I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
14
, N
o.
2
,
A
pr
il
2025
, pp.
1461
~
1470
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
2
.pp
1461
-
1470
1461
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
E
n
h
a
n
c
i
n
g f
al
l
d
e
t
e
c
t
i
on
an
d
c
l
ass
i
f
i
c
at
i
on
u
si
n
g
J
ar
r
at
t
‐
b
u
t
t
e
r
f
l
y op
t
i
m
i
z
at
i
on
al
gor
i
t
h
m
w
i
t
h
d
e
e
p
l
e
ar
n
i
n
g
K
ak
ir
al
a
D
u
r
ga B
h
avan
i,
M
e
lk
ia
s
F
e
r
n
i
U
k
r
it
D
e
pa
r
t
m
e
nt
of
C
om
put
a
t
i
ona
l
I
nt
e
l
l
i
ge
nc
e
, S
R
M
I
ns
t
i
t
ut
e
of
S
c
i
e
nc
e
a
nd T
e
c
h
nol
ogy, K
a
t
t
a
nkul
a
t
hur
, I
ndi
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
A
pr
6, 2024
R
e
vi
s
e
d
N
ov 3, 2024
A
c
c
e
pt
e
d
N
ov 14, 2024
Falls
pose
significant
risk
to
the
health
and
safety
of
individuals,
speci
fically
for
vulnerable
populations
as
the
elderly
and
those
with
specific
medical
conditi
ons.
The
repercussio
ns
of
falls
can
be
severe,
leading
to
injurie
s,
loss
of
independence,
and
incre
ased
healthcare
costs.
Consequentl
y,
the
development
of
effective
fall
detection
systems
is
crucial
for
pro
viding
timely
assistance
and
enhancing
the
overall
well
-
being
of
affected
individuals.
Recent
advance
ments
in
deep
learning
(DL)
have
opene
d
ne
w
avenues
for
automati
ng
fall
detection
through
the
analysis
of
sensor
d
ata
and
video
footage.
DL
algorithms
are
especially
well
-
suited
for
this
task
b
ecause
they
can
automatically
learn
complex
features
and
patterns
from
ra
w
data,
elimin
ating
the
need
for
extensiv
e
manual
feature
engineerin
g.
This
article
introduces
a
novel
approac
h
to
fall
detection
and
classifica
tion,
term
ed
the
fall detecti
on and class
ification
using
Jarratt‐butter
fly optimization alg
orithm
with
deep
learning
(FDC
-
JBOADL)
algorithm.
The
FDC
-
JB
OADL
technique
employs
a
median
filtering
(MF)
method
to
mitigate
noi
se
and
utilizes
the
EfficientNet
model
for
robust
feature
extraction,
capturin
g
both
motion
patterns
and
appearance
characteristics
of
individuals.
Furthe
rmore,
the
classification
of
fall
events
is
achieved
through
a
long
sho
rt
-
term
memory
(LSTM)
classifi
er,
with
hyperparamet
er
optimi
zation
facilita
ted
by
Jarratt‐butter
fly
optimization
algorithm
(
JBOA
)
.
Through
a
compreh
ensive
series
of
experiments,
the
efficacy
of
FDC
-
JBOADL
techni
que
is
val
idated,
demonstrating
superior
performance
compared
to
existing
methodolo
gies
in
the domain of
fall dete
ction.
K
e
y
w
o
r
d
s
:
C
om
put
e
r
vi
s
io
n
D
e
e
p l
e
a
r
ni
ng
F
a
ll
de
te
c
ti
on
M
a
c
hi
ne
l
e
a
r
ni
ng
M
e
ta
he
ur
is
ti
c
s
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
M
e
lk
ia
s
F
e
r
ni
U
kr
it
D
e
pa
r
tm
e
nt
of
C
om
put
a
ti
ona
l
I
nt
e
ll
ig
e
nc
e
,
S
R
M
I
ns
ti
tu
te
of
S
c
ie
nc
e
a
nd T
e
c
hnol
ogy
K
a
tt
a
nkul
a
th
ur
-
603203, C
he
nna
i,
I
ndi
a
E
m
a
il
:
f
e
r
ni
ukm
@
s
r
m
is
t.
e
du.i
n
1.
I
N
T
R
O
D
U
C
T
I
O
N
T
he
in
c
r
e
a
s
in
g
a
gi
ng
of
th
e
popula
ti
on,
pa
r
ti
c
ul
a
r
ly
in
de
ve
lo
pi
ng
na
ti
ons
,
pos
e
s
a
s
ig
ni
f
ic
a
nt
c
ha
ll
e
nge
to
th
e
s
u
s
ta
in
a
bi
li
ty
of
m
e
di
c
a
l
tr
e
a
tm
e
nt
s
[
1]
.
T
he
p
r
opor
ti
on
of
pe
opl
e
of
w
or
ki
ng
a
ge
(
15
to
64)
a
m
ong
th
e
to
ta
l
popula
ti
on
in
E
ur
ope
a
n
na
ti
ons
is
p
r
oj
e
c
te
d
t
o
de
c
li
ne
f
r
om
65.16%
in
2016
to
56.15%
by
2070,
w
hi
le
li
f
e
e
xpe
c
ta
nc
y
a
t
bi
r
th
is
a
nt
ic
ip
a
te
d
to
r
is
e
by
a
n
a
ddi
ti
ona
l
7
ye
a
r
s
f
or
bot
h
w
om
e
n
a
nd
m
e
n
ove
r
th
e
s
a
m
e
ti
m
e
f
r
a
m
e
[
2]
.
I
n
th
is
in
s
ta
nc
e
,
f
a
ll
s
ha
ve
be
e
n
a
s
ig
ni
f
ic
a
nt
c
a
us
e
of
lo
s
s
of
a
ut
onomy
a
nd
a
c
c
id
e
nt
s
a
m
ong
th
e
e
ld
e
r
ly
.
T
h
e
W
or
ld
H
e
a
lt
h
O
r
ga
ni
z
a
ti
on
(
W
H
O
)
r
e
s
e
a
r
c
h
in
di
c
a
te
s
th
a
t
th
e
a
nnua
l
f
a
ll
r
a
te
f
or
th
os
e
a
ge
d
64
to
70
is
a
r
ound
28
to
35%
a
nd
32
t
o
40%
,
r
e
s
pe
c
ti
ve
ly
[
3]
.
D
e
s
pi
te
th
e
la
c
k
of
s
ig
ni
f
ic
a
nt
in
ju
r
ie
s
,
47%
o
f
in
di
v
id
ua
ls
w
ho
f
a
ll
a
r
e
una
bl
e
to
r
is
e
a
lo
ne
pos
t
-
f
a
ll
[
4]
.
M
or
e
ove
r
,
pr
ol
onge
d
pe
r
io
ds
of
ly
in
g
on
th
e
g
r
ound
pr
io
r
to
f
a
ll
in
g
a
r
e
s
ig
ni
f
ic
a
nt
l
y
a
s
s
oc
ia
te
d
w
it
h
c
om
or
bi
di
ti
e
s
a
nd
pr
e
s
s
ur
e
s
or
e
s
,
w
hi
c
h
in
c
r
e
a
s
e
th
e
li
ke
li
hood
of
m
or
ta
li
ty
w
it
hi
n
s
ix
m
o
nt
hs
to
50%
[
5]
.
F
r
om
th
is
vi
e
w
poi
nt
,
pr
om
pt
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 2, A
pr
il
2025
:
1461
-
1470
1462
r
e
s
pons
e
a
f
te
r
a
f
a
ll
i
s
c
r
uc
ia
l
f
or
m
it
ig
a
ti
ng
th
e
phys
ic
a
l
a
n
d
ps
yc
hol
ogi
c
a
l
e
f
f
e
c
t
s
(
f
e
a
r
of
f
a
ll
in
g
(
F
oF
)
s
yndr
om
e
)
,
w
hi
c
h
unde
r
m
in
e
th
e
w
e
ll
-
be
in
g
of
e
ld
e
r
ly
in
di
vi
dua
ls
a
nd
th
e
ir
c
onf
id
e
nc
e
in
li
vi
ng
in
de
pe
nde
nt
ly
a
nd s
e
lf
-
s
uf
f
ic
ie
nt
ly
[
6]
.
F
a
ll
de
te
c
ti
on
s
y
s
te
m
(
F
D
S
)
,
is
pr
of
ic
ie
nt
in
di
s
c
r
im
in
a
ti
ve
f
a
ll
s
f
r
om
a
c
ti
vi
ti
e
s
of
da
il
y
li
vi
ng
(
ADL
)
th
e
r
e
by
a
n
a
la
r
m
to
r
e
m
ot
e
m
oni
to
r
in
g
poi
nt
ha
s
be
e
n
a
ut
om
a
ti
c
a
ll
y
pr
oduc
e
d
im
m
e
di
a
te
ly
,
a
nd
th
e
us
e
r
or
pa
ti
e
nt
unde
r
obs
e
r
va
ti
on
is
s
u
s
pe
c
te
d
to
h
a
ve
f
a
ll
e
n
[
7]
.
T
he
tr
a
di
ti
ona
l
m
a
c
hi
ne
le
a
r
ni
ng
(
M
L
)
a
ppr
oa
c
he
s
e
ndur
e
th
e
s
hor
ta
ge
of
la
be
ll
e
d
tr
a
in
in
g
da
ta
ba
s
e
s
a
nd
gr
e
a
tl
y
de
pe
nd
on
th
e
e
xt
r
a
c
te
d
f
e
a
tu
r
e
s
by
hum
a
ns
w
hi
c
h
c
r
e
a
te
s
it
ha
r
d
to
ut
i
li
z
e
on
m
a
s
s
iv
e
pl
a
tf
or
m
s
[
8]
.
D
e
e
p
le
a
r
ni
ng
(
D
L
)
ha
s
a
ne
w
pa
r
a
di
gm
i
n
th
e
M
L
dom
a
in
pr
im
a
r
il
y
de
te
r
m
in
e
d
by
u
ti
li
z
in
g
a
r
ti
f
ic
ia
l
ne
ur
a
l
ne
twor
ks
(
A
N
N
s
)
a
nd
gr
e
a
te
r
pe
r
f
or
m
a
nc
e
th
a
n t
he
ot
he
r
s
ta
nda
r
d M
L
a
ppr
oa
c
he
s
. T
he
D
L
i
nc
lu
de
s
di
f
f
e
r
e
nt
ne
twor
ks
na
m
e
ly
r
e
c
ur
r
e
nt
ne
ur
a
l
ne
twor
k
(
R
N
N
)
,
r
e
s
tr
ic
te
d
B
ol
tz
m
a
nn
m
a
c
hi
ne
s
(
R
B
M
s
)
,
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
(
C
N
N
)
,
de
e
p
be
li
e
f
ne
twor
k
(
D
B
N
)
,
w
hi
c
h
ha
ve
va
r
io
us
f
e
a
tu
r
e
s
a
nd
a
bi
li
ti
e
s
[
9]
.
T
h
e
s
e
ne
twor
ks
c
a
n
p
e
r
f
or
m
th
e
le
a
r
ni
ng
pr
oc
e
s
s
in
uns
upe
r
vi
s
e
d,
s
e
m
i
-
s
up
e
r
vi
s
e
d,
or
s
upe
r
vi
s
e
d
be
ha
vi
or
s
.
A
ls
o,
it
a
dva
nt
a
ge
s
f
r
om
th
e
hi
e
r
a
r
c
hi
c
a
l
la
ye
r
s
ta
r
ge
te
d
f
or
f
in
d
in
g
a
ppr
opr
ia
te
hi
ghe
r
-
le
ve
l
f
e
a
tu
r
e
s
f
r
om
th
e
r
a
w
in
put
da
ta
in
pl
a
c
e
of
u
ti
li
z
in
g
m
a
nua
l
f
e
a
tu
r
e
s
[
10]
.
