I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
15
,
No
.
2
,
A
p
r
il
20
26
,
p
p
.
1
4
6
4
~
1
4
7
2
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
2
.
p
p
1
4
6
4
-
1
4
7
2
1464
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Explain
a
ble so
cia
l media
disa
ster
i
ma
g
e clas
sifica
tion usin
g
a
lig
htweight a
tt
en
t
io
n
-
ba
sed deep
le
a
rning
appro
a
ch
Ra
s
hm
i K
a
ng
o
ka
r
T
a
ra
na
t
h
,
G
ee
t
a
Chid
a
na
nd
a
pp
a
M
a
ra
S
c
h
o
o
l
o
f
C
o
m
p
u
t
i
n
g
a
n
d
I
n
f
o
r
m
a
t
i
o
n
Te
c
h
n
o
l
o
g
y
,
R
EV
A
U
n
i
v
e
r
s
i
t
y
,
B
e
n
g
a
l
u
r
u
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Au
g
23
,
2
0
2
5
R
ev
is
ed
J
an
10
,
2
0
2
6
Acc
ep
ted
J
an
25
,
2
0
2
6
In
re
c
e
n
t
y
e
a
rs,
t
h
e
ra
p
id
d
isse
m
in
a
ti
o
n
o
f
so
c
ial
m
e
d
ia
c
o
n
te
n
t
d
u
ri
n
g
n
a
tu
ra
l
a
n
d
m
a
n
-
m
a
d
e
d
isa
ste
rs
h
a
s
c
re
a
ted
a
n
e
e
d
f
o
r
a
u
t
o
m
a
ted
a
n
d
a
c
c
u
ra
te
d
isa
ste
r
im
a
g
e
c
las
sifica
ti
o
n
sy
ste
m
s.
Th
is
p
a
p
e
r
p
ro
p
o
se
s
li
g
h
twe
i
g
h
t
e
x
p
lain
a
b
le
a
tt
e
n
ti
o
n
-
b
a
se
d
d
isa
ste
r
n
e
tw
o
rk
(
LE
A
D
-
Ne
t
),
a
d
e
e
p
lea
rn
in
g
(DL)
m
o
d
e
l
d
e
sig
n
e
d
fo
r
c
las
sify
in
g
d
isa
ste
r
-
re
late
d
ima
g
e
s
with
h
ig
h
a
c
c
u
ra
c
y
a
n
d
i
n
te
rp
re
tab
il
it
y
.
Th
e
sy
ste
m
i
n
teg
ra
tes
a
n
Eff
icie
n
tNe
t
-
B0
b
a
c
k
b
o
n
e
e
n
h
a
n
c
e
d
with
sq
u
e
e
z
e
-
a
n
d
-
e
x
c
it
a
ti
o
n
(
S
E)
a
tt
e
n
ti
o
n
m
o
d
u
les
a
n
d
a
li
g
h
twe
ig
h
t
n
e
u
ra
l
a
rc
h
i
tec
tu
re
se
a
rc
h
(NA
S
-
li
te)
stra
teg
y
f
o
r
t
u
n
in
g
t
h
e
c
las
sifier
h
e
a
d
a
n
d
train
in
g
h
y
p
e
rp
a
ra
m
e
ters
.
Th
e
m
o
d
e
l
wa
s
e
v
a
l
u
a
ted
o
n
two
b
e
n
c
h
m
a
rk
d
a
tas
e
ts
c
o
m
p
re
h
e
n
si
v
e
d
isa
ste
r
d
a
tas
e
t
(CDD
)
a
n
d
d
a
m
a
g
e
m
u
lt
imo
d
a
l
d
a
tas
e
t
(DMD)
a
c
h
iev
i
n
g
9
6
%
a
n
d
8
7
%
a
c
c
u
ra
c
y
,
re
sp
e
c
ti
v
e
ly
,
o
u
tp
e
rfo
rm
in
g
se
v
e
ra
l
e
sta
b
li
sh
e
d
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
(
CNN
)
b
a
se
li
n
e
s.
To
e
n
su
re
tran
s
p
a
re
n
c
y
,
g
ra
d
ien
t
-
we
ig
h
ted
c
las
s
a
c
ti
v
a
ti
o
n
m
a
p
p
in
g
(
G
ra
d
-
CAM
)
wa
s
e
m
p
lo
y
e
d
to
g
e
n
e
ra
te
v
isu
a
l
e
x
p
lan
a
ti
o
n
s
o
f
t
h
e
m
o
d
e
l’s
d
e
c
isio
n
s,
c
o
n
firmi
n
g
i
ts
fo
c
u
s
o
n
se
m
a
n
ti
c
a
ll
y
re
lev
a
n
t
ima
g
e
re
g
i
o
n
s.
K
ey
w
o
r
d
s
:
Atten
tio
n
m
ec
h
an
is
m
C
o
m
p
u
tatio
n
al
-
ef
f
icien
c
y
Dee
p
lear
n
in
g
Dis
aster
im
ag
e
class
if
icatio
n
E
x
p
lain
ab
le
AI
R
ea
l
-
tim
e
d
ep
lo
y
m
en
t
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
:
R
ash
m
i K
an
g
o
k
ar
T
ar
a
n
ath
Sch
o
o
l o
f
C
o
m
p
u
tin
g
an
d
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
,
R
E
VA
Un
iv
er
s
ity
R
u
k
m
in
i K
n
o
wled
g
e
Par
k
,
Ka
ttig
en
ah
alli,
B
en
g
alu
r
u
,
Kar
n
a
tak
a
5
6
0
0
6
4
,
I
n
d
ia
E
m
ail: r
ash
m
ik
t1
9
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Natu
r
al
d
is
aster
s
s
u
ch
as
ea
r
t
h
q
u
ak
es,
f
lo
o
d
s
,
wild
f
ir
es,
an
d
h
u
r
r
ica
n
es
b
r
in
g
d
is
tr
ess
in
g
co
s
ts
f
o
r
h
u
m
an
life
,
in
f
r
astru
ctu
r
e,
an
d
th
e
en
v
ir
o
n
m
e
n
t.
T
h
e
tr
en
d
o
f
p
ast
f
ew
d
ec
ad
es
s
h
o
ws
th
at
th
e
f
r
eq
u
en
c
y
an
d
in
ten
s
ity
o
f
n
atu
r
al
d
is
aster
s
h
av
e
in
cr
ea
s
ed
d
u
e
to
clim
ate
c
h
an
g
e
a
n
d
u
r
b
a
n
e
x
p
an
s
io
n
[
1
]
.
Du
r
in
g
d
is
aster
s
,
g
ettin
g
tim
ely
an
d
ac
cu
r
ate
in
f
o
r
m
atio
n
b
ec
o
m
es
v
er
y
cr
u
cia
l
f
o
r
tak
in
g
s
tr
ateg
ies
f
o
r
an
e
f
f
ec
tiv
e
em
er
g
en
cy
r
esp
o
n
s
e,
r
eso
u
r
ce
allo
ca
tio
n
an
d
d
am
a
g
e
ass
ess
m
en
t
[
2
]
.
Ho
wev
er
,
in
s
u
ch
s
i
tu
atio
n
s
,
tr
ad
itio
n
al
co
m
m
u
n
icatio
n
s
s
y
s
tem
s
o
f
te
n
b
ec
o
m
e
o
v
er
lo
a
d
ed
o
r
f
ail,
s
ev
er
ely
d
elay
in
g
d
is
aster
r
esp
o
n
s
e
o
p
e
r
atio
n
s
[
3
]
.
I
n
r
ec
en
t
y
ea
r
s
,
s
o
cial
m
ed
ia
ch
an
n
els
s
u
ch
as
T
witter
,
Face
b
o
o
k
,
a
n
d
I
n
s
tag
r
am
h
a
v
e
p
r
o
v
en
to
b
e
u
s
ef
u
l
s
o
u
r
ce
s
o
f
r
ea
l
-
tim
e
in
f
o
r
m
at
io
n
d
u
r
in
g
d
is
aster
s
[
4
]
,
[
5
]
.
Peo
p
le
o
f
ten
s
h
a
r
e
b
r
ea
k
in
g
n
ews,
p
h
o
to
s
,
a
n
d
v
id
eo
s
o
f
d
is
aster
-
af
f
ec
ted
a
r
ea
s
,
wh
ich
ca
n
s
er
v
e
as
d
is
a
s
ter
war
n
in
g
s
,
g
r
o
u
n
d
r
ef
er
en
ce
s
,
an
d
ev
i
d
en
ce
.
Am
o
n
g
th
ese,
p
h
o
t
o
s
p
r
o
v
id
e
r
ich
v
is
u
al
in
f
o
r
m
atio
n
th
at
h
elp
s
ass
es
s
th
e
s
ev
er
ity
an
d
ty
p
e
o
f
d
is
aster
s
[
6
]
.
Ho
wev
er
,
m
an
u
al
an
aly
s
is
o
f
p
h
o
to
s
is
tim
e
-
co
n
s
u
m
in
g
,
a
n
d
it
is
n
o
t
f
ea
s
ib
le
to
an
aly
ze
t
h
o
u
s
an
d
s
o
f
p
h
o
to
s
d
aily
,
wh
ich
d
ela
y
s
im
m
ed
iate
r
esp
o
n
s
e.
