I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
5
,
Octo
b
er
20
25
,
p
p
.
4
8
1
3
~
4
8
2
8
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
5
.
pp
4
8
1
3
-
4
8
2
8
4813
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
O
ptimi
zing
int
er
net
o
f
things
ba
se
d ga
s sens
o
rs: dee
p learning
a
nd perf
o
rma
nce
o
ptimiza
tion stra
t
eg
ies
M
a
ria
m
M
.
Abdella
t
if
1
,6
,
M
ehm
et
Ak
if
Çif
çi
2
,
3
,
Asm
a
a
A.
I
bra
him
1
,
H
a
ny
M
.
H
a
rb
4
,
Abeer
S.
Desuk
y
1,
5
1
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
,
A
l
-
A
z
h
a
r
U
n
i
v
e
r
si
t
y
,
C
a
i
r
o,
E
g
y
p
t
2
En
g
i
n
e
e
r
i
n
g
a
n
d
I
n
f
o
r
mat
i
c
s
D
e
p
a
r
t
men
t
,
K
l
a
i
p
ė
d
o
s V
a
l
s
t
y
b
i
n
ė
K
o
l
e
g
i
j
a
/
H
i
g
h
e
r
E
d
u
c
a
t
i
o
n
I
n
st
i
t
u
t
i
o
n
,
K
l
a
i
p
ė
d
a
,
L
i
t
h
u
a
n
i
a
3
D
e
p
a
r
t
m
e
n
t
o
f
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
,
F
a
c
u
l
t
y
o
f
E
n
g
i
n
e
e
r
i
n
g
a
n
d
N
a
t
u
r
a
l
S
c
i
e
n
c
e
s
,
B
a
n
d
ı
r
m
a
O
n
y
e
d
i
E
y
l
ü
l
U
n
i
v
e
r
s
i
t
y
,
B
a
l
ı
k
e
s
i
r
,
T
ü
r
k
i
y
e
4
F
a
c
u
l
t
y
o
f
En
g
i
n
e
e
r
i
n
g
,
A
l
-
A
z
h
a
r
U
n
i
v
e
r
si
t
y
(
B
o
y
s)
,
C
a
i
r
o
,
E
g
y
p
t
5
H
i
g
h
e
r
I
n
st
i
t
u
t
e
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
I
n
f
o
r
m
a
t
i
o
n
S
y
st
e
ms,
C
a
i
r
o
,
E
g
y
p
t
6
El
S
e
w
e
d
y
U
n
i
v
e
r
s
i
t
y
o
f
T
e
c
h
n
o
l
o
g
y
,
C
a
i
r
o
,
E
g
y
p
t
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 1
0
,
2
0
2
4
R
ev
is
ed
J
u
n
1
8
,
2
0
2
5
Acc
ep
ted
J
u
n
3
0
,
2
0
2
5
Th
e
ra
p
id
g
ro
wt
h
o
f
i
n
d
u
strializ
a
ti
o
n
a
n
d
in
tern
e
t
o
f
t
h
in
g
s
(Io
T)
d
riv
e
n
a
d
v
a
n
c
e
m
e
n
ts
in
I
n
d
u
stry
5
.
0
n
e
c
e
ss
it
a
tes
e
fficie
n
t
a
n
d
u
s
e
r
-
frien
d
ly
e
n
g
in
e
e
rin
g
s
o
lu
ti
o
n
s.
G
a
s
lea
k
a
g
e
i
n
c
id
e
n
ts
in
c
o
a
l
m
in
e
s,
c
h
e
m
ica
l
e
n
terp
rise
s,
a
n
d
h
o
u
se
h
o
ld
s
p
o
se
sig
n
ifi
c
a
n
t
risk
s
to
e
c
o
sy
ste
m
s
a
n
d
h
u
m
a
n
sa
fe
ty
,
e
m
p
h
a
siz
in
g
t
h
e
n
e
e
d
f
o
r
a
u
to
m
a
ted
a
n
d
ra
p
id
g
a
s
-
ty
p
e
d
e
tec
ti
o
n
.
Trad
it
io
n
a
l
d
e
tec
ti
o
n
m
e
th
o
d
s
r
e
ly
o
n
si
n
g
le
-
so
u
rc
e
d
a
ta
a
n
d
fo
c
u
s
o
n
iso
late
d
sp
a
ti
a
l
o
r
tem
p
o
ra
l
fe
a
tu
re
s,
li
m
it
in
g
a
c
c
u
ra
c
y
.
T
h
is
p
a
p
e
r
p
ro
p
o
se
s
a
m
u
lt
imo
d
a
l
a
rti
ficia
l
in
telli
g
e
n
c
e
(AI)
fu
sio
n
tec
h
n
i
q
u
e
c
o
m
b
i
n
in
g
p
re
-
train
e
d
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(CNN
s),
su
c
h
a
s
VG
G
1
6
,
wi
th
a
d
e
e
p
n
e
u
ra
l
n
e
two
rk
(DN
N)
m
o
d
e
l.
Th
e
p
a
rti
c
le
sw
a
rm
o
p
t
imiz
a
ti
o
n
(P
S
O
)
a
lg
o
rit
h
m
o
p
t
imiz
e
s
CNN
h
y
p
e
rp
a
ra
m
e
ters
,
o
u
tp
e
rfo
rm
in
g
trad
it
i
o
n
a
l
tri
a
l
-
a
n
d
-
e
rro
r
m
e
t
h
o
d
s.
T
h
e
sy
ste
m
a
d
d
re
ss
e
s
c
h
a
ll
e
n
g
e
s
p
o
se
d
b
y
g
a
se
s
b
e
in
g
o
d
o
rles
s,
c
o
lo
rles
s,
a
n
d
tas
tele
ss
,
wh
ich
li
m
it
c
o
n
v
e
n
ti
o
n
a
l
h
u
m
a
n
d
e
tec
ti
o
n
m
e
th
o
d
s.
By
le
v
e
ra
g
in
g
se
n
so
r
f
u
sio
n
,
th
e
late
fu
sio
n
tec
h
n
iq
u
e
in
teg
ra
tes
d
isti
n
c
t
n
e
two
r
k
a
rc
h
it
e
c
tu
re
s
fo
r
u
n
ifi
e
d
g
a
s
i
d
e
n
ti
fica
ti
o
n
.
Ex
p
e
rime
n
t
a
l
re
su
lt
s
d
e
m
o
n
stra
te
9
5
%
a
c
c
u
ra
c
y
u
sin
g
DN
N
with
g
a
s
se
n
so
r
d
a
ta,
9
6
%
with
o
p
ti
m
ize
d
VG
G
1
6
u
sin
g
th
e
rm
a
l
ima
g
in
g
,
a
n
d
9
9
.
5
%
th
ro
u
g
h
m
u
lt
imo
d
a
l
late
fu
si
o
n
.
Th
is
Io
T
-
e
n
h
a
n
c
e
d
so
l
u
ti
o
n
o
u
t
p
e
rfo
r
m
s
sin
g
le
-
se
n
so
r
a
p
p
r
o
a
c
h
e
s,
o
ffe
rin
g
a
ro
b
u
st
a
n
d
re
li
a
b
le
g
a
s
lea
k
a
g
e
d
e
tec
ti
o
n
sy
ste
m
su
it
a
b
le fo
r
in
d
u
strial
a
n
d
sm
a
rt
c
it
y
a
p
p
li
c
a
ti
o
n
s
.
K
ey
w
o
r
d
s
:
Ar
tific
ial
in
tellig
en
ce
Dee
p
lear
n
in
g
I
n
ter
n
et
o
f
th
in
g
s
L
ate
f
u
s
io
n
Mu
ltimo
d
al
d
ata
f
u
s
io
n
Par
ticle
s
war
m
o
p
tim
izatio
n
T
r
an
s
f
er
lear
n
i
n
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Ma
r
iam
M.
Ab
d
ellatif
Facu
lty
o
f
Scien
ce
,
Al
-
Azh
ar
Un
iv
er
s
ity
C
air
o
,
E
g
y
p
t
E
m
ail:
m
ar
iem
m
o
h
am
m
ed
9
7
4
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
I
n
th
e
r
ap
id
ly
a
d
v
an
ci
n
g
in
te
r
n
et
o
f
th
in
g
s
(
I
o
T
)
e
r
a,
g
as
s
en
s
o
r
s
h
av
e
b
ec
o
m
e
cr
u
cial
in
v
ar
io
u
s
s
ettin
g
s
,
f
r
o
m
h
o
m
es
to
h
ea
lth
ca
r
e
f
ac
ilit
ies
[
1
]
,
[
2
]
.
T
h
ese
s
en
s
o
r
s
ar
e
v
ital
f
o
r
d
etec
tin
g
h
ar
m
f
u
l
g
ases
an
d
en
s
u
r
in
g
air
q
u
ality
,
b
u
t
f
ac
e
ch
allen
g
es
lik
e
cr
o
s
s
-
s
en
s
itiv
ity
an
d
lim
ited
s
elec
tiv
ity
[
3
]
.
T
o
o
v
er
co
m
e
th
ese,
s
m
ar
t
g
as
s
en
s
in
g
in
teg
r
ates
s
en
s
o
r
ar
r
ay
s
,
s
ig
n
al
p
r
o
ce
s
s
in
g
,
an
d
m
ac
h
in
e
lear
n
in
g
,
en
h
a
n
cin
g
p
r
ec
is
io
n
a
n
d
s
en
s
itiv
ity
.
Desp
ite
tech
n
o
lo
g
ical
ad
v
an
ce
m
en
ts
,
g
as
leak
ag
e
r
em
ain
s
a
s
ig
n
if
ican
t
r
is
k
,
ca
u
s
in
g
in
d
u
s
tr
ial
d
is
aster
s
an
d
en
v
ir
o
n
m
e
n
tal
h
ar
m
[
4
]
.
E
f
f
icien
t
an
d
au
to
m
ated
g
as
leak
d
etec
tio
n
s
y
s
tem
s
ar
e
ess
en
tial
f
o
r
s
af
ety
.
R
ec
en
t
in
n
o
v
atio
n
s
in
clu
d
e
m
etal
o
x
id
e
s
em
ico
n
d
u
cto
r
(
MO
X)
s
en
s
o
r
s
in
“e
lectr
o
n
ic
n
o
s
e”
(
e
-
n
o
s
e
)
s
etu
p
s
,
wh
ich
u
s
e
g
as sen
s
o
r
a
r
r
ay
s
,
s
ig
n
al
p
r
o
ce
s
s
in
g
m
o
d
u
les,
an
d
p
atter
n
r
ec
o
g
n
itio
n
s
y
s
tem
s
[
5
]
.
Ar
tific
ial
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
8
1
3
-
4
8
2
8
4814
in
tellig
en
ce
(
AI
)
tech
n
iq
u
es
h
av
e
f
u
r
th
er
im
p
r
o
v
ed
g
as
d
etec
tio
n
,
with
ap
p
licatio
n
s
in
m
u
ltip
le
in
d
u
s
tr
ies
[
6
]
–
[
8
]
.
AI
e
n
h
an
c
es
th
e
e
-
n
o
s
e
's
ab
ilit
y
to
ac
cu
r
ately
id
en
tify
a
n
d
class
if
y
g
ases
,
cr
u
cial
f
o
r
p
r
e
v
en
tin
g
ac
cid
en
ts
an
d
h
ea
lth
r
is
k
s
.
P
r
ev
io
u
s
m
et
h
o
d
s
r
elied
o
n
s
i
n
g
le
d
ee
p
lear
n
in
g
(
DL
)
m
o
d
els,
b
u
t
co
m
b
i
n
in
g
m
u
ltip
le
DL
s
tr
u
ctu
r
es
h
as
im
p
r
o
v
e
d
d
etec
tio
n
ac
cu
r
ac
y
[
5
]
.
I
n
teg
r
atin
g
s
p
atial,
s
p
ec
tr
al,
an
d
tem
p
o
r
al
d
at
a
en
h
an
ce
s
p
e
r
f
o
r
m
an
ce
,
an
d
r
ed
u
cin
g
attr
ib
u
te
s
izes
in
cr
ea
s
es
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
[
9
]
.
A
n
AI
-
d
r
iv
e
n
ap
p
r
o
ac
h
u
s
in
g
DL
f
r
am
ew
o
r
k
s
n
o
w
in
teg
r
ates d
ata
f
r
o
m
d
i
v
er
s
e
s
o
u
r
ce
s
,
co
m
b
i
n
in
g
g
as sen
s
o
r
s
an
d
th
er
m
al
im
ag
in
g
f
o
r
ef
f
ec
tiv
e
g
as
d
et
ec
tio
n
.
T
h
ese
ad
v
a
n
ce
m
en
ts
p
r
o
v
id
e
v
al
u
ab
le
i
n
s
ig
h
ts
in
t
o
en
h
an
cin
g
s
af
ety
m
ea
s
u
r
es
an
d
d
ev
elo
p
in
g
r
o
b
u
s
t
g
as
-
s
en
s
in
g
s
o
lu
tio
n
s
.
I
n
th
is
ap
p
r
o
ac
h
,
two
in
d
ep
en
d
e
n
t
m
o
d
els
ar
e
f
ir
s
t
tr
ain
ed
o
n
t
h
eir
r
esp
ec
tiv
e
d
atasets
.
T
h
e
g
as
s
en
s
o
r
s
m
o
d
el
p
r
o
ce
s
s
es
th
e
s
eq
u
en
tial
d
ata
co
llected
f
r
o
m
v
ar
io
u
s
g
as
s
en
s
o
r
s
o
v
er
tim
e,
ex
tr
ac
tin
g
v
al
u
ab
le
tem
p
o
r
al
p
atter
n
s
an
d
tr
e
n
d
s
r
elate
d
to
g
as
co
n
ce
n
tr
atio
n
s
.
