ti
on
ne
e
d
to
be
opti
mi
z
e
d.
2.
5.
1.
Hi
e
r
ar
c
h
ical
B
aye
s
ian
o
p
t
im
izat
io
n
T
he
B
a
ye
s
ian
opti
mi
z
a
ti
on
is
a
n
e
f
f
e
c
ti
ve
tec
hnique
f
or
global
opti
mi
z
a
ti
on
of
objec
ti
ve
f
unc
ti
ons
whic
h
is
e
xpe
ns
ive
f
or
e
va
luation.
I
n
thi
s
r
e
s
e
a
r
c
h,
HB
O
is
p
r
opos
e
d
a
s
a
n
e
xtens
ion
of
s
tanda
r
d
B
a
ye
s
ian
opti
mi
z
a
ti
on
by
incor
po
r
a
ti
ng
the
hie
r
a
r
c
hica
l
s
tr
uc
tur
e
of
hype
r
pa
r
a
mete
r
s
that
a
ll
ow
a
n
e
f
f
e
c
ti
ve
e
xplor
a
ti
on
a
nd
opti
mi
z
a
ti
on
of
d
if
f
icult
s
e
a
r
c
h
s
pa
c
e
s
.
T
he
HB
O
c
a
ptur
e
s
the
de
pe
nde
nc
ies
a
m
ong
the
hype
r
pa
r
a
mete
r
s
by
uti
li
z
ing
pr
oba
bil
is
ti
c
method
s
(
Ga
us
s
ian
pr
oc
e
s
s
)
a
t
e
ve
r
y
hier
a
r
c
hica
l
leve
l
.
T
he
HB
O
is
de
ployed
f
o
r
hype
r
pa
r
a
mete
r
tun
ing
o
f
the
C
NN
model
whic
h
e
nha
nc
e
s
the
p
r
oc
e
s
s
of
the
C
NN
m
ode
l
a
nd
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
Hie
r
ar
c
hical
B
ay
e
s
ian
opti
miz
ati
on
bas
e
d
c
onv
olut
ional
ne
ur
al
…
(
B
har
ath
K
umar
Gow
r
u
)
575
im
pr
ove
s
the
c
he
s
t
X
-
r
a
y
c
las
s
if
ica
ti
on
pe
r
f
or
man
c
e
.
I
n
HB
O,
the
e
xpe
c
ted
im
p
r
ove
ment
(
E
I
)
is
uti
l
ize
d
due
to
it
s
s
im
pli
c
it
y.
F
igu
r
e
3
r
e
pr
e
s
e
nts
the
pr
oc
e
s
s
of
HB
O
ba
s
e
d
C
NN
method.
C
ons
ider
ing
that
the
opti
m
iza
ti
on
p
r
oblem
is
opti
mi
z
ing
the
a
r
g
(
)
,
the
pr
e
s
e
nt
be
s
t
is
a
t
+
=
∈
1
:
(
)
.
T
he
numer
ica
l
e
xpr
e
s
s
ion
f
or
the
im
p
r
ove
d
f
u
nc
ti
on
is
g
iven
a
s
(
14)
.
T
he
numer
ica
l
e
xpr
e
s
s
ion
f
or
HB
O
on
the
e
xpe
c
ted
v
a
lue
of
(
)
is
given
a
s
(
15)
.
He
r
e
,
1
:
=
{
1
:
,
1
:
}
,
the
numer
ica
l
e
xpr
e
s
s
ion
f
or
the
(
(
)
)
is
f
or
mul
a
ted
in
(
16)
.
(
)
=
m
a
x
{
0
,
(
)
−
(
+
)
}
(
14)
a
r
g
(
(
)
|
1
:
)
(
15)
(
(
)
)
=
{
(
(
)
−
(
+
)
)
Φ
(
)
+
(
)
(
)
,
(
)
>
0
0
,
(
)
=
0
(
16)
T
he
p
r
opos
e
d
HB
O
-
ba
s
e
d
C
NN
method
e
f
f
e
c
ti
ve
ly
c
las
s
if
ies
the
c
he
s
t
X
-
r
a
y
c
las
s
e
s
with
high
a
c
c
ur
a
c
y.
T
he
hype
r
pa
r
a
mete
r
s
of
C
NN
opti
mi
z
e
d
a
r
e
o
f
the
ba
tch
s
ize
:
32,
Optim
ize
r
:
Ada
m,
los
s
:
c
a
tegor
i
c
a
l
c
r
os
s
e
ntr
opy,
lea
r
ning
r
a
te:
0.
001
,
e
poc
h:
10
a
nd
a
c
ti
va
ti
on
f
unc
ti
on:
R
e
L
U.
T
he
da
ta
p
r
e
-
pr
oc
e
s
s
ing
of
im
a
ge
r
e
s
izing
a
nd
mi
n
-
max
nor
maliza
ti
on
methods
r
e
s
i
z
e
the
im
a
ge
in
the
s
a
me
dim
e
ns
ion
a
nd
s
c
a
les
th
e
im
a
ge
withi
n
the
r
a
nge
of
[
0
,
1]
.
T
he
n,
the
hier
a
r
c
hica
l
f
e
a
tur
e
s
a
r
e
e
xtr
a
c
ted
f
r
om
the
p
r
e
-
pr
oc
e
s
s
e
d
i
mage
s
by
us
ing
the
R
e
s
Ne
t
50
method,
a
nd
the
r
e
leva
nt
f
e
a
tur
e
s
a
r
e
s
e
lec
ted
f
r
om
the
whole
f
e
a
tur
e
s
ubs
e
t
by
us
ing
M
OA
a
nd
T
S
O
a
lgor
it
hms
.