T
o
ad
d
r
ess
th
is
ch
al
len
g
e,
r
esear
ch
er
s
h
a
v
e
b
eg
u
n
to
u
s
e
d
ata
-
d
r
iv
e
n
an
d
d
ee
p
lear
n
in
g
(
DL
)
m
o
d
el
s
th
at
ca
n
id
en
tify
ty
p
es
o
f
d
is
aster
im
ag
es
an
d
ac
ce
ler
ate
d
ec
is
io
n
m
ak
in
g
[
7
]
.
Alth
o
u
g
h
th
ese
m
o
d
els
ar
e
v
er
y
ef
f
ec
tiv
e,
b
u
t
s
ev
er
al
ch
allen
g
es
s
till
r
em
ain
s
u
ch
as
d
is
aster
im
ag
es
o
f
ten
co
n
tain
in
g
c
o
m
p
lex
a
n
d
o
v
er
l
ap
p
in
g
v
is
u
al
in
f
o
r
m
atio
n
,
s
u
ch
th
at
th
e
s
am
e
im
ag
e
m
ay
s
h
o
w
m
u
ltip
le
ty
p
es
o
f
d
is
aster
s
o
r
ev
en
lo
o
k
lik
e
o
th
er
e
v
en
ts
;
d
atasets
ar
e
im
b
alan
ce
d
;
an
d
s
o
cial
n
o
is
e
an
d
u
n
lab
eled
d
ata
p
o
s
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
xp
la
in
a
b
le
s
o
ci
a
l m
ed
ia
d
is
a
s
ter ima
g
e
cla
s
s
ifica
tio
n
u
s
in
g
a
lig
h
tw
eig
h
t …
(
R
a
s
h
mi
K
a
n
g
o
ka
r
Ta
r
a
n
a
th
)
1465
p
r
o
b
lem
s
f
o
r
m
o
d
el
tr
ain
in
g
[
8
]
.
T
h
e
r
ef
o
r
e
,
d
ev
elo
p
in
g
a
n
e
f
f
ec
tiv
e
s
o
lu
tio
n
r
e
q
u
ir
es a
r
o
b
u
s
t f
ea
tu
r
e
lear
n
in
g
s
ch
em
e
an
d
ev
en
t
-
awa
r
e
class
if
icatio
n
s
tr
ateg
ies
to
m
ak
e
co
r
r
ec
t
p
r
ed
ictio
n
s
o
n
n
atu
r
al
d
is
aster
im
ag
es
f
r
o
m
s
o
cial
m
ed
ia
p
o
s
ts
.
T
h
e
r
ec
e
n
t
r
esear
ch
h
as
ex
p
lo
r
ed
d
if
f
er
en
t
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NNs)
ar
ch
ite
ctu
r
es
[
9
]
tr
an
s
f
er
an
d
h
y
b
r
id
lear
n
in
g
te
ch
n
iq
u
es
[
1
0
]
to
class
if
y
d
is
aster
im
ag
es.
Ah
m
ed
et
a
l.
[
1
1
]
ad
o
p
ted
f
ed
e
r
ated
lear
n
in
g
ap
p
r
o
ac
h
to
r
ed
u
ce
d
ep
en
d
e
n
cy
o
f
m
an
u
al
a
n
n
o
t
atio
n
an
d
waste
class
if
icatio
n
.
T
h
is
wo
r
k
s
h
o
ws
th
eir
m
o
d
el
e
n
ab
les
lo
ca
l
cli
en
ts
to
au
to
m
atica
lly
s
elec
t
in
f
o
r
m
ativ
e
u
n
lab
ele
d
d
ata,
a
n
d
ac
h
iev
e
3
%
–
4
%
im
p
r
o
v
em
e
n
ts
in
ac
cu
r
ac
y
to
f
u
lly
lab
eled
d
atasets
,
b
u
t
it
m
ay
f
ac
e
is
s
u
es
wh
en
s
u
b
ject
ed
to
h
eter
o
g
en
e
o
u
s
s
o
u
r
ce
s
.
Ma
et
a
l.
[
1
2
]
ad
d
r
e
s
s
ed
th
e
s
ca
le
v
ar
iatio
n
p
r
o
b
lem
in
n
atu
r
al
d
is
aster
im
ag
er
y
b
y
d
ev
elo
p
i
n
g
a
h
y
b
r
id
v
is
io
n
tr
a
n
s
f
o
r
m
er
m
o
d
el
b
ased
o
n
th
e
jo
in
t
a
p
p
r
o
ac
h
o
f
co
n
v
o
lu
tio
n
o
p
er
atio
n
with
tim
e
s
er
ies
d
o
wn
s
am
p
lin
g
m
o
d
u
le.
T
h
e
r
e
s
u
l
t
a
n
a
l
y
s
is
s
h
o
ws
b
e
n
e
f
i
t
s
i
n
p
r
e
d
i
c
t
i
o
n
i
m
p
r
o
v
e
m
e
n
t
b
u
t
a
t
c
o
s
t
o
f
h
i
g
h
e
r
c
o
m
p
l
e
x
i
t
y
t
h
a
t
m
a
y
h
i
n
d
e
r
r
e
a
l
-
t
i
m
e
d
e
p
l
o
y
m
e
n
t
.
S
h
et
h
e
t
a
l
.
[
1
3
]
in
tr
o
d
u
ce
d
an
en
s
em
b
le
m
o
d
el
o
f
I
n
ce
p
tio
n
V3
an
d
a
cu
s
to
m
C
NN
to
class
if
y
d
is
aster
im
ag
es
with
ac
cu
r
ac
y
o
f
9
2
.
7
9
%.
T
h
is
ap
p
r
o
ac
h
lack
s
n
o
v
elty
an
d
n
o
s
ig
n
if
ican
t
c
o
n
tr
ib
u
tio
n
s
h
o
wn
.
C
o
o
p
en
an
d
Pu
d
a
r
u
th
[
1
4
]
d
ev
elo
p
e
d
a
c
u
s
to
m
d
ataset
co
m
b
in
in
g
ex
is
tin
g
d
ataset
to
in
cr
ea
s
e
th
e
d
ata
d
iv
er
s
ity
a
n
d
class
es,
wh
ich
is
ev
alu
ated
with
p
r
e
-
tr
ain
e
d
m
o
d
els
lik
e
Mo
b
ileNetV2
,
V
GG1
6
,
an
d
I
n
ce
p
tio
n
V3
.
T
h
eir
r
esu
lts
s
h
o
wed
I
n
ce
p
tio
n
V3
an
d
Mo
b
ileNetV2
ac
h
iev
ed
9
6
.
8
6
%
ac
cu
r
ac
y
b
u
t
u
n
d
er
s
p
ec
if
ic
tr
ain
in
g
c
o
n
d
itio
n
s
.
Mu
s
taf
a
et
a
l.
[
1
5
]
p
r
esen
ted
a
DL
f
r
am
ewo
r
k
,
wh
er
e
th
ey
h
av
e
ex
p
lo
r
ed
d
if
f
e
r
en
t
ex
p
lain
a
b
le
AI
m
eth
o
d
s
s
u
ch
as
g
r
ad
ien
t
-
weig
h
ted
class
ac
tiv
atio
n
m
ap
p
in
g
(
Gr
ad
-
C
AM
)
,
Gr
ad
-
C
AM
++
,
an
d
l
o
ca
l
in
ter
p
r
etab
le
m
o
d
el
-
ag
n
o
s
tic
ex
p
lan
atio
n
s
(
LIME
)
f
o
r
d
is
aster
im
ag
e
class
if
icatio
n
.
B
a
s
h
ir
et
a
l.
[
1
6
]
in
tr
o
d
u
ce
d
a
d
is
aster
m
o
n
ito
r
in
g
s
ch
em
e
th
at
u
s
es
ae
r
ial
im
ag
es
an
d
tr
a
n
s
m
its
th
em
to
a
b
ase
s
tatio
n
wh
er
e
a
DL
m
o
d
el
with
f
ea
tu
r
e
co
n
ca
ten
atio
n
is
u
s
ed
to
class
if
y
d
is
as
ter
ty
p
es.
Yash
i
et
a
l.
[
1
7
]
p
r
esen
ted
a
n
en
s
em
b
le
f
r
am
ewo
r
k
b
ased
o
n
th
e
c
o
m
b
in
atio
n
o
f
eig
h
t
C
NN
m
o
d
els
f
o
r
f
lo
o
d
class
if
icatio
n
u
s
in
g
Flo
o
d
Net
d
atasets
,
wh
ich
s
h
o
ws
tr
ain
in
g
ac
cu
r
ac
y
o
f
9
8
.
9
%
an
d
test
ac
cu
r
ac
y
o
f
9
7
.
4
%.
Ho
w
ev
er
,
th
ey
h
av
e
n
o
t
d
is
cu
s
s
ed
th
e
tim
e
co
m
p
lex
ity
ass
o
ciate
d
with
en
s
em
b
le
p
r
ed
ictio
n
f
o
r
m
8
C
NNs.