On
th
e
o
th
er
h
an
d
,
th
e
t
h
er
m
a
l
ca
m
er
a
m
o
d
el
an
aly
ze
s
th
e
t
h
er
m
al
im
ag
es
ca
p
tu
r
ed
b
y
th
e
ca
m
er
a,
d
etec
tin
g
tem
p
er
atu
r
e
v
ar
iatio
n
s
th
at
m
a
y
in
d
icate
g
as
leak
s
o
r
an
o
m
al
ies.
Af
ter
tr
ain
in
g
,
b
o
th
m
o
d
el
s
ar
e
u
s
ed
to
m
a
k
e
p
r
ed
ictio
n
s
o
n
t
h
e
s
am
e
s
et
o
f
test
s
am
p
les
o
r
d
ata
in
s
tan
c
es.
T
h
e
g
as
s
en
s
o
r
s
m
o
d
el
p
r
o
v
id
es
p
r
ed
ictio
n
s
b
ased
o
n
its
s
eq
u
en
ce
d
ata
a
n
aly
s
is
,
wh
ile
th
e
th
er
m
al
ca
m
er
a
m
o
d
el
g
e
n
er
ates
p
r
e
d
ictio
n
s
b
ased
o
n
its
in
ter
p
r
etatio
n
o
f
th
e
th
er
m
al
i
m
ag
es.
T
h
e
p
r
ed
ictio
n
s
f
r
o
m
b
o
th
m
o
d
els
a
r
e
th
e
n
c
o
m
b
in
ed
o
r
f
u
s
ed
u
s
in
g
a
late
f
u
s
io
n
tech
n
iq
u
e.
L
ate
f
u
s
io
n
b
etwe
en
g
as
s
en
s
o
r
s
s
eq
u
en
ce
d
ata
a
n
d
th
e
r
m
al
ca
m
er
a
im
ag
es
d
ata
is
p
ar
ticu
lar
ly
v
alu
ab
le
in
r
ea
l
-
wo
r
ld
s
ce
n
ar
i
o
s
wh
er
e
m
u
lti
p
le
s
o
u
r
ce
s
o
f
in
f
o
r
m
atio
n
ar
e
av
ailab
le,
an
d
th
e
f
u
s
io
n
o
f
th
ese
d
iv
e
r
s
e
d
ata
s
o
u
r
ce
s
ca
n
p
r
o
v
id
e
a
m
o
r
e
co
m
p
r
eh
e
n
s
iv
e
a
n
d
r
eliab
le
g
as
leak
d
etec
tio
n
s
o
lu
tio
n
.
Ho
wev
er
,
th
e
e
f
f
ec
tiv
en
ess
o
f
th
e
late
f
u
s
io
n
[
1
0
]
ap
p
r
o
ac
h
r
elies
o
n
ca
r
ef
u
lly
d
esig
n
in
g
th
e
f
u
s
io
n
s
tr
ateg
y
an
d
en
s
u
r
in
g
th
at
t
h
e
m
o
d
els
ar
e
well
-
tr
ain
e
d
o
n
th
e
ir
r
esp
ec
tiv
e
d
atasets
to
ca
p
tu
r
e
r
elev
an
t
p
atter
n
s
an
d
f
ea
tu
r
es sp
ec
if
ic
to
ea
ch
m
o
d
ality
.
T
h
e
s
ig
n
if
ican
t c
o
n
t
r
ib
u
tio
n
s
o
f
th
is
r
esear
ch
ar
e
a
s
f
o
llo
ws:
a.
T
h
e
p
a
r
t
i
c
l
e
s
wa
r
m
o
p
t
i
m
i
za
t
i
o
n
(
P
S
O
)
al
g
o
r
i
t
h
m
is
u
t
il
i
z
e
d
to
f
i
n
e
-
t
u
n
e
v
a
r
i
o
u
s
h
y
p
e
r
p
a
r
am
e
t
e
r
s
ass
o
c
i
a
te
d
w
i
t
h
t
h
e
VG
G
1
6
m
o
d
e
l
.
b.
A
p
io
n
ee
r
i
n
g
AI
-
d
r
i
v
en
f
r
am
ewo
r
k
th
at
in
teg
r
ates
two
d
is
tin
ct
m
o
d
alities
is
in
tr
o
d
u
ce
d
to
b
o
ls
ter
th
e
r
o
b
u
s
tn
ess
an
d
d
e
p
en
d
a
b
ilit
y
o
f
g
as d
etec
tio
n
.
c.
T
h
e
u
s
e
o
f
late
f
u
s
io
n
tech
n
i
q
u
es
f
o
r
g
as
d
etec
tio
n
an
d
t
h
e
i
d
en
tific
atio
n
o
f
leak
ed
g
ases
ar
e
d
em
o
n
s
tr
ated
b
y
lev
er
a
g
in
g
th
e
o
u
tp
u
ts
o
f
d
ee
p
lear
n
in
g
ar
ch
itectu
r
es VG
G1
6
an
d
d
ee
p
n
eu
r
al
n
etwo
r
k
(
DNN)
.
d.
Mo
r
e
ac
cu
r
ate
a
n
d
ef
f
ec
tiv
e
d
etec
tio
n
o
f
g
as is
ac
h
iev
e
d
b
as
ed
o
n
t
h
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
e
s
.
T
h
e
s
tr
u
ctu
r
e
o
f
t
h
is
p
ap
er
is
o
r
g
an
ized
as
f
o
llo
ws:
s
ec
tio
n
2
p
r
o
v
id
es
a
r
ev
iew
o
f
r
elate
d
r
esear
ch
.
Sectio
n
3
co
v
e
r
s
th
e
f
u
n
d
am
en
tals
an
d
b
ac
k
g
r
o
u
n
d
o
f
D
NN
,
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
et
wo
r
k
s
(
C
NNs)
,
an
d
PS
O.
Sectio
n
4
d
escr
ib
es
th
e
d
ataset
u
s
ed
in
th
e
ex
p
er
im
e
n
ts
.
Sectio
n
5
p
r
esen
ts
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
f
o
r
d
iag
n
o
s
in
g
p
lan
t
d
is
ea
s
es.
Se
ctio
n
6
f
o
cu
s
es
o
n
th
e
a
n
aly
s
is
an
d
d
is
cu
s
s
io
n
o
f
ex
p
er
im
e
n
tal
r
esu
lts
.
L
astl
y
,
s
ec
tio
n
7
o
f
f
e
r
s
th
e
co
n
cl
u
d
in
g
r
em
ar
k
s
.
2.
RE
L
AT
E
D
WO
RK
S
N
u
m
e
r
o
u
s
s
t
u
d
i
e
s
h
a
v
e
i
n
v
es
t
ig
a
t
e
d
t
h
e
i
n
t
e
g
r
a
t
i
o
n
o
f
c
h
e
m
ic
a
l
a
n
d
s
u
b
s
t
a
n
ce
-
d
e
t
e
c
ti
n
g
s
en
s
o
r
s
i
n
t
o
I
o
T
p
l
a
t
f
o
r
m
s
.
F
o
r
e
x
a
m
p
l
e
,
V
e
n
k
a
t
as
u
b
r
a
m
a
n
i
a
n
e
t
a
l
.
[
1
1
]
e
n
h
a
n
c
e
s
d
e
t
e
c
ti
o
n
a
n
d
m
a
n
a
g
e
m
e
n
t
o
f
b
r
e
a
k
d
o
w
n
s
i
n
i
n
d
u
s
t
r
i
a
l
I
o
T
(
I
I
o
T
)
d
e
v
i
c
e
s
u
s
i
n
g
r
e
al
-
t
im
e
s
e
n
s
o
r
d
a
t
a
f
o
r
d
e
ci
s
i
o
n
-
m
a
k
i
n
g
,
a
d
d
r
e
s
s
i
n
g
c
h
a
l
l
e
n
g
e
s
l
i
k
e
d
e
n
o
is
i
n
g
a
n
d
m
i
s
s
i
n
g
d
a
t
a
wi
t
h
p
r
e
p
r
o
c
e
s
s
i
n
g
a
n
d
d
a
t
a
f
u
s
i
o
n
te
c
h
n
i
q
u
e
s
.
M
o
d
e
ls
l
i
k
e
P
r
o
p
e
n
s
i
t
y
N
e
t
,
D
NN
,
a
n
d
C
NN
l
o
n
g
s
h
o
r
t
-
t
e
r
m
m
e
m
o
r
y
(
C
N
N
-
L
S
T
M
)
a
r
e
u
s
e
d
,
p
r
o
v
i
n
g
e
f
f
e
c
t
i
v
e
w
it
h
C
ase
W
es
t
e
r
n
R
e
s
e
r
v
e
U
n
i
v
e
r
s
it
y
(
C
W
R
U
)
d
a
t
a
f
o
r
f
a
u
l
t
d
e
t
e
c
ti
o
n
.
I
n
[
1
2
]
,
t
h
e
f
o
c
u
s
i
s
o
n
h
a
z
a
r
d
o
u
s
g
a
s
l
e
a
k
a
g
e
,
e
m
p
h
a
s
i
z
i
n
g
s
a
f
et
y
i
n
c
h
e
m
i
c
al
i
n
d
u
s
t
r
i
e
s
.
T
h
e
s
t
u
d
y
p
r
o
p
o
s
es
a
n
I
o
T
-
b
as
e
d
s
t
r
at
e
g
y
u
s
i
n
g
an
A
r
d
u
i
n
o
U
NO
R
3
a
n
d
s
e
n
s
o
r
s
t
o
m
o
n
i
t
o
r
g
as
e
s
l
i
k
e
NO
2
a
n
d
c
a
r
b
o
n
m
o
n
o
x
i
d
e
.
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
e
n
h
a
n
c
e
s
d
a
t
a
ac
c
u
r
a
c
y
,
a
n
d
h
y
b
r
i
d
h
i
d
d
e
n
M
a
r
k
o
v
a
n
d
A
I
m
o
d
e
l
s
i
m
p
r
o
v
e
e
r
r
o
r
d
e
t
e
c
t
i
o
n
.
S
t
u
d
i
e
s
a
ls
o
c
o
m
b
i
n
e
g
a
s
s
e
n
s
o
r
s
a
n
d
t
h
e
r
m
a
l
i
m
a
g
i
n
g
f
o
r
g
a
s
d
e
t
e
ct
i
o
n
.
F
o
r
i
n
s
t
a
n
c
e
,
P
e
n
g
e
t
a
l
.
[
1
3
]
u
s
es
C
N
N
t
o
c
a
t
e
g
o
r
i
z
e
g
a
s
es
u
s
i
n
g
e
i
g
h
t
s
e
n
s
o
r
s
,
w
h
i
l
e
[
1
4
]
e
m
p
l
o
y
s
v
a
r
i
o
u
s
D
L
m
o
d
e
l
s
,
i
n
c
l
u
d
i
n
g
C
NN
,
L
S
T
M
,
an
d
a
u
t
o
e
n
c
o
d
e
r
s
f
o
r
g
as
l
e
a
k
d
e
t
e
c
t
i
o
n
,
a
c
h
i
e
v
i
n
g
n
o
t
a
b
l
e
a
c
c
u
r
a
c
y
.
R
es
e
a
r
c
h
es
l
i
k
e
in
[
1
5
]
a
n
d
[
1
6
]
u
t
i
l
i
z
e
C
N
N
a
n
d
L
S
T
M
f
o
r
e
x
t
r
a
c
ti
n
g
s
p
a
t
i
a
l
-
t
e
m
p
o
r
a
l
i
n
f
o
r
m
a
t
i
o
n
f
r
o
m
s
p
a
r
s
e
d
at
a
.
R
es
e
a
r
c
h
e
s
l
i
k
e
[
1
7
]
a
n
d
[
1
8
]
e
x
p
l
o
r
e
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
te
c
h
n
i
q
u
e
s
f
o
r
a
n
a
l
y
z
i
n
g
i
n
f
r
a
r
e
d
(
I
R
)
t
h
e
r
m
a
l
i
m
a
g
es
i
n
g
a
s
d
et
e
c
ti
o
n
.
B
i
n
et
a
l
.
[
1
9
]
in
t
r
o
d
u
c
e
s
a
t
e
n
s
o
r
-
b
as
e
d
le
a
k
ag
e
d
e
t
e
c
ti
o
n
m
e
t
h
o
d
w
i
t
h
R
e
s
N
e
t
5
0
f
o
r
g
a
s
i
d
e
n
t
if
i
c
a
t
i
o
n
,
a
n
d
L
i
u
e
t
a
l
.
[
2
0
]
u
s
e
s
a
n
I
R
t
h
e
r
m
a
l
i
m
a
g
i
n
g
s
y
s
te
m
w
i
t
h
d
i
v
e
r
s
e
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
al
g
o
r
i
t
h
m
s
f
o
r
d
e
t
e
c
t
i
n
g
f
la
m
m
a
b
l
e
g
as
lea
k
s
.
I
n
[
2
1
]
,
p
a
t
t
e
r
n
r
ec
o
g
n
i
t
io
n
t
e
c
h
n
i
q
u
e
s
l
i
k
e
r
a
n
d
o
m
f
o
r
e
s
t
,
s
u
p
p
o
r
t
v
e
c
t
o
r
m
a
c
h
i
n
e
(
SV
M
)
,
a
n
d
s
h
a
l
l
o
w
m
u
l
t
i
-
l
a
y
e
r
p
e
r
ce
p
t
r
o
n
a
l
g
o
r
i
t
h
m
s
i
d
e
n
ti
f
y
g
a
s
es
a
t
l
o
w
c
o
n
c
e
n
t
r
a
t
i
o
n
s
u
s
i
n
g
a
Sn
O
2
g
a
s
s
e
n
s
o
r
,
w
it
h
a
r
t
i
f
i
cia
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
a
c
h
i
e
v
i
n
g
h
i
g
h
e
f
f
e
c
t
i
v
e
n
e
s
s
.