F
inally
,
the
c
he
s
t
X
-
r
a
y
c
las
s
e
s
a
r
e
c
las
s
if
ied
by
us
ing
the
HB
O
-
C
NN
method
with
high
a
c
c
ur
a
c
y.
3.
E
XP
E
RI
M
E
NT
AL
AN
A
L
YSI
S
T
he
pe
r
f
o
r
manc
e
of
the
pr
opos
e
d
HB
O
-
C
NN
is
s
im
ulate
d
with
a
P
y
thon
e
nvir
onment
a
nd
the
s
ys
tem
r
e
quir
e
ments
a
r
e
a
n
i5
pr
oc
e
s
s
or
,
16
GB
R
AM
a
nd
a
W
indows
10
ope
r
a
ti
ng
s
ys
tem.
T
he
e
va
luation
mea
s
ur
e
s
us
e
d
to
a
na
lyze
the
pr
opos
e
d
HB
O
-
C
NN
a
r
e
,
a
c
c
ur
a
c
y,
pr
e
c
is
ion,
r
e
c
a
ll
,
a
nd
F
1
-
s
c
or
e
.
I
n
e
qua
ti
ons
,
T
P
de
notes
the
c
or
r
e
c
tl
y
c
las
s
if
ied
c
he
s
t
X
-
r
a
y.
T
N
de
notes
the
c
or
r
e
c
tl
y
c
las
s
if
ied
non
-
c
he
s
t
X
-
r
a
y
dis
e
a
s
e
or
nor
mal
c
las
s
.
W
he
r
e
a
s
,
F
P
s
igni
f
ies
the
incor
r
e
c
tl
y
c
las
s
if
ied
c
he
s
t
X
-
r
a
y
c
las
s
a
nd
s
igni
f
ies
the
incor
r
e
c
tl
y
c
las
s
if
ied
a
s
the
non
-
c
h
e
s
t
X
-
r
a
y
dis
e
a
s
e
.
T
he
mathe
matica
l
f
or
mul
a
f
or
e
v
a
luation
metr
ics
is
given
f
r
om
(
17)
to
(
20
)
.
=
+
+
+
+
(
17)
=
+
(
18)
=
+
(
19)
1
−
=
2
2
+
+
(
20)
3.
1.
Qu
an
t
it
a
t
ive
an
d
q
u
al
it
at
ive
an
alys
is
T
he
pe
r
f
o
r
manc
e
of
HB
O
-
C
NN
method
is
e
s
ti
mate
d
with
a
c
he
s
t
X
-
r
a
y
da
tas
e
t
with
a
c
c
ur
a
c
y,
pr
e
c
is
ion,
r
e
c
a
ll
,
a
nd
F
1
-
s
c
or
e
.
I
n
thi
s
s
e
c
ti
on,
the
pr
opos
e
d
method
is
e
va
luate
d
with
dif
f
e
r
e
nt
f
e
a
tur
e
e
xtr
a
c
ti
on
tec
hniques
,
dif
f
e
r
e
nt
opti
mi
z
a
ti
on
a
lgor
i
thm
s
a
nd
dif
f
e
r
e
nt
c
las
s
if
ier
s
.
I
n
T
a
ble
2
a
nd
F
igu
r
e
4,
the
pe
r
f
or
manc
e
of
the
f
e
a
tur
e
e
xtr
a
c
ti
on
method
is
e
s
ti
mate
d
with
the
c
he
s
t
X
-
r
a
y
da
tas
e
t
in
ter
ms
of
the
dif
f
e
r
e
nt
e
va
luation
mea
s
ur
e
s
.
T
he
c
onve
nti
ona
l
f
e
a
tur
e
e
xtr
a
c
ti
on
tec
hniques
of
R
e
s
Ne
t
18,
VG
G
16,
a
nd
C
NN
a
r
e
c
ons
ider
e
d
f
or
e
va
luation
of
the
R
e
s
Ne
t
50
method.
T
he
R
e
s
Ne
t
50
method
a
c
hieve
s
the
highes
t
a
c
c
ur
a
c
y
97.
94%
,
pr
e
c
is
ion
92
.
00%
,
r
e
c
a
ll
89
.
0
0%
,
a
nd
F1
-
s
c
or
e
92.
00%
,
while
the
R
e
s
Ne
t
50
method
a
tt
a
ins
maximi
z
e
d
a
c
c
ur
a
c
y,
a
s
oppos
e
d
to
other
methods
.
I
n
T
a
ble
3
a
nd
F
igur
e
5
,
the
outcome
s
of
the
c
las
s
if
ica
ti
on
method
-
ba
s
e
d
M
AO
+
T
S
O
a
r
e
e
va
luate
d
with
the
c
he
s
t
X
-
r
a
y
da
tas
e
t
de
s
c
r
ibed.
T
he
c
onve
nti
ona
l
opti
mi
z
a
ti
on
tec
hniques
take
n
f
or
e
va
luation
of
the
M
AO
+
T
S
O
method
is
gr
e
y
wolf
opti
mi
z
a
ti
on
(
GW
O)
,
M
AO
a
nd
T
S
O.
T
he
M
A
O+
T
S
O
method
a
c
hieve
s
the
highes
t
a
c
c
ur
a
c
y
97.
94%
,
pr
e
c
is
ion
92.
00%
,
r
e
c
a
ll
89.
00
%
a
nd
F1
-
s
c
or
e
92.