Sath
ian
ar
ay
an
a
n
et
a
l.
[
1
8
]
a
d
d
r
e
s
s
ed
th
e
lack
o
f
g
eo
l
o
ca
tio
n
d
ata
in
s
o
cial
m
ed
ia
im
ag
es
f
o
r
d
is
aster
r
esp
o
n
s
e
b
y
s
u
g
g
esti
n
g
a
p
h
o
n
e
-
n
u
m
b
er
-
b
ased
lo
ca
tio
n
ex
tr
ac
ti
o
n
f
r
am
ewo
r
k
.
T
h
e
au
th
o
r
s
h
av
e
u
s
ed
R
etin
aNe
t CNN
m
o
d
el
th
at
d
etec
ts
d
ig
its
,
an
d
u
s
ed
Go
o
g
le
Ma
p
s
API
to
r
etr
iev
e
lo
ca
tio
n
s
,
an
d
th
e
o
u
tco
m
e
s
h
o
ws
an
av
er
ag
e
p
r
ec
is
io
n
o
f
8
2
%.
T
h
o
u
g
h
th
e
au
th
o
r
s
h
av
e
p
r
esen
ted
in
n
o
v
ativ
e
a
p
p
r
o
ac
h
b
u
t
m
ay
lim
ited
b
y
th
e
ass
u
m
p
tio
n
th
at
p
h
o
n
e
n
u
m
b
er
s
a
r
e
alwa
y
s
v
is
ib
le
an
d
r
ea
d
ab
l
e
in
d
is
aster
im
ag
es.
Van
E
x
el
et
a
l.
[
1
9
]
in
t
r
o
d
u
c
ed
a
DL
a
p
p
r
o
ac
h
f
o
r
d
etec
t
in
g
f
lo
o
d
in
g
an
d
d
eser
tific
ati
o
n
f
r
o
m
u
n
m
a
n
n
ed
ae
r
ial
v
eh
icle
(
UAV
)
a
n
d
s
atellite
im
ag
er
y
u
s
in
g
a
cli
m
ate
ch
an
g
e
d
ataset,
wh
er
e
Den
s
eNe
t2
0
1
an
d
R
esNet5
0
ac
h
iev
ed
ac
cu
r
ac
ies
o
f
9
9
.
3
7
% a
n
d
9
9
.
2
1
%,
r
esp
e
ctiv
ely
.
T
h
e
wo
r
k
o
f
Du
b
ey
an
d
Kata
r
y
a
[
2
0
]
d
esig
n
e
d
a
h
y
b
r
id
f
l
o
o
d
d
etec
tio
n
m
o
d
el
b
ased
o
n
th
e
jo
in
t
ap
p
r
o
ac
h
o
f
tr
an
s
f
o
r
m
er
an
d
a
d
ap
tiv
e
f
ir
ef
ly
alg
o
r
ith
m
f
o
r
f
ea
tu
r
e
o
p
tim
izatio
n
an
d
ac
h
iev
ed
an
ac
cu
r
ac
y
o
f
9
7
.
8
5
%
o
v
e
r
b
e
n
ch
m
a
r
k
s
.
T
h
is
wo
r
k
d
o
es
n
o
t
d
is
cu
s
s
m
u
ch
ab
o
u
t
t
h
e
ev
o
lu
tio
n
a
r
y
p
ar
a
m
eter
s
tu
n
in
g
as
it
m
ay
in
cr
ea
s
e
tr
ain
in
g
an
d
i
n
f
e
r
en
ce
o
v
e
r
h
ea
d
.
W
en
et
a
l.
[
2
1
]
s
u
g
g
ested
im
p
r
o
v
e
m
en
t
o
v
er
R
esNet
-
5
0
C
NN
ar
ch
itectu
r
e
to
class
if
y
s
ev
en
ty
p
es
o
f
n
atu
r
al
d
is
aster
s
o
n
co
m
p
lex
d
ataset
an
d
o
n
ly
ac
h
iev
ed
8
7
%
ac
cu
r
ac
y
.
T
h
is
s
h
o
ws
ad
d
r
ess
ed
g
r
ad
i
en
t
v
an
is
h
in
g
is
s
u
es,
an
d
i
m
p
r
o
v
e
d
tr
ai
n
in
g
s
tab
ilit
y
b
u
t
lef
t
a
s
co
p
e
f
o
r
im
p
r
o
v
em
e
n
t
o
n
class
if
icatio
n
ac
cu
r
ac
y
an
d
m
o
d
el
s
en
s
itiv
ity
.
Sh
ao
an
d
Xu
[
2
2
]
in
tr
o
d
u
ce
d
a
m
u
ltimo
d
al
d
is
aster
r
ec
o
g
n
itio
n
m
o
d
el
b
ased
o
n
p
r
e
-
tr
ain
ed
lan
g
u
ag
e
m
o
d
els
with
v
is
u
al
b
ac
k
b
o
n
es
b
y
in
teg
r
atin
g
q
u
esti
o
n
-
an
s
wer
in
g
a
p
p
r
o
ac
h
.
T
h
ey
co
m
b
in
ed
m
o
d
alities
an
d
th
e
Ma
m
b
a
m
e
ch
an
is
m
,
an
d
o
b
tain
e
d
an
ac
cu
r
ac
y
o
f
9
7
.
8
2
%
ag
ain
s
t
C
NNs
an
d
v
is
io
n
tr
a
n
s
f
o
r
m
er
s
.
Ho
wev
er
,
it
m
ay
lack
s
ca
lab
i
lity
an
d
ass
o
ciate
d
with
lar
g
e
co
m
p
u
tatio
n
al
co
s
t
s
d
u
e
to
th
e
n
atu
r
e
o
f
m
o
d
el
im
p
lem
en
tatio
n
.
T
h
e
s
tu
d
y
o
f
W
an
g
et
a
l.
[
2
3
]
p
r
esen
ted
a
m
u
lti
-
s
tag
e
DL
ar
ch
itectu
r
e
b
ased
o
n
th
e
in
teg
r
atio
n
o
f
U2
Net
m
o
d
el
an
d
s
win
tr
an
s
f
o
r
m
er
f
o
r
d
am
ag
e
ass
ess
m
en
t
f
r
o
m
s
ate
llit
e
im
ag
es.
T
h
eir
m
o
d
el
s
h
o
wed
s
ig
n
if
ican
t
p
er
f
o
r
m
an
ce
i
m
p
r
o
v
e
m
en
ts
o
v
er
ex
is
tin
g
m
eth
o
d
s
o
n
th
e
x
B
D
d
ataset.
Ho
wev
er
,
it
h
as
h
i
g
h
er
d
ep
e
n
d
en
cies
o
n
ac
cu
r
ate
d
is
aster
-
ty
p
e
in
p
u
t
an
d
p
r
e
-
d
is
aster
s
eg
m
en
tatio
n
p
er
f
o
r
m
an
ce
,
wh
ic
h
m
ay
li
m
it
its
ad
ap
tiv
en
ess
to
co
m
p
l
ex
im
ag
es
o
b
tain
e
d
f
r
o
m
s
o
cial
m
ed
ia
p
o
s
ts
.
Alth
o
u
g
h
m
a
n
y
s
ch
em
es
h
av
e
b
ee
n
p
r
esen
ted
in
th
e
liter
atu
r
e,
b
u
t
m
o
s
t
o
f
th
em
p
r
io
r
itized
ac
cu
r
ac
y
an
d
n
e
g
lect
co
m
p
u
tatio
n
al
ef
f
icien
cy
.
Als
o
,
m
an
y
wo
r
k
s
lack
n
o
v
elty
as
th
er
e
a
r
e
s
im
ilar
wo
r
k
s
b
y
d
if
f
e
r
en
t
r
esear
ch
er
s
with
s
lig
h
t
ch
an
g
es
in
th
e
ad
o
p
tio
n
o
f
s
tan
d
ar
d
C
NNs,
an
d
en
s
em
b
le
m
eth
o
d
s
wh
ich
in
cr
ea
s
es
co
m
p
lex
ity
,
tr
ain
in
g
tim
e,
an
d
in
f
er
en
ce
co
s
t.
Mo
s
t
DL
m
o
d
els
o
p
er
ate
as
b
lack
b
o
x
es,
with
lim
ited
f
o
cu
s
o
n
in
ter
p
r
etab
ilit
y
,
th
at
m
a
k
es it d
if
f
icu
lt f
o
r
f
ir
s
t r
esp
o
n
d
er
s
to
u
n
d
er
s
tan
d
an
d
t
r
u
s
t th
e
s
y
s
tem
’
s
p
r
ed
ictio
n
s
.
T
h
is
p
ap
er
i
n
tr
o
d
u
ce
s
lig
h
tweig
h
t
ex
p
lain
a
b
le
atten
tio
n
-
b
ased
d
is
aster
n
etwo
r
k
(
L
E
AD
-
Net
)
a
DL
ar
ch
itectu
r
e
d
esig
n
e
d
f
o
r
ef
f
i
cien
t,
in
ter
p
r
eta
b
le,
an
d
h
ig
h
-
ac
cu
r
ac
y
d
is
aster
im
ag
e
cla
s
s
if
icatio
n
.