R
e
l
at
e
d
s
t
u
d
i
es
[
2
2
]
a
n
d
[
2
3
]
u
s
e
p
a
t
t
e
r
n
r
e
c
o
g
n
i
t
i
o
n
m
e
t
h
o
d
s
f
o
r
c
l
a
s
s
i
f
y
i
n
g
g
r
a
v
i
t
a
ti
o
n
a
l
s
e
a
r
c
h
a
l
g
o
r
i
t
h
m
(
G
S
A
)
d
at
a
.
M
o
r
e
o
v
e
r
,
Ni
k
o
l
o
p
o
u
l
o
s
e
t
a
l
.
[
2
4
]
i
n
v
es
t
i
g
at
e
s
t
h
e
r
m
a
l
i
m
a
g
i
n
g
f
o
r
g
as
l
e
a
k
ag
e
d
e
t
e
c
t
i
o
n
,
u
s
i
n
g
VGG
-
1
6
a
n
d
k
-
N
N
f
o
r
cl
as
s
i
f
y
i
n
g
g
a
s
e
s
w
it
h
h
i
g
h
a
c
c
u
r
a
c
y
,
d
e
p
e
n
d
i
n
g
o
n
t
h
e
s
p
e
c
i
f
i
c
a
p
p
l
i
c
a
t
i
o
n
a
n
d
t
r
a
d
e
-
o
f
f
s
b
e
t
w
e
e
n
c
o
m
p
u
t
a
t
i
o
n
a
l
c
o
s
t
a
n
d
p
e
r
f
o
r
m
a
n
c
e
.
T
h
e
s
e
s
t
u
d
i
e
s
i
ll
u
s
t
r
a
t
e
t
h
e
s
y
n
e
r
g
y
b
e
t
w
ee
n
a
d
v
a
n
c
e
d
s
e
n
s
o
r
t
e
c
h
n
o
l
o
g
y
a
n
d
a
n
a
l
y
t
i
c
a
l
t
e
c
h
n
i
q
u
e
s
f
o
r
e
n
h
a
n
c
i
n
g
i
n
d
u
s
t
r
i
al
s
a
f
e
t
y
a
n
d
g
a
s
d
et
e
c
ti
o
n
c
a
p
a
b
i
l
i
ti
e
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Op
timiz
in
g
in
tern
et
o
f th
in
g
s
b
a
s
ed
g
a
s
s
en
s
o
r
s
:
d
ee
p
le
a
r
n
in
g
a
n
d
…
(
Ma
r
ia
m
M.
A
b
d
ell
a
tif
)
4815
3.
B
AS
I
CS
AND
B
ACK
G
RO
U
ND
3
.
1
.
I
nte
rnet
o
f
t
hin
g
s
T
h
e
in
ter
n
et
o
f
th
i
n
g
s
(
I
o
T
)
c
o
n
n
ec
ts
d
ev
ices
an
d
s
y
s
tem
s
,
en
ab
lin
g
s
ea
m
less
co
m
m
u
n
icatio
n
,
d
at
a
ex
ch
an
g
e
,
an
d
in
tellig
en
t
d
e
cisi
o
n
-
m
ak
in
g
b
y
lin
k
in
g
p
h
y
s
ical
o
b
jects
wi
th
s
en
s
o
r
s
a
n
d
ac
tu
ato
r
s
to
th
e
in
ter
n
et.
I
n
ter
m
s
o
f
ar
ch
itect
u
r
e,
I
o
T
co
n
s
is
ts
o
f
s
ev
er
al
in
ter
co
n
n
ec
ted
co
m
p
o
n
en
ts
[
2
5
]
.
At
th
e
f
o
u
n
d
atio
n
ar
e
s
en
s
o
r
s
an
d
ac
tu
ato
r
s
,
wh
ich
ca
p
tu
r
e
d
ata
s
u
ch
as
tem
p
er
atu
r
e
an
d
m
o
tio
n
,
an
d
en
ab
le
p
h
y
s
ical
ac
tio
n
s
b
ased
o
n
th
is
d
ata.
T
h
ese
d
ev
ices
r
ely
o
n
c
o
n
n
ec
tiv
ity
m
ec
h
an
is
m
s
lik
e
W
i
-
Fi,
B
lu
eto
o
th
,
o
r
ce
llu
la
r
n
etwo
r
k
s
to
co
m
m
u
n
icate
an
d
tr
an
s
m
it
d
ata
e
f
f
icien
tly
.
On
ce
tr
an
s
m
itted
,
clo
u
d
c
o
m
p
u
tin
g
p
latf
o
r
m
s
s
to
r
e
an
d
p
r
o
ce
s
s
th
is
d
ata,
p
r
o
v
id
in
g
th
e
co
m
p
u
tatio
n
al
p
o
wer
n
ec
ess
ar
y
f
o
r
r
ea
l
-
tim
e
in
s
ig
h
ts
.
Data
an
aly
tics
p
lay
s
a
cr
u
cial
r
o
le
at
th
is
s
tag
e,
wh
er
e
m
ac
h
i
n
e
lear
n
i
n
g
tech
n
iq
u
es
ar
e
em
p
lo
y
e
d
to
ex
tr
ac
t
m
ea
n
in
g
f
u
l
p
atter
n
s
an
d
s
u
p
p
o
r
t
p
r
ed
icti
v
e
ac
tio
n
s
.
T
h
e
f
in
al
lay
er
in
v
o
lv
es
u
s
er
in
ter
f
ac
es,
s
u
ch
as
ap
p
licatio
n
s
an
d
d
ash
b
o
ar
d
s
,
wh
ich
f
ac
ilit
ate
r
em
o
te
m
o
n
ito
r
in
g
an
d
m
a
n
ag
em
en
t
o
f
I
o
T
s
y
s
tem
s
.
I
o
T
a
p
p
licatio
n
s
s
p
an
a
wid
e
r
an
g
e
o
f
in
d
u
s
tr
ies,
in
clu
d
in
g
s
m
ar
t
cities
—
wh
er
e
th
ey
s
u
p
p
o
r
t
tr
af
f
ic
m
an
a
g
em
en
t
an
d
en
er
g
y
o
p
tim
izatio
n
—
an
d
h
ea
lth
ca
r
e
,
en
ab
lin
g
r
ea
l
-
tim
e
h
ea
lth
m
o
n
ito
r
in
g
an
d
im
p
r
o
v
ed
p
atie
n
t
o
u
tco
m
es.
T
h
is
wo
r
k
b
u
ild
s
u
p
o
n
th
ese
ca
p
a
b
i
liti
es
b
y
u
tili
zin
g
m
u
ltimo
d
al
d
ata
f
o
r
g
as d
etec
tio
n
.
T
h
r
o
u
g
h
th
e
ap
p
licatio
n
o
f
f
u
s
io
n
tech
n
i
q
u
es,
d
is
cu
s
s
ed
in
th
e
f
o
llo
win
g
s
ec
tio
n
,
th
e
s
y
s
tem
aim
s
to
en
h
an
ce
d
etec
tio
n
ac
cu
r
ac
y
an
d
r
eliab
ilit
y
in
s
af
ety
-
cr
itical
en
v
ir
o
n
m
e
n
ts
.
3
.
2
.
M
ultim
o
da
l
da
t
a
f
us
io
n a
pp
ro
a
ches
Fu
s
io
n
s
tr
ateg
ies
in
m
u
ltimo
d
al
m
ac
h
in
e
lear
n
in
g
ca
n
b
e
eith
er
m
o
d
el
-
b
ased
o
r
m
o
d
e
l
-
ag
n
o
s
tic.
Mo
d
el
-
ag
n
o
s
tic
f
u
s
io
n
m
ix
es
m
o
d
alities
,
s
u
ch
as
th
er
m
al
i
m
ag
in
g
an
d
s
en
s
o
r
d
ata,
em
p
l
o
y
in
g
ea
r
ly
,
late,
o
r
in
ter
m
ed
iate
f
u
s
io
n
tec
h
n
iq
u
es.
E
ar
ly
f
u
s
io
n
in
co
r
p
o
r
ate
s
r
aw
d
ata
d
u
r
i
n
g
th
e
in
itia
l
p
r
o
ce
s
s
in
g
s
tag
e,
ca
p
tu
r
in
g
in
ter
ac
tio
n
s
b
etwe
en
m
o
d
alities
,
wh
er
ea
s
late
f
u
s
io
n
,
o
r
d
ec
is
io
n
-
lev
el
f
u
s
io
n
,
m
er
g
es
in
d
ep
en
d
en
t
p
r
ed
ictio
n
s
u
s
in
g
m
eth
o
d
s
s
u
ch
as
m
ajo
r
ity
v
o
tin
g
.
I
n
ter
m
ed
iate
f
u
s
io
n
m
ix
es
f
ea
tu
r
es
f
r
o
m
d
if
f
er
e
n
t
m
o
d
alities
at
h
ig
h
er
a
b
s
tr
ac
tio
n
lev
els,
r
esu
ltin
g
i
n
im
p
r
o
v
ed
p
er
f
o
r
m
a
n
ce
.
Mo
d
el
-
b
ased
f
u
s
io
n
,
also
k
n
o
w
n
as
m
u
ltit
ask
f
u
s
io
n
,
tr
ain
s
m
o
d
els
o
n
m
a
n
y
task
s
at
th
e
s
am
e
tim
e
wh
ile
s
h
ar
in
g
r
ep
r
esen
tatio
n
s
ac
r
o
s
s
m
o
d
alities
to
p
r
o
m
o
te
g
en
e
r
aliza
tio
n
.
T
h
e
n
ex
t sectio
n
d
is
cu
s
s
es d
ee
p
lear
n
in
g
m
o
d
els f
o
r
g
as d
etec
tio
n
[
2
6
]
.
3
.
3
.
Deep
lea
rning
m
o
dels
Dee
p
lear
n
in
g
,
a
k
e
y
s
u
b
s
et
o
f
m
ac
h
in
e
lear
n
in
g
,
ex
ce
ls
in
p
r
o
ce
s
s
in
g
co
m
p
le
x
d
ata.
T
w
o
ess
en
tia
l
m
o
d
els
ar
e
DNNs
an
d
C
NNs,
o
f
ten
en
h
an
ce
d
b
y
tr
a
n
s
f
er
lea
r
n
in
g
tech
n
iq
u
es
[
2
7
]
.
T
h
e
f
o
llo
win
g
s
u
b
s
ec
tio
n
s
p
r
o
v
id
e
a
b
r
ief
o
v
e
r
v
iew
o
f
D
NNs,
C
NN
s
,
an
d
th
e
r
o
le
o
f
tr
an
s
f
er
lear
n
in
g
in
en
h
an
cin
g
t
h
eir
p
er
f
o
r
m
a
n
ce
.
3
.
3
.
1
.
Dee
p
neura
l
net
wo
rk
s
DNNs
ar
e
ad
v
an
ce
d
ar
tific
ial
n
eu
r
al
n
etwo
r
k
s
with
m
u
ltip
le
h
id
d
en
lay
er
s
th
at
au
to
m
atica
lly
lear
n
co
m
p
lex
p
atter
n
s
f
r
o
m
r
aw
d
ata.
T
h
ey
e
x
ce
l
in
task
s
s
u
c
h
as
im
ag
e
r
ec
o
g
n
itio
n
,
s
p
ee
ch
p
r
o
ce
s
s
in
g
,
an
d
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
s
.
Desp
ite
r
eq
u
ir
in
g
s
ig
n
i
f
ican
t
c
o
m
p
u
tatio
n
al
r
eso
u
r
ce
s
,
ad
v
a
n
ce
m
en
ts
in
h
ar
d
war
e
an
d
alg
o
r
ith
m
s
h
av
e
e
n
ab
led
t
h
eir
wid
esp
r
ea
d
a
d
o
p
tio
n
[
2
8
]
.
3
.
3
.
2
.
T
ra
ns
f
er
lea
rning
wit
h
CNNs
C
NNs
co
n
s
is
t
o
f
co
n
v
o
lu
tio
n
al,
p
o
o
lin
g
,
an
d
f
u
lly
co
n
n
ec
t
ed
lay
er
s
,
ea
ch
s
er
v
in
g
d
is
tin
ct
r
o
les
in
im
ag
e
p
r
o
ce
s
s
in
g
.
T
r
an
s
f
er
le
ar
n
in
g
lev
er
a
g
es
p
r
e
-
tr
ain
ed
m
o
d
els
lik
e
VGG1
6
,
k
n
o
wn
f
o
r
its
s
u
cc
es
s
in
task
s
s
u
ch
as
I
m
ag
eNe
t
class
if
ica
tio
n
.
B
y
r
eu
s
in
g
VGG1
6
'
s
co
n
v
o
l
u
tio
n
al
lay
er
s
,
tr
an
s
f
e
r
lear
n
in
g
en
ab
les
ef
f
ec
tiv
e
f
ea
tu
r
e
ex
tr
ac
tio
n
o
r
f
in
e
-
tu
n
in
g
f
o
r
n
ew
task
s
with
lim
ited
d
ata.
T
h
e
n
ex
t
s
ec
tio
n
ad
d
r
ess
es
h
y
p
er
p
ar
am
eter
o
p
tim
izatio
n
f
o
r
th
ese
m
o
d
els u
s
in
g
p
ar
ticle
s
war
m
o
p
tim
izatio
n
[
2
9
]
.
3
.
4
.
P
a
rt
icle
s
wa
rm
o
pti
m
iza
t
io
n
a
lg
o
rit
hm
PS
O
s
im
u
late
s
th
e
b
eh
av
io
r
o
f
b
ir
d
a
n
d
f
is
h
g
r
o
u
p
s
,
with
p
ar
ticles
d
is
tr
ib
u
ted
in
a
s
ea
r
ch
s
p
ac
e
an
d
ev
alu
ated
b
ased
o
n
a
n
o
b
ject
iv
e.