00%
,
whe
r
e
a
s
the
R
e
s
Ne
t
50
method
a
tt
a
ins
a
maximi
z
e
d
a
c
c
ur
a
c
y
whe
n
c
ompar
e
d
to
the
othe
r
methods
.
I
n
T
a
ble
4
a
nd
F
igur
e
6,
the
r
e
s
ult
s
of
the
p
r
opos
e
d
HB
O
-
C
NN
method
is
e
s
ti
mate
d
with
the
c
he
s
t
X
-
r
a
y
da
tas
e
t
in
ter
ms
of
dif
f
e
r
e
nt
e
va
luation
m
e
a
s
ur
e
s
is
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
569
-
579
576
de
s
c
r
ibed.
T
he
c
onve
nti
ona
l
c
las
s
if
ier
s
take
n
int
o
c
ons
ider
a
ti
on
f
or
the
e
va
luation
of
the
p
r
opos
e
d
HB
O
-
C
NN
method
a
r
e
I
nc
e
pti
onV2,
VG
G
19
a
nd
C
NN
tec
hniques
.
T
he
pr
opos
e
d
HB
O
-
C
NN
method
a
tt
a
ins
a
s
upe
r
ior
a
c
c
ur
a
c
y
of
97.
94%
,
p
r
e
c
is
ion
of
92.
00
%
,
r
e
c
a
ll
of
89
.
00%
a
nd
F
1
-
s
c
or
e
of
92.
00%
,
wh
il
e
on
the
other
ha
nd,
the
R
e
s
Ne
t
50
method
a
c
hieve
s
a
maxi
mi
z
e
d
a
c
c
ur
a
c
y
whe
n
c
ompar
e
d
to
the
o
ther
meth
ods
.
T
he
hybr
id
opti
mi
z
a
ti
on
-
ba
s
e
d
f
e
a
tur
e
s
e
lec
ti
on
met
hod
is
de
ployed
in
the
p
r
opos
e
d
method
to
c
h
oos
e
the
r
e
leva
nt
f
e
a
tur
e
s
f
r
om
the
whole
f
e
a
tur
e
s
ubs
e
t,
whic
h
f
ur
the
r
r
e
duc
e
s
the
dim
e
ns
ionalit
y
o
f
f
e
a
tur
e
s
by
e
nha
nc
ing
the
c
las
s
if
ier
pe
r
f
or
manc
e
.
T
he
n,
th
e
pa
r
a
mete
r
s
of
C
NN
a
r
e
opti
mi
z
e
d
by
uti
l
izing
HB
O
a
lgor
it
hm
by
tuni
ng
the
opti
mi
z
e
d
pa
r
a
mete
r
s
o
f
C
NN
to
im
pr
ove
the
c
las
s
if
ica
ti
on
pe
r
f
o
r
manc
e
.
T
a
ble
2.
P
e
r
f
o
r
manc
e
of
f
e
a
tur
e
e
xt
r
a
c
ti
on
method
M
e
th
ods
A
c
c
ur
a
c
y (
%
)
P
r
e
c
is
io
n (
%
)
R
e
c
a
ll
(
%
)
F1
-
s
c
or
e
(
%
)
R
e
s
N
e
t
18
81.10
80.50
80.00
81.00
V
G
G
16
79.20
75.80
73.40
70.00
C
N
N
73.00
72.89
72.65
72.90
R
e
s
N
e
t
50
97.94
92.00
89.00
92.00
F
igur
e
4.
P
e
r
f
or
manc
e
of
f
e
a
tur
e
e
xtr
a
c
ti
on
metho
d
T
a
ble
3.
P
e
r
f
o
r
manc
e
of
c
las
s
if
ica
ti
on
ba
s
e
d
on
opti
mi
z
a
ti
on
a
lgor
it
hm
A
lg
or
it
hms
A
c
c
ur
a
c
y (
%
)
P
r
e
c
is
io
n (
%
)
R
e
c
a
ll
(
%
)
F1
-
s
c
or
e
(
%
)
G
W
O
81.10
80.50
80.00
81.00
M
A
O
79.20
75.80
73.40
70.00
T
S
O
73.00
72.89
72.65
72.90
M
A
O
+
T
S
O
97.94
92.00
89.00
92.00
F
igur
e
5.
P
e
r
f
or
manc
e
of
c
las
s
if
ica
ti
on
ba
s
e
d
on
o
pti
mi
z
a
ti
on
a
lgor
it
hm
T
a
ble
4.
P
e
r
f
o
r
manc
e
of
c
las
s
if
ica
ti
on
ba
s
e
d
on
dif
f
e
r
e
nt
c
las
s
if
ier
s
.
M
e
th
ods
A
c
c
ur
a
c
y (
%
)
P
r
e
c
is
io
n (
%
)
R
e
c
a
ll
(
%
)
F1
-
s
c
or
e
(
%
)
I
nc
e
pt
io
nV
2
80.05
82.00
80.00
80.00
V
G
G
19
87.90
87.00
87.00
88.00
C
N
N
74.73
75.00
72.00
73.50
P
r
opos
e
d H
B
O
-
C
N
N
97.94
92.00
89.00
92.00
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
Hie
r
ar
c
hical
B
ay
e
s
ian
opti
miz
ati
on
bas
e
d
c
onv
olut
ional
ne
ur
al
…
(
B
har
ath
K
umar
Gow
r
u
)
577
F
igur
e
6.
P
e
r
f
or
manc
e
of
c
las
s
if
ica
ti
on
ba
s
e
d
on
di
f
f
e
r
e
nt
c
las
s
if
ier
s
3.