T
hi
s
a
r
ti
c
le
pr
e
s
e
nt
s
a
ne
w
f
a
ll
de
te
c
ti
on
a
nd
c
la
s
s
if
ic
a
ti
on
us
in
g
J
a
r
r
a
tt
‐
b
ut
te
r
f
ly
opt
im
iz
e
r
a
lg
or
it
hm
w
it
h
d
e
e
p
le
a
r
ni
ng
(
F
D
C
-
J
B
O
A
D
L
)
te
c
hni
que
.
T
he
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
a
ppl
ie
s
m
e
di
a
n
f
il
te
r
in
g
(
M
F
)
a
ppr
oa
c
h
to
r
e
m
ove
th
e
noi
s
e
.
I
n
a
ddi
ti
on,
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
qu
e
m
a
ke
s
us
e
of
E
f
f
ic
ie
nt
N
e
t
m
ode
l
f
o
r
th
e
e
xt
r
a
c
ti
on
o
f
r
e
le
va
nt
f
e
a
tu
r
e
s
f
r
om
bot
h
m
ot
io
n
pa
tt
e
r
ns
a
nd
a
ppe
a
r
a
nc
e
c
ha
r
a
c
te
r
is
ti
c
s
of
in
di
vi
dua
ls
.
M
or
e
ove
r
,
th
e
c
l
a
s
s
if
ic
a
ti
on
of
f
a
ll
e
ve
nt
s
oc
c
ur
s
u
s
in
g
lo
ng
s
hor
t
-
te
r
m
m
e
m
or
y
(
L
S
T
M
)
ne
twor
k.
F
in
a
ll
y,
th
e
J
a
r
r
a
tt
‐
bu
tt
e
r
f
ly
opt
i
m
i
z
a
ti
on
a
lg
or
it
hm
(
J
B
O
A
)
c
a
n
be
e
m
pl
oye
d
f
or
th
e
opt
im
a
hype
r
pa
r
a
m
e
te
r
c
hoi
c
e
of
th
e
L
S
T
M
m
ode
l.
A
w
id
e
r
a
nge
of
e
xpe
r
im
e
nt
s
w
a
s
pe
r
f
or
m
e
d
to
va
li
da
te
t
he
s
upe
r
io
r
r
e
c
ogni
ti
on r
e
s
ul
ts
of
t
he
F
D
C
-
J
B
O
A
D
L
t
e
c
hni
que
.
2.
R
E
L
A
T
E
D
WORKS
I
n
r
e
c
e
nt
ye
a
r
s
,
th
e
in
te
gr
a
ti
on
of
in
nova
ti
ve
D
L
te
c
hni
que
s
a
nd
a
dva
nc
e
d
s
e
ns
or
te
c
hnol
ogi
e
s
ha
s
s
ig
ni
f
ic
a
nt
ly
e
nha
nc
e
d
th
e
c
a
pa
bi
li
ti
e
s
of
c
la
s
s
if
ic
a
ti
on
a
nd
m
oni
to
r
in
g
s
ys
te
m
s
in
va
r
io
us
dom
a
in
s
.
F
or
in
s
ta
nc
e
,
L
i
e
t
al
.
[
11
]
in
tr
oduc
e
d
a
nove
l
D
L
f
r
a
m
e
w
or
k
t
ha
t
in
te
gr
a
te
s
te
m
por
a
l
c
onvolut
io
n
ne
twor
ks
(
T
C
N
)
w
it
h ga
te
d r
e
c
ur
r
e
nt
uni
ts
(
G
R
U
)
, a
im
e
d a
t
e
xt
r
a
c
ti
ng hi
ghe
r
-
le
ve
l
f
e
a
tu
r
e
s
f
or
i
m
p
r
ove
d c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y. T
h
e
ir
r
e
s
e
a
r
c
h i
nvol
ve
d a
c
om
pa
r
a
ti
ve
a
na
ly
s
is
a
ga
in
s
t
2 e
xt
e
ns
iv
e
ly
ut
il
iz
e
d M
L
c
la
s
s
if
ie
r
s
a
nd s
ix
e
xi
s
ti
ng
D
L
a
ppr
oa
c
he
s
,
le
v
e
r
a
gi
ng
2
w
e
ll
-
e
s
ta
bl
is
h
e
d
ope
n
-
s
our
c
e
da
ta
s
e
t
s
c
ol
le
c
te
d
f
r
om
in
e
r
ti
a
l
s
e
ns
or
s
.
C
onc
ur
r
e
nt
ly
,
R
a
e
ve
e
t
al
.
[
12]
pr
opos
e
d
a
n
in
nova
ti
ve
f
a
ul
t
de
te
c
ti
on
a
nd
a
le
r
t
s
ys
t
e
m
ta
il
or
e
d
f
or
c
a
r
e
c
e
nt
e
r
s
,
w
hi
c
h
ut
il
iz
e
s
bl
ue
to
ot
h
lo
w
e
ne
r
gy
(
B
L
E
)
f
or
w
ir
e
le
s
s
c
om
m
uni
c
a
ti
on.
T
hi
s
s
tu
dy
a
ls
o
e
m
pha
s
iz
e
s
th
e
de
ve
lo
pm
e
nt
of
a
r
e
a
l
-
ti
m
e
da
ta
f
il
te
r
in
g
m
e
th
od
to
e
nha
nc
e
th
e
a
c
c
ur
a
c
y
of
m
e
a
s
ur
e
m
e
nt
s
.
A
ddi
ti
ona
ll
y,
th
e
e
xpl
or
a
ti
on
of
m
il
li
m
e
te
r
-
w
a
ve
(
m
m
W
a
ve
)
r
a
da
r
te
c
hnol
ogy
f
or
unobtr
us
iv
e
hum
a
n
f
a
ll
de
te
c
ti
on
ha
s
be
e
n
hi
ghl
ig
ht
e
d
in
[
13]
.
T
hi
s
r
e
s
e
a
r
c
h
in
vol
ve
d
c
ol
le
c
ti
ng
da
ta
f
r
om
he
a
lt
hy
young
vol
unt
e
e
r
s
,
w
it
h
r
a
da
r
s
ys
te
m
s
s
tr
a
te
gi
c
a
ll
y
pos
it
io
ne
d
e
it
he
r
on
th
e
s
id
e
w
a
ll
or
ove
r
he
a
d
w
it
hi
n
a
de
s
ig
na
te
d
a
r
e
a
.
T
o
a
ddr
e
s
s
th
e
unde
r
ly
in
g
f
a
ul
t
de
te
c
ti
on
c
ha
ll
e
nge
s
,
a
C
N
N
-
ba
s
e
d
D
L
a
ppr
oa
c
h
w
a
s
a
ls
o
d
e
ve
lo
pe
d.
C
ol
le
c
ti
ve
ly
,
th
e
s
e
s
tu
di
e
s
unde
r
s
c
or
e
th
e
tr
a
ns
f
or
m
a
ti
ve
pot
e
nt
ia
l
of
in
te
gr
a
ti
ng
a
dva
nc
e
d
D
L
m
e
th
odol
ogi
e
s
w
it
h
c
ut
ti
ng
-
e
dge
s
e
ns
or
t
e
c
hnol
ogi
e
s
t
o i
m
pr
ove
m
oni
to
r
in
g a
nd c
la
s
s
if
ic
a
ti
on t
a
s
ks
a
c
r
os
s
va
r
io
us
a
ppl
ic
a
ti
ons
.
T
hi
s
s
tu
dy
a
ls
o
de
ve
lo
ps
a
C
N
N
ba
s
e
d
a
ppr
oa
c
h
to
a
ddr
e
s
s
un
de
r
ly
in
g
F
D
c
ha
ll
e
nge
s
.
B
ui
ld
in
g
o
n
th
e
s
e
a
dva
nc
e
m
e
nt
s
,
Y
a
o
e
t
al
.
[
14]
pr
opos
e
d
a
n
e
f
f
ic
ie
nt
F
D
te
c
hni
que
ut
il
iz
in
g
a
jo
in
t
m
ot
io
n
m
a
p
c
ons
tr
uc
te
d
f
r
om
two
pa
r
a
ll
e
l
C
N
N
s
,
in
nova
ti
ve
ly
e
m
pl
oyi
ng
th
e
r
e
d,
g
r
e
e
n,
a
nd
bl
ue
(
R
G
B
)
c
ha
nne
ls
of
pi
xe
ls
to
c
a
pt
ur
e
r
e
la
ti
ve
m
ot
io
n
da
ta
.
T
he
ir
m
e
th
od
pr
e
di
c
ts
th
e
li
m
it
s
of
s
ta
bi
li
ty
a
nd
a
c
c
ur
a
te
ly
id
e
nt
if
ie
s
th
e
i
ni
ti
a
l
a
nd f
in
a
l
ke
y f
r
a
m
e
s
pr
e
c
e
di
ng a
pot
e
nt
ia
l
f
a
ll
. F
ur
th
e
r
m
or
e
, t
he
hybr
id
de
e
p C
N
N
m
ode
l
known a
s
s
que
e
z
e
a
nd
e
xc
it
a
ti
on
(
SE
)
-
D
e
e
p
C
onvNe
t,
de
s
ig
ne
d
by
M
e
kr
uks
a
va
ni
c
h
e
t
al
.
[
15]
,
e
nha
nc
e
s
f
a
ll
de
te
c
ti
on
c
a
pa
bi
li
ti
e
s
w
it
h
th
e
im
pl
e
m
e
nt
a
ti
on
of
s
que
e
z
e
a
nd
e
xc
it
a
ti
on
te
c
hni
que
s
.
C
ol
le
c
ti
ve
ly
,
th
e
s
e
s
tu
di
e
s
unde
r
s
c
or
e
th
e
tr
a
ns
f
or
m
a
ti
ve
pot
e
nt
ia
l
of
D
L
a
nd
a
dva
nc
e
d
s
e
ns
or
te
c
hnol
ogi
e
s
in
im
pr
ovi
ng
F
D
S
,
ul
ti
m
a
te
ly
c
ont
r
ib
ut
in
g t
o e
nha
nc
e
d s
a
f
e
ty
a
nd moni
to
r
in
g i
n va
r
io
us
e
nvi
r
onm
e
nt
s
.
I
n
r
e
c
e
nt
ye
a
r
s
,
th
e
a
dva
nc
e
m
e
nt
of
f
a
ul
t
d
ia
gnos
is
te
c
hni
que
s
ha
s
ga
in
e
d
s
ig
ni
f
ic
a
nt
a
tt
e
nt
io
n
in
va
r
io
us
f
ie
ld
s
,
m
os
tl
y
in
th
e
c
ont
e
xt
of
hi
gh
vol
ta
ge
di
r
e
c
t
c
ur
r
e
nt
(
H
V
D
C
)
m
ode
ls
a
nd
f
a
ll
de
te
c
ti
on
f
or
th
e
e
ld
e
r
ly
.
J
a
w
a
d
a
nd
A
bi
d
[
16]
pr
opos
e
d
a
nov
e
l
a
ppr
oa
c
h
f
or
H
V
D
C
f
a
ul
t
di
a
gnos
is
th
a
t
in
te
gr
a
te
s
a
pr
oba
bi
li
s
ti
c
ge
ne
r
a
ti
ve
a
lg
or
it
hm
ba
s
e
d
on
f
e
a
tu
r
e
s
e
le
c
ti
on
(
F
S
)
a
nd
w
a
ve
le
t
tr
a
ns
f
or
m
m
e
th
ods
.
T
he
ir
m
e
th
odol
ogy
in
vol
ve
s
th
e
e
xt
r
a
c
ti
on
of
noi
s
e
f
r
om
bot
h
non
-
f
a
ul
t
a
nd
f
a
ul
t
s
ig
na
ls
,
f
ol
lo
w
e
d
by
th
e
a
ppl
ic
a
ti
on of
a
nt
c
ol
ony opti
m
iz
a
ti
on
(
A
C
O
)
t
o e
li
m
in
a
te
i
r
r
e
l
e
va
nt
a
tt
r
ib
ut
e
s
w
it
hi
n t
he
f
e
a
tu
r
e
ve
c
to
r
s
. T
he
r
e
f
in
e
d
f
e
a
tu
r
e
s
a
r
e
s
ubs
e
que
nt
ly
ut
il
iz
e
d
f
or
tr
a
in
in
g
a
n
A
N
N
to
e
f
f
e
c
ti
ve
ly
di
s
ti
ngui
s
h
be
twe
e
n
non
-
f
a
ul
t
a
nd
f
a
ul
t
c
ondi
ti
ons
.