E
ac
h
p
a
r
ticle
ad
ju
s
ts
its
p
o
s
itio
n
u
s
in
g
its
cu
r
r
e
n
t
lo
ca
tio
n
,
b
est
-
k
n
o
wn
p
o
s
itio
n
,
an
d
n
ei
g
h
b
o
r
s
'
p
o
s
iti
o
n
s
.
T
h
is
iter
ativ
e
p
r
o
ce
s
s
co
n
tin
u
es
u
n
til
th
e
d
esire
d
o
u
tco
m
e
is
ac
h
iev
ed
[
3
0
]
.
T
h
e
v
elo
city
a
n
d
p
o
s
itio
n
u
p
d
ates f
o
r
ea
ch
p
a
r
ticle
at
iter
atio
n
t+1
ar
e
g
iv
en
b
y
:
v
t
+
1
=
ω
×
v
t
+
c
1
×
r
1
× (
p
(
b
es
t
)
t
-
x
t
)
+
c
2
×
r
2
×
(
G
(
b
es
t
)
t
-
x
t
)
(
1
)
x
t
+
1
=
x
t
+
v
t
+
1
(
2
)
W
h
er
e,
t
is
th
e
iter
atio
n
n
u
m
b
er
,
ω
(
o
m
e
g
a
)
is
th
e
weig
h
t,
1
an
d
2
ar
e
co
g
n
itiv
e
an
d
s
o
cial
p
ar
am
eter
s
,
a
n
d
1
,
2
ar
e
r
a
n
d
o
m
n
u
m
b
er
s
b
etwe
en
0
an
d
1
.
Fo
r
th
is
r
esear
c
h
,
th
e
co
s
t
f
u
n
ctio
n
is
th
e
m
ea
n
s
q
u
ar
e
er
r
o
r
(
MSE
)
o
f
a
n
e
u
r
al
n
etwo
r
k
with
o
n
e
h
id
d
en
lay
er
a
n
d
1
0
n
eu
r
o
n
s
.
T
h
e
PS
O
alg
o
r
ith
m
s
y
s
tem
atica
lly
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
8
1
3
-
4
8
2
8
4816
m
in
im
izes
th
is
co
s
t
b
y
r
ef
in
i
n
g
in
p
u
ts
,
s
tar
tin
g
with
a
r
an
d
o
m
s
u
b
s
et
o
f
f
ea
tu
r
es
an
d
g
r
ad
u
ally
ac
h
iev
in
g
o
p
tim
al
r
esu
lts
,
as sh
o
wn
in
Fig
u
r
e
1
.
Fig
u
r
e
1
.
Flo
wch
ar
t
o
f
PS
O
alg
o
r
ith
m
4.
DATAS
E
T
DE
SCRI
P
T
I
O
N
T
h
e
m
u
ltimo
d
al
g
as
d
ata
is
a
co
m
p
r
eh
e
n
s
iv
e
d
ataset
th
at
a
m
alg
am
ates
v
ital
in
f
o
r
m
atio
n
f
r
o
m
s
ev
en
d
is
tin
ct
g
as
s
en
s
o
r
s
an
d
th
e
r
m
al
im
ag
es
ca
p
tu
r
ed
u
s
in
g
a
th
er
m
al
ca
m
er
a
.
T
h
e
d
a
taset
en
co
m
p
ass
es
m
ea
s
u
r
em
en
ts
o
f
two
d
if
f
er
e
n
t
g
ases
,
cr
ea
tin
g
f
o
u
r
well
-
d
ef
in
ed
class
es:
Per
f
u
m
e
,
N
o
G
as,
Sm
o
k
e
,
an
d
Mix
tu
r
e
o
f
p
er
f
u
m
e
a
n
d
Sm
o
k
e
.
T
h
e
d
ata
co
llectio
n
p
r
o
ce
s
s
in
v
o
lv
ed
u
tili
zin
g
s
ev
en
m
etal
o
x
i
d
e
g
as
s
en
s
o
r
s
,
MQ
2
,
MQ
5
,
MQ
3
,
MQ
8
,
MQ
6
,
MQ
7
,
an
d
MQ
1
3
5
,
alo
n
g
w
ith
a
s
o
p
h
is
ticated
th
er
m
al
im
ag
in
g
ca
m
er
a.
T
h
is
m
u
ltimo
d
al
ap
p
r
o
ac
h
en
a
b
led
th
e
s
im
u
ltan
eo
u
s
ac
q
u
is
itio
n
o
f
n
u
m
er
ical
v
alu
es
f
r
o
m
th
e
g
as
s
en
s
o
r
s
an
d
th
er
m
al
im
ag
es,
p
r
o
v
i
d
in
g
a
d
iv
er
s
e
an
d
in
f
o
r
m
ativ
e
d
ataset
[
3
1
]
.
T
h
e
co
m
p
r
e
h
en
s
iv
e
d
etails
o
f
th
e
d
ataset
ar
e
elab
o
r
ated
in
s
u
b
s
eq
u
e
n
t sectio
n
s
.
4
.
1
.
G
a
s
s
ens
o
rs
Gas
s
en
s
o
r
s
d
etec
t
g
ases
b
y
co
n
v
er
tin
g
ch
em
ical
d
ata
in
t
o
elec
tr
ical
s
ig
n
als,
with
MO
X
s
en
s
o
r
s
in
ter
f
ac
ed
to
a
m
icr
o
co
n
tr
o
lle
r
f
o
r
p
r
o
ce
s
s
in
g
an
d
d
ata
co
m
m
u
n
icatio
n
.
An
alo
g
-
to
-
d
ig
ital
co
n
v
er
ter
s
(
ADCs
)
tr
an
s
f
o
r
m
an
al
o
g
o
u
tp
u
ts
in
t
o
d
ig
ital
d
ata,
wh
ile
wir
ed
o
r
wir
eless
co
m
m
u
n
icatio
n
tr
a
n
s
m
its
th
e
d
ata
f
o
r
s
to
r
ag
e
an
d
an
aly
s
is
.
T
h
e
d
at
aset
in
clu
d
es
s
ev
en
MO
X
s
en
s
o
r
s
(
MQ
2
,
MQ
3
,
MQ
5
,
MQ
6
,
MQ
7
,
MQ
8
,
an
d
MQ
1
3
5
)
in
teg
r
ated
in
to
an
I
o
T
s
y
s
tem
f
o
r
au
to
m
ated
d
ata
c
o
llectio
n
,
as
s
h
o
wn
in
Fig
u
r
e
2
.
T
h
ese
s
en
s
o
r
s
ar
e
co
m
p
ac
t,
d
u
r
ab
le,
an
d
r
esp
o
n
s
iv
e
to
g
ases
lik
e
C
O,
m
eth
a
n
e,
L
PG,
an
d
alco
h
o
l
as
s
h
o
wn
in
T
ab
le
1
,
with
s
en
s
itiv
ity
,
s
elec
tiv
ity
,
an
d
r
e
s
p
o
n
s
e
tim
e
cr
itical
to
th
eir
p
er
f
o
r
m
an
ce
.
D
u
r
in
g
d
ata
co
l
lectio
n
,
th
e
s
en
s
o
r
s
wer
e
p
lace
d
1
m
m
ap
a
r
t.
T
h
e
o
b
j
ec
t
i
v
e
f
u
n
ct
i
o
n
mee
t
s
t
h
e
cr
i
t
eri
a.
Dete
r
m
in
e
th
e
n
ew
d
i
r
ec
tio
n
f
o
r
th
e
s
ea
r
ch
Ra
n
d
o
m
ly
se
t
th
e
i
n
stati
stica
l
P
S
O p
a
ra
m
e
ters
NO
Ad
ju
st t
h
e
v
e
l
o
c
it
y
a
n
d
c
o
ll
e
c
ti
v
e
p
o
sit
io
n
o
f
th
e
sw
a
rm
b
y
a
p
p
ly
i
n
g
th
e
u
p
d
a
ted
P
S
O v
e
l
o
c
it
y
to
e
a
c
h
p
a
rti
c
le
Co
m
p
u
te
t
h
e
o
b
je
c
tiv
e
f
u
n
c
ti
o
n
f
o
r t
h
e
in
it
ia
l
p
o
p
u
la
tio
n
As
se
ss
th
e
o
b
jec
ti
v
e
f
u
n
c
ti
o
n
f
o
r
e
a
c
h
in
d
i
v
id
u
a
l
p
a
rt
icle
S
et
th
e
v
elo
citi
es an
d
p
o
sit
i
o
n
s
o
f
th
e sw
arm
to
t
h
eir
i
n
it
ial
v
alu
es
Co
m
p
a
re
th
e
p
a
rti
c
le'
s p
e
rso
n
a
l
b
e
st (
p
b
e
st)
wit
h
th
e
g
l
o
b
a
l
b
e
st (
g
b
e
st)
R
etr
iev
e
th
e
g
lo
b
al
s
o
lu
tio
n
Sta
r
t
Iter=
it
e
r+
1
Sto
p
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Op
timiz
in
g
in
tern
et
o
f th
in
g
s
b
a
s
ed
g
a
s
s
en
s
o
r
s
:
d
ee
p
le
a
r
n
in
g
a
n
d
…
(
Ma
r
ia
m
M.
A
b
d
ell
a
tif
)
4817
Fig
u
r
e
2
.
I
o
T
-
en
ab
led
th
e
p
r
o
c
ess
o
f
g
ath
er
in
g
d
ata
T
ab
le
1
.
Sen
s
in
g
g
ases
an
d
g
a
s
s
en
s
o
r
s
S
e
n
s
o
r
S
e
n
s
i
t
i
v
i
t
y
g
as
M
Q
2
Pr
o
p
a
n
e
,
B
u
t
a
n
e
,
M
e
t
h
a
n
e
,
LPG
,
S
m
o
k
e
M
Q
8
H
y
d
r
o
g
e
n
g
as
M
Q
3
S
mo
k
e
,
Et
h
a
n
o
l
,
A
l
c
o
h
o
l
M
Q
1
3
5
A
i
r
Q
u
a
l
i
t
y
(
B
e
n
z
e
n
e
,
S
mo
k
e
)
M
Q
6
B
u
t
a
n
e
g
a
s,
LPG
M
Q
5
N
a
t
u
r
a
l
g
a
s,
LPG
M
Q
7
C
a
r
b
o
n
M
o
n
o
x
i
d
e
4
.
2
.
T
herm
a
l
ca
m
er
a
i
m
a
g
es
T
h
e
d
ataset
u
s
es
a
th
er
m
al
c
am
er
a
th
at
ca
p
tu
r
es
tem
p
er
at
u
r
e
f
lu
ctu
atio
n
s
v
ia
in
f
r
ar
ed
lig
h
t,
with
ea
ch
p
ix
el
ac
tin
g
as
an
in
f
r
ar
e
d
tem
p
er
atu
r
e
s
en
s
o
r
.
I
m
a
g
es
ar
e
o
u
tp
u
t
in
R
G
B
f
o
r
m
at,
en
ab
lin
g
v
is
u
aliza
tio
n
ir
r
esp
ec
tiv
e
o
f
lig
h
tin
g
co
n
d
itio
n
s
.
T
h
e
Seek
th
er
m
al
ca
m
er
a
,
with
a
2
0
6
×
1
5
6
s
en
s
o
r
,
a
-
4
0
°C
to
3
3
0
°C
r
an
g
e,
an
d
3
2
,
1
3
6
p
ix
els,
was
u
s
ed
.
Gas
s
en
s
o
r
s
an
d
th
e
th
er
m
al
ca
m
er
a
co
llected
d
ata
s
im
u
ltan
eo
u
s
ly
,
as
n
o
p
u
b
lic
d
ataset
co
m
b
in
in
g
th
e
r
m
al
im
ag
es a
n
d
g
as sen
s
o
r
d
ata
ex
is
ted
.
Data
was g
ath
er
ed
b
y
p
o
s
itio
n
in
g
s
ev
en
g
as
s
en
s
o
r
s
1
m
m
ap
ar
t,
m
o
n
ito
r
in
g
g
ases
f
r
o
m
p
er
f
u
m
es
an
d
i
n
ce
n
s
e
at
in
ter
v
als
o
v
e
r
1
.
5
h
o
u
r
s
.
T
h
r
ee
class
es
—
No
Gas
,
Per
f
u
m
e
,
an
d
Sm
o
k
e
—
wer
e
s
am
p
led
,
wit
h
6
,
4
0
0
to
tal
s
am
p
les
(
1
,
6
0
0
p
er
class
)
.
Sen
s
o
r
o
u
tp
u
ts
wer
e
co
n
v
er
ted
in
to
1
0
-
b
it
d
ig
ital v
alu
es f
o
r
a
n
aly
s
is
,
as sh
o
wn
in
T
ab
le
2
.
T
ab
le
2
.
E
x
am
p
les o
f
t
h
e
d
ata
f
o
r
th
e
t
h
er
m
al
im
ag
es a
n
d
th
e
r
elate
d
g
as a
r
r
a
y
r
ea
d
in
g
s
C
l
a
s
s
t
y
p
e
G
a
s
Th
e
r
m
a
l
i
m
a
g
e
G
a
s
Th
e
r
m
a
l
i
m
a
g
e
N
o
G
a
s
[
7
3
3
,
5
3
0
,
4
0
5
,
4
1
4
,
5
8
9
,
6
2
8
,
4
5
6
]
[
5
5
9
,
5
1
6
,
3
7
4
,
3
3
5
,
6
6
4
,
4
4
8
,
4
1
5
]
P
e
r
f
u
me
[
7
3
8
,
5
2
9
,
3
9
4
,
3
9
5
,
5
6
6
,
5
7
7
,
4
4
2
]
[
7
9
4
,
5
1
8
,
4
9
4
,
4
4
7
,
6
8
6
,
6
5
8
,
4
9
0
]
S
mo
k
e
[
6
8
2
,
4
2
8
,
2
9
9
,
3
3
3
,
5
9
2
,
5
9
6
,
3
3
5
]
[
6
8
6
,
4
2
9
,
2
9
9
,
3
3
3
,
5
9
1
,
5
9
8
,
3
3
5
]
M
i
x
t
u
r
e
[
6
3
2
,
4
4
3
,
4
4
4
,
4
0
5
,
4
0
1
,
3
0
9
,
4
3
0
]
[
5
0
6
,
3
9
2
,
3
4
4
,
3
1
1
,
3
9
5
,
2
2
2
,
3
0
2
]
5.