2.
Com
p
ar
a
t
ive
a
n
alys
is
T
he
pr
opos
e
d
HB
O
-
C
NN
method
is
c
ompar
e
d
with
the
e
xis
ti
ng
methods
,
C
he
XN
e
t
[
16]
,
VG
GN
e
t
19
[
17]
,
DC
NN
[
18]
a
nd
De
ns
e
Ne
t
121
[
19]
,
a
s
pr
e
s
e
nted
in
T
a
ble
5.
T
he
pr
opos
e
d
HB
O
-
ba
s
e
d
C
NN
method
a
c
hieve
s
a
c
omm
e
nda
ble
a
c
c
ur
a
c
y
of
97
.
94%
,
p
r
e
c
is
ion
of
92.
00
%
,
r
e
c
a
ll
of
89
.
0
0%
a
nd
F
1
-
s
c
or
e
of
92.
00%
,
whic
h
is
mor
e
pr
e
f
e
r
a
ble
than
the
other
tec
hniques
.
T
he
hybr
id
opti
mi
z
a
ti
on
-
ba
s
e
d
f
e
a
tur
e
s
e
lec
ti
on
method
is
us
e
d
in
the
pr
opos
e
d
method
to
c
hoos
e
the
r
e
leva
nt
f
e
a
tur
e
s
f
r
om
th
e
whole
f
e
a
tur
e
s
ubs
e
t,
whic
h
f
ur
ther
r
e
duc
e
s
the
dim
e
ns
ionalit
y
of
f
e
a
tur
e
s
a
nd
im
pr
ove
s
the
c
las
s
if
ier
pe
r
f
o
r
manc
e
.
T
he
n,
the
pa
r
a
mete
r
s
of
C
NN
a
r
e
opti
mi
z
e
d
by
uti
li
z
ing
the
HB
O
a
lgor
it
hm
whic
h
tunes
the
opti
mi
z
e
d
pa
r
a
mete
r
s
of
C
NN
whic
h
f
ur
ther
e
nha
nc
e
the
c
las
s
if
ica
ti
on
pe
r
f
or
manc
e
.
T
a
ble
5.
C
ompar
a
ti
ve
a
na
lys
is
of
the
pr
opos
e
d
met
hod
M
e
th
ods
A
c
c
ur
a
c
y (
%
)
P
r
e
c
is
io
n (
%
)
R
e
c
a
ll
(
%
)
F1
-
s
c
or
e
(
%
)
C
he
X
N
e
t
[
16]
87.88
N
/A
N
/A
N
/A
V
G
G
N
e
t
19
[
17]
90.5
91.5
90.3
87
D
C
N
N
[
18]
95.20
95.60
95.20
95.20
D
e
ns
e
N
e
t
121
[
19]
97
N
/A
N
/A
N
/A
P
r
opos
e
d H
B
O
-
C
N
N
97.94
92.00
89.00
92.00
3.
3.
Dis
c
u
s
s
ion
T
he
e
xis
ti
ng
methods
C
he
XN
e
t
[
16]
,
VG
GN
e
t
19
[
17]
,
DC
NN
[
18
]
a
nd
De
ns
e
Ne
t
121
[
19]
ha
ve
the
dr
a
wba
c
ks
of
the
ove
r
f
it
ti
ng
is
s
ue
s
,
high
dim
e
n
s
ionalit
y
of
f
e
a
tur
e
s
,
a
nd
no
c
ons
ider
a
ti
on
of
r
e
s
izing
of
im
a
ge
s
.
I
n
thi
s
r
e
s
e
a
r
c
h,
the
opti
mi
z
a
ti
on
o
f
C
NN
hype
r
pa
r
a
mete
r
s
tac
kles
the
is
s
ue
of
ove
r
f
it
ti
ng
,
a
nd
then,
by
us
ing
a
n
opt
im
iza
ti
on
-
ba
s
e
d
f
e
a
tur
e
s
e
lec
ti
on
method,
the
high
d
im
e
ns
ional
f
e
a
tur
e
s
a
r
e
mi
nim
ize
d
by
s
e
lec
ti
ng
r
e
leva
nt
f
e
a
tur
e
s
f
r
om
the
whole
f
e
a
tur
e
s
ubs
e
t.
T
he
im
a
ge
r
e
s
izing
is
pe
r
f
o
r
me
d
in
the
pr
e
-
pr
oc
e
s
s
ing
s
tag
e
to
unif
or
ml
y
pe
r
f
o
r
m
i
mage
r
e
s
izing,
e
ns
ur
ing
a
n
a
ugmente
d
c
las
s
if
ica
ti
on
out
put.
T
he
pr
opos
e
d
HB
O
-
C
N
N
method
e
xhibi
ts
a
n
e
f
f
e
c
ti
ve
c
las
s
if
ica
ti
on
of
c
he
s
t
X
-
r
a
ys
with
s
upe
r
ior
a
c
c
ur
a
c
y.
4.
CONC
L
USI
ON
I
n
thi
s
r
e
s
e
a
r
c
h,
the
HB
O
-
C
NN
method
is
pr
opos
e
d
to
e
f
f
e
c
ti
ve
ly
c
las
s
if
y
the
c
he
s
t
X
-
r
a
y
dis
e
a
s
e
s
.