C
onc
ur
r
e
nt
ly
,
L
e
e
e
t
al
.
[
17]
in
tr
oduc
e
d
a
dua
l
ve
r
if
ic
a
ti
on
s
tr
a
te
gy
f
or
f
a
ll
de
te
c
ti
on
in
ol
de
r
a
dul
ts
,
e
m
pl
oyi
ng
a
c
om
bi
na
ti
on
of
R
G
B
c
a
m
e
r
a
s
a
nd
in
e
r
ti
a
l
m
e
a
s
ur
e
m
e
nt
uni
t
-
lo
c
a
ti
on
(
I
M
U
-
L
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
nhanc
in
g f
al
l
de
te
c
ti
on and c
la
s
s
if
ic
at
io
n us
in
g j
ar
r
at
t‐
but
te
r
fl
y
opt
imi
z
at
io
n
…
(
K
ak
ir
al
a
D
u
r
ga B
hav
ani
)
1463
s
e
ns
or
s
.
T
hi
s
a
ppr
oa
c
h
is
pa
r
ti
c
ul
a
r
ly
in
nova
ti
ve
a
s
it
le
v
e
r
a
ge
s
w
e
a
r
a
bl
e
te
c
hnol
ogy
to
m
oni
to
r
f
a
ll
s
,
w
it
h
th
e
I
M
U
-
L
s
e
ns
or
pr
ovi
d
in
g
r
e
a
l
-
t
im
e
de
te
c
ti
on
c
a
pa
bi
li
ti
e
s
.
T
o
e
nha
nc
e
th
e
a
c
c
ur
a
c
y
of
f
a
ll
c
la
s
s
if
ic
a
ti
on,
a
D
L
m
e
th
od
ut
il
iz
in
g
R
N
N
is
e
m
pl
oye
d,
m
a
r
ki
ng
a
s
ig
ni
f
ic
a
nt
s
te
p
f
or
w
a
r
d
in
th
e
c
ons
is
te
n
c
y
of
F
D
S
.
T
hi
s
r
e
s
e
a
r
c
h
hi
ghl
ig
ht
s
th
e
in
te
r
s
e
c
ti
on
of
a
dva
nc
e
d
s
ig
na
l
pr
oc
e
s
s
in
g
t
e
c
hni
que
s
a
nd
ML
m
e
th
odol
ogi
e
s
in
a
ddr
e
s
s
in
g c
r
it
ic
a
l
s
a
f
e
ty
c
onc
e
r
ns
i
n both e
le
c
tr
ic
a
l
e
ngi
ne
e
r
in
g a
nd ge
r
ia
tr
ic
c
a
r
e
.
3.
T
H
E
P
R
O
P
O
S
E
D
M
O
D
E
L
F
a
ll
s
r
e
pr
e
s
e
nt
a
m
a
jo
r
h
e
a
lt
h
r
i
s
k
, pa
r
ti
c
ul
a
r
l
y
a
m
on
g
th
e
e
ld
e
r
l
y pop
ul
a
ti
on
, l
e
a
d
in
g
t
o
s
e
ve
r
e
i
nj
ur
ie
s
a
nd
i
nc
r
e
a
s
e
d
m
or
t
a
l
it
y
r
a
te
s
.
A
s
s
u
c
h,
th
e
pr
o
gr
e
s
s
of
a
ut
om
a
t
e
d
s
y
s
t
e
m
s
f
or
f
a
ll
de
te
c
t
io
n
a
nd
c
l
a
s
s
if
i
c
a
ti
o
n
ha
s
ga
r
n
e
r
e
d
c
o
n
s
id
e
r
a
bl
e
a
t
te
nt
i
on
ov
e
r
r
e
c
e
nt
y
e
a
r
s
.
T
hi
s
r
e
s
e
a
r
c
h
in
tr
o
du
c
e
s
th
e
F
D
C
-
J
B
O
A
D
L
te
c
h
ni
qu
e
,
a
nov
e
l
a
ppr
o
a
c
h d
e
s
ig
n
e
d
t
o
d
e
v
e
l
op t
h
e
a
c
c
ur
a
c
y
a
nd e
f
f
e
c
ti
v
e
n
e
s
s
of
f
a
ll
e
v
e
n
t
i
de
nt
if
ic
a
ti
on
a
nd c
l
a
s
s
if
i
c
a
ti
on
th
r
o
ugh
t
he
i
nt
e
gr
a
ti
on
of
D
L
m
od
e
l
s
.
T
h
e
F
D
C
-
J
B
O
A
D
L
te
c
h
n
iq
u
e
e
m
pl
o
y
s
a
m
ul
ti
f
a
c
e
t
e
d
m
e
t
hod
ol
o
gy
t
ha
t
in
c
lu
d
e
s
noi
s
e
r
e
m
ov
a
l
us
in
g
a
M
F
,
f
e
a
t
ur
e
e
xt
r
a
c
ti
o
n
le
ve
r
a
g
i
ng
t
he
E
f
f
i
c
i
e
n
tN
e
t
a
r
c
hi
te
c
t
ur
e
,
f
a
ll
d
e
t
e
c
ti
o
n
us
in
g
L
S
T
M
,
a
nd
h
yp
e
r
p
a
r
a
m
e
t
e
r
op
ti
m
iz
a
ti
on
th
r
o
ug
h
th
e
J
B
O
A
a
l
gor
i
th
m
.
T
h
e
w
or
kf
l
ow
of
th
e
F
D
C
-
J
B
O
A
D
L
m
e
t
ho
d
i
s
e
xe
m
pl
if
i
e
d
in
F
i
g
ur
e
1,
d
e
m
on
s
tr
a
ti
ng
th
e
s
y
s
t
e
m
a
ti
c
pr
o
c
e
s
s
i
nvo
lv
e
d
in
r
e
c
og
ni
z
in
g
a
n
d
c
a
t
e
g
or
i
z
in
g
f
a
ll
e
ve
nt
s
.
T
hi
s
in
nov
a
t
iv
e
te
c
hn
iq
u
e
a
i
m
s
to
a
s
s
is
t
th
e
pr
ogr
e
s
s
of
a
ut
o
m
a
t
e
d
F
D
S
,
a
dv
a
n
c
e
d
e
n
ha
nc
e
m
e
n
t
t
he
s
a
f
e
t
y
a
nd
w
e
l
l
-
b
e
in
g
of
vul
ne
r
a
bl
e
po
pul
a
ti
on
s
.
F
ig
ur
e
1
.
W
or
kf
lo
w
of
F
D
C
-
J
B
O
A
D
L
a
ppr
oa
c
h
3.1.
I
m
age
p
r
e
-
p
r
oc
e
s
s
in
g
T
o
pr
e
-
pr
oc
e
s
s
th
e
in
put
im
a
ge
s
,
th
e
M
F
te
c
hni
que
is
e
xpl
oi
te
d
in
th
is
s
tu
dy.
I
t
is
a
nonl
in
e
a
r
di
gi
ta
l
im
a
ge
pr
oc
e
s
s
in
g
m
e
th
od
us
e
d
f
or
pr
e
s
e
r
vi
ng
e
dge
s
a
nd
r
e
duc
in
g
noi
s
e
in
im
a
g
e
s
.
I
t
is
ve
r
y
e
f
f
ic
ie
nt
a
t
e
li
m
in
a
ti
ng
s
a
lt
-
a
nd
-
pe
ppe
r
noi
s
e
s
,
w
he
r
e
r
a
ndom
w
hi
te
a
nd
bl
a
c
k
pi
xe
ls
a
ppe
a
r
th
r
oughout
th
e
im
a
ge
.
T
he
M
F
m
e
th
od
in
c
lu
de
s
s
ubs
ti
tu
ti
ng
th
e
va
lu
e
of
a
ll
th
e
pi
xe
l
s
w
it
h
m
e
di
a
n
va
lu
e
of
it
s
ne
ig
hbor
in
g
pi
xe
ls
w
it
hi
n t
he
gi
ve
n ke
r
ne
l
or
w
in
dow
.
3.2.
F
e
at
u
r
e
e
xt
r
ac
t
io
n
u
s
in
g E
f
f
ic
ie
n
t
N
e
t
m
od
e
l
F
or
e
f
f
e
c
tu
a
l
id
e
nt
if
ic
a
ti
on
of
th
e
f
e
a
tu
r
e
ve
c
to
r
s
,
th
e
E
f
f
ic
ie
nt
N
e
t
m
e
th
od
c
a
n
b
e
e
m
pl
oye
d.
E
f
f
ic
ie
nt
N
e
t
is
a
f
a
m
il
y
of
de
e
p
ne
ur
a
l
ne
twor
k
(
D
N
N
)
a
r
c
hi
te
c
tu
r
e
th
a
t
ha
s
be
e
n
in
te
nde
d
to
a
c
c
om
pl
is
h
r
e
m
a
r
ka
bl
e
pe
r
f
or
m
a
nc
e
w
hi
le
be
in
g
c
om
put
a
ti
ona
ll
y
e
f
f
e
c
ti
v
e
[
18]
.
T
he
ba
s
ic
c
onc
e
pt
be
hi
nd
E
f
f
ic
ie
nt
N
e
t
is
to
s
im
ul
ta
ne
ous
ly
ba
la
nc
e
th
e
m
ode
l'
s
de
pt
h,
w
id
th
,
a
nd
r
e
s
ol
ut
io
n
to
a
tt
a
in
be
s
t
out
c
om
e
s
w
it
h
le
s
s
c
om
put
a
ti
ona
l
pa
r
a
m
e
te
r
s
a
nd.
T
r
a
di
ti
ona
l
m
ode
l
s
c
a
li
ng
t
e
c
hni
que
s
f
oc
u
s
m
a
in
ly
on
in
c
r
e
a
s
in
g
th
e
di
m
e
ns
io
n
(
f
or
e
xa
m
pl
e
:
w
id
th
or
de
pt
h)
,
r
e
s
ul
ti
ng
in
s
ubopti
m
um
pe
r
f
or
m
a
nc
e
.
E
f
f
ic
ie
nt
N
e
t
m
a
ke
s
us
e
of
a
c
om
pound
c
oe
f
f
ic
ie
nt
to
uni
f
or
m
ly
s
c
a
le
th
e
3D
,
w
hi
c
h
is
de
r
iv
e
d
f
r
om
a
s
e
r
ie
s
of
e
xpe
r
im
e
nt
s
.
T
he
c
om
pound
s
c
a
li
ng
c
oe
f
f
ic
ie
nt
is
u
s
e
d
f
or
s
c
a
li
ng
th
e
r
e
s
ol
ut
io
n,
de
pt
h
(
num
be
r
of
la
ye
r
s
)
,
a
nd
w
id
th
(
num
be
r
of
c
ha
nne
ls
)
of
th
e
ne
twor
ks
.
T
he
E
f
f
ic
ie
nt
N
e
t
m
ode
l
ha
s
a
c
c
om
pl
is
he
d
out
s
ta
ndi
ng
pe
r
f
or
m
a
nc
e
s
on
c
om
put
e
r
vi
s
io
n
ta
s
k
w
hi
le
be
in
g
m
or
e
e
f
f
e
c
ti
ve
th
a
n
ot
he
r
D
e
ns
e
N
e
t
a
nd
R
e
s
N
e
t
a
r
c
hi
te
c
tu
r
e
s
.