T
H
E
P
RO
P
O
SE
D
T
H
E
RM
AL
-
G
AS F
US
I
O
N
DE
T
E
CT
I
O
N
M
O
D
E
L
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
8
1
3
-
4
8
2
8
4818
T
h
e
th
er
m
al
-
g
as
f
u
s
io
n
d
etec
ti
o
n
m
o
d
el
co
m
b
in
es
g
as
s
en
s
o
r
s
an
d
a
t
h
er
m
al
ca
m
e
r
a
f
o
r
ac
c
u
r
ate
g
as
d
etec
tio
n
,
with
a
b
lo
c
k
d
ia
g
r
a
m
as
s
h
o
wn
i
n
Fig
u
r
e
3
an
d
th
r
ee
s
ce
n
ar
io
s
d
etailin
g
th
e
d
ata
co
llectio
n
a
n
d
tr
ain
in
g
p
r
o
ce
s
s
.
T
h
e
f
o
llo
win
g
s
u
b
s
ec
tio
n
s
o
u
tlin
e
th
e
s
y
s
tem
ar
ch
itectu
r
e,
d
ata
c
o
llectio
n
s
ce
n
ar
io
s
,
an
d
th
e
tr
ain
in
g
m
eth
o
d
o
lo
g
y
u
s
ed
in
th
e
p
r
o
p
o
s
ed
m
o
d
el.
T
h
is
in
teg
r
ated
ap
p
r
o
ac
h
is
d
esig
n
ed
t
o
im
p
r
o
v
e
d
etec
tio
n
ac
cu
r
ac
y
an
d
r
o
b
u
s
tn
ess
ac
r
o
s
s
v
ar
y
in
g
e
n
v
ir
o
n
m
en
tal
co
n
d
i
tio
n
s
.
Fig
u
r
e
3
.
T
h
e
s
ch
em
atic
d
iag
r
am
r
ep
r
esen
ts
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
5
.
1
.
G
a
s
s
ens
o
rs o
f
s
ce
na
rio
I
5
.
1
.
1.
P
re
pro
ce
s
s
ing
ph
a
s
e
f
o
r
g
a
s
s
ens
o
rs
I
n
th
e
p
r
ep
r
o
ce
s
s
in
g
p
h
ase
o
f
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
,
s
ev
er
al
cr
u
cial
d
ata
p
r
ep
a
r
atio
n
s
tep
s
ar
e
p
er
f
o
r
m
ed
to
ef
f
ec
tiv
ely
p
r
e
p
ar
e
th
e
d
ataset
f
o
r
tr
ain
in
g
th
e
m
ac
h
in
e
lear
n
in
g
m
o
d
el.
I
n
i
tially
,
th
e
d
ataset
i
s
d
iv
id
ed
in
t
o
f
ea
tu
r
e
v
ec
to
r
s
an
d
th
eir
c
o
r
r
esp
o
n
d
in
g
tar
g
e
ts
.
Su
b
s
eq
u
en
tly
,
f
ea
t
u
r
e
s
elec
tio
n
is
p
er
f
o
r
m
e
d
u
s
in
g
th
e
SelectKBe
s
t a
lg
o
r
ith
m
with
th
e
_
s
co
r
in
g
f
u
n
ctio
n
,
wh
ich
s
elec
ts
th
e
to
p
6
m
o
s
t
s
i
g
n
if
ican
t
f
ea
tu
r
es
f
r
o
m
th
e
o
r
ig
in
al
f
ea
t
u
r
e
s
et
b
ased
o
n
th
eir
r
elev
an
ce
to
th
e
tar
g
et
v
ar
iab
le.
Af
ter
th
at,
lab
el
en
co
d
er
is
em
p
lo
y
ed
to
co
n
v
er
t
th
e
c
ateg
o
r
ical
lab
els
o
f
th
e
tar
g
et
v
ar
iab
le
in
to
n
u
m
er
ical
r
ep
r
esen
tatio
n
s
.
Fu
r
th
er
en
h
an
cin
g
th
e
d
ata,
th
e
en
c
o
d
ed
tar
g
et
v
ar
iab
le
is
tr
an
s
f
o
r
m
ed
in
to
b
in
ar
y
v
ec
to
r
s
u
s
in
g
o
n
e
-
h
o
t
en
c
o
d
in
g
,
r
esu
ltin
g
in
a
b
in
ar
y
m
atr
i
x
r
e
p
r
esen
tin
g
t
h
e
class
es.
T
o
en
s
u
r
e
c
o
n
s
is
ten
cy
an
d
u
n
if
o
r
m
it
y
in
f
ea
tu
r
e
s
ca
les,
th
e
n
u
m
er
ical
c
o
lu
m
n
s
o
f
th
e
f
ea
tu
r
e
m
atr
ix
ar
e
s
ca
led
u
s
in
g
Min
Ma
x
Scaler
,
r
escalin
g
th
eir
v
alu
es
b
etwe
en
0
an
d
1
.
Fin
ally
,
th
e
p
r
e
p
r
o
ce
s
s
ed
d
ata
is
s
p
li
t
in
to
tr
ain
in
g
an
d
test
in
g
s
ets,
wi
th
2
0
%
r
eser
v
ed
f
o
r
test
in
g
,
wh
ile
m
ain
tain
in
g
r
ep
r
o
d
u
cib
ilit
y
b
y
s
ettin
g
a
r
an
d
o
m
s
ta
te
o
f
0
.
T
h
ese
m
eticu
lo
u
s
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Op
timiz
in
g
in
tern
et
o
f th
in
g
s
b
a
s
ed
g
a
s
s
en
s
o
r
s
:
d
ee
p
le
a
r
n
in
g
a
n
d
…
(
Ma
r
ia
m
M.
A
b
d
ell
a
tif
)
4819
o
p
tim
ally
eq
u
ip
th
e
d
ata
f
o
r
tr
ain
in
g
an
d
ev
alu
atin
g
th
e
d
ee
p
lear
n
in
g
m
o
d
el
in
t
h
e
s
u
b
s
e
q
u
en
t
s
tag
es
o
f
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
.
5.
1.
2
.
L
ea
rning
ph
a
s
e
I
n
th
is
p
h
ase,
g
iv
en
th
e
s
eq
u
e
n
tial
n
atu
r
e
o
f
s
en
s
o
r
m
ea
s
u
r
e
m
en
ts
,
a
s
u
itab
le
ch
o
ice
f
o
r
a
s
eq
u
en
ce
m
o
d
el
is
th
e
d
ee
p
n
eu
r
al
n
et
wo
r
k
(
DNN)
.
I
n
th
e
lear
n
in
g
p
h
ase
o
f
th
e
p
r
o
p
o
s
ed
a
p
p
r
o
a
ch
,
th
e
DNN
m
o
d
el
co
n
s
is
ts
o
f
m
u
ltip
le
lay
er
s
:
in
p
u
t,
h
id
d
en
,
an
d
o
u
t
p
u
t.
T
h
e
in
p
u
t
lay
er
is
d
ef
in
e
d
u
s
in
g
t
h
e
'
Den
s
e
'
f
u
n
ctio
n
with
1
,
5
0
0
n
eu
r
o
n
s
an
d
a
R
eL
U
ac
tiv
atio
n
f
u
n
ctio
n
,
en
s
u
r
in
g
n
o
n
-
lin
ea
r
ity
an
d
ef
f
ec
tiv
e
f
ea
tu
r
e
r
ep
r
esen
tatio
n
.
T
h
e
in
p
u
t
s
h
ap
e
o
f
(
6
)
in
d
icate
s
th
at
th
e
in
p
u
t
d
ata
h
as
s
ix
f
ea
tu
r
es.
Su
b
s
eq
u
en
tly
,
two
h
id
d
e
n
lay
er
s
ar
e
ad
d
ed
with
1
,
5
0
0
an
d
1
,
0
0
0
n
e
u
r
o
n
s
,
r
esp
ec
tiv
ely
,
em
p
lo
y
in
g
R
eL
U
ac
tiv
atio
n
f
u
n
ctio
n
s
to
en
h
an
ce
f
ea
tu
r
e
ex
tr
ac
tio
n
f
u
r
th
er
an
d
in
t
r
o
d
u
ce
n
o
n
-
lin
ea
r
ity
.
B
atch
No
r
m
aliza
tio
n
la
y
er
s
ar
e
i
n
teg
r
ated
af
ter
ea
ch
h
id
d
e
n
lay
er
to
n
o
r
m
alize
th
e
o
u
tp
u
t,
a
v
o
id
in
g
o
v
er
f
itti
n
g
an
d
i
m
p
r
o
v
in
g
co
n
v
er
g
e
n
ce
.
Dr
o
p
o
u
t
lay
er
s
ar
e
in
co
r
p
o
r
ated
af
ter
t
h
e
f
ir
s
t
two
h
id
d
en
lay
er
s
to
m
itig
ate
o
v
er
f
itti
n
g
d
u
r
in
g
tr
a
in
in
g
,
with
d
r
o
p
o
u
t
r
ates
o
f
0
.
7
a
n
d
0
.
1
,
r
esp
e
ctiv
ely
,
r
an
d
o
m
ly
d
ea
cti
v
atin
g
n
eu
r
o
n
s
d
u
r
in
g
t
r
ain
in
g
.
T
h
is
r
eg
u
lar
izatio
n
tech
n
iq
u
e
im
p
r
o
v
es
th
e
g
e
n
e
r
aliza
tio
n
ca
p
ab
ilit
ies
o
f
th
e
m
o
d
el
a
n
d
p
r
ev
en
ts
it
f
r
o
m
b
ec
o
m
in
g
o
v
er
ly
d
ep
en
d
e
n
t
o
n
s
p
ec
if
ic
n
e
u
r
o
n
s
.
Af
ter
th
e
last
h
id
d
en
lay
er
,
th
e
o
u
tp
u
t
lay
er
is
ad
d
ed
with
f
o
u
r
n
eu
r
o
n
s
,
r
ep
r
esen
tin
g
th
e
f
o
u
r
class
es,
an
d
u
tili
ze
s
th
e
So
f
tMa
x
ac
tiv
atio
n
f
u
n
ctio
n
,
en
a
b
lin
g
p
r
o
b
a
b
ilis
tic
p
r
ed
ictio
n
s
.
T
h
e
m
o
d
el
is
th
en
co
m
p
iled
u
s
in
g
th
e
Ad
am
o
p
tim
izer
with
a
lear
n
in
g
r
ate
1
e
-
4
a
n
d
ca
teg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
as th
e
lo
s
s
f
u
n
ctio
n
to
o
p
tim
iz
e
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
.
5.
2
.
T
herm
a
l
im
a
g
es o
f
s
ce
na
rio
II
5.
2
.
1
.
Da
t
a
prepa
ra
t
io
n pha
s
e
f
o
r
t
herma
l ima
g
e
T
h
e
th
er
m
al
ca
m
e
r
a
is
a
n
o
n
-
in
v
asiv
e
in
s
tr
u
m
en
t
f
o
r
ass
ess
in
g
tem
p
er
atu
r
e
v
ar
iatio
n
s
b
y
d
etec
tin
g
in
f
r
ar
ed
lig
h
t.
I
ts
im
ag
e
s
en
s
o
r
'
s
p
ix
els
s
er
v
e
as
in
f
r
ar
ed
tem
p
er
atu
r
e
s
en
s
o
r
s
,
s
im
u
ltan
eo
u
s
ly
r
ec
o
r
d
in
g
tem
p
er
atu
r
es
in
s
ev
er
al
p
lace
s
.
I
m
ag
es
r
ep
r
esen
tin
g
th
e
o
u
tc
o
m
e
ar
e
s
h
o
wn
in
R
GB
f
o
r
m
at,
co
r
r
elatin
g
with
tem
p
er
atu
r
e
d
ata.
T
h
e
t
h
er
m
a
l
ca
m
er
a
h
as
a
s
u
b
s
tan
tial
ad
v
an
tag
e
o
v
er
tr
a
d
itio
n
al
ca
m
er
as
in
th
at
it
ca
n
o
p
er
ate
ef
f
icien
tly
in
v
ar
io
u
s
s
itu
atio
n
s
,
r
eg
ar
d
less
o
f
th
ei
r
s
h
ap
e
o
r
r
o
u
g
h
n
ess
.
T
h
e
c
h
o
ice
o
f
th
e
Seek
th
er
m
al
ca
m
er
a
f
o
r
th
is
s
tu
d
y
was
b
ased
o
n
its
co
m
p
ac
t
s
ize,
a
th
er
m
al
s
en
s
o
r
r
eso
lu
tio
n
o
f
2
0
6
×
1
5
6
p
ix
els,
a
wid
e
3
6
-
d
eg
r
ee
f
ield
o
f
v
iew,
tem
p
er
atu
r
e
m
ea
s
u
r
em
e
n
t
ca
p
ab
ilit
ies
r
an
g
in
g
f
r
o
m
-
4
0
°C
to
3
3
0
°C
,
a
f
r
am
er
ate
o
f
b
elo
w
9
Hz,
an
d
a
h
ig
h
th
er
m
al
p
ix
el
co
u
n
t
o
f
3
2
,
1
3
6
,
wh
ich
allo
ws
f
o
r
s
m
o
o
th
v
is
u
aliza
tio
n
o
f
th
er
m
al
im
ag
es.