Un
lik
e
ex
is
tin
g
ap
p
r
o
ac
h
es
th
at
ad
o
p
t
ed
s
tan
d
ar
d
b
u
t
co
m
p
le
x
C
NN
s
,
th
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
in
teg
r
ates
a
m
o
d
if
ied
atten
tio
n
b
lo
ck
with
d
u
al
p
o
o
lin
g
,
d
y
n
am
ic
r
ed
u
ctio
n
,
an
d
Swis
h
ac
tiv
atio
n
to
en
h
an
ce
f
e
atu
r
e
lear
n
in
g
.
T
h
e
m
o
d
el
also
p
r
o
v
id
es
v
is
u
al
e
x
p
lan
atio
n
s
v
ia
Gr
ad
-
C
AM
,
to
en
s
u
r
e
tr
an
s
p
a
r
en
cy
in
h
ig
h
-
s
tak
es
d
ec
is
io
n
s
.
T
h
e
m
o
d
el
is
ex
ten
s
iv
ely
v
alid
ated
o
n
two
d
iv
er
s
if
ied
r
ea
l
-
wo
r
ld
d
is
aster
d
atasets
to
p
r
o
v
e
r
o
b
u
s
tn
ess
an
d
g
en
er
aliza
b
ilit
y
,
s
o
m
eth
in
g
o
f
ten
m
is
s
in
g
in
p
r
io
r
wo
r
k
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
4
6
4
-
1
4
7
2
1466
2.
M
E
T
H
O
D
T
h
e
p
r
im
e
aim
o
f
th
e
p
r
o
p
o
s
ed
wo
r
k
is
to
d
e
v
elo
p
a
r
o
b
u
s
t
im
ag
e
class
if
icatio
n
m
o
d
el
th
at
ca
n
ac
cu
r
ately
id
en
tify
d
is
aster
ev
en
ts
f
r
o
m
s
o
cial
m
ed
ia
im
ag
e
s
,
wh
ich
ar
e
o
f
ten
v
is
u
ally
co
m
p
lex
an
d
c
o
n
tain
s
am
b
ig
u
o
u
s
n
atu
r
e
o
f
d
is
aster
s
ce
n
es
wh
er
e
a
s
in
g
le
im
ag
e
m
ay
illu
s
tr
ate
m
u
ltip
le
o
v
e
r
la
p
p
in
g
p
h
en
o
m
en
a.
T
h
is
p
ap
er
in
tr
o
d
u
ce
s
L
E
AD
-
Net
to
ef
f
ec
tiv
ely
lear
n
d
ee
p
v
is
u
al
p
atter
n
s
,
g
en
e
r
a
lize
ac
r
o
s
s
d
is
as
ter
ca
teg
o
r
ies,
an
d
p
r
o
v
id
e
in
ter
p
r
etab
le
p
r
ed
ictio
n
s
.
T
h
e
s
y
s
tem
d
esig
n
an
d
m
eth
o
d
o
lo
g
y
wo
r
k
f
lo
w
o
f
th
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
is
illu
s
tr
ated
in
Fig
u
r
e
1
.
Fig
u
r
e
1
.
S
y
s
tem
d
esig
n
an
d
m
eth
o
d
o
l
o
g
y
wo
r
k
f
lo
w
o
f
L
E
AD
-
Net
T
h
e
in
p
u
t
c
o
n
s
id
er
ed
in
th
e
p
r
o
p
o
s
ed
s
y
s
tem
ar
e
th
e
n
atu
r
al
d
is
aster
im
ag
e
f
r
o
m
th
e
s
o
cial
m
ed
ia
p
o
s
ts
o
b
tain
ed
f
r
o
m
th
e
co
m
p
r
eh
en
s
iv
e
d
is
aster
d
ataset
(
C
DD)
[
2
4
]
an
d
d
am
ag
e
m
u
ltim
o
d
al
d
ataset
(
DM
D)
o
b
tain
ed
f
r
o
m
ME
DI
C
b
en
c
h
m
ar
k
[
2
5
]
,
[
2
6
]
.
T
h
e
C
DD
d
ataset
co
n
s
is
ts
o
f
to
tal
1
3
5
6
5
im
a
g
e
s
am
p
les
d
is
tr
ib
u
ted
ac
r
o
s
s
1
2
class
es,
wh
er
ea
s
DM
D
h
as
to
tal
5
8
7
8
im
ag
es
s
am
p
les
with
in
6
cla
s
s
es.
B
o
th
d
ataset
s
co
n
s
is
t
o
f
v
is
u
al
s
ce
n
es
o
f
n
a
tu
r
al
d
is
aster
ev
en
ts
wh
ich
is
n
o
t
o
n
ly
co
n
tain
s
v
is
u
ally
a
m
b
ig
u
o
u
s
n
atu
r
e
o
f
d
is
aster
s
ce
n
es
b
u
t
also
s
u
b
j
ec
ted
to
m
u
lti
-
f
ac
eted
c
o
n
ten
t.
T
h
e
f
ir
s
t
co
m
p
u
tin
g
m
o
d
u
le
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
tem
is
th
e
p
r
e
p
r
o
ce
s
s
in
g
wh
ich
p
er
f
o
r
m
s
im
ag
e
r
esizin
g
(
2
2
4
×
2
2
4
p
ix
els)
an
d
d
at
a
n
o
r
m
aliza
tio
n
to
s
tan
d
ar
d
ize
p
ix
el
v
alu
es
to
m
atch
th
e
in
p
u
t
r
e
q
u
ir
em
e
n
t
o
f
th
e
L
E
AD
-
Net
DL
m
o
d
el
.
T
h
e
d
atasets
with
p
r
ep
r
o
ce
s
s
ed
im
ag
es
ar
e
th
en
s
u
b
jecte
d
to
tr
ai
n
-
test
s
p
lit
o
p
er
atio
n
wh
er
e
th
r
ee
d
if
f
e
r
en
t
s
u
b
-
s
ets
ar
e
cr
ea
te
d
s
u
ch
as
tr
ain
-
s
et
(
8
0
%),
v
alid
atio
n
-
s
et
(
1
0
%),
a
n
d
test
-
s
et
(
1
0
%).
I
n
o
r
d
er
to
p
r
o
v
id
e
b
etter
lear
n
i
n
g
ex
p
er
ien
ce
to
L
E
AD
-
Net
m
o
d
el,
th
e
d
iv
er
s
ity
o
f
tr
ain
-
s
et
is
en
r
ich
ed
v
ia
d
ata
au
g
m
en
tatio
n
s
tr
ateg
ies
s
u
ch
as
h
o
r
izo
n
tal
a
n
d
v
er
tical
f
lip
p
in
g
,
zo
o
m
i
n
g
,
r
o
tatin
g
an
d
r
an
d
o
m
b
r
ig
h
tn
ess
v
ar
iatio
n
.
T
h
e
d
ata
au
g
m
en
tatio
n
is
n
o
t a
p
p
lied
to
v
alid
atio
n
an
d
test
s
ets to
av
o
id
an
y
d
ata
leak
ag
e
an
d
m
o
d
el
b
iasn
ess
.
T
h
e
n
ex
t
co
m
p
u
tin
g
m
o
d
u
le
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
tem
is
L
E
AD
-
Net
m
o
d
el,
wh
ich
is
b
a
s
ed
o
n
th
e
in
teg
r
atio
n
o
f
DL
m
o
d
el
i.e
.
,
E
f
f
icien
tNet
-
B
0
C
NN
with
tr
an
s
f
er
lear
n
in
g
ap
p
r
o
ac
h
[
2
7
]
a
n
d
atten
tio
n
m
ec
h
an
is
m
.
T
h
e
E
f
f
icien
tNet
-
B
0
C
NN
is
a
lig
h
tweig
h
t
an
d
r
o
b
u
s
t
C
NN
ar
ch
itectu
r
e
a
n
d
in
th
e
p
r
o
p
o
s
ed
s
y
s
tem
it
is
im
p
lem
en
ted
as
a
f
ea
tu
r
e
e
x
tr
ac
to
r
m
o
d
u
le
with
o
u
t
u
s
in
g
its
o
r
i
g
in
al
t
o
p
class
if
ier
lay
er
.
Af
ter
ex
tr
ac
tin
g
f
ea
tu
r
es
with
E
f
f
icien
tNet
-
B
0
,
a
cu
s
to
m
class
if
ier
h
ea
d
i
s
d
ev
elo
p
ed
wh
ich
in
clu
d
es
in
teg
r
atio
n
o
f
atten
tio
n
lay
e
r
,
g
lo
b
al
av
e
r
ag
e
p
o
o
lin
g
(
G
AP)
lay
er
an
d
d
en
s
e
lay
er
f
o
r
th
e
m
u
lti
-
class
class
if
icatio
n
lay
er
.