T
he
pr
opos
e
d
HB
O
a
lgor
it
hm
opti
mi
z
e
s
the
pa
r
a
mete
r
s
of
C
NN
,
a
nd
mi
nim
ize
s
the
ove
r
f
i
tt
ing
is
s
ue
a
nd
e
nha
nc
e
s
the
pe
r
f
o
r
manc
e
of
c
las
s
if
ica
ti
on.
T
he
hybr
id
M
AO
a
nd
T
S
O
-
ba
s
e
d
f
e
a
tur
e
s
e
lec
ti
on
m
e
thod
is
e
mpl
oye
d
f
or
s
e
lec
ti
ng
the
r
e
leva
nt
f
e
a
tur
e
s
f
or
c
las
s
if
ica
ti
on,
ther
e
by
mi
nim
izing
the
high
dim
e
ns
ional
f
e
a
tur
e
s
.
T
he
p
r
opos
e
d
HB
O
-
C
NN
method
e
f
f
e
c
ti
ve
ly
c
las
s
if
ies
c
he
s
t
X
-
r
a
y
dis
e
a
s
e
with
high
a
c
c
ur
a
c
y.
I
t
a
ls
o
a
c
c
ompl
is
he
s
the
highes
t
a
c
c
ur
a
c
y
of
97.
95%
,
pr
e
c
is
ion
of
92
.
00%
,
r
e
c
a
ll
of
89
.
00%
a
nd
F
1
-
s
c
or
e
of
92.
00%
,
a
s
oppos
e
d
to
the
c
onve
nti
ona
l
methods
.
I
n
the
f
u
tur
e
,
hybr
id
c
las
s
if
ier
s
c
a
n
be
us
e
d
f
or
c
he
s
t
X
-
r
a
y
c
las
s
if
ica
ti
on
to
f
ur
ther
im
pr
ove
the
pe
r
f
or
manc
e
.
RE
F
E
RE
NC
E
S
[
1]
D
.
R
.
B
e
dd
ia
r
,
M
.
O
u
s
s
a
l
a
h
,
U
.
M
uh
a
m
ma
d,
a
nd
T
.
S
e
pp
ä
n
e
n
,
“
A
D
e
e
p
l
e
a
r
ni
ng
b
a
s
e
d
da
ta
a
ug
me
nt
a
ti
o
n
m
e
th
od
to
im
pr
o
v
e
C
O
V
I
D
-
19
d
e
t
e
c
ti
on
f
r
om
m
e
d
ic
a
l
i
m
a
gi
ng
,
”
K
no
w
l
e
dg
e
-
B
as
e
d S
y
s
te
m
s
,
v
ol
.
28
0,
N
ov
.
20
23
,
do
i:
1
0.
10
16
/j
.
kn
o
s
y
s
.2
02
3.
11
09
85.
[
2]
C
.
Z
ha
ng,
J
.
H
e
,
a
n
d
L
.
S
ha
ng,
“
A
n
X
-
r
a
y
im
a
ge
c
la
s
s
if
ic
a
ti
on
me
t
hod
w
it
h
f
in
e
-
gr
a
in
e
d
f
e
a
tu
r
e
s
f
or
e
xp
la
i
na
bl
e
di
a
gn
os
i
s
o
f
pne
um
oc
o
ni
o
s
i
s
,”
P
e
r
s
o
nal
and
U
bi
q
ui
to
u
s
C
om
put
i
ng
,
vol
.
28,
no.
2, p
p.
4
03
–
4
15,
A
pr
.
2024
,
doi
:
10
.100
7/
s
0077
9
-
023
-
0173
0
-
3.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
569
-
579
578
[
3]
M
.
I
.
R
a
ja
b,
“
C
la
s
s
if
ic
a
ti
on
of
C
O
V
I
D
-
19
c
he
s
t
X
-
r
a
y
im
a
ge
s
ba
s
e
d
on
s
pe
e
de
d
up
r
obus
t
f
e
a
tu
r
e
s
a
nd
c
lu
s
te
r
in
g
-
ba
s
e
d
s
uppor
t
ve
c
to
r
ma
c
hi
ne
s
,”
A
ppl
ie
d C
om
put
e
r
Sy
s
te
m
s
, vol
. 28, no. 1, pp
. 163
–
169, J
un. 2023, doi:
10.2478/ac
s
s
-
2023
-
0016.
[
4]
S
.
A
s
if
,
Y
.
W
e
nhui
,
K
.
A
mj
a
d,
H
.
J
in
,
Y
.
T
a
o,
a
nd
S
.
J
in
ha
i,
“
D
e
te
c
ti
on
of
C
O
V
I
D
‐
19
f
r
om
c
he
s
t
X
‐
r
a
y
im
a
ge
s
:
boo
s
ti
ng
th
e
pe
r
f
or
ma
nc
e
w
it
h
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
a
nd
tr
a
n
s
f
e
r
le
a
r
ni
ng,”
E
x
pe
r
t
Sy
s
te
m
s
,
vol
.
40,
no.
1,
J
a
n.
2023,
doi
:
10.1111/e
xs
y.13099.
[
5]
A
.
K
.
D
ube
y
a
nd
K
.
K
.
M
ohbe
y,
“
E
na
bl
in
g
C
T
-
s
c
a
n
s
f
or
c
ovi
d
de
te
c
ti
on
us
in
g
tr
a
ns
f
e
r
le
a
r
ni
ng
-
ba
s
e
d
ne
ur
a
l
ne
twor
ks
,”
J
ou
r
nal
of
B
io
m
ol
e
c
ul
ar
St
r
uc
tu
r
e
and D
y
nam
ic
s
, vol
. 41, no. 6, pp. 25
28
–
2539, Apr
. 2023, doi:
10.1080/07391102.