T
he
y
a
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 2, A
pr
il
2025
:
1461
-
1470
1464
w
id
e
ly
a
ppl
ie
d
f
or
th
e
ta
s
ks
in
c
lu
di
ng
obj
e
c
t
d
e
te
c
ti
on,
im
a
ge
c
la
s
s
if
ic
a
ti
on,
a
nd
s
e
gm
e
nt
a
ti
on.
T
he
r
e
e
xi
s
t
num
e
r
ous
va
r
ia
nt
s
of
E
f
f
ic
ie
nt
N
e
t,
na
m
e
ly
E
f
f
ic
ie
nt
N
e
t
-
B0
-
B
7,
w
it
h
va
r
io
us
le
ve
ls
of
pe
r
f
or
m
a
nc
e
a
nd
c
om
pl
e
xi
ty
.
B
0
i
s
th
e
s
im
pl
e
s
t
a
nd
s
m
a
ll
e
s
t
ve
r
s
io
n,
w
h
e
r
e
a
s
B
7
is
th
e
m
os
t
c
om
pl
e
x
a
nd
la
r
ge
s
t
one
.
B
a
s
e
d
on t
he
c
om
put
a
ti
ona
l
r
e
s
our
c
e
a
va
il
a
bl
e
a
nd t
he
t
a
s
k r
e
qui
r
e
m
e
nt
, us
e
r
s
c
a
n
s
e
le
c
t
th
e
s
ui
ta
bl
e
va
r
ia
nt
.
3.3.
F
al
l
d
e
t
e
c
t
io
n
u
s
in
g
lo
n
g s
h
or
t
-
t
e
r
m
m
e
m
or
y
I
n
th
is
w
or
k,
th
e
L
S
T
M
f
r
a
m
e
w
or
k
c
a
n
be
ut
il
iz
e
d
f
or
th
e
id
e
nt
if
ic
a
ti
on
a
nd
c
la
s
s
if
ic
a
ti
on
of
f
a
ll
e
ve
nt
s
. T
he
L
S
T
M
ne
twor
k
i
s
a
n
im
pr
ove
d
m
e
th
od
of
a
n
R
N
N
[
19
]
,
[
20
]
.
D
if
f
e
r
e
nt
C
N
N
s
na
m
e
ly
m
ul
ti
la
ye
r
pe
r
c
e
pt
r
on
(
M
L
P
)
a
nd
R
N
N
s
c
oul
d
not
be
r
e
s
tr
ic
te
d
to
a
uni
d
ir
e
c
ti
ona
l
f
lo
w
of
da
ta
.
I
t
is
lo
op
done
m
a
ny
la
ye
r
s
a
nd
te
m
por
a
r
il
y
m
e
m
or
iz
e
s
da
t
a
w
hi
c
h
is
e
m
pl
oye
d
la
te
r
.
I
n
th
e
m
e
a
nt
im
e
,
a
n
e
a
s
y
R
N
N
is
vul
ne
r
a
bl
e
to
gr
a
di
e
nt
di
s
a
ppe
a
r
in
g
pr
obl
e
m
s
,
a
nd
th
e
G
R
U
a
nd
L
S
T
M
a
r
e
e
s
ta
bl
is
he
d
f
or
s
ol
vi
ng
th
e
pr
obl
e
m
.
T
he
L
S
T
M
le
a
r
ns
lo
ng‐
te
r
m
de
pe
nde
nc
ie
s
,
e
ndur
in
g
a
ppr
opr
ia
te
to
c
la
s
s
if
y
s
e
que
nt
ia
l
da
ta
li
ke
c
r
e
di
t
c
a
r
d
in
f
or
m
a
ti
on.
L
S
T
M
ne
twor
k
c
om
pr
is
e
s
m
e
m
or
y
c
e
ll
,
w
it
h
in
p
ut
ga
te
,
out
put
ga
te
,
a
nd
f
or
ge
t
ga
te
.
T
he
3 ga
te
s
c
ont
r
ol
t
ha
t
th
e
da
ta
ha
s
be
e
n m
a
na
ge
d a
nd e
m
pl
oye
d. F
ig
ur
e
2 i
ll
us
tr
a
te
s
t
he
s
tr
uc
tu
r
e
of
L
S
T
M
.
T
he
s
ub
s
e
que
nt
m
a
th
e
m
a
ti
c
a
l
e
qua
ti
ons
d
e
f
in
e
t
he
da
ta
f
lo
w
i
n t
he
L
S
T
M
l
a
ye
r
s
a
s
i
n (
1)
to
(
6)
.
=
(
+
ℎ
−
1
+
)
(
1)
=
(
+
ℎ
(
−
1
)
+
)
(
2)
̃
=
ℎ
(
+
ℎ
(
−
1
)
+
)
(
3)
=
⊗
(
−
1
)
+
⊗
̃
(
4)
=
(
+
ℎ
(
−
1
)
+
)
(
5)
ℎ
=
⊗
ℎ
(
)
(
6)
W
he
r
e
a
s
∗
,
∗
,
a
nd
∗
s
ig
ni
f
ie
s
th
e
le
a
r
na
bl
e
pa
r
a
m
e
te
r
,
ℎ
∗
de
not
e
s
th
e
hi
dde
n
la
ye
r
,
w
hi
c
h,
∗
is
e
m
pl
oye
d
in
pl
a
c
e
of
,
,
,
or
to
s
ig
ni
f
y
th
e
pr
ovi
de
d
m
e
m
or
y
c
e
ll
a
nd
ga
te
s
.
I
n
th
e
m
e
a
nt
im
e
,
⊗
de
not
e
s
th
e
e
le
m
e
nt
‐
by
-
e
le
m
e
nt
pr
oduc
t;
ℎ
a
nd
de
not
e
t
he
t
a
n
h
a
c
ti
va
ti
on a
nd s
ig
m
oi
d f
unc
ti
ons
.
F
ig
ur
e
2.
L
S
T
M
a
r
c
hi
te
c
tu
r
e
3.4.
P
ar
am
e
t
e
r
t
u
n
in
g
u
s
in
g
Jar
r
at
t
‐
b
u
t
t
e
r
f
ly
op
t
im
iz
at
io
n
al
gor
it
h
m
I
n
r
e
c
e
nt
ye
a
r
s
,
th
e
opt
im
iz
a
ti
on
of
L
S
T
M
ha
s
ga
r
ne
r
e
d
s
ig
ni
f
ic
a
nt
a
tt
e
nt
io
n
be
c
a
us
e
of
o
th
e
ir
e
f
f
ic
a
c
y i
n ha
ndl
in
g s
e
que
nt
ia
l
da
ta
. O
ne
of
t
he
pr
o
m
is
in
g s
tr
a
te
gi
e
s
f
or
e
nha
nc
in
g t
he
pe
r
f
or
m
a
nc
e
of
L
S
T
M
ne
twor
ks
is
th
e
in
te
gr
a
ti
on
of
opt
im
iz
a
ti
on
a
lg
or
it
h
m
s
,
s
uc
h
a
s
th
e
but
te
r
f
ly
opt
i
m
iz
a
ti
on
a
lg
or
it
hm
(
B
O
A
)
.
W
hi
le
B
O
A
ha
s
r
e
ve
a
le
d
th
a
t
a
pow
e
r
f
ul
to
ol
f
or
va
r
io
us
a
ppl
ic
a
ti
ons
,
it
is
not
w
i
th
out
it
s
c
ha
ll
e
nge
s
,
pa
r
ti
c
ul
a
r
ly
w
it
h
is
s
ue
s
of
di
ve
r
ge
nc
e
a
nd
th
e
pr
ope
ns
it
y
to
be
c
om
e
tr
a
ppe
d
in
lo
c
a
l
opt
im
a
dur
in
g
th
e
r
e
s
ol
ut
io
n
of
nonl
in
e
a
r
s
ys
te
m
s
of
e
qu
a
ti
ons
(
N
S
E
)
.
T
o
m
it
ig
a
te
th
e
s
e
li
m
it
a
ti
ons
,
th
e
J
a
r
r
a
tt
'
s
m
ode
l
is
in
c
or
por
a
te
d
in
to
th
e
B
O
A
f
r
a
m
e
w
or
k,
r
e
s
ul
ti
ng
in
th
e
d
e
ve
l
opm
e
nt
of
th
e
J
B
O
A
.
T
hi
s
hybr
id
a
ppr
oa
c
h
le
ve
r
a
ge
s
th
e
s
tr
e
ngt
hs
of
bot
h
a
lg
or
it
hm
s
,
s
ig
ni
f
ic
a
nt
ly
im
pr
ovi
ng
th
e
a
c
c
ur
a
c
y
a
nd
c
onve
r
ge
nc
e
r
a
te
in
s
ol
vi
ng
N
S
E
.
S
pe
c
if
ic
a
ll
y,
J
a
r
r
a
tt
'
s
te
c
hni
que
is
e
m
pl
oye
d
it
e
r
a
ti
ve
ly
w
it
hi
n
th
e
B
O
A
pr
oc
e
s
s
,
e
nha
n
c
in
g
th
e
c
a
ndi
da
te
but
te
r
f
ly
pos
it
io
ns
id
e
nt
if
ie
d
by
B
O
A
a
nd
e
n
s
ur
in
g
th
a
t
th
e
m
o
s
t
opt
im
a
l
s
ol
ut
io
ns
a
r
e
s
e
le
c
te
d
ba
s
e
d
on
f
it
ne
s
s
c
r
it
e
r
ia
.
T
he
in
te
gr
a
ti
on
of
J
a
r
r
a
tt
'
s
m
e
th
od
not
onl
y
a
c
c
e
le
r
a
te
s
th
e
c
onve
r
ge
n
c
e
pr
oc
e
s
s
but
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
nhanc
in
g f
al
l
de
te
c
ti
on and c
la
s
s
if
ic
at
io
n us
in
g j
ar
r
at
t‐
but
te
r
fl
y
opt
imi
z
at
io
n
…
(
K
ak
ir
al
a
D
u
r
ga B
hav
ani
)
1465
a
ls
o
e
nha
nc
e
s
th
e
ove
r
a
ll
e
f
f
e
c
ti
ve
ne
s
s
of
J
B
O
A
in
r
e
s
ol
vi
ng
c
om
pl
e
x
opt
im
iz
a
ti
on
pr
obl
e
m
s
.
T
he
f
ol
lo
w
in
g
s
e
c
ti
ons
w
il
l
de
ta
il
th
e
a
lg
or
it
hm
ic
f
r
a
m
e
w
or
k
of
J
B
O
A
,
il
lu
s
t
r
a
te
d
th
r
ough
it
s
ps
e
udoc
ode
in
A
lg
or
it
hm
1
,
a
nd dis
c
us
s
i
ts
i
m
pl
ic
a
ti
ons
f
or
pa
r
a
m
e
te
r
t
uni
ng i
n L
S
T
M
ne
tw
or
ks
.
A
lg
or
it
hm
1
.