A
v
alid
atio
n
s
et
was
cr
ea
ted
to
en
s
u
r
e
p
r
ec
is
e
m
o
n
ito
r
in
g
o
f
th
e
m
o
d
els'
p
er
f
o
r
m
an
ce
d
u
r
in
g
th
e
tr
ain
in
g
p
h
ase
b
y
r
an
d
o
m
l
y
ass
ig
n
in
g
2
0
%
o
f
th
e
s
am
p
l
es
f
r
o
m
th
e
o
r
ig
i
n
al
tr
ain
in
g
s
et.
T
h
e
d
ata
s
et
was
s
p
lit
in
to
th
r
ee
s
u
b
s
ets:
th
e
tr
a
in
in
g
,
v
alid
atio
n
,
an
d
test
s
ets.
T
ab
le
3
p
r
esen
ts
th
e
n
u
m
b
er
o
f
s
am
p
les
in
ea
ch
ca
teg
o
r
y
,
alo
n
g
with
th
e
to
tal
co
u
n
t
o
f
s
am
p
les
in
b
o
th
th
e
tr
ain
in
g
an
d
v
alid
atio
n
s
ets
af
ter
th
e
s
p
lit.
T
h
is
d
iv
is
io
n
f
ac
ilit
ates
p
r
ac
tical
tr
ain
in
g
a
n
d
e
v
alu
atio
n
o
f
th
e
m
o
d
els,
en
a
b
lin
g
th
em
t
o
lear
n
f
r
o
m
d
iv
er
s
e
d
ata
wh
ile
allo
win
g
f
o
r
u
n
b
iased
ass
e
s
s
m
en
t
o
f
th
eir
ac
cu
r
ac
y
an
d
g
en
er
aliza
tio
n
ca
p
a
b
ilit
ies.
Var
io
u
s
d
ata
au
g
m
en
tatio
n
m
eth
o
d
s
wer
e
ap
p
lied
to
th
e
tr
ain
in
g
s
am
p
les
to
p
r
ev
en
t
o
v
er
f
itti
n
g
an
d
im
p
r
o
v
e
g
en
er
aliza
tio
n
.
T
h
ese
tech
n
iq
u
es
in
clu
d
e
r
o
tatio
n
,
wid
th
a
d
ju
s
tm
en
t,
h
eig
h
t
a
d
ju
s
tm
en
t,
s
h
ea
r
,
zo
o
m
,
an
d
h
o
r
izo
n
tal
f
lip
p
in
g
.
Su
b
s
eq
u
e
n
tly
,
b
o
th
th
e
v
alid
atio
n
an
d
test
d
atasets
wer
e
r
esized
to
a
r
eso
lu
tio
n
o
f
2
5
6
x
2
5
6
a
n
d
n
o
r
m
alize
d
b
y
d
i
v
id
i
n
g
th
e
co
lo
r
v
alu
e
o
f
ea
ch
p
ix
el
b
y
2
5
5
,
en
s
u
r
i
n
g
th
at
all
p
ix
el
v
alu
es
f
all
with
in
th
e
r
an
g
e
o
f
[
0
,
1
]
.
Du
r
in
g
th
e
tr
ain
in
g
p
h
ase,
th
e
au
g
m
en
tat
io
n
s
ettin
g
s
in
v
o
lv
e
r
escalin
g
th
e
p
ix
el
v
alu
es
to
th
e
d
esire
d
r
an
g
e
o
f
0
to
1
,
r
o
tatin
g
im
ag
es
b
y
u
p
to
3
0
d
eg
r
ee
s
,
s
h
if
tin
g
th
e
h
eig
h
t
a
n
d
wid
th
b
y
2
0
%,
ap
p
ly
in
g
s
h
ea
r
t
r
an
s
f
o
r
m
ati
o
n
s
,
h
o
r
iz
o
n
tal
f
lip
p
in
g
,
an
d
zo
o
m
in
g
th
e
im
ag
es
b
y
3
0
%.
T
h
e
′
_
′
p
ar
am
eter
is
also
s
et
to
'
n
ea
r
est
'
to
h
a
n
d
le
a
n
y
p
o
ten
tial
m
is
s
in
g
p
ix
els
a
f
ter
th
e
tr
a
n
s
f
o
r
m
atio
n
s
.
I
n
co
n
tr
ast,
th
e
test
d
ata
u
n
d
er
we
n
t r
escalin
g
f
r
o
m
0
t
o
1
.
T
ab
le
3
.
T
h
e
co
u
n
ts
o
f
s
am
p
le
s
f
o
r
ea
ch
ca
teg
o
r
y
an
d
t
h
e
to
t
al
co
m
b
in
ed
s
am
p
les in
b
o
th
th
e
tr
ain
in
g
a
n
d
v
alid
atio
n
s
ets ar
e
p
r
o
v
id
ed
C
r
o
p
c
a
t
e
g
o
r
y
Tr
a
i
n
i
n
g
se
t
i
ma
g
e
s
V
a
l
i
d
a
t
i
o
n
se
t
i
ma
g
e
s
N
o
G
a
s
1
,
2
8
0
1
6
0
M
i
x
t
u
r
e
1
,
2
8
0
1
6
0
P
e
r
f
u
me
1
,
2
8
0
1
6
0
S
mo
k
e
1
,
2
8
0
1
6
0
To
t
a
l
5
,
1
2
0
6
4
0
5
.
2
.
2
.
O
ptim
iza
t
io
n
ph
a
s
e
f
o
r
t
herm
a
l ima
g
e
T
h
e
u
s
e
o
f
p
r
e
-
tr
ain
e
d
C
NN
ar
ch
itectu
r
es
p
lace
s
s
ev
er
al
r
estrictio
n
s
o
n
th
e
o
p
tim
izatio
n
p
h
ase.
T
h
e
m
ajo
r
ity
o
f
th
ese
ar
ch
itectu
r
e
s
'
h
y
p
er
p
ar
am
eter
s
ar
e
f
ix
ed
.
Sti
ll,
a
f
ew
im
p
o
r
tan
t
o
n
es
n
ee
d
to
b
e
ad
ju
s
ted
,
in
clu
d
in
g
th
e
_
ℎ
s
ize,
lear
n
in
g
r
a
te,
q
u
an
tity
o
f
u
n
its
in
d
en
s
e
lay
er
s
,
an
d
d
r
o
p
o
u
t r
ate.
T
o
o
v
er
co
m
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
8
1
3
-
4
8
2
8
4820
th
is
,
th
e
s
tu
d
y
u
s
es
th
e
o
p
ti
m
ized
lin
ea
r
p
ar
ticle
s
war
m
o
p
tim
izatio
n
(
OL
PS
O
)
alg
o
r
ith
m
to
tu
n
e
th
e
h
y
p
er
p
ar
am
eter
.
Acc
o
r
d
i
n
g
to
T
ab
le
4
,
th
e
OL
PS
O
alg
o
r
it
h
m
u
s
es
two
co
o
r
d
in
ates
as
p
ar
ticles
to
in
d
icate
th
e
b
atch
s
ize
an
d
lear
n
in
g
r
ate
o
f
th
e
Ad
am
o
p
tim
izer
.
R
an
d
o
m
in
itializatio
n
cr
ea
tes
th
e
in
i
tial
s
war
m
p
ar
ticles.
I
n
p
ar
ticu
lar
,
th
e
i
n
itializatio
n
u
s
es
in
teg
er
v
alu
es
f
r
o
m
2
5
to
1
2
8
an
d
d
ec
im
al
v
alu
es
f
r
o
m
0
.
0
to
1
.
0
.
T
h
e
OL
PS
O
alg
o
r
ith
m
f
in
d
s
th
e
b
est
v
alu
es
f
o
r
th
ese
h
y
p
er
p
a
r
a
m
eter
s
th
r
o
u
g
h
iter
ativ
e
o
p
tim
izatio
n
,
u
ltima
tely
im
p
r
o
v
in
g
th
e
p
er
f
o
r
m
an
ce
a
n
d
ac
cu
r
ac
y
o
f
t
h
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
.
T
ab
le
4
d
etails
th
e
u
s
e
o
f
th
e
OL
PS
O
alg
o
r
ith
m
f
o
r
tu
n
i
n
g
h
y
p
er
p
ar
am
eter
s
in
a
th
er
m
al
im
ag
e
an
aly
s
is
m
o
d
el
u
s
in
g
p
r
e
-
tr
ain
ed
C
NN
ar
ch
itectu
r
es.
Sp
ec
if
i
ca
lly
,
it
s
h
o
ws
th
at
OL
PS
O
u
s
es
two
co
o
r
d
in
ates
as
p
ar
ticles:
o
n
e
f
o
r
b
atch
s
iz
e
(
P1
)
r
an
g
in
g
f
r
o
m
2
5
to
1
2
8
,
an
d
an
o
th
er
f
o
r
lear
n
in
g
r
ate
(
P2
)
r
an
g
in
g
f
r
o
m
0
.
0
to
1
.
0
.
T
h
r
o
u
g
h
iter
ativ
e
o
p
tim
izatio
n
,
OL
PS
O
id
en
ti
f
ies
th
e
b
est
v
alu
es
f
o
r
th
ese
h
y
p
er
p
a
r
am
eter
s
,
en
h
an
cin
g
th
e
p
e
r
f
o
r
m
an
ce
an
d
ac
cu
r
ac
y
o
f
t
h
e
m
o
d
el.
T
ab
le
4
.
Par
ticle
s
p
ec
if
icatio
n
f
o
r
h
y
p
er
p
ar
am
eter
o
p
tim
izatio
n
in
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
o
lo
g
y
P
a
r
t
i
c
l
e
(
p
)
B
a
t
c
h
si
z
e
Le
a
r
n
i
n
g
r
a
t
e
P
1
2
5
t
o
1
2
8
N
o
t
A
p
p
l
i
c
a
b
l
e
P
2
N
o
t
A
p
p
l
i
c
a
b
l
e
0
.
0
t
o
1
.
0
5
.
2
.
3
.
L
ea
rning
ph
a
s
e
T
h
i
s
p
h
a
s
e
p
r
o
p
o
s
ed
t
h
e
c
l
as
s
i
f
i
c
a
t
i
o
n
o
f
g
a
s
s
a
m
p
l
e
s
i
n
t
h
i
s
i
n
v
e
s
t
i
g
a
t
io
n
u
s
i
n
g
th
e
o
p
t
i
m
iz
e
d
V
G
G
1
6
m
o
d
e
l.
A
d
e
e
p
l
e
ar
n
i
n
g
m
o
d
e
l
i
s
co
n
s
t
r
u
c
t
ed
to
p
e
r
f
o
r
m
i
m
ag
e
c
l
a
s
s
i
f
i
c
a
t
io
n
u
s
i
n
g
t
r
an
s
f
er
l
e
a
r
n
in
g
.
T
h
e
m
o
d
e
l
i
s
b
a
s
ed
o
n
th
e
V
G
G1
6
a
r
ch
i
t
e
c
tu
r
e,
a
p
r
e
-
tr
a
i
n
ed
C
N
N
m
e
n
t
io
n
ed
i
n
s
e
c
t
i
o
n
3
.
3
.
2
.
T
h
e
m
o
d
e
l
a
r
ch
i
t
e
c
tu
r
e
co
n
s
is
t
s
o
f
a
p
r
e
-
t
r
a
i
n
ed
V
G
G
1
6
m
o
d
e
l
a
s
t
h
e
b
a
s
e
,
f
o
l
l
o
w
ed
b
y
ad
d
i
t
io
n
a
l
l
ay
er
s
.
T
h
e
V
G
G1
6
b
a
s
e
i
s
f
r
o
z
e
n
(
n
o
n
-
tr
a
in
a
b
l
e)
t
o
r
e
t
a
in
i
t
s
p
r
e
-
t
r
a
i
n
ed
w
e
ig
h
t
s
an
d
f
e
a
tu
r
e
s
.
A
d
e
n
s
e
la
y
er
w
i
t
h
2
5
6
u
n
i
t
s
an
d
R
e
L
U
a
c
t
iv
a
t
io
n
i
s
a
d
d
e
d
,
f
o
l
lo
w
e
d
b
y
d
r
o
p
o
u
t
an
d
b
a
t
ch
n
o
r
m
a
l
i
za
t
i
o
n
l
ay
e
r
s
to
p
r
ev
e
n
t
o
v
er
f
i
t
t
in
g
.
T
h
e
o
u
t
p
u
t
l
ay
er
co
m
p
r
i
s
e
s
f
o
u
r
u
n
i
t
s
w
i
th
a
So
f
t
M
ax
a
c
t
i
v
a
t
i
o
n
f
u
n
c
ti
o
n
f
o
r
m
u
l
ti
c
l
a
s
s
c
l
a
s
s
i
f
i
c
a
t
i
o
n
.
T
h
e
m
o
d
e
l
i
s
o
p
t
i
m
i
z
ed
u
s
i
n
g
t
h
e
Ad
a
m
o
p
ti
m
i
z
e
r
w
i
t
h
a
l
e
a
r
n
i
n
g
r
a
t
e
2
e
-
5
a
n
d
c
a
t
eg
o
r
i
c
a
l
c
r
o
s
s
-
en
t
r
o
p
y
a
s
t
h
e
l
o
s
s
f
u
n
c
t
i
o
n
.
T
h
e
m
o
d
e
l
c
h
e
c
k
p
o
in
t
a
n
d
e
a
r
ly
s
t
o
p
p
in
g
c
a
l
lb
a
ck
s
a
r
e
e
m
p
l
o
y
e
d
to
i
m
p
r
o
v
e
m
o
d
e
l
p
e
r
f
o
r
m
a
n
c
e
a
n
d
p
r
ev
e
n
t
o
v
er
f
i
t
t
i
n
g
.