2022.2034668.
[
6]
C
.
C
.
U
kw
uoma
e
t
al
.
,
“
A
ut
oma
te
d
L
ung
-
r
e
la
te
d
pne
umon
ia
a
nd
C
O
V
I
D
-
19
de
te
c
ti
on
ba
s
e
d
on
nove
l
f
e
a
tu
r
e
e
xt
r
a
c
t
io
n
f
r
a
me
w
or
k
a
nd
vi
s
io
n
tr
a
ns
f
or
me
r
a
ppr
oa
c
he
s
u
s
in
g
c
he
s
t
X
-
r
a
y
im
a
ge
s
,”
B
io
e
ngi
ne
e
r
in
g
,
vol
.
9,
no.
11,
N
ov.
2
022,
doi
:
10.3390/bi
oe
ngi
ne
e
r
in
g9110709.
[
7]
P
.
P
a
ti
l
a
nd
V
.
N
a
r
a
w
a
de
,
“
D
e
e
p
c
onvolut
io
n
ne
ur
a
l
n
e
tw
or
k
f
or
r
e
s
pi
r
a
to
r
y
di
s
e
a
s
e
s
de
t
e
c
ti
on
us
in
g
r
a
di
ol
ogy
im
a
g
e
s
,”
I
nt
e
r
nat
io
nal
J
our
nal
of
I
nt
e
ll
ig
e
nt
Sy
s
te
m
s
and A
ppl
ic
at
io
ns
i
n
E
ngi
ne
e
r
in
g
, vol
. 12, no. 2, pp. 686
–
704, 2024.
[
8]
B
.
K
r
is
hna
,
T
.
C
ha
li
ka
nt
i,
B
.
S
.
R
e
ddy,
a
nd
C
.
N
.
R
a
j,
“
E
nha
nc
in
g
th
or
a
x
di
s
e
a
s
e
c
la
s
s
if
ic
a
ti
on
in
c
he
s
t
X
-
r
a
y
im
a
g
e
s
th
r
o
ugh
a
dva
nc
e
de
e
p
le
a
r
ni
ng
te
c
hni
que
s
,”
I
nt
e
r
nat
io
nal
J
our
nal
of
I
nnov
at
iv
e
Sc
ie
nc
e
and
R
e
s
e
ar
c
h
T
e
c
hnol
ogy
(
I
J
I
SR
T
)
,
vol
.
8,
no
.
8,
2023.
[
9]
R.
-
K
.
S
he
u,
M
.
S
.
P
a
r
de
s
hi
,
K
.
-
C
.
P
a
i,
L
.
-
C
.
C
he
n,
C
.
-
L
.
W
u,
a
nd
W
.
-
C
.
C
he
n,
“
I
nt
e
r
pr
e
ta
bl
e
c
la
s
s
if
ic
a
ti
on
of
pne
umoni
a
in
f
e
c
ti
on us
in
g e
X
pl
a
in
a
bl
e
A
I
(
X
A
I
-
I
C
P
)
,”
I
E
E
E
A
c
c
e
s
s
, vol
. 11, pp. 28896
–
28919, 2023,
doi
:
10.1109/AC
C
E
S
S
.2023.3255
403.
[
10]
C
.
C
.
U
kw
uoma
e
t
al
.
,
“
D
ua
l_
P
a
c
hi
:
a
tt
e
nt
io
n
-
ba
s
e
d
dua
l
pa
th
f
r
a
me
w
or
k
w
it
h
in
te
r
me
di
a
te
s
e
c
ond
or
de
r
-
pool
in
g
f
or
C
ovi
d
-
19
de
te
c
ti
on
f
r
om
c
h
e
s
t
X
-
r
a
y
im
a
ge
s
,”
C
om
put
e
r
s
in
B
io
lo
gy
and
M
e
di
c
in
e
,
vol
.
151,
D
e
c
.
2
022,
doi
:
10.1016/j
.c
ompbi
ome
d.2022.106324.
[
11]
U
.
C
.
A
yt
a
ç
,
A
.
G
üne
ş
,
a
nd
N
.
A
jl
ouni
,
“
A
nove
l
a
da
pt
i
ve
mom
e
nt
um
me
th
od
f
or
me
di
c
a
l
im
a
ge
c
la
s
s
if
ic
a
ti
on
us
in
g
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k,”
B
M
C
M
e
di
c
al
I
m
agi
ng
, vol
. 22,
no. 1, De
c
. 2022, doi:
10.1186/s
12880
-
022
-
00755
-
z.
[
12]
H
.
A
s
gha
r
ne
z
ha
d
e
t
al
.
,
“
O
bj
e
c
ti
ve
e
va
lu
a
ti
on
of
de
e
p
unc
e
r
ta
in
ty
pr
e
di
c
ti
ons
f
or
C
O
V
I
D
-
19
de
te
c
ti
on,”
Sc
ie
nt
if
ic
R
e
por
ts
,
vol
. 12, no. 1, J
a
n. 2022, doi:
10.1038/s
41598
-
022
-
05052
-
x.
[
13]
M
.
M
uj
a
hi
d,
F
.
R
us
t
a
m,
R
.
Á
lv
a
r
e
z
,
J
.
L
ui
s
V
id
a
l
M
a
z
ón,
I
.
d
e
la
T
.
D
íe
z
,
a
nd
I
.