P
s
e
udoc
ode
of
J
B
O
A
O
b
j
e
c
t
i
v
e
f
u
n
c
t
i
o
n
(
)
,
=
(
1
,
2
,
…
,
)
,
=
No
.
of
d
i
m
e
n
s
i
on
s
C
r
e
a
t
e
p
o
p
u
l
a
t
i
o
n
i
n
i
t
i
a
l
i
z
a
t
i
o
n
o
f
n
B
u
t
t
e
r
f
l
i
e
s
=
(
1
,
2
,
…
,
)
S
t
i
m
u
l
u
s
I
n
t
e
n
s
i
t
y
at
i
s
d
e
t
e
r
m
i
n
e
d
(
)
D
e
s
c
r
i
b
e
s
w
i
t
c
h
i
n
g
p
r
o
b
a
b
i
l
i
t
y
,
s
e
n
s
o
r
m
o
d
a
l
i
t
y
a
n
d
p
o
w
e
r
e
x
p
o
n
e
n
t
W
h
i
l
e
e
n
d
i
n
g
c
o
n
d
i
t
i
o
n
i
s
n
o
t
m
e
t
d
o
F
o
r
a
l
l
t
h
e
b
u
t
t
e
r
f
l
i
e
s
i
n
t
h
e
p
o
p
u
l
a
t
i
o
n
d
o
E
v
a
l
u
a
t
e
f
r
a
g
r
a
n
c
e
f
o
r
End for
F
i
n
d
t
h
e
b
e
t
t
e
r
f
o
r
e
v
e
r
y
b
u
t
t
e
r
f
l
y
i
n
t
h
e
p
o
p
u
l
a
t
i
o
n
d
o
G
e
n
e
r
a
t
e
a
m
o
d
e
m
f
r
o
m
[
0
,
1
]
If
<
then
M
o
v
e
t
o
w
a
r
d
s
o
l
u
t
i
o
n
o
r
b
u
t
t
e
r
f
l
y
Else
M
o
v
e
r
a
n
d
o
m
l
y
End if
End for
U
p
g
r
a
d
e
t
h
e
v
a
l
u
e
o
f
a
e
n
d
w
h
i
l
e
C
o
m
p
u
t
e
J
a
r
r
a
t
t
’
s
l
o
c
a
t
i
o
n
+
1
u
s
i
n
g
E
v
a
l
u
a
t
e
t
h
e
f
i
t
n
e
s
s
o
f
+
1
and
I
f
F
i
t
n
e
s
s
(
+
1
)
<
F
i
t
n
e
s
s
(
)
then
=
+
1
End if
O
u
t
p
u
t
t
h
e
b
e
t
t
e
r
s
o
l
u
t
i
o
n
f
o
u
n
d
(
)
J
B
O
A
u
s
e
s
th
e
m
odi
f
ic
a
ti
on
gi
ve
n
in
th
e
r
e
d
box
a
t
th
e
it
e
r
a
t
io
n
e
nd.
B
a
s
e
d
on
f
it
ne
s
s
v
a
lu
e
,
th
is
c
om
pa
r
is
on
w
a
s
m
a
de
be
tw
e
e
n
th
e
B
O
A
but
te
r
f
ly
’
s
lo
c
a
ti
on
(
)
a
nd
J
a
r
r
a
tt
’
s
te
c
hni
que
’
s
lo
c
a
ti
on
(
+
1
)
.
A
t
la
s
t,
t
he
be
s
t
po
s
it
io
n t
ha
t
e
va
lu
a
te
s
b
e
tt
e
r
f
it
ne
s
s
i
s
c
hos
e
n a
s
a
n opti
m
um
s
ol
ut
io
n.
T
he
f
it
ne
s
s
c
hoi
c
e
is
a
n
im
por
ta
nt
e
l
e
m
e
nt
in
th
e
J
B
O
A
c
la
s
s
if
ie
r
.
E
nc
ode
r
pe
r
f
or
m
a
nc
e
c
a
n
be
e
xe
c
ut
e
d
f
or
m
e
a
s
ur
in
g
th
e
goodne
s
s
of
c
a
ndi
da
t
e
out
c
om
e
s
.
T
he
a
c
c
ur
a
c
y
v
a
lu
e
is
th
e
b
a
s
ic
pr
e
m
is
e
e
nga
ge
d f
or
de
ve
lo
pi
ng a
f
it
ne
s
s
f
unc
ti
on
(
FF
)
.
=
m
a
x
(
)
(
7)
=
+
(
8)
W
he
r
e
a
nd
r
e
pr
e
s
e
nt
t
he
t
r
ue
pos
it
iv
e
r
a
ti
o a
nd t
he
f
a
ls
e
pos
it
i
ve
r
a
ti
o.
4.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
hi
s
s
e
c
ti
on
e
va
lu
a
t
e
s
th
e
pe
r
f
or
m
a
nc
e
of
th
e
F
D
C
-
J
B
O
A
D
L
m
e
th
odol
ogy
ut
il
iz
in
g
th
e
K
F
a
ll
a
nd
C
A
U
C
A
F
a
ll
da
t
a
s
e
t
s
.
T
he
K
F
a
ll
d
a
ta
s
e
t
[
21
]
,
[
22]
in
c
lu
de
s
va
r
io
us
c
l
a
s
s
e
s
s
uc
h
a
s
f
or
w
a
r
d
f
a
ll
w
hi
le
a
tt
e
m
pt
in
g
to
s
it
dow
n
(
20)
,
ba
c
kw
a
r
d
f
a
ll
w
hi
le
a
tt
e
m
pt
in
g
t
o
s
it
dow
n
(
21
)
,
la
te
r
a
l
f
a
ll
w
hi
le
tr
yi
ng
to
s
it
dow
n (
22)
,
f
or
w
a
r
d
f
a
ll
w
hi
le
t
r
y
in
g t
o ge
t
up (
23
)
, l
a
te
r
a
l
f
a
ll
w
hi
le
ge
tt
in
g up (
24)
,
f
o
r
w
a
r
d
f
a
ll
w
hi
le
s
it
ti
n
g
due
to
f
a
in
ti
ng
(
25)
,
la
te
r
a
l
f
a
ll
w
hi
le
s
it
ti
ng
due
to
f
a
in
ti
ng
(
2
6)
,
B
a
c
kw
a
r
d
f
a
ll
due
to
f
a
in
ti
ng
(
27)
,
ve
r
ti
c
a
l
(
f
or
w
a
r
d)
f
a
ll
w
hi
le
w
a
lk
in
g
due
to
f
a
in
ti
ng
(
28)
,
f
a
ll
w
hi
le
w
a
l
ki
ng
w
it
h
ha
nds
us
e
d
to
s
of
te
n
th
e
im
pa
c
t
due
to
f
a
in
ti
ng
(
29)
,
f
or
w
a
r
d
f
a
ll
w
hi
le
w
a
lk
in
g
due
to
tr
ip
pi
ng
(
3
0)
,
f
or
w
a
r
d
f
a
ll
w
hi
le
jo
ggi
ng
due
to
tr
ip
pi
n
g
(
31)
,
f
or
w
a
r
d
f
a
ll
w
hi
le
w
a
lk
in
g
due
to
s
li
ppi
ng
(
32)
,
la
te
r
a
l
f
a
ll
w
hi
le
w
a
lk
in
g
du
e
to
s
li
ppi
ng
(
33)
,
a
nd
ba
c
kw
a
r
d
f
a
ll
w
hi
le
w
a
lk
in
g
due
to
s
li
ppi
ng
(
34)
.
M
e
a
nw
hi
le
,
th
e
C
A
U
C
A
F
a
ll
d
a
ta
s
e
t
[
23]
c
ons
is
ts
of
13,581 AD
L
l
a
be
le
d a
s
"
nof
a
ll
"
a
nd 6,421 labe
le
d a
s
"
f
a
ll
"
.
S
a
m
pl
e
i
m
a
ge
s
a
r
e
i
ll
us
tr
a
te
d i
n F
ig
ur
e
3.
T
a
bl
e
1
a
nd
F
ig
ur
e
4
de
m
on
s
tr
a
te
s
th
e
ove
r
a
ll
c
la
s
s
if
ie
r
r
e
s
ul
t
s
of
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
on
th
e
K
F
a
ll
da
ta
s
e
t.
T
he
out
c
om
e
s
in
di
c
a
te
th
a
t
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
r
e
a
c
he
s
e
f
f
e
c
tu
a
l
out
c
om
e
s
on
bot
h
tr
a
in
in
g
s
e
t
(
T
R
S
)
a
nd
te
s
ti
ng
s
e
t
(
T
S
S
)
.
O
n
th
e
a
ppl
i
e
d
T
R
S
,
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
of
f
e
r
s
,
,
,
,
a
nd
M
a
tt
he
w
s
c
or
r
e
la
ti
on
c
oe
f
f
ic
ie
nt
(
M
C
C
)
of
99.39,
99
.39,
99.41,
99.30,
a
nd
99.03%
r
e
s
pe
c
ti
ve
ly
.
A
t
th
e
s
a
m
e
ti
m
e
,
on
th
e
a
ppl
ie
d
T
S
S
,
th
e
F
D
C
-
J
B
O
A
D
L
m
e
th
od
pr
ovi
de
s
,
,
,
, a
nd M
C
C
of
99.16, 99.32, 99.38, 99.25, a
nd 99.01%
c
or
r
e
s
pondingl
y.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 2, A
pr
il
2025
:
1461
-
1470
1466
F
ig
ur
e
3.
S
a
m
pl
e
i
m
a
ge
s
T
a
bl
e
1
.
C
la
s
s
if
ie
r
out
c
om
e
of
F
D
C
-
J
B
O
A
D
L
a
lg
or
it
hm
on KF
a
ll
da
ta
s
e
t
M
e
t
r
i
c
s
T
r
a
i
ni
ng
s
et
(%)
T
e
s
t
i
ng
s
et
(%)
A
c
c
ur
a
c
y
99.39
99.16
P
r
e
c
i
s
i
on
99.39
99.32
R
e
c
a
l
l
99.41
99.38
F
-
s
c
or
e
99.30
99.25
M
C
C
99.03
99.01
F
ig
ur
e
4. C
la
s
s
if
ie
r
out
c
om
e
of
F
D
C
-
J
B
O
A
D
L
a
ppr
oa
c
h on KF
a
ll
da
ta
ba
s
e
T
he
pe
r
f
or
m
a
nc
e
e
va
lu
a
ti
on
of
ML
m
e
th
odol
ogi
e
s
i
s
c
r
uc
ia
l
f
o
r
unde
r
s
ta
ndi
ng
th
e
ir
e
f
f
e
c
ti
ve
ne
s
s
in
va
r
io
us
a
ppl
ic
a
ti
ons
.
T
hi
s
s
tu
dy
f
oc
us
e
s
on
th
e
F
D
C
-
J
B
O
A
D
L
m
e
th
od,
s
pe
c
if
ic
a
ll
y
it
s
a
ppl
ic
a
ti
on
to
th
e
K
F
a
ll
da
ta
s
e
t,
to
a
na
ly
z
e
tr
a
in
in
g
a
nd
va
li
da
ti
on
a
c
c
ur
a
c
ie
s
a
s
w
e
ll
a
s
lo
s
s
m
e
tr
ic
s
.
A
s
il
lu
s
tr
a
te
d
in
F
ig
ur
e
5,
th
e
tr
a
in
in
g
a
c
c
ur
a
c
y
(
T
R
_a
c
c
u_y)
a
nd
va
li
da
ti
on
a
c
c
ur
a
c
y
(
V
L
_a
c
c
u_y)
e
xhi
bi
t
a
pos
it
iv
e
c
or
r
e
la
ti
on
w
it
h
th
e
num
be
r
of
tr
a
in
in
g
e
poc
hs
,
in
di
c
a
ti
ng
th
a
t
in
c
r
e
a
s
e
d
e
poc
hs
c
ont
r
ib
ut
e
to
e
nha
nc
e
d
m
ode
l
e
f
f
ic
a
c
y
on
bot
h t
he
t
r
a
in
in
g a
nd t
e
s
ti
ng da
ta
s
e
ts
. F
ur
th
e
r
m
or
e
,
F
ig
ur
e
6 p
r
e
s
e
nt
s
t
he
t
r
e
nds
i
n t
r
a
in
in
g l
os
s
(
T
R
_l
os
s
)
a
nd
va
li
da
ti
on
lo
s
s
(
V
R
_l
os
s
)
a
s
s
oc
ia
te
d
w
it
h
th
e
F
D
C
-
J
B
O
A
D
L
a
ppr
oa
c
h.
T
he
r
e
s
ul
ts
r
e
ve
a
l
a
c
ons
is
te
nt
de
c
r
e
a
s
e
in
bot
h
T
R
_l
os
s
a
nd
V
R
_l
os
s
a
s
th
e
e
poc
h
s
pr
ogr
e
s
s
,
unde
r
s
c
or
in
g
th
e
m
ode
l'
s
c
a
pa
bi
li
ty
to
m
in
im
iz
e
pr
e
di
c
ti
on
di
s
c
r
e
pa
nc
ie
s
a
nd
im
pr
ove
c
la
s
s
if
ic
a
ti
on
pr
e
c
is
io
n.