D
u
r
i
n
g
t
h
e
tr
a
i
n
in
g
p
h
a
s
e
,
t
h
e
m
o
d
e
l
w
a
s
tr
a
in
e
d
u
s
i
n
g
t
h
e
tr
a
i
n
in
g
s
e
t
a
n
d
v
a
li
d
a
t
e
d
u
s
i
n
g
th
e
v
a
l
id
a
t
i
o
n
s
e
t
.
B
a
t
ch
e
s
o
f
au
g
m
e
n
t
ed
im
a
g
e
s
w
er
e
u
s
ed
f
o
r
b
o
t
h
t
r
a
in
i
n
g
an
d
v
a
l
id
a
t
i
o
n
.
T
h
e
m
o
d
e
l
u
n
d
e
r
w
en
t
tr
a
i
n
in
g
f
o
r
1
0
0
e
p
o
c
h
s
,
a
n
d
th
e
b
e
s
t
m
o
d
e
l
we
i
g
h
t
s
w
e
r
e
s
a
v
e
d
f
o
r
f
u
tu
r
e
u
s
e
an
d
e
v
a
lu
a
t
i
o
n
af
t
e
r
t
h
e
s
e
p
r
e
-
tr
ai
n
e
d
m
o
d
e
l
s
h
a
d
f
in
i
s
h
e
d
t
h
e
ir
t
r
a
in
i
n
g
.
5
.
2
.
4
.
E
v
a
lua
t
io
n
ph
a
s
e
f
o
r
g
a
s
s
ens
o
rs a
nd
t
herm
a
l im
a
g
es
Du
r
in
g
th
is
p
h
ase,
th
e
p
r
ed
ic
tiv
e
ca
p
ab
ilit
ies
o
f
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
wer
e
ass
ess
ed
u
s
in
g
f
iv
e
co
m
m
o
n
l
y
em
p
lo
y
e
d
ev
alu
atio
n
m
etr
ics in
class
if
icatio
n
p
r
o
b
lem
s
: th
e
co
n
f
u
s
io
n
m
atr
ix
,
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
F1
-
s
co
r
e,
a
n
d
r
ec
all.
Acc
u
r
a
cy
q
u
an
tifie
s
th
e
r
atio
o
f
co
r
r
ec
t
p
r
ed
ictio
n
s
t
o
all
p
r
e
d
ictio
n
s
m
ad
e
an
d
is
ty
p
ically
ex
p
r
ess
ed
as
a
p
er
c
en
tag
e,
ca
lc
u
lated
u
s
in
g
(
3
)
.
Pre
cisi
o
n
ass
ess
e
s
th
e
m
o
d
el's
ab
ilit
y
to
p
r
e
d
ict
v
alu
es f
o
r
s
p
ec
if
ic
ca
teg
o
r
ies
ac
cu
r
ately
; its
ca
lcu
latio
n
is
d
etailed
in
(
4
)
.
R
ec
all
[
3
1
]
g
a
u
g
es th
e
p
r
o
p
o
r
tio
n
o
f
co
r
r
ec
tly
class
if
ied
p
o
s
itiv
e
p
atter
n
s
an
d
(
5
)
o
u
tlin
es
its
d
er
iv
atio
n
.
T
h
e
F1
-
s
co
r
e
r
ep
r
e
s
en
ts
th
e
weig
h
te
d
av
er
ag
e
o
f
p
r
ec
is
io
n
an
d
r
ec
all,
as
co
m
p
u
ted
i
n
(
6
)
.
B
o
th
m
ac
r
o
a
n
d
m
icr
o
av
e
r
ag
es
wer
e
em
p
lo
y
e
d
to
ev
alu
ate
p
er
f
o
r
m
an
ce
,
ex
clu
d
in
g
th
e
co
n
f
u
s
io
n
m
atr
ix
co
m
p
r
eh
en
s
iv
ely
.
T
h
e
co
n
f
u
s
io
n
m
atr
ix
,
a
wid
ely
u
s
e
d
tab
u
lar
r
e
p
r
esen
tatio
n
,
illu
s
tr
ates
th
e
class
if
icatio
n
m
o
d
e
l's
p
er
f
o
r
m
an
ce
o
n
th
e
test
s
et.
I
t
e
n
ab
les
a
co
m
p
ar
is
o
n
b
etwe
en
p
r
ed
icte
d
an
d
ac
tu
al
o
u
tco
m
es,
o
f
f
e
r
i
n
g
v
alu
ab
le
in
s
ig
h
ts
in
to
th
e
m
o
d
el'
s
ac
cu
r
ac
y
in
co
r
r
ec
tly
id
e
n
tify
in
g
v
al
u
es.
B
y
alig
n
in
g
p
r
ed
icted
a
n
d
ac
tu
al
o
u
tco
m
es,
th
e
co
n
f
u
s
io
n
m
atr
ix
ai
d
s
in
ass
es
s
in
g
th
e
o
v
er
all
p
er
f
o
r
m
a
n
ce
o
f
th
e
m
o
d
el
an
d
id
en
tify
i
n
g
an
y
m
is
class
if
icatio
n
s
o
r
er
r
o
r
s
th
at
m
ay
h
a
v
e
o
cc
u
r
r
e
d
d
u
r
in
g
th
e
class
if
icat
io
n
p
r
o
ce
s
s
.
=
T
otal
pr
e
dicti
on
(
3
)
P
r
ec
is
io
n
=
C
or
r
e
c
t
P
r
e
diction
s
for
a
P
ar
ti
c
ular
C
ategor
y
T
otal
P
r
e
dict
ion
s
for
that
C
ategor
y
(
4
)
R
ec
a
ll =
C
or
r
e
c
tl
y
P
r
e
dicte
d
I
ns
tance
s
of
a
C
ategor
y
T
otal
I
ns
ta
nc
e
s
of
that
C
ategor
y
(
5
)
F
-
Mea
s
u
r
e
=
2×P
r
e
c
is
ion×
R
e
c
all
P
r
e
c
is
ion+
R
e
c
all
(
6
)
5
.
3
.
F
us
io
n
ph
a
s
e
in
s
ce
na
rio
I
I
I
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Op
timiz
in
g
in
tern
et
o
f th
in
g
s
b
a
s
ed
g
a
s
s
en
s
o
r
s
:
d
ee
p
le
a
r
n
in
g
a
n
d
…
(
Ma
r
ia
m
M.
A
b
d
ell
a
tif
)
4821
I
n
th
is
p
h
ase,
th
e
p
r
im
ar
y
o
b
j
ec
tiv
e
is
to
m
ak
e
p
r
ec
is
e
d
ec
is
io
n
s
b
y
am
alg
am
atin
g
f
ea
tu
r
es
d
er
iv
ed
f
r
o
m
th
er
m
al
im
ag
es a
n
d
g
as sen
s
o
r
m
ea
s
u
r
em
en
ts
.
T
h
e
d
esig
n
s
o
f
th
e
m
o
d
els f
o
r
f
u
s
in
g
i
m
ag
e
an
d
s
eq
u
en
ce
d
ata
ar
e
d
ep
icted
in
Fig
u
r
e
1
.
T
h
e
ce
n
tr
al
aim
is
co
n
s
tr
u
ctin
g
a
u
n
if
ied
class
if
ier
th
at
ad
ep
tly
m
er
g
es
in
f
o
r
m
atio
n
f
r
o
m
th
er
m
al
im
a
g
es
an
d
th
e
g
as
s
en
s
o
r
s
eq
u
e
n
ce
ar
r
ay
.
T
h
e
o
u
tp
u
ts
o
f
th
e
DNN
an
d
VGG1
6
m
o
d
els
m
u
s
t
b
e
in
th
e
s
am
e
f
ea
tu
r
e
s
p
ac
e
f
o
r
th
e
f
u
s
io
n
to
b
e
ef
f
ec
tiv
e.
A
L
ate
Fu
s
i
o
n
m
o
d
el
th
at
u
s
es
d
ec
is
io
n
-
lev
el
f
u
s
io
n
is
also
u
s
ed
to
ac
co
m
p
lis
h
th
is
.
T
h
e
in
d
iv
id
u
al
p
r
ed
ictio
n
s
o
f
th
e
DNN
an
d
VGG1
6
m
o
d
els
ar
e
o
b
tain
ed
f
ir
s
t.
T
h
e
f
in
al
r
esu
lt
o
f
f
u
s
io
n
,
k
n
o
wn
as
av
e
r
ag
e
f
u
s
io
n
,
is
th
en
co
n
s
id
er
ed
th
e
ar
ith
m
etic
av
er
ag
e
o
f
ea
ch
m
o
d
el
p
r
e
d
ictio
n
d
u
r
in
g
th
e
lat
e
f
u
s
io
n
p
r
o
ce
s
s
.
T
h
e
av
ailab
l
e
d
ataset
is
u
s
ed
to
d
ev
elo
p
an
d
v
alid
ate
th
e
late
f
u
s
io
n
m
o
d
el.
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
f
u
s
io
n
m
o
d
els
is
p
r
esen
ted
in
th
e
f
o
llo
win
g
s
ec
tio
n
.
T
h
e
f
u
s
io
n
p
r
o
ce
s
s
aim
s
to
m
ak
e
th
e
m
o
s
t
o
f
th
e
ad
v
an
tag
es
o
f
b
o
th
th
er
m
al
p
ictu
r
es
an
d
g
as sen
s
o
r
d
ata,
en
h
a
n
cin
g
class
if
icatio
n
p
r
ec
is
io
n
f
o
r
g
as sam
p
le
an
aly
s
is
.
6.
RE
SU
L
T
AND
DI
SCUS
SI
O
N
T
h
is
s
ec
tio
n
p
r
esen
ts
an
d
an
a
ly
ze
s
th
e
o
u
tco
m
es
o
f
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
a
p
p
lied
to
t
h
r
ee
d
is
tin
ct
g
as
d
etec
tio
n
s
ce
n
ar
io
s
.
L
ev
er
ag
in
g
th
e
Ker
as
lib
r
ar
y
,
a
Py
t
h
o
n
-
b
ased
h
i
g
h
-
lev
el
API
r
en
o
wn
ed
f
o
r
b
u
ild
in
g
an
d
im
p
lem
e
n
tin
g
d
ee
p
lear
n
in
g
ar
ch
itectu
r
es.
T
h
e
f
r
am
ew
o
r
k
o
f
th
e
a
p
p
r
o
ac
h
is
s
ea
m
le
s
s
ly
co
n
s
tr
u
cted
an
d
in
teg
r
ated
with
p
o
wer
f
u
l
n
u
m
er
ical
co
m
p
u
tatio
n
al
lib
r
a
r
ies
lik
e
T
en
s
o
r
Flo
w.
A
DNN
is
tr
ain
ed
in
s
ce
n
ar
io
I
u
s
in
g
o
n
ly
th
e
s
eq
u
en
ce
g
as
s
en
s
o
r
m
o
d
ality
.
I
n
s
ce
n
ar
i
o
I
I
,
VGG1
6
is
u
tili
ze
d
t
o
p
r
o
ce
s
s
th
er
m
al
im
a
g
es
o
f
g
as
s
en
s
o
r
s
.
Fin
ally
,
late
f
u
s
io
n
is
em
p
l
o
y
ed
in
s
ce
n
a
r
io
I
I
I
,
b
y
co
m
b
i
n
in
g
f
ea
tu
r
es
f
r
o
m
th
e
DNN
an
d
VGG1
6
m
o
d
els
th
r
o
u
g
h
d
ec
is
io
n
-
lev
el
f
u
s
io
n
.
I
n
th
is
ap
p
r
o
ac
h
,
in
d
iv
id
u
al
p
r
e
d
ictio
n
s
ar
e
o
b
tain
ed
f
r
o
m
th
e
DNN
an
d
VGG1
6
m
o
d
els b
ef
o
r
e
p
er
f
o
r
m
i
n
g
th
e
late
f
u
s
io
n
to
ac
h
iev
e
g
as d
etec
tio
n
o
b
jectiv
es.
6
.
1
.
G
a
s
s
equence
s
re
s
ult
o
f
s
ce
n
a
rio
I
T
h
e
o
u
tco
m
es d
er
iv
e
d
f
r
o
m
th
is
s
ce
n
ar
io
ar
e
n
o
t ju
s
t
s
h
o
wca
s
ed
,
b
u
t m
eticu
lo
u
s
ly
ex
am
in
e
d
th
r
o
u
g
h
a
co
m
p
r
e
h
en
s
iv
e
an
al
y
s
is
co
n
d
u
cted
in
th
r
ee
d
is
tin
ct
an
d
c
r
u
cial
p
h
ases
.
T
h
e
s
u
b
s
eq
u
en
t
s
u
b
s
ec
tio
n
s
d
etail
th
e
th
r
ee
k
e
y
p
h
ases
o
f
an
aly
s
is
: p
r
ep
r
o
ce
s
s
in
g
,
lear
n
in
g
,
a
n
d
ev
alu
atio
n
o
f
th
e
DNN
m
o
d
el
.
6
.
1
.
1
.
P
re
pro
ce
s
s
ing
ph
a
s
e
I
n
th
e
p
a
p
er
'
s
f
ea
tu
r
e
s
elec
tio
n
an
d
s
ca
lin
g
s
ec
tio
n
,
th
e
au
t
h
o
r
s
u
tili
ze
d
th
e
s
cik
it
-
lear
n
lib
r
ar
y
to
p
er
f
o
r
m
th
ese
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
o
n
th
e
d
ataset.
First,
th
e
f
ea
tu
r
e
s
elec
tio
n
was
ap
p
lied
u
s
in
g
th
e
SelectKBe
s
t
m
eth
o
d
f
r
o
m
s
cik
it
-
lear
n
'
s
f
ea
tu
r
e
s
elec
tio
n
m
o
d
u
le.