A
s
hr
a
f
,
“
P
ne
umoni
a
c
la
s
s
if
ic
a
ti
on
f
r
om
X
-
r
a
y
im
a
ge
s
w
it
h
I
nc
e
pt
io
n
-
V
3
a
nd
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k,”
D
ia
gnos
ti
c
s
,
vol
.
12,
no.
5,
M
a
y
2022,
doi
:
10.3390/di
a
gnos
ti
c
s
12051280.
[
14]
S
.
G
oya
l
a
nd
R
.
S
in
gh,
“
D
e
te
c
ti
on
a
nd
c
la
s
s
if
ic
a
ti
on
of
lu
ng
di
s
e
a
s
e
s
f
or
pne
umoni
a
a
nd
C
ovi
d
-
19
us
in
g
ma
c
hi
ne
a
nd
d
e
e
p
le
a
r
ni
ng
te
c
hni
que
s
,”
J
our
nal
of
A
m
bi
e
nt
I
nt
e
ll
ig
e
nc
e
and
H
um
ani
z
e
d
C
om
put
in
g
,
vol
.
14,
no.
4,
pp.
3239
–
3259,
A
pr
.
2
023,
doi
:
10.1007/s
12652
-
021
-
03464
-
7.
[
15]
H
.
A
.
S
a
nghvi,
R
. H
.
P
a
te
l,
A
. A
ga
r
w
a
l,
S
.
G
upt
a
,
V
.
S
a
w
hne
y
,
a
nd
A
.
S
.
P
a
ndya
,
“
A
de
e
p
le
a
r
ni
ng
a
ppr
oa
c
h
f
or
c
la
s
s
if
ic
a
ti
on
of
C
O
V
I
D
a
nd
pn
e
umoni
a
u
s
in
g
D
e
n
s
e
N
e
t‐
201,”
I
nt
e
r
nat
io
nal
J
our
nal
of
I
m
agi
ng
Sy
s
te
m
s
and
T
e
c
hnol
ogy
,
vol
.
33,
no.
1,
pp. 18
–
38, J
a
n. 2023, doi:
10.1002/i
ma
.22812.
[
16]
A
.
H
a
gha
ni
f
a
r
,
M
.
M
.
M
a
jd
a
ba
di
,
Y
.
C
hoi
,
S
.
D
e
iv
a
la
ks
hmi
,
a
nd
S
.
K
o,
“
C
O
V
I
D
-
C
X
N
e
t:
de
te
c
ti
ng
C
O
V
I
D
-
19
in
f
r
ont
a
l
c
he
s
t
X
-
r
a
y
im
a
ge
s
us
in
g
de
e
p
le
a
r
ni
ng,”
M
ul
ti
m
e
di
a
T
ool
s
and
A
ppl
ic
at
io
ns
,
vol
.
81,
no.
21,
pp.
30615
–
30645,
S
e
p.
20
22,
doi
:
10.1007/s
11042
-
022
-
12156
-
z.
[
17]
I
.
C
houa
t,
A
.
E
c
ht
io
ui
,
R
.
K
h
e
ma
khe
m,
W
.
Z
ouc
h,
M
.
G
hor
be
l,
a
nd
A
.
B
e
n
H
a
mi
da
,
“
C
O
V
I
D
-
19
de
te
c
ti
on
in
C
T
a
nd
C
X
R
im
a
ge
s
us
in
g de
e
p l
e
a
r
ni
ng mode
ls
,”
B
io
ge
r
ont
ol
ogy
, vol
. 23, n
o. 1, pp. 65
–
84, F
e
b. 2022, doi:
10.1007/s
10522
-
021
-
09946
-
7.
[
18]
T
.
A
gr
a
w
a
l
a
nd
P
.
C
houdha
r
y,
“
F
oc
us
C
ovi
d:
a
ut
oma
te
d
C
O
V
I
D
-
19
de
te
c
ti
on
us
in
g
de
e
p
le
a
r
ni
ng
w
it
h
c
he
s
t
X
-
r
a
y
im
a
g
e
s
,”
E
v
ol
v
in
g Sy
s
te
m
s
, vol
. 13, no. 4, pp. 519
–
533, Aug. 2022, doi:
10.1007/s
12530
-
021
-
09385
-
2.
[
19]
S
.
A
gga
r
w
a
l,
S
.
G
upt
a
,
A
.
A
lh
udha
if
,
D
.
K
ounda
l,
R
.
G
upt
a
,
a
nd
K
.
P
ol
a
t,
“
A
ut
oma
te
d
C
O
V
I
D
‐
19
de
te
c
ti
on
in
c
he
s
t
X
‐
r
a
y
im
a
ge
s
us
in
g f
in
e
‐
tu
ne
d de
e
p l
e
a
r
ni
ng a
r
c
hi
te
c
tu
r
e
s
,”
E
x
pe
r
t
Sy
s
te
m
s
, vol
. 39, no. 3, M
a
r
. 2022, doi:
10.1111/exs
y.12749.
[
20]
Z
.
M
ous
a
vi
,
N
.
S
ha
hi
ni
,
S
.
S
he
ykhi
va
nd,
S
.
M
oj
ta
h
e
di
,
a
nd A
. A
r
s
ha
di
,
“
C
O
V
I
D
-
19
de
t
e
c
ti
on
us
in
g c
he
s
t
X
-
r
a
y
im
a
ge
s
ba
s
e
d
on
a
de
ve
lo
pe
d de
e
p n
e
ur
a
l
ne
twor
k,”
SL
A
S T
e
c
hnol
ogy
, vol
. 27, no. 1, pp. 63
–
75, F
e
b. 2022, doi:
10.1016/j
.s
la
s
t.