T
h
e
atten
tio
n
lay
er
u
s
ed
in
th
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
m
o
d
el
is
a
m
o
d
if
ied
v
er
s
io
n
o
f
t
h
e
s
q
u
ee
ze
-
an
d
-
e
x
citatio
n
(
SE)
a
tten
tio
n
b
lo
ck
to
r
e
f
in
e
th
e
c
h
an
n
el
-
wis
e
f
ea
tu
r
e
m
a
p
s
,
wh
ic
h
en
ab
les
m
o
d
el
to
em
p
h
asize
d
is
aster
-
r
elev
an
t
f
e
atu
r
es
an
d
ig
n
o
r
e
ir
r
elev
an
t
o
n
es.
I
n
th
e
m
o
d
if
ied
v
er
s
io
n
o
f
SE
atten
tio
n
,
th
e
s
tu
d
y
h
as
in
tr
o
d
u
ce
d
d
u
al
-
p
o
o
lin
g
in
th
e
s
q
u
ee
ze
p
h
ase,
a
d
ee
p
er
ex
citatio
n
n
etwo
r
k
,
Swis
h
ac
tiv
atio
n
,
an
d
a
d
y
n
am
ically
d
eter
m
in
e
d
r
ed
u
ctio
n
r
atio
.
T
h
e
atten
tio
n
lay
e
r
co
n
s
id
er
s
th
e
f
ea
tu
r
e
m
ap
∈
ℝ
×
×
,
wh
er
e
H
an
d
W
r
ep
r
esen
t
th
e
d
im
en
s
io
n
o
f
th
e
f
ea
tu
r
e
m
ap
o
b
tain
ed
f
r
o
m
c
o
n
v
o
lu
tio
n
al
lay
e
r
s
o
f
E
f
f
icien
tNet
-
B
0
an
d
C
p
r
esen
ts
th
e
n
u
m
b
er
o
f
f
ea
tu
r
e
ch
a
n
n
el
o
r
d
ep
th
.
Af
ter
war
d
s
,
two
g
l
o
b
al
d
escr
ip
t
o
r
s
ar
e
co
m
p
u
ted
v
i
a
GAP
an
d
g
lo
b
al
m
a
x
p
o
o
lin
g
(
GM
P)
as in
(
1
)
an
d
th
e
f
i
n
al
s
q
u
ee
ze
d
v
ec
to
r
is
o
b
tain
e
d
b
y
co
n
ca
ten
atin
g
b
o
th
d
escr
ip
to
r
s
as g
iv
en
in
(
2
)
.
=
1
×
∑
∑
F
(
,
)
=
1
=
1
,
=
ma
x
,
F
(
,
)
(
1
)
=
[
;
]
∈
ℝ
2
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
xp
la
in
a
b
le
s
o
ci
a
l m
ed
ia
d
is
a
s
ter ima
g
e
cla
s
s
ifica
tio
n
u
s
in
g
a
lig
h
tw
eig
h
t …
(
R
a
s
h
mi
K
a
n
g
o
ka
r
Ta
r
a
n
a
th
)
1467
Af
ter
ex
tr
ac
tin
g
a
co
m
p
ac
t
ch
an
n
el
d
escr
ip
to
r
v
ec
to
r
f
r
o
m
th
e
ab
o
v
e
s
q
u
ee
z
s
tep
,
th
e
n
ex
t
ex
citatio
n
o
p
er
atio
n
tr
an
s
f
o
r
m
s
to
p
r
o
d
u
ce
ch
a
n
n
el
-
wis
e
atten
tio
n
weig
h
ts
f
o
llo
wed
b
y
d
ee
p
f
ee
d
f
o
r
war
d
with
Swis
h
ac
tiv
atio
n
an
d
d
y
n
am
ic
r
ed
u
ctio
n
s
tr
ateg
y
as
in
(
3
)
with
k
co
n
s
tan
t
a
n
d
t
h
e
f
in
al
ex
citatio
n
v
ec
t
o
r
is
co
m
p
u
ted
as e
x
p
r
ess
ed
in
(
4
)
.
=
(
1
,
⌊
⌋
)
(
3
)
=
(
W
3
×
(
W
2
×
(
1
×
)
)
)
(
4
)
W
h
er
e
is
weig
h
t
m
atr
ices
wh
er
e
1
∈
ℝ
×
2
is
weig
h
t
o
f
th
e
f
i
r
s
t
lay
er
wh
ich
r
ed
u
ce
s
th
e
d
im
en
s
o
n
a
lity
,
2
∈
ℝ
×
is
th
e
weig
h
ts
o
f
th
e
s
ec
o
n
d
lay
er
th
at
m
ain
tain
s
th
e
s
am
e
s
ize
,
an
d
3
∈
ℝ
×
d
en
o
t
e
s
th
ir
d
lay
er
weig
h
t
wh
ich
ex
p
a
n
d
s
b
ac
k
to
th
e
o
r
i
g
in
al
n
u
m
b
er
o
f
ch
an
n
els.
Her
e
=
/
,
wh
ich
r
e
p
r
esen
ts
th
e
n
u
m
b
er
o
f
u
n
its
in
ea
ch
h
id
d
e
n
lay
er
s
an
d
is
a
s
ig
m
o
id
f
u
n
ctio
n
th
at
s
ca
les
v
alu
es b
etwe
e
n
0
an
d
1
.
Fin
ally
,
th
e
o
u
tp
u
t
atten
tio
n
is
ap
p
lie
d
ch
an
n
el
-
wis
e
to
th
e
o
r
ig
in
al
f
e
atu
r
e
m
ap
as
g
i
v
en
in
(
5
)
,
h
el
p
s
th
e
m
o
d
el
f
o
cu
s
o
n
d
is
aster
-
s
p
ec
if
ic
f
ea
tu
r
es w
h
ile
s
u
p
p
r
ess
in
g
n
o
is
y
o
r
ir
r
el
ev
an
t a
ctiv
atio
n
s
.
′
=
×
∀
∈
{
1
,
2
,
⋯
,
}
(
5
)
T
o
f
u
r
th
er
o
p
tim
ize
m
o
d
el
p
e
r
f
o
r
m
an
ce
,
a
lig
h
tweig
h
t
n
eu
r
a
l
ar
ch
itectu
r
e
s
ea
r
ch
(
NAS
-
lite)
s
tr
ateg
y
is
u
s
ed
to
f
in
e
-
tu
n
e
th
e
class
if
ier
h
ea
d
an
d
tr
ain
in
g
h
y
p
er
p
ar
am
eter
s
p
r
o
p
o
s
ed
L
E
AD
-
Net
m
o
d
el.
A
NAS
r
ef
er
s
to
a
tech
n
iq
u
e
o
f
au
to
m
atica
lly
s
ea
r
ch
in
g
f
o
r
an
o
p
tim
al
n
eu
r
al
n
etwo
r
k
ar
ch
itectu
r
e
f
o
r
a
g
iv
en
task
b
y
ex
p
lo
r
in
g
a
lar
g
e
s
ea
r
ch
s
p
ac
e
o
f
p
o
s
s
ib
le
ar
ch
itectu
r
al
co
n
f
i
g
u
r
atio
n
s
an
d
tr
ain
in
g
h
y
p
e
r
p
ar
am
eter
s
.
Fig
u
r
e
2
illu
s
tr
ates
th
e
wo
r
k
f
lo
w
o
f
p
r
o
p
o
s
ed
NAS
d
r
iv
e
n
ef
f
icien
t
class
if
icatio
n
m
o
d
el
s
etu
p
with
aim
o
f
ac
h
iev
in
g
ef
f
icien
cy
in
tr
ai
n
in
g
p
er
f
o
r
m
an
ce
.
T
h
e
p
r
o
p
o
s
ed
s
tu
d
y
co
n
s
id
er
s
a
lar
g
e
s
ea
r
ch
s
p
ac
e
o
f
tr
ain
i
n
g
h
y
p
er
p
ar
am
eter
s
as
d
em
o
n
s
tr
ated
o
n
th
e
lef
t
s
id
e
o
f
Fig
u
r
e
2
,
w
h
ich
in
clu
d
es
v
ar
io
u
s
d
en
s
e
lay
er
co
n
f
ig
u
r
atio
n
s
,
d
r
o
p
o
u
t
r
at
es,
o
p
tim
izer
s
,
an
d
lear
n
in
g
r
ates,
with
o
u
t
th
e
o
v
er
h
ea
d
o
f
ex
h
au
s
tiv
e
g
r
id
-
b
as
ed
ev
al
u
atio
n
.
T
h
e
s
ea
r
ch
p
r
o
ce
s
s
will
co
n
tin
u
e
an
d
iter
ativ
ely
e
v
alu
ates
s
am
p
led
ar
ch
itectu
r
es
an
d
u
p
d
ate
t
h
e
s
tr
ateg
y
u
n
til
th
e
co
n
v
er
g
e
n
ce
cr
iter
ia
is
m
ee
t.
T
h
e
f
in
al
s
elec
ted
ar
ch
itectu
r
e
in
clu
d
es
an
o
p
tim
al
p
ar
am
e
ter
as
s
h
o
wn
o
n
r
ig
h
t
h
an
d
s
i
d
e
o
f
Fig
u
r
e
2
with
f
ix
ed
ch
o
ice
o
f
b
atch
n
o
r
m
ali
za
tio
n
an
d
GAP
to
en
h
an
ce
c
o
n
v
er
g
en
ce
s
tab
ilit
y
an
d
f
ea
t
u
r
e
s
u
m
m
ar
izatio
n
.