C
ol
le
c
ti
ve
ly
,
th
e
s
e
f
in
di
ngs
a
f
f
ir
m
th
e
F
D
C
-
J
B
O
A
D
L
m
e
th
od'
s
pot
e
nt
ia
l
in
e
f
f
e
c
ti
ve
ly
id
e
nt
if
yi
ng
pa
tt
e
r
ns
a
nd
r
e
la
ti
ons
hi
ps
w
it
hi
n
da
ta
,
th
e
r
e
by e
s
ta
bl
is
hi
ng i
ts
s
ig
ni
f
ic
a
nc
e
i
n t
he
r
e
a
lm
of
pr
e
di
c
ti
ve
a
na
ly
ti
c
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
nhanc
in
g f
al
l
de
te
c
ti
on and c
la
s
s
if
ic
at
io
n us
in
g j
ar
r
at
t‐
but
te
r
fl
y
opt
imi
z
at
io
n
…
(
K
ak
ir
al
a
D
u
r
ga B
hav
ani
)
1467
F
ig
ur
e
5.
c
ur
ve
of
F
D
C
-
J
B
O
A
D
L
a
lg
or
it
hm
on KF
a
ll
da
ta
ba
s
e
F
ig
ur
e
6. L
os
s
c
ur
ve
of
F
D
C
-
J
B
O
A
D
L
a
lg
or
it
hm
on KF
a
ll
da
ta
s
e
t
T
a
bl
e
2
a
nd
F
ig
ur
e
7
s
ig
ni
f
ie
s
th
e
c
la
s
s
if
ic
a
ti
on
out
c
om
e
s
o
f
th
e
F
D
C
-
J
B
O
A
D
L
m
e
th
od
on
th
e
C
A
U
C
A
F
a
ll
da
ta
s
e
t.
T
he
out
c
om
e
s
s
pe
c
if
y
th
a
t
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
r
e
a
c
he
s
e
f
f
e
c
tu
a
l
out
c
om
e
s
on
bot
h
T
R
S
a
nd
T
S
S
.
O
n
th
e
a
ppl
ie
d
T
R
S
,
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
of
f
e
r
s
,
,
,
,
a
nd
M
C
C
of
98.81,
98.72,
98.54,
98.50,
a
nd
98.18%
r
e
s
pe
c
ti
ve
ly
.
A
t
th
e
s
a
m
e
ti
m
e
,
on
th
e
a
ppl
ie
d
T
S
S
,
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
of
f
e
r
s
,
,
,
,
a
nd
M
C
C
of
98.33,
98.27,
98.12,
98.20,
a
nd
98.04%
r
e
s
pe
c
ti
ve
ly
.
T
he
e
va
lu
a
ti
on
of
ML
m
e
th
odol
ogi
e
s
of
te
n
hi
nge
s
on
th
e
ir
a
bi
li
ty
to
a
c
c
ur
a
te
ly
c
la
s
s
if
y
a
nd
pr
e
di
c
t
out
c
om
e
s
ba
s
e
d
on
tr
a
in
in
g
a
nd
va
li
da
ti
on
d
a
ta
s
e
t
s
.
I
n
th
is
c
ont
e
xt
,
th
e
F
D
C
-
J
B
O
A
D
L
m
e
th
od
ha
s
be
e
n
a
ppl
ie
d
to
th
e
C
A
U
C
A
F
a
ll
da
ta
s
e
t,
r
e
ve
a
li
ng
s
ig
ni
f
ic
a
nt
in
s
ig
ht
s
in
to
it
s
pe
r
f
or
m
a
nc
e
m
e
tr
ic
s
.
A
s
il
lu
s
tr
a
te
d
in
F
ig
ur
e
8,
bot
h
T
R
_a
c
c
u_y
a
nd
V
L
_a
c
c
u_y
e
xhi
bi
ts
a
po
s
it
iv
e
c
or
r
e
la
ti
on
w
it
h
th
e
num
be
r
of
tr
a
in
in
g
e
poc
hs
,
s
ugg
e
s
ti
ng
th
a
t
pr
ol
onge
d
tr
a
in
in
g
e
nha
nc
e
s
th
e
m
o
de
l'
s
e
f
f
ic
a
c
y
on
bot
h
th
e
tr
a
in
in
g
(
T
R
)
a
nd
te
s
ti
ng
(
T
S
)
da
ta
s
e
ts
.
T
hi
s
tr
e
nd
unde
r
s
c
or
e
s
th
e
im
por
ta
nc
e
of
e
poc
h
dur
a
ti
on
in
opt
im
iz
in
g
th
e
e
f
f
ic
ie
nc
y
of
M
L
a
ppr
oa
c
he
s
.
M
or
e
ov
e
r
,
F
ig
ur
e
9
pr
e
s
e
nt
s
th
e
lo
s
s
m
e
tr
ic
s
a
s
s
oc
ia
t
e
d
w
it
h
th
e
F
D
C
-
J
B
O
A
D
L
a
ppr
oa
c
h,
s
pe
c
if
ic
a
ll
y
th
e
T
R
_l
o
s
s
a
nd
V
R
_l
os
s
.
T
h
e
s
e
m
e
tr
ic
s
pr
ovi
d
e
a
qua
nt
it
a
ti
ve
m
e
a
s
ur
e
of
th
e
di
s
c
r
e
pa
nc
y
a
m
ong
pr
e
di
c
te
d
out
c
om
e
s
a
nd
a
c
tu
a
l
out
c
om
e
s
,
w
it
h
f
in
di
ngs
in
di
c
a
ti
ng
a
c
on
s
is
te
nt
de
c
li
ne
in
bot
h
T
R
_l
os
s
a
nd
V
R
_l
os
s
a
s
e
poc
h
s
pr
ogr
e
s
s
.
T
hi
s
r
e
duc
ti
on
in
lo
s
s
va
lu
e
s
f
ur
th
e
r
c
or
r
obor
a
te
s
th
e
m
ode
l'
s
in
c
r
e
a
s
in
g
pr
of
ic
ie
nc
y
in
id
e
nt
if
yi
ng
unde
r
ly
in
g
pa
tt
e
r
ns
a
nd
r
e
la
ti
ons
hi
ps
w
it
hi
n
th
e
da
ta
.
C
ol
le
c
ti
ve
ly
,
th
e
s
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 2, A
pr
il
2025
:
1461
-
1470
1468
out
c
om
e
s
e
m
pha
s
iz
e
th
e
s
ol
ut
io
n
of
th
e
F
D
C
-
J
B
O
A
D
L
m
e
th
od
in
a
c
hi
e
vi
ng
pr
e
c
is
e
c
la
s
s
if
ic
a
ti
ons
,
th
e
r
e
by
c
ont
r
ib
ut
in
g t
o t
he
br
oa
de
r
di
s
c
our
s
e
on t
he
opt
im
iz
a
ti
on of
ML
te
c
hni
que
s
i
n c
om
pl
e
x da
ta
s
e
ts
.
T
he
c
om
pa
r
is
on
s
tu
dy
of
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
in
e
xi
s
t
in
g
a
ppr
oa
c
he
s
ta
ke
pl
a
c
e
in
T
a
bl
e
3
.
T
he
r
e
s
ul
ts
in
di
c
a
te
th
a
t
th
e
F
D
C
-
J
B
O
A
D
L
m
e
th
od
a
c
hi
e
ve
s
e
nr
ic
he
d
pe
r
f
or
m
a
nc
e
ove
r
ot
he
r
m
ode
ls
[
24
]
,
[
25]
.
B
a
s
e
d
on
,
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
a
c
c
om
pl
is
he
s
hi
ghe
r
of
99.39%
but
th
e
C
N
N
c
la
s
s
if
ie
r
, L
S
T
M
a
lg
or
it
hm
, C
N
N
-
L
S
T
M
a
ppr
oa
c
h, a
nd F
D
S
N
e
X
t
m
ode
ls
a
tt
a
in
m
in
im
a
l
va
lu
e
s
of
85.69,
90.12,
84.04,
a
nd
91.87%
r
e
s
pe
c
ti
ve
ly
.
I
n
a
ddi
ti
on,
ba
s
e
d
on
,
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
a
c
c
om
pl
is
he
s
hi
ghe
r
of
91.44%
w
he
r
e
a
s
th
e
C
N
N
c
la
s
s
if
ie
r
,
L
S
T
M
a
lg
or
it
hm
,
C
N
N
-
L
S
T
M
a
ppr
oa
c
h,
a
nd
F
D
S
N
e
X
t
a
lg
or
it
hm
a
tt
a
in
lo
w
e
r
va
lu
e
s
of
91.28,
89.90,
91.07,
a
nd
99.39%
r
e
s
pe
c
ti
ve
ly
.
N
e
xt
to
th
a
t,
ba
s
e
d
on
,
th
e
F
D
C
-
J
B
O
A
D
L
m
e
th
od
a
c
c
om
pl
is
h
e
s
hi
ghe
r
of
89.72%
w
hi
le
th
e
C
N
N
c
la
s
s
if
ie
r
,
L
S
T
M
a
lg
or
it
hm
,
C
N
N
-
L
S
T
M
a
ppr
oa
c
h,
a
nd
F
D
S
N
e
X
t
s
ys
te
m
a
tt
a
in
r
e
duc
e
va
lu
e
s
of
90.67,
89,
90.05,
a
nd
99.41%
c
or
r
e
s
pondingl
y.
A
t
la
s
t,
ba
s
e
d
on
,
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
a
c
c
om
pl
is
he
s
hi
ghe
r
of
89.69%
w
hi
le
th
e
C
N
N
c
la
s
s
if
ie
r
,
L
S
T
M
a
lg
or
it
hm
,
C
N
N
-
L
S
T
M
a
ppr
oa
c
h,
a
nd
F
D
S
N
e
X
t
m
e
th
odol
ogy
a
tt
a
in
le
s
s
e
r
va
lu
e
s
of
91.39,
90.
44,
90.75,
a
nd
99.30%
c
or
r
e
s
pondingl
y.
T
he
s
e
p
e
r
f
or
m
a
nc
e
s
gua
r
a
nt
e
e
d t
h
e
e
xc
e
ll
e
nt
s
ol
ut
io
n
of
t
he
F
D
C
-
J
B
O
A
D
L
t
e
c
hni
que
.
T
a
bl
e
2
.
C
la
s
s
if
ie
r
out
c
om
e
of
F
D
C
-
J
B
O
A
D
L
a
lg
or
it
hm
on C
A
U
C
A
F
a
ll
da
ta
ba
s
e
M
e
t
r
i
c
s
T
r
a
i
ni
ng
s
et
(%)
T
e
s
t
i
ng
s
et
(%)
A
c
c
ur
a
c
y
98.81
98.33
P
r
e
c
i
s
i
on
98.72
98.27
R
e
c
a
l
l
98.54
98.12
F
-
S
c
or
e
98.50
98.20
M
C
C
98.18
98.04
F
ig
ur
e
7. C
la
s
s
if
ie
r
out
c
om
e
of
F
D
C
-
J
B
O
A
D
L
a
lg
or
it
hm
on C
A
U
C
A
F
a
ll
da
ta
s
e
t
F
ig
ur
e
8.
c
ur
ve
of
F
D
C
-
J
B
O
A
D
L
a
lg
or
it
hm
on C
A
U
C
A
F
a
ll
da
ta
s
e
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
E
nhanc
in
g f
al
l
de
te
c
ti
on and c
la
s
s
if
ic
at
io
n us
in
g j
ar
r
at
t‐
but
te
r
fl
y
opt
imi
z
at
io
n
…
(
K
ak
ir
al
a
D
u
r
ga B
hav
ani
)
1469
F
ig
ur
e
9. L
os
s
c
ur
ve
of
F
D
C
-
J
B
O
A
D
L
a
lg
or
it
hm
on C
A
U
C
A
F
a
ll
da
ta
s
e
t
T
a
bl
e
3
.