2021.10.011.
[
21]
P
.
P
a
te
l,
“
C
he
s
t
X
-
r
a
y
(
C
ovi
d
-
19
&
P
ne
umoni
a
)
,”
K
aggl
e
,
ht
tp
s
:/
/ww
w
.ka
ggl
e
.c
om/
da
ta
s
e
ts
/p
r
a
s
ha
nt
268/
c
h
e
s
t
-
xr
a
y
-
c
ovi
d19
-
pne
umoni
a
/d
a
ta
(
a
c
c
e
s
s
e
d S
e
p 1, 2024)
.
[
22]
S
.
R
.
N
a
ya
k,
J
.
N
a
y
a
k,
U
.
S
in
ha
,
V
.
A
r
or
a
,
U
.
G
hos
h,
a
nd
S
.
C
.
S
a
ta
pa
th
y,
“
A
n
a
ut
oma
te
d
li
ght
w
e
ig
ht
de
e
p
ne
ur
a
l
ne
twor
k
f
or
di
a
gnos
is
of
C
O
V
I
D
-
19
f
r
om
c
he
s
t
X
-
r
a
y
im
a
ge
s
,”
A
r
a
bi
an
J
our
nal
fo
r
Sc
ie
nc
e
and
E
ngi
ne
e
r
in
g
,
vol
.
48,
no.
8,
pp. 11085
–
11102, Aug. 2023, d
oi
:
10.1007/s
13369
-
021
-
05956
-
2.
[
23]
D
.
A
lS
a
e
e
d
a
nd
S
.
F
.
O
ma
r
,
“
B
r
a
in
M
R
I
a
na
ly
s
i
s
f
or
A
lz
he
i
me
r
’
s
di
s
e
a
s
e
di
a
gno
s
is
u
s
in
g
C
N
N
-
ba
s
e
d
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
a
nd
ma
c
hi
ne
l
e
a
r
ni
ng,”
Se
ns
o
r
s
, vol
. 22, no. 8, Apr
. 2022, doi:
10.3
390/
s
22082911.
[
24]
Y
.
V
il
lu
e
nda
s
-
R
e
y,
J
.
L
.
V
e
lá
z
que
z
-
R
odr
íg
ue
z
,
M
.
D
.
A
la
ni
s
-
T
a
me
z
,
M
.
-
A
.
M
or
e
no
-
I
ba
r
r
a
,
a
nd
C
.
Y
á
ñe
z
-
M
á
r
que
z
,
“
M
e
xi
c
a
n
a
xol
ot
l
opt
im
iz
a
ti
on:
a
nove
l
bi
oi
ns
pi
r
e
d he
ur
is
ti
c
,”
M
at
he
m
at
ic
s
, vol
. 9, no. 7, Apr
. 2021, doi:
10.3390/m
a
th
9070781.
[
25]
L
.
X
ie
,
T
.
H
a
n,
H
.
Z
hou,
Z
.
-
R
.
Z
ha
ng,
B
.
H
a
n,
a
nd
A
.
T
a
ng,
“
T
una
s
w
a
r
m
opt
im
iz
a
ti
on:
a
nove
l
s
w
a
r
m
-
ba
s
e
d
me
ta
he
ur
is
ti
c
a
lg
or
it
hm
f
or
gl
oba
l
opt
im
iz
a
ti
on,”
C
om
put
at
io
nal
I
nt
e
l
li
ge
nc
e
and
N
e
ur
o
s
c
ie
nc
e
,
vol
.
2021,
no.
1,
J
a
n.
20
21,
doi
:
10.1155/2021/
9210050.
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
Bha
ra
th
Kum
a
r
Go
w
ru
Pu
rs
u
i
n
g
Ph
.
D
.
i
n
co
mp
u
t
er
s
ci
en
ce
an
d
en
g
i
n
eeri
n
g
at
G
IT
A
M
D
eemed
t
o
b
e
u
n
i
v
er
s
i
t
y
,
V
i
zag
,
A
n
d
h
ra
P
rad
es
h
,
In
d
i
a.
H
e
s
ec
u
red
mas
t
er
o
f
t
ech
n
o
l
o
g
y
i
n
co
mp
u
t
er
s
c
i
en
ce
an
d
t
ech
n
o
l
o
g
y
at
V
R
Si
d
d
h
ar
t
h
a
E
n
g
i
n
eer
i
n
g
Co
l
l
eg
e,
V
i
j
ay
a
w
ad
a,
A
n
d
h
ra
Prad
es
h
,
In
d
i
a.
H
e
g
rad
u
at
e
d
en
g
i
n
eer
i
n
g
i
n
i
n
f
o
rmat
i
o
n
t
ech
n
o
l
o
g
y
at
Bap
at
l
a
E
n
g
i
n
eeri
n
g
C
o
l
l
eg
e,
Bap
a
t
l
a,
A
n
d
h
ra
Prad
e
s
h
,
In
d
i
a.
H
i
s
res
earc
h
area
i
s
b
i
g
d
at
a
an
al
y
t
i
cs
,
art
i
fi
ci
a
l
i
n
t
e
l
l
i
g
e
n
ce,
mach
i
n
e
l
earn
i
n
g
a
n
d
c
l
o
u
d
co
m
p
u
t
i
n
g
.
H
e
can
b
e
c
o
n
t
act
e
d
at
emai
l
:
b
h
ara
t
h
.
k
u
mar
4
3
6
@
g
mai
l
.
co
m.
Evaluation Warning : The document was created with Spire.PDF for Python.