Ad
d
itio
n
ally
,
a
b
atch
s
ize
o
f
1
6
was
u
s
ed
d
u
r
in
g
m
o
d
el
tr
ain
in
g
t
o
p
r
ev
en
t
m
em
o
r
y
o
v
er
h
ea
d
,
an
d
class
weig
h
tin
g
was
ap
p
lied
to
ad
d
r
ess
th
e
im
b
alan
ce
p
r
esen
t
in
th
e
d
ataset.
T
h
e
m
o
d
el
is
tr
ain
ed
f
o
r
3
0
ep
o
c
h
s
o
n
th
e
p
r
ep
a
r
ed
d
ataset.
Af
ter
tr
ai
n
in
g
,
th
e
m
o
d
el
is
ev
alu
ate
d
u
s
in
g
b
o
th
q
u
an
titativ
e
an
d
q
u
a
litativ
e
m
eth
o
d
s
.
Fig
u
r
e
2
.
NAS
wo
r
k
f
lo
w
f
o
r
E
f
f
icien
tNetB
0
o
p
tim
izatio
n
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
is
s
ec
tio
n
d
is
cu
s
s
es
th
e
ex
p
er
im
en
tal
o
u
tc
o
m
es
o
f
th
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
m
o
d
el,
an
d
v
alid
ates
its
p
er
f
o
r
m
an
ce
with
two
b
en
c
h
m
ar
k
ed
p
u
b
licly
a
v
ailab
le
d
a
tasets
n
am
ely
C
DD
an
d
DM
D
.
T
h
e
d
ev
el
o
p
m
en
t
an
d
p
er
f
o
r
m
a
n
ce
ass
ess
m
en
t
o
f
th
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
is
ca
r
r
ied
o
u
t
u
s
in
g
p
y
t
h
o
n
wit
h
GPU
s
u
p
p
o
r
t
o
n
win
d
o
ws
1
1
6
4
-
b
it
m
ac
h
in
e.
T
h
e
v
alid
atio
n
o
f
m
o
d
el
is
d
o
n
e
co
n
s
id
er
in
g
class
i
f
icatio
n
ac
cu
r
ac
y
,
co
m
p
u
tatio
n
al
ef
f
icien
c
y
,
m
o
d
el
in
ter
p
r
etab
ilit
y
,
an
d
co
m
p
a
r
is
o
n
with
b
aselin
e
ap
p
r
o
ac
h
es.
Fig
u
r
e
3
p
r
o
v
id
es
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
4
6
4
-
1
4
7
2
1468
a
co
m
p
ar
ativ
e
v
is
u
aliza
tio
n
o
f
th
e
tr
ain
in
g
an
d
v
alid
atio
n
lo
s
s
cu
r
v
es
f
o
r
th
e
L
E
AD
-
Net
m
o
d
el
o
n
b
o
th
C
DD
an
d
DM
D
d
ataset
.
T
h
e
lo
s
s
cu
r
v
e
illu
s
tr
ates
th
e
m
o
d
el
’
s
lear
n
in
g
d
y
n
am
ics
o
v
er
3
0
tr
ain
in
g
ep
o
c
h
s
,
th
at
s
h
o
ws
th
e
m
o
d
el
is
o
p
tim
izin
g
b
y
m
in
im
izin
g
er
r
o
r
d
u
r
in
g
its
tr
ain
in
g
p
h
ase.
B
ase
o
n
th
e
an
aly
s
is
o
f
th
e
Fig
u
r
e
3
(
a)
,
it
ca
n
b
e
s
ee
n
th
at
th
e
lo
s
s
cu
r
v
e
o
f
t
h
e
m
o
d
el
o
n
th
e
C
DD
ex
h
ib
its
a
s
tab
le
tr
en
d
th
at
s
u
g
g
est
ef
f
ec
tiv
e
co
n
v
e
r
g
en
ce
with
g
o
o
d
g
en
e
r
aliza
tio
n
o
n
th
e
v
ali
d
atio
n
s
et.
On
th
e
o
th
er
h
an
d
,
Fig
u
r
e
3
(
b
)
s
h
o
ws
s
lig
h
t
d
iv
er
g
en
ce
b
etwe
en
t
h
e
v
alid
atio
n
an
d
tr
ain
i
n
g
lo
s
s
es
o
n
DM
D
d
ataset,
wh
ich
m
ay
b
e
d
u
e
t
o
th
e
s
m
aller
d
ata
s
am
p
les
an
d
clas
s
im
b
alan
ce
is
s
u
es
a
s
s
o
ciate
d
with
DM
D.
Ho
wev
er
,
b
o
th
cu
r
v
es
p
latea
u
b
y
th
e
f
in
al
ep
o
ch
s
,
t
h
at
d
em
o
n
s
tr
ates stab
le
co
n
v
er
g
en
ce
ac
h
iev
ed
b
y
th
e
L
E
AD
-
Net
m
o
d
el.
Fig
u
r
e
4
illu
s
tr
ates
th
e
co
n
f
u
s
i
o
n
m
atr
ices
o
f
th
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
m
o
d
el
o
n
th
e
C
DD
an
d
DM
D
d
atasets
,
r
esp
ec
tiv
ely
.
T
h
e
co
n
f
u
s
io
n
m
atr
ix
in
Fig
u
r
e
4
(
a)
p
r
o
v
id
e
s
class
-
wis
e
p
r
ed
ictio
n
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
to
th
e
C
DD
d
ataset.
I
t
ca
n
b
e
s
ee
n
th
at
th
e
m
o
d
el
h
as
ac
h
iev
ed
h
ig
h
tr
u
e
p
o
s
itiv
e
co
u
n
ts
f
o
r
m
u
ltip
le
class
es
s
u
ch
as
n
o
n
_
d
am
ag
e_
b
u
ild
in
g
s
_
s
tr
ee
t,
n
o
n
_
d
am
ag
e
_
wild
life
_
f
o
r
est,
a
n
d
n
o
n
_
d
am
ag
e
_
s
ea
,
d
em
o
n
s
tr
atin
g
th
e
m
o
d
el’
s
ab
ilit
y
to
d
is
tin
g
u
is
h
d
is
aste
r
ty
p
es
f
r
o
m
s
im
ilar
v
is
u
al
co
n
tex
ts
.
On
th
e
DM
D
d
ataset
in
Fig
u
r
e
4
(
b
)
,
th
e
co
n
f
u
s
io
n
m
atr
ix
s
h
o
ws
r
e
lativ
ely
s
tr
o
n
g
p
er
f
o
r
m
an
ce
f
o
r
th
e
n
o
n
_
d
am
ag
e
an
d
d
am
ag
ed
_
in
f
r
astru
ct
u
r
e
class
es,
b
o
th
o
f
wh
ich
h
av
e
h
ig
h
tr
u
e
p
o
s
itiv
e
r
ates
(
T
PR
)
.
Ho
wev
er
,
m
o
r
e
co
n
f
u
s
io
n
is
o
b
s
er
v
ed
am
o
n
g
v
is
u
ally
s
im
ilar
class
e
s
s
u
ch
as
d
am
ag
ed
_
n
atu
r
e,
f
ir
e
s
,
an
d
f
lo
o
d
,
lik
el
y
d
u
e
to
o
v
er
lap
p
i
n
g
f
ea
tu
r
es
in
r
ea
l
-
wo
r
ld
s
ce
n
ar
io
s
a
n
d
d
u
e
to
lack
o
f
lar
g
e
d
ata
s
am
p
les
i
n
DM
D.
Ho
wev
e
r
,
L
E
AD
-
Net
m
ain
tain
s
a
h
ig
h
lev
el
o
f
ac
cu
r
ac
y
ac
r
o
s
s
m
o
s
t
ca
teg
o
r
ies,
with
p
ar
ticu
lar
ly
s
tr
o
n
g
g
e
n
er
aliza
tio
n
in
id
en
tify
in
g
n
o
n
_
d
am
ag
e
c
o
n
ten
t,
wh
ich
is
cr
itical
f
o
r
f
ilter
in
g
ir
r
elev
a
n
t
d
ata
i
n
e
m
er
g
en
cy
r
esp
o
n
s
e
p
ip
elin
es.
T
ab
le
1
p
r
o
v
id
es
q
u
an
titativ
e
an
aly
s
is
o
f
th
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
o
n
C
DD
d
ataset
co
n
s
id
er
in
g
p
r
ec
is
io
n
,
r
ec
all
,
an
d
F1
-
s
co
r
e
.
(a
)
(b
)
Fig
u
r
e
3
.
T
r
ain
in
g
a
n
d
v
alid
atio
n
lo
s
s
cu
r
v
es o
f
th
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
m
o
d
el
o
n
(
a)
C
D
D
d
ataset
an
d
(
b
)
DM
D
d
atasets
(
a)
(
b
)
Fig
u
r
e
4
.