C
om
pa
r
a
ti
ve
out
c
om
e
of
F
D
C
-
J
B
O
A
D
L
a
lg
or
it
hm
w
it
h ot
he
r
m
e
th
ods
M
ode
l
(%)
(%)
(%)
(%)
C
N
N
a
l
gor
i
t
hm
85.69
91.44
89.72
89.69
L
S
T
M
a
l
gor
i
t
hm
90.12
91.28
90.67
91.39
C
N
N
-
L
S
T
M
84.04
89.90
89.00
90.44
F
D
S
N
e
X
t
91.87
91.07
90.05
90.75
F
D
C
-
J
B
O
A
D
L
99.39
99.39
99.41
99.30
5.
C
O
N
C
L
U
S
I
O
N
T
hi
s
r
e
s
e
a
r
c
h
hi
ghl
ig
ht
s
th
e
e
f
f
ic
a
c
y
of
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
in
a
ut
om
a
ti
ng
th
e
r
e
c
ogni
ti
on
a
nd c
la
s
s
if
ic
a
ti
on of
f
a
ll
e
ve
nt
s
t
hr
ough the
i
nt
e
gr
a
ti
on
of
a
dva
nc
e
d
DL
m
ode
ls
. B
y e
m
pl
oyi
ng a
m
ul
ti
f
a
c
e
te
d
a
ppr
oa
c
h
th
a
t
in
c
lu
de
s
M
F
-
ba
s
e
d
noi
s
e
r
e
m
ova
l,
E
f
f
ic
ie
nt
N
e
t
f
or
f
e
a
tu
r
e
e
xt
r
a
c
ti
on,
L
S
T
M
f
or
f
a
ll
de
te
c
ti
on,
a
nd
J
B
O
A
f
or
hype
r
pa
r
a
m
e
te
r
opt
im
iz
a
ti
on,
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
de
m
ons
tr
a
te
s
s
ig
ni
f
ic
a
nt
im
pr
ove
m
e
nt
s
i
n r
e
c
ogni
z
in
g a
nd c
la
s
s
if
yi
ng f
a
ll
i
nc
id
e
nt
s
. T
he
m
e
th
odol
ogy not onl
y l
e
ve
r
a
ge
s
t
he
s
tr
e
ngt
hs
of
E
f
f
ic
ie
nt
N
e
t
in
c
a
pt
ur
in
g
bot
h
m
ot
io
n
pa
tt
e
r
ns
a
nd
in
di
vi
du
a
l
a
ppe
a
r
a
nc
e
c
h
a
r
a
c
te
r
is
ti
c
s
but
a
ls
o
e
n
s
ur
e
s
opt
im
a
l
pe
r
f
or
m
a
nc
e
th
r
ough
m
e
ti
c
ul
ous
hype
r
pa
r
a
m
e
te
r
tu
ni
n
g.
T
he
c
om
pr
e
h
e
ns
iv
e
e
xpe
r
im
e
nt
a
l
va
li
da
ti
on
unde
r
s
c
or
e
s
th
e
e
xc
e
ll
e
nt
s
ol
ut
io
n
of
th
e
F
D
C
-
J
B
O
A
D
L
te
c
hni
que
c
om
pa
r
e
d
to
ot
he
r
s
ys
te
m
s
,
r
e
in
f
or
c
in
g
it
s
pot
e
nt
ia
l
a
s
a
r
obus
t
s
ol
ut
io
n
f
or
f
a
ll
de
te
c
ti
on
in
va
r
io
us
a
ppl
ic
a
ti
ons
.
F
ut
ur
e
w
or
k
c
on
c
e
nt
r
a
te
s
on
f
ur
th
e
r
r
e
f
in
in
g
th
e
te
c
hni
que
a
nd
e
xpl
or
in
g
it
s
a
ppl
ic
a
bi
li
ty
in
r
e
a
l
-
ti
m
e
m
oni
to
r
in
g
s
ys
te
m
s
,
ul
ti
m
a
te
ly
c
ont
r
ib
ut
in
g
to
e
nha
nc
e
d s
a
f
e
ty
a
nd w
e
ll
-
be
in
g f
or
in
di
vi
dua
ls
a
t
r
is
k of
f
a
ll
s
.
R
E
F
E
R
E
N
C
E
S
[
1]
X
. J
i
a
ng, L
. Z
ha
ng, a
nd
L
. L
i
,
“
M
ul
t
i
-
t
a
s
k l
e
a
r
ni
ng r
a
da
r
t
r
a
ns
f
or
m
e
r
(
M
L
R
T
)
:
a
pe
r
s
ona
l
i
de
nt
i
f
i
c
a
t
i
on a
nd f
a
l
l
de
t
e
c
t
i
on ne
t
w
or
k
ba
s
e
d on I
R
-
U
W
B
r
a
da
r
,”
Se
n
s
or
s
, vol
. 23, no. 12, 2023, doi
:
10.3390/
s
231256
32.
[
2]
H
.
S
a
dr
e
a
z
a
m
i
,
M
.
B
ol
i
c
,
a
nd
S
.
R
a
j
a
n,
“
C
ont
a
c
t
l
e
s
s
f
a
l
l
de
t
e
c
t
i
on
us
i
ng
t
i
m
e
-
f
r
e
que
nc
y
a
na
l
ys
i
s
a
nd
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
ks
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on I
ndus
t
r
i
al
I
nf
or
m
at
i
c
s
, vol
. 17, no. 10, pp. 68
42
–
6851, 2021, doi
:
10.1109/
T
I
I
.2021.3049342.
[
3]
A
.
D
e
,
A
.
S
a
ha
,
P
.
K
um
a
r
,
a
nd
G
.
P
a
l
,
“
F
a
l
l
de
t
e
c
t
i
on
a
ppr
oa
c
h
b
a
s
e
d
on
c
o
m
bi
ne
d
t
w
o
-
c
ha
nne
l
body
a
c
t
i
vi
t
y
c
l
a
s
s
i
f
i
c
a
t
i
on
f
or
i
nnova
t
i
ve
i
ndoor
e
nvi
r
onm
e
nt
,”
J
our
nal
of
A
m
bi
e
nt
I
nt
e
l
l
i
ge
nc
e
and
H
um
an
i
z
e
d
C
om
put
i
ng
,
vol
.
14,
no.
9,
pp.
11407
–
11418,
2023, doi
:
10.1007/
s
12652
-
022
-
03714
-
2.
[
4]
S
.
M
obs
i
t
e
,
N
.
A
l
a
oui
,
a
nd
M
.
B
oul
m
a
l
f
,
“
A
f
r
a
m
e
w
or
k
f
or
e
l
de
r
s
f
a
l
l
d
e
t
e
c
t
i
on
us
i
ng
de
e
p
l
e
a
r
ni
ng,”
i
n
C
ol
l
oqui
um
i
n
I
nf
or
m
at
i
on Sc
i
e
nc
e
and T
e
c
hnol
ogy
, C
I
ST
, 2020, pp. 69
–
74
, doi
:
10.1109/
C
i
S
t
49399.2021.9357184.
[
5]
D
.
K
r
a
f
t
,
K
.
S
r
i
ni
va
s
a
n,
a
nd
G
.
B
i
e
b
e
r
,
“
D
e
e
p
l
e
a
r
ni
ng
ba
s
e
d
f
a
l
l
de
t
e
c
t
i
on
a
l
g
or
i
t
hm
s
f
or
e
m
be
dde
d
s
ys
t
e
m
s
,
s
m
a
r
t
w
a
t
c
he
s
,
a
n
d
I
oT
de
vi
c
e
s
us
i
ng a
c
c
e
l
e
r
om
e
t
e
r
s
,
”
T
e
c
hnol
ogi
e
s
, vol
. 8, no. 4, 2020, doi
:
10.33
90/
t
e
c
hnol
ogi
e
s
8040072.
[
6]
K
.
C
.
L
i
u,
K
.
H
.
H
ung,
C
.
Y
.
H
s
i
e
h,
H
.
Y
.
H
ua
ng,
C
.
T
.
C
h
a
n,
a
nd
Y
.
T
s
a
o,
“
D
e
e
p
-
l
e
a
r
ni
ng
-
ba
s
e
d
s
i
gna
l
e
nha
n
c
e
m
e
nt
of
l
ow
-
r
e
s
ol
ut
i
on
a
c
c
e
l
e
r
om
e
t
e
r
f
or
f
a
l
l
de
t
e
c
t
i
on
s
y
s
t
e
m
s
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
C
o
gni
t
i
v
e
and
D
e
v
e
l
opm
e
nt
al
S
y
s
t
e
m
s
,
vol
.
14,
no.
3
,
pp. 1270
–
1281, 2022, doi
:
10.1109/
T
C
D
S
.2021.3116228.
[
7]
R
.
D
e
l
ga
do
-
E
s
c
a
ño,
F
.
M
.
C
a
s
t
r
o,
J
.
R
.
C
óz
a
r
,
M
.
J
.
M
a
r
í
n
-
J
i
m
é
ne
z
,
N
.
G
ui
l
,
a
nd
E
.
C
a
s
i
l
a
r
i
,
“
A
c
r
os
s
-
da
t
a
s
e
t
de
e
p
l
e
a
r
ni
ng
-
ba
s
e
d
c
l
a
s
s
i
f
i
e
r
f
or
pe
opl
e
f
a
l
l
de
t
e
c
t
i
on
a
nd
i
de
nt
i
f
i
c
a
t
i
on,”
C
om
put
e
r
M
e
t
ho
ds
and
P
r
ogr
am
s
i
n
B
i
om
e
di
c
i
ne
,
vol
.
184,
2020
,
doi
:
10.1016/
j
.c
m
pb.2019.105265.
[
8]
X
.
C
a
i
,
X
.
L
i
u,
M
.
A
n,
a
nd
G
.
H
a
n,
“
V
i
s
i
on
-
ba
s
e
d
f
a
l
l
de
t
e
c
t
i
on
us
i
ng
de
n
s
e
bl
oc
k
w
i
t
h
m
ul
t
i
-
c
ha
nne
l
c
onvol
ut
i
ona
l
f
us
i
o
n
s
t
r
a
t
e
gy,”
I
E
E
E
A
c
c
e
s
s
, vol
. 9, pp. 18318
–
18325, 2021, doi
:
10.1109/
A
C
C
E
S
S
.2021.3054469.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 2, A
pr
il
2025
:
1461
-
1470
1470
[
9]
C
.
Y
a
o,
J
.
H
u,
W
.
M
i
n,
Z
.
D
e
ng,
S
.
Z
ou,
a
nd
W
.
M
i
n,
“
A
nove
l
r
e
a
l
-
t
i
m
e
f
a
l
l
de
t
e
c
t
i
on
m
e
t
hod
ba
s
e
d
on
he
a
d
s
e
gm
e
nt
a
t
i
on
a
n
d
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k,”
J
ou
r
nal
of
R
e
al
-
T
i
m
e
I
m
age
P
r
oc
e
s
s
i
ng
,
vol
.
17,
no.
6,
pp.
1939
–
1949,
2020,
doi
:
10.1007/
s
11554
-
020
-
00982
-
z.
[
10]
G
.
V
.
L
e
i
t
e
,
G
.
P
.
da
S
i
l
va
,
a
nd
H
.
P
e
dr
i
ni
,
“
T
hr
e
e
-
s
t
r
e
a
m
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k
f
or
hum
a
n
f
a
l
l
de
t
e
c
t
i
on,”
A
dv
anc
e
s
i
n
I
nt
e
l
l
i
ge
nt
Sy
s
t
e
m
s
and C
om
put
i
ng
, vol
. 1232, pp. 49
–
80, 2021, doi
:
10.1007/
9
78
-
981
-
15
-
6759
-
9_3.
[
11]
Y
.
L
i
,
Z
.
Z
uo,
a
nd
J
.