T
h
is
m
eth
o
d
u
tili
ze
s
th
e
ANOV
A
F
-
v
alu
e
(
_
)
to
ev
alu
ate
th
e
im
p
o
r
tan
ce
o
f
f
ea
tu
r
es
an
d
s
elec
t
th
e
to
p
′
′
f
ea
tu
r
es.
I
n
th
is
ca
s
e,
′
′
was
s
et
to
6
,
in
d
icatin
g
th
at
th
e
alg
o
r
ith
m
will
s
elec
t
th
e
s
ix
m
o
s
t
r
elev
an
t
f
ea
tu
r
es
f
o
r
f
u
r
th
er
an
aly
s
is
.
T
h
e
s
elec
ted
f
ea
tu
r
es
wer
e
th
e
n
o
b
tai
n
ed
b
y
ca
llin
g
_
(
)
o
n
t
h
e
d
ata
m
atr
ix
′
′
an
d
th
e
co
r
r
esp
o
n
d
in
g
tar
g
et
′
′
.
Nex
t,
t
o
h
a
n
d
le
th
e
tar
g
et
v
a
r
iab
le
′
′
,
th
e
a
u
th
o
r
s
u
s
ed
L
a
b
e
l
E
n
co
d
e
r
f
r
o
m
s
cik
it
-
lear
n
'
s
p
r
ep
r
o
ce
s
s
in
g
m
o
d
u
le
to
co
n
v
er
t
ca
teg
o
r
ica
l
class
lab
els
in
to
n
u
m
er
ical
r
ep
r
esen
tatio
n
s
.
T
h
is
s
tep
is
ess
en
tial
f
o
r
ce
r
tain
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
th
at
r
e
q
u
ir
e
n
u
m
er
ical
in
p
u
ts
.
Af
ter
c
o
n
v
e
r
tin
g
class
l
ab
els
to
n
u
m
er
ical
r
ep
r
esen
tatio
n
s
,
th
e
a
u
th
o
r
s
a
p
p
lied
o
n
e
-
h
o
t
en
co
d
in
g
.
M
o
v
in
g
o
n
to
f
ea
tu
r
e
s
ca
lin
g
,
th
e
au
th
o
r
s
em
p
lo
y
ed
Min
Ma
x
Scaler
f
r
o
m
s
cik
it
-
lear
n
'
s
p
r
ep
r
o
ce
s
s
in
g
m
o
d
u
le.
Featu
r
e
s
ca
lin
g
is
ess
en
tial
to
en
s
u
r
e
th
at
all
f
ea
tu
r
es
ar
e
o
n
th
e
s
am
e
s
ca
le,
wh
ich
h
elp
s
p
r
ev
e
n
t
ce
r
tain
f
ea
tu
r
es
f
r
o
m
d
o
m
in
atin
g
th
e
lear
n
in
g
p
r
o
ce
s
s
d
u
r
in
g
m
o
d
el
tr
ai
n
in
g
.
I
n
th
is
co
d
e,
th
e
n
u
m
er
ical
f
ea
t
u
r
es in
′
′
wer
e
s
ca
led
b
etwe
en
0
a
n
d
1
u
s
in
g
Min
-
Ma
x
s
ca
lin
g
,
m
ak
in
g
t
h
e
d
ataset
s
u
itab
le
f
o
r
v
ar
io
u
s
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
.
6
.
1
.
2
.
L
ea
rning
ph
a
s
e
T
h
e
m
o
d
el
is
d
esig
n
e
d
as
a
Seq
u
en
tial
s
tack
o
f
lay
e
r
s
,
f
ea
t
u
r
in
g
d
en
s
e
lay
er
s
with
R
eL
U
ac
tiv
atio
n
f
o
r
h
id
d
en
lay
e
r
s
an
d
So
f
tM
ax
f
o
r
th
e
o
u
tp
u
t
lay
er
.
B
atch
n
o
r
m
aliza
tio
n
lay
er
s
ar
e
ad
d
ed
f
o
r
tr
ain
in
g
s
tab
ilit
y
,
an
d
d
r
o
p
o
u
t
la
y
er
s
a
r
e
u
tili
ze
d
f
o
r
r
eg
u
la
r
izatio
n
a
n
d
o
v
er
f
itti
n
g
p
r
e
v
en
tio
n
.
T
h
e
m
o
d
el
is
co
m
p
iled
with
ca
teg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
an
d
ac
cu
r
ac
y
m
etr
ics
f
o
r
m
u
lti
-
class
class
if
icat
io
n
.
Du
r
in
g
tr
ain
in
g
,
th
r
ee
ca
llb
ac
k
s
ar
e
em
p
lo
y
e
d
:
m
o
d
el
ch
ec
k
p
o
in
t
s
av
es
th
e
b
est
m
o
d
el
weig
h
ts
b
ased
o
n
v
a
lid
atio
n
lo
s
s
,
ea
r
ly
s
to
p
p
in
g
h
a
lts
tr
ain
in
g
if
v
alid
atio
n
lo
s
s
s
tag
n
ates,
an
d
r
ed
u
ce
L
R
On
Plateau
ad
ju
s
ts
th
e
lear
n
in
g
r
ate
wh
en
v
alid
atio
n
lo
s
s
p
latea
u
s
.
T
h
e
m
o
d
el
is
tr
ain
ed
f
o
r
3
0
0
ep
o
c
h
s
u
s
in
g
th
e
Ad
a
m
o
p
tim
izer
with
a
lear
n
in
g
r
ate
o
f
1
e
-
4
.
Af
ter
th
o
r
o
u
g
h
test
in
g
,
it wa
s
f
o
u
n
d
t
h
at
th
e
m
o
d
els h
ad
r
ea
ch
e
d
th
eir
o
p
tim
al
v
ali
d
atio
n
r
esu
lts
.
6
.
1
.
3
.
E
v
a
lua
t
io
n o
f
DNN
m
o
del
T
o
ass
ess
its
ef
f
ec
tiv
en
ess
,
th
e
DNN
m
o
d
el
was
ev
alu
ated
o
n
th
e
test
s
et
u
s
in
g
th
e
f
iv
e
p
er
f
o
r
m
a
n
ce
in
d
icato
r
s
.
A
d
etailed
co
m
p
ar
is
o
n
o
f
t
h
ese
r
esu
lts
is
p
r
o
v
id
ed
in
T
a
b
le
5
.
As
s
h
o
wn
,
th
e
m
o
d
el
u
tili
zin
g
th
e
Ad
am
o
p
tim
izer
ac
h
iev
ed
a
n
i
m
p
r
ess
iv
e
ac
cu
r
ac
y
o
f
9
5
%.
Fig
u
r
e
4
p
r
esen
ts
th
e
p
r
o
v
id
e
d
co
n
f
u
s
io
n
m
atr
ix
as
a
4
x
4
m
atr
ix
th
at
e
v
alu
ates
th
e
p
er
f
o
r
m
an
ce
o
f
a
class
if
ier
in
a
m
u
lti
-
class
clas
s
if
icatio
n
p
r
o
b
lem
with
f
o
u
r
cl
ass
es:
'
No
Ga
s
,
'
'Per
f
u
m
e,
'
'
Sm
o
k
e,
'
an
d
'
Mix
tu
r
e.
'
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
8
1
3
-
4
8
2
8
4822
T
h
e
d
iag
o
n
al
elem
en
ts
(
1
,
1
)
,
(
2
,
2
)
,
(
3
,
3
)
,
an
d
(
4
,
4
)
r
ep
r
esen
t
th
e
n
u
m
b
er
o
f
in
s
tan
ce
s
co
r
r
ec
tly
class
if
ied
f
o
r
ea
ch
r
esp
ec
tiv
e
class
.
Pre
cisely
,
1
8
5
in
s
tan
ce
s
o
f
'
No
Gas,'
1
3
3
in
s
tan
ce
s
o
f
'Per
f
u
m
e,
'
1
4
5
in
s
tan
ce
s
o
f
'
Sm
o
k
e,
'
an
d
1
4
8
in
s
tan
ce
s
o
f
'
Mix
tu
r
e'
wer
e
co
r
r
ec
tly
p
r
ed
ic
ted
.
T
h
e
o
f
f
-
d
ia
g
o
n
al
elem
en
ts
(
2
,
3
)
a
n
d
(
3
,
2
)
r
ep
r
esen
t
m
is
class
if
icatio
n
s
b
etwe
en
th
e
'
Per
f
u
m
e'
an
d
'
Sm
o
k
e
'
clas
s
es,
wh
er
e
1
0
in
s
tan
ce
s
o
f
'
Per
f
u
m
e'
wer
e
in
co
r
r
ec
tly
class
if
ied
as
'
Sm
o
k
e,
'
an
d
1
9
in
s
tan
ce
s
o
f
'
Sm
o
k
e
'
wer
e
in
co
r
r
ec
tly
clas
s
if
ied
as
'Per
f
u
m
e.
'
Ho
wev
er
,
n
o
m
is
class
if
icatio
n
s
wer
e
o
b
s
er
v
ed
b
etwe
en
th
e
o
th
er
class
es.
T
ab
le
5
.
T
h
e
p
er
f
o
r
m
a
n
ce
m
et
r
ics (
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
-
s
co
r
e)
f
o
r
th
e
o
p
tim
ized
DNN
(
T
r
ain
in
g
Acc
u
r
ac
y
:
0
.
9
7
,
T
esti
n
g
Acc
u
r
ac
y
:
0
.
9
5
)
af
ter
A
d
a
m
o
p
tim
izer
C
l
a
s
s
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F
-
S
c
o
r
e
N
o
G
a
s
1
.
0
0
1
.
0
0
1
.
0
0
P
e
r
f
u
me
0
.
8
8
0
.
9
3
0
.
9
0
S
mo
k
e
0
.
9
4
0
.
8
8
0
.
9
1
M
i
x
t
u
r
e
1
.
0
0
1
.
0
0
1
.
0
0
Fig
u
r
e
4
.
T
h
e
co
n
f
u
s
io
n
m
atr
ix
o
f
th
e
DNN
m
o
d
el
6
.
2
.
T
herm
a
l
im
a
g
es r
esu
lt
s
o
f
s
ce
na
rio
I
I
T
h
e
r
esu
lts
o
b
tain
ed
f
r
o
m
th
is
s
ce
n
ar
io
a
r
e
p
r
esen
ted
an
d
an
aly
ze
d
in
th
r
ee
s
tr
u
ctu
r
e
d
p
h
a
s
es.
T
h
ese
in
clu
d
e
th
e
d
ata
au
g
m
en
tatio
n
p
r
o
ce
s
s
,
th
e
o
p
tim
izatio
n
o
f
th
e
m
o
d
el,
an
d
th
e
ev
alu
ati
o
n
o
f
th
e
o
p
tim
ized
VGG1
6
m
o
d
el.
E
ac
h
p
h
ase
p
lay
s
a
cr
itical
r
o
le
in
en
h
an
cin
g
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
c
e
an
d
en
s
u
r
in
g
th
e
r
eliab
ilit
y
o
f
th
e
d
etec
tio
n
s
y
s
tem
.
6
.
2
.
1
.
Da
t
a
a
ug
m
ent
a
t
io
n
T
h
e
im
ag
e
d
ata
g
e
n
er
ato
r
f
u
n
c
tio
n
f
r
o
m
th
e
Ker
as
lib
r
ar
y
is
u
tili
ze
d
to
im
p
lem
e
n
t
d
ata
au
g
m
en
tatio
n
an
d
to
r
esize
a
n
d
r
escale
th
e
s
am
p
les.
T
ab
le
6
p
r
o
v
id
es
a
s
u
m
m
ar
y
o
f
th
e
d
ata
a
u
g
m
e
n
tatio
n
tech
n
iq
u
es
u
s
ed
,
alo
n
g
with
th
eir
co
r
r
esp
o
n
d
i
n
g
v
alu
es.
Ad
d
itio
n
ally
,
Fig
u
r
e
5
p
r
esen
ts
v
is
u
al
ex
am
p
les
o
f
th
e
au
g
m
en
tatio
n
p
r
o
ce
s
s
.
I
t
in
clu
d
es
two
p
ar
ts
:
Fig
u
r
e
5
(
a
)
in
w
h
ich
th
e
o
r
i
g
in
al
im
ag
e
o
f
a
c
o
m
m
o
n
r
u
s
t
-
af
f
ec
ted
s
am
p
le.
Fig
u
r
e
5
(
b
)
,
au
g
m
en
ted
v
er
s
io
n
s
o
f
th
e
s
am
e
im
a
g
e
g
e
n
er
ated
u
s
in
g
th
e
s
p
ec
if
ied
d
ata
au
g
m
e
n
tatio
n
tech
n
iq
u
es,
d
e
m
o
n
s
tr
atin
g
th
e
d
iv
er
s
e
tr
an
s
f
o
r
m
atio
n
s
a
p
p
lied
.
T
h
ese
d
ata
au
g
m
en
tat
io
n
tech
n
i
q
u
es
ar
e
ap
p
lied
d
u
r
in
g
th
e
tr
ain
i
n
g
p
r
o
ce
s
s
to
en
h
an
ce
th
e
tr
ain
in
g
d
ataset's
d
iv
er
s
ity
an
d
s
ize,
w
h
ich
h
elp
s
im
p
r
o
v
e
th
e
g
en
er
aliza
tio
n
ca
p
ab
ilit
y
o
f
th
e
d
ee
p
lear
n
in
g
m
o
d
el.
T
ab
le
6
.
Data
au
g
m
en
tatio
n
m
eth
o
d
s
an
d
t
h
eir
ass
o
ciate
d
p
ar
am
eter
s
D
a
t
a
a
u
g
m
e
n
t
a
t
i
o
n
me
t
h
o
d
A
sso
c
i
a
t
e
d
p
a
r
a
m
e
t
e
r
Evaluation Warning : The document was created with Spire.PDF for Python.