T
h
e
co
n
f
u
s
io
n
m
atr
ix
o
f
th
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
m
o
d
el
o
n
(
a)
C
DD
d
atasets
an
d
(
b
)
DM
D
d
atasets
T
h
e
q
u
a
n
tifie
d
o
u
tc
o
m
e
d
e
m
o
n
s
tr
ated
in
T
ab
le
1
r
ev
ea
ls
th
at
th
e
m
o
d
el
h
as
ac
h
iev
ed
an
o
v
er
all
class
if
icatio
n
ac
cu
r
ac
y
with
9
6
%,
th
at
in
d
icate
s
s
tr
o
n
g
p
e
r
f
o
r
m
a
n
ce
ac
r
o
s
s
all
th
e
1
2
c
lass
es.
I
n
ter
m
s
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
xp
la
in
a
b
le
s
o
ci
a
l m
ed
ia
d
is
a
s
ter ima
g
e
cla
s
s
ifica
tio
n
u
s
in
g
a
lig
h
tw
eig
h
t …
(
R
a
s
h
mi
K
a
n
g
o
ka
r
Ta
r
a
n
a
th
)
1469
p
r
ec
is
io
n
an
d
r
ec
all,
a
co
n
s
id
er
ab
le
p
er
f
o
r
m
a
n
ce
is
s
h
o
wn
b
y
th
e
m
o
d
el
f
o
r
all
m
o
s
t
ev
er
y
class
ex
ce
p
t
ea
r
th
q
u
ak
e
d
u
e
to
v
er
y
less
in
s
tan
ce
s
in
th
e
tes
t
d
ataset.
T
h
e
class
h
u
m
an
_
d
am
ag
e
ac
h
iev
e
d
F1
-
s
co
r
e
o
f
0
.
9
4
,
an
d
wate
r
_
d
is
aster
an
d
wild
_
f
ir
e
b
o
t
h
ex
ce
e
d
ed
0
.
9
3
,
th
e
r
e
b
y
s
h
o
win
g
s
tr
o
n
g
class
if
icatio
n
ca
p
a
b
ilit
y
ev
e
n
f
o
r
d
is
aster
ca
teg
o
r
ies.
Ov
er
al
l
o
u
tco
m
e
d
em
o
n
s
tr
ate
th
at
L
E
AD
-
Net
p
er
f
o
r
m
s
r
eliab
ly
a
cr
o
s
s
m
o
s
t
d
is
aster
class
es in
C
DD
d
ataset,
with
e
x
ce
p
tio
n
al
s
tr
en
g
t
h
in
h
i
g
h
-
s
u
p
p
o
r
t c
ateg
o
r
ies an
d
n
o
n
-
d
am
ag
e
d
etec
tio
n
.
T
ab
le
2
p
r
esen
ts
q
u
an
tifie
d
o
u
tco
m
e
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
o
n
DM
D
d
ataset
wh
ich
is
co
n
s
id
er
ed
to
b
e
m
o
r
e
ch
allen
g
in
g
d
u
e
s
m
a
ller
s
am
p
le
s
ize,
clas
s
im
b
alan
ce
,
an
d
g
r
ea
ter
v
is
u
al
am
b
ig
u
ity
am
o
n
g
class
es.
T
h
e
an
aly
s
is
o
f
th
e
n
u
m
er
ical
o
u
tco
m
e
in
T
ab
le
2
s
h
o
ws
th
at
o
n
DM
D
d
ataset
th
e
m
o
d
el
h
as
ac
h
iev
ed
an
o
v
er
all
ac
cu
r
ac
y
o
f
8
7
%,
with
s
tr
o
n
g
p
er
f
o
r
m
an
ce
p
a
r
ticu
lar
ly
in
th
e
n
o
n
_
d
a
m
ag
e
(
F1
-
s
co
r
e:
0
.
9
4
)
,
h
u
m
an
_
d
am
ag
e
(
F1
-
s
co
r
e:
0
.
8
7
)
,
a
n
d
f
ir
es
(
F1
-
s
co
r
e:
0
.
8
6
)
class
es,
ex
ce
p
t
d
am
ag
e
d
_
n
atu
r
e
class
with
F1
-
s
co
r
e
0
.
6
5
,
s
u
g
g
ests
th
e
m
o
d
el
f
ac
ed
d
if
f
icu
lty
d
is
tin
g
u
i
s
h
in
g
it
f
r
o
m
o
th
e
r
d
am
ag
e
-
r
e
lated
class
es,
lik
ely
d
u
e
to
o
v
er
lap
p
in
g
v
is
u
al
c
h
a
r
ac
ter
is
tics
in
d
is
aster
im
ag
er
y
.
T
h
is
lev
el
o
f
p
er
f
o
r
m
an
ce
i
s
n
o
tewo
r
th
y
g
iv
en
th
e
n
atu
r
e
o
f
th
e
d
ataset.
I
n
t
h
e
ex
is
tin
g
liter
atu
r
e,
m
o
d
els
tr
ain
ed
o
n
im
a
g
e
-
o
n
ly
d
is
aster
d
atasets
ty
p
ically
r
ep
o
r
t
class
if
icatio
n
ac
cu
r
ac
i
es
b
elo
w
9
0
%.
Acc
u
r
ac
y
lev
els
ab
o
v
e
9
0
%
ar
e
u
s
u
ally
a
ch
iev
ed
o
n
ly
wh
en
tex
tu
al
m
etad
ata
(
e.
g
.
,
twee
ts
o
r
ca
p
tio
n
s
)
is
co
m
b
in
ed
with
im
ag
es
in
a
m
u
ltimo
d
al
s
etu
p
.
T
h
er
ef
o
r
e,
ac
h
iev
in
g
8
7
%
ac
cu
r
ac
y
u
s
in
g
im
ag
e
d
ata
alo
n
e
h
ig
h
lig
h
ts
th
e
r
o
b
u
s
tn
ess
o
f
th
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
m
o
d
el,
ev
en
u
n
d
er
d
ata
co
n
s
tr
ain
ts
an
d
r
ea
l
-
wo
r
ld
v
ar
iab
ilit
y
,
wh
ich
is
f
u
r
th
er
ev
id
en
t
b
y
r
ec
eiv
er
o
p
er
atin
g
ch
ar
ac
ter
is
tic
(
R
OC
)
cu
r
v
e
in
Fig
u
r
e
5
.
T
ab
le
1
.
C
lass
if
icatio
n
r
ep
o
r
t
o
n
C
DD
d
ataset
C
l
a
s
s
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
s
c
o
r
e
S
u
p
p
o
r
t
d
r
o
u
g
h
t
1
.
0
0
0
.
9
5
0
.
9
8
21
e
a
r
t
h
q
u
a
k
e
0
.
1
0
.
1
0
.
1
4
h
u
m
a
n
_
d
a
m
a
g
e
0
.
8
9
1
.
0
0
0
.
9
4
24
i
n
f
r
a
s
t
r
u
c
t
u
r
e
0
.
9
3
0
.
9
0
0
.
9
1
1
4
3
l
a
n
d
_
sl
i
d
e
0
.
6
4
0
.
7
8
0
.
7
1
46
n
o
n
_
d
a
ma
g
e
_
b
u
i
l
d
i
n
g
s_
st
r
e
e
t
1
.
0
0
0
.
9
9
0
.
9
9
4
5
8
n
o
n
_
d
a
ma
g
e
_
h
u
m
a
n
1
.
0
0
0
.
8
3
0
.
9
1
12
n
o
n
_
d
a
ma
g
e
_
se
a
0
.
9
8
0
.
9
9
0
.
9
9
2
2
9
n
o
n
_
d
a
ma
g
e
_
w
i
l
d
l
i
f
e
_
f
o
r
e
s
t
1
.
0
0
1
.
0
0
1
.
0
0
2
2
8
u
r
b
a
n
_
f
i
r
e
0
.
8
5
0
.
7
7
0
.
8
0
43
w
a
t
e
r
_
d
i
sas
t
e
r
0
.
9
4
0
.
9
2
0
.
9
3
1
0
4
w
i
l
d
_
f
i
r
e
0
.
9
1
0
.
9
4
0
.
9
3
53
O
v
e
r
a
l
l
t
e
s
t
a
c
c
u
r
a
c
y
=
9
6
%
1
,
3
6
5
T
ab
le
2
.
C
lass
if
icatio
n
r
ep
o
r
t
o
n
DM
D
d
ataset
C
l
a
s
s
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
s
c
o
r
e
S
u
p
p
o
r
t
d
a
m
a
g
e
d
_
i
n
f
r
a
s
t
r
u
c
t
u
r
e
0
.
8
4
0
.
8
6
0
.
8
5
2
1
3
d
a
m
a
g
e
d
_
n
a
t
u
r
e
0
.
6
8
0
.
6
3
0
.
6
5
78
f
i
r
e
s
0
.
8
7
0
.
8
5
0
.
8
6
53
f
l
o
o
d
0
.
7
6
0
.
7
6
0
.
7
6
58
h
u
m
a
n
_
d
a
m
a
g
e
0
.
9
1
0
.
8
3
0
.
8
7
36
n
o
n
_
d
a
ma
g
e
0
.
9
3
0
.
9
4
0
.