P
a
n,
“
S
e
ns
or
-
ba
s
e
d
f
a
l
l
de
t
e
c
t
i
on
us
i
ng
a
c
om
bi
na
t
i
on
m
ode
l
of
a
t
e
m
por
a
l
c
onvol
ut
i
ona
l
ne
t
w
or
k
a
nd
a
ga
t
e
d r
e
c
ur
r
e
nt
uni
t
,”
F
ut
ur
e
G
e
ne
r
at
i
on C
om
put
e
r
Sy
s
t
e
m
s
, vol
. 139, pp. 53
–
6
3, 2023, doi
:
10.1016/
j
.f
ut
ur
e
.2022.09.011.
[
12]
N
.
D
.
R
a
e
ve
e
t
al
.
,
“
B
l
ue
t
oot
h
-
l
ow
-
e
ne
r
gy
-
ba
s
e
d
f
a
l
l
de
t
e
c
t
i
on
a
nd
w
a
r
ni
ng
s
y
s
t
e
m
f
or
e
l
de
r
l
y
pe
opl
e
i
n
nur
s
i
ng
hom
e
s
,”
J
our
nal
of
Se
ns
or
s
, vol
. 2022, 2022, doi
:
10.1155/
2022/
9930681.
[
13]
A
.
R
e
z
a
e
i
e
t
al
.
,
“
U
nobt
r
us
i
ve
hum
a
n
f
a
l
l
de
t
e
c
t
i
on
s
ys
t
e
m
us
i
ng
m
m
w
a
ve
r
a
da
r
a
nd
da
t
a
d
r
i
ve
n
m
e
t
hods
,”
I
E
E
E
Se
ns
or
s
J
our
nal
, vol
. 23, no. 7, pp. 7968
–
7976, 2023, doi
:
10.1109/
J
S
E
N
.2023.324506
3.
[
14]
L
.
Y
a
o,
W
.
Y
a
ng,
a
nd
W
.
H
ua
ng,
“
A
f
a
l
l
de
t
e
c
t
i
on
m
e
t
hod
ba
s
e
d
on
a
j
oi
nt
m
ot
i
on
m
a
p
us
i
ng
doubl
e
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
ks
,”
M
ul
t
i
m
e
di
a T
ool
s
and A
ppl
i
c
at
i
ons
, vol
. 81, no. 4, pp. 4551
–
4568, 2022, doi
:
10.1007/
s
11042
-
020
-
09181
-
1.
[
15]
S
.
M
e
kr
uks
a
va
ni
c
h,
P
.
J
a
nt
a
w
ong,
A
.
C
ha
r
oe
nphol
,
a
nd
A
.
J
i
t
pa
t
t
a
na
kul
,
“
F
a
l
l
de
t
e
c
t
i
on
f
r
om
s
m
a
r
t
w
e
a
r
a
bl
e
s
e
ns
or
s
us
i
ng
de
e
p
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k
w
i
t
h
s
que
e
z
e
-
a
nd
-
e
xc
i
t
a
t
i
on
m
odul
e
,”
i
n
I
C
SE
C
2021
-
25t
h
I
nt
e
r
nat
i
onal
C
om
put
e
r
Sc
i
e
nc
e
an
d
E
ngi
ne
e
r
i
ng C
onf
e
r
e
nc
e
, 2021, pp. 448
–
453
, doi
:
10.1109/
I
C
S
E
C
53205.2021.9684626.
[
16]
R
.
S
.
J
a
w
a
d
a
nd
H
.
A
bi
d,
“
H
V
D
C
f
a
ul
t
de
t
e
c
t
i
on
a
nd
c
l
a
s
s
i
f
i
c
a
t
i
on
w
i
t
h
a
r
t
i
f
i
c
i
a
l
ne
ur
a
l
ne
t
w
or
k
ba
s
e
d
on
A
C
O
-
D
W
T
m
e
t
hod,
”
E
ne
r
gi
e
s
, vol
. 16, no. 3, 2023, doi
:
10.3390/
e
n16031064.
[
17]
D
. W
. L
e
e
,
K
. J
un, K
. N
a
he
e
m
, a
nd
M
. S
. K
i
m
, “
D
e
e
p ne
ur
a
l
ne
t
w
or
k
–
ba
s
e
d
d
oubl
e
-
c
he
c
k m
e
t
hod f
or
f
a
l
l
de
t
e
c
t
i
on us
i
ng I
M
U
-
L
s
e
ns
or
a
nd R
G
B
c
a
m
e
r
a
da
t
a
,
”
I
E
E
E
A
c
c
e
s
s
, vol
. 9, pp. 48064
–
48079, 2021, do
i
:
10.1109/
A
C
C
E
S
S
.2021.3065105.
[
18]
R
.
C
ha
ga
nt
i
,
V
.
R
a
vi
,
a
nd
T
.
D
.
P
ha
m
,
“
I
m
a
ge
-
ba
s
e
d
m
a
l
w
a
r
e
r
e
pr
e
s
e
nt
a
t
i
on
a
ppr
oa
c
h
w
i
t
h
E
f
f
i
c
i
e
nt
N
e
t
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
ks
f
or
e
f
f
e
c
t
i
ve
m
a
l
w
a
r
e
c
l
a
s
s
i
f
i
c
a
t
i
on,”
J
our
nal
of
I
nf
or
m
at
i
on
Se
c
ur
i
t
y
and
A
ppl
i
c
at
i
ons
,
vol
.
69,
2022,
doi
:
10.1016/
j
.j
i
s
a
.2022.103306.
[
19]
I
.
D
.
M
i
e
nye
a
nd
Y
.
S
un,
“
A
de
e
p
l
e
a
r
ni
ng
e
n
s
e
m
bl
e
w
i
t
h
da
t
a
r
e
s
a
m
pl
i
ng
f
or
c
r
e
di
t
c
a
r
d
f
r
a
ud
de
t
e
c
t
i
on,”
I
E
E
E
A
c
c
e
s
s
,
vol
.
11,
pp. 30628
–
30638, 2023, doi
:
10.1109/
A
C
C
E
S
S
.2023.3262020.
[
20]
R
. S
i
hw
a
i
l
, O
. S
a
i
d
S
ol
a
i
m
a
n, a
nd
K
. A
.
Z
a
i
nol
A
r
i
f
f
i
n, “
N
e
w
r
obus
t
hybr
i
d J
a
r
r
a
t
t
-
B
ut
t
e
r
f
l
y opt
i
m
i
z
a
t
i
on a
l
gor
i
t
hm
f
or
nonl
i
ne
a
r
m
ode
l
s
,”
J
our
nal
of
K
i
ng
Saud
U
ni
v
e
r
s
i
t
y
-
C
om
put
e
r
and
I
nf
or
m
at
i
on
Sc
i
e
nc
e
s
,
vol
.
34,
no.
10,
pp.
8207
–
8220,
2022,
doi
:
10.1016/
j
.j
ks
uc
i
.2022.08.004.
[
21]
X
.
Y
u,
J
.
J
a
ng,
a
nd
S
.
X
i
ong,
“
K
F
a
l
l
:
A
c
om
pr
e
he
ns
i
ve
m
ot
i
on
da
t
a
s
e
t
t
o
de
t
e
c
t
pr
e
-
i
m
pa
c
t
f
a
l
l
f
o
r
t
he
e
l
de
r
l
y
ba
s
e
d
on
w
e
a
r
a
bl
e
i
ne
r
t
i
a
l
s
e
ns
or
s
,”
K
F
al
l
D
at
as
e
t
. 2021. [
O
nl
i
ne
]
. A
va
i
l
a
bl
e
:
ht
t
ps
:
/
/
s
i
t
e
s
.googl
e
.
c
om
/
vi
e
w
/
kf
a
l
l
da
t
a
s
e
t
[
22]
X
.
Y
u,
J
.
J
a
ng,
a
nd
S
.
X
i
ong,
“
A
l
a
r
ge
-
s
c
a
l
e
ope
n
m
ot
i
on
da
t
a
s
e
t
(
K
F
a
l
l
)
a
nd
be
nc
hm
a
r
k
a
l
gor
i
t
hm
s
f
or
de
t
e
c
t
i
ng
pr
e
-
i
m
pa
c
t
f
a
l
l
of
t
he
e
l
de
r
l
y us
i
ng w
e
a
r
a
bl
e
i
ne
r
t
i
a
l
s
e
ns
or
s
,”
F
r
ont
i
e
r
s
i
n A
gi
ng N
e
u
r
os
c
i
e
n
c
e
, vol
. 13, 2021, doi
:
10.3389/
f
na
gi
.2021.692865.
[
23]
J
.
C
.
E
.
G
ue
r
r
e
r
o,
E
.
M
.
E
s
pa
ña
,
M
.
M
.
A
ñ
a
s
c
o,
a
nd
J
.
E
.
P
.
L
ope
r
a
,
“
D
a
t
a
s
e
t
f
or
hum
a
n
f
a
l
l
r
e
c
ogni
t
i
on
i
n
a
n
unc
ont
r
ol
l
e
d
e
nvi
r
onm
e
nt
,”
D
at
a i
n B
r
i
e
f
, vol
. 45, 2022, doi
:
10.1016/
j
.di
b.2022.108610.
[
24]
N
. H
noohom
, S
.
M
e
kr
uks
a
va
ni
c
h, a
nd
A
. J
i
t
pa
t
t
a
na
kul
, “
P
r
e
-
i
m
pa
c
t
a
nd i
m
pa
c
t
f
a
l
l
de
t
e
c
t
i
on ba
s
e
d
on a
m
ul
t
i
m
oda
l
s
e
n
s
or
us
i
n
g
a
de
e
p
r
e
s
i
dua
l
ne
t
w
or
k,”
I
nt
e
l
l
i
ge
nt
A
ut
om
at
i
on
and
Sof
t
C
om
put
i
ng
,
vol
.
36,
no.
3,
pp.
3371
–
3385,
2023,
doi
:
10.32604/
i
a
s
c
.2023.036551.
[
25]
M
.
K
.
Y
i
,
K
.
H
a
n,
a
nd
S
.
O
.
H
w
a
ng,
“
F
a
l
l
de
t
e
c
t
i
on
of
t
he
e
l
de
r
l
y
us
i
ng
de
noi
s
i
ng
L
S
T
M
-
ba
s
e
d
c
onvol
ut
i
ona
l
va
r
i
a
n
t
a
ut
oe
nc
ode
r
,”
I
E
E
E
Se
ns
or
s
J
our
nal
, vol
. 24, no. 11, pp. 18556
–
18567, 2024,
doi
:
10.1109/
J
S
E
N
.2024.3388478.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
S
Kakirala
Durga
Bhavani
is
doing
research
on
deep
learning
in
Department
of
Computer
Science
and
Enginee
ring
at
SRM
Institute
of
Science
and
Technol
ogy,
India.
She
did
her
post
-
graduation
in
computer
science
and
engineering
at
Amrit
a
Vishwa
Vidyapeetham
Coimbatore
.
She
has
published
few
papers
in
top
journals
.
She
ca
n
be
contacted
at
email:
durgabhavaniresea
rchscholar@gmail.com.
Melkias
Ferni
Ukrit
holds
a
Ph.D
.
in
computer
science
and
en
gineering
from
Sathyab
ama
Institute
of
Scienc
e
and
Techn
ology
Chenna
i,
India.
She
is
workin
g
as
Associa
t
e
Profes
sor
in
SRM
Institute
of
Scienc
e
and
Techn
ology,
Kattan
kul
athur.
Her
main
area
of
research
includes
image
processing,
machine
learning,
deep
learnin
g,
and
I
o
T.
She
is
a
life
member
of
the
Indian
Society
for
Technical
Education
(ISTE).
Sh
e
has
published
several
papers
in
well
-
known
peer
-
reviewed
journals.
She
can
be
contacted
at
email:
ferniukm@
srmist.ed
u.in.
Evaluation Warning : The document was created with Spire.PDF for Python.