9
4
4
4
6
O
v
e
r
a
l
l
t
e
s
t
a
c
c
u
r
a
c
y
=
8
7
%
8
8
4
T
h
e
R
OC
cu
r
v
es
ar
e
b
asica
lly
a
th
r
esh
o
ld
-
in
d
ep
e
n
d
en
t
ass
ess
m
en
t
m
etr
ic
th
at
p
lo
ts
th
e
T
P
R
ag
ain
s
t
th
e
f
alse
p
o
s
itiv
e
r
ate
(
FP
R
)
at
v
ar
io
u
s
class
if
icatio
n
th
r
esh
o
ld
s
.
I
t
o
f
f
e
r
s
a
m
o
r
e
c
o
m
p
let
e
v
iew
o
f
a
m
o
d
el’
s
ab
ilit
y
to
d
is
tin
g
u
is
h
b
etwe
en
class
es
th
an
a
s
in
g
le
-
p
o
i
n
t
m
e
tr
ic
lik
e
ac
cu
r
ac
y
o
r
F1
-
s
co
r
e.
As
it
ca
n
b
e
s
ee
n
th
at
th
e
p
r
o
p
o
s
ed
m
o
d
el
d
em
o
n
s
tr
ated
d
is
cr
im
in
ativ
e
ca
p
a
city
o
f
its
p
r
ed
ictio
n
b
y
ac
h
ie
v
in
g
ar
ea
u
n
d
er
th
e
cu
r
v
e
(
AUC)
s
co
r
es a
b
o
v
e
0
.
9
7
f
o
r
m
o
s
t o
f
th
e
class
es o
n
th
e
b
o
th
d
atasets
.
On
th
e
C
DD
d
ata
s
et
in
F
ig
u
r
e
5
(
a
)
,
m
o
s
t
class
es
ac
h
iev
ed
p
er
f
ec
t
o
r
n
ea
r
-
p
er
f
ec
t
AUCs
o
f
1
.
0
0
,
an
d
th
e
o
th
e
r
class
es
s
u
ch
as
ea
r
th
q
u
ak
e
an
d
u
r
b
an
_
f
ir
e
s
h
o
ws
AUC
o
f
0
.
9
7
an
d
0
.
9
8
,
r
esp
ec
tiv
ely
.
I
n
ter
esti
n
g
ly
,
wh
ile
th
e
clas
s
if
icatio
n
ac
cu
r
ac
y
o
n
th
e
DM
D
d
ataset
i
s
s
lig
h
tly
lo
wer
,
th
e
R
OC
cu
r
v
es
in
Fig
u
r
e
5
(
b
)
r
ev
ea
l
th
at
L
E
AD
-
Net
ac
h
iev
es
s
tr
o
n
g
AUC
s
co
r
es
r
an
g
in
g
f
r
o
m
0
.
9
3
to
0
.
9
9
,
wh
ich
s
h
o
ws
s
tr
o
n
g
ev
id
en
ce
t
h
at
th
e
L
E
AD
-
Net
m
o
d
el
ef
f
ec
tiv
el
y
d
if
f
er
en
tiates
b
etwe
en
d
is
aster
ca
teg
o
r
ies,
r
ein
f
o
r
cin
g
th
e
v
alid
ity
o
f
th
e
class
if
icatio
n
r
esu
lts
p
r
esen
t
ed
ea
r
lier
.
T
h
e
n
ex
t
an
aly
s
i
s
in
T
ab
le
3
is
ca
r
r
ied
o
u
t
f
o
r
ev
alu
atin
g
th
e
ex
p
lain
ab
ilit
y
o
f
th
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
m
o
d
el.
T
ab
le
3
illu
s
tr
ates
th
e
ex
p
lai
n
ab
ilit
y
o
f
L
E
AD
-
Net
u
s
in
g
Gr
ad
-
C
AM
v
is
u
aliza
tio
n
s
o
n
s
elec
ted
s
am
p
les
f
r
o
m
th
e
C
DD
an
d
DM
D
d
ataset
s
.
T
h
e
ac
tiv
atio
n
m
ap
s
clea
r
ly
s
h
o
w
th
at
th
e
m
o
d
el
f
o
cu
s
es
o
n
co
n
tex
tu
ally
r
elev
an
t
r
eg
i
o
n
s
s
u
ch
as
f
lam
es
in
wild
f
ir
e
s
ce
n
es,
d
a
m
ag
ed
in
f
r
astru
ctu
r
e,
o
r
af
f
ec
ted
in
d
iv
id
u
als
wh
e
n
m
a
k
in
g
its
p
r
ed
ictio
n
s
.
T
h
is
in
d
icate
s
th
at
th
e
m
o
d
el
is
lear
n
in
g
s
em
a
n
tically
m
ea
n
in
g
f
u
l
f
ea
tu
r
es r
ath
er
th
a
n
r
ely
i
n
g
o
n
ir
r
elev
an
t b
ac
k
g
r
o
u
n
d
n
o
is
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
4
6
4
-
1
4
7
2
1470
(
a)
(
b)
Fig
u
r
e
5
.
T
h
e
R
OC
cu
r
v
e
o
f
t
h
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
m
o
d
e
l o
n
(
a)
C
DD
d
atasets
an
d
(
b
)
DM
D
d
atasets
T
ab
le
3
.
Gr
a
d
-
C
AM
b
ased
m
o
d
el
ex
p
lain
ab
ilit
y
o
n
C
DD
an
d
DM
D
d
atasets
C
D
D
D
M
D
Te
st
i
m
a
g
e
s
A
c
t
i
v
a
t
i
o
n
ma
p
Te
st
i
m
a
g
e
s
A
c
t
i
v
a
t
i
o
n
ma
p
T
ab
le
4
p
r
o
v
id
es
a
co
m
p
ar
ativ
e
an
aly
s
is
o
f
th
e
p
r
o
p
o
s
ed
L
E
AD
-
Net
m
o
d
el
ag
ain
s
t
s
ev
er
al
well
-
estab
lis
h
ed
C
NN
ar
ch
itectu
r
es
s
u
ch
as
I
n
ce
p
tio
n
,
Den
s
eNe
t1
2
1
,
R
esNet
-
5
0
,
VGG
-
1
6
,
an
d
Alex
Net
b
ased
o
n
th
eir
p
er
f
o
r
m
a
n
ce
o
n
th
e
C
DD
an
d
DM
D
d
atasets
,
as
well
as
th
eir
m
o
d
el
c
o
m
p
lex
ity
m
ea
s
u
r
e
d
b
y
th
e
n
u
m
b
er
o
f
p
ar
am
ete
r
s
.
On
th
e
C
DD
d
ataset,
L
E
AD
-
Net
ac
h
iev
ed
th
e
h
ig
h
est
ac
cu
r
ac
y
o
f
9
6
%,
o
u
tp
er
f
o
r
m
in
g
all
b
aselin
e
m
o
d
els,
in
clu
d
in
g
De
n
s
eNe
t
(
9
3
%)
an
d
R
esNet
(
9
1
%).
Similar
ly
,
o
n
th
e
m
o
r
e
ch
allen
g
in
g
DM
D
d
ataset,
L
E
AD
-
Net
m
ain
tain
ed
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
with
a
n
ac
cu
r
ac
y
o
f
8
7
%,
co
m
p
ar
ed
to
8
4
%
f
o
r
b
o
th
Den
s
eNe
t
a
n
d
R
esNet,
an
d
s
ig
n
if
ican
tly
h
ig
h
er
th
an
VGG
(
7
3
%)
a
n
d
Alex
Net
(
7
2
%).
I
n
a
d
d
i
t
i
o
n
t
o
i
t
s
s
t
r
o
n
g
p
r
e
d
i
ct
i
v
e
p
e
r
f
o
r
m
a
n
c
e
,
L
E
A
D
-
N
et
i
s
a
ls
o
c
o
m
p
u
t
a
t
i
o
n
al
l
y
e
f
f
ic
i
e
n
t
m
o
d
e
l
,
w
i
t
h
a
p
p
r
o
x
i
m
a
t
e
l
y
6
m
i
ll
i
o
n
p
a
r
am
e
t
e
r
s
t
h
a
n
o
t
h
e
r
m
o
d
e
ls
,
t
h
e
r
e
b
y
m
a
k
i
n
g
i
t
m
u
c
h
li
g
h
t
e
r
t
o
b
e
d
e
p
l
o
y
e
d
i
n
t
h
e
r
e
a
l
-
w
o
r
l
d
s
c
e
n
a
r
i
o
s
.
T
h
e
a
b
o
v
e
o
u
t
c
o
m
e
s
u
g
g
e
s
t
s
t
h
a
t
L
E
A
D
-
N
e
t
s
u
c
c
ess
f
u
l
l
y
b
a
l
a
n
ce
s
h
i
g
h
c
l
a
s
s
i
f
ic
a
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
w
i
t
h
l
o
w
m
o
d
e
l
co
m
p
l
e
x
i
t
y
f
o
r
i
d
e
n
t
i
f
y
i
n
g
d
i
s
a
s
t
e
r
e
v
e
n
t
s
f
r
o
m
t
h
e
s
o
c
i
al
m
e
d
ia
p
o
s
t
s
.
T
ab
le
4
.
C
o
m
p
a
r
ativ
e
an
aly
s
is
with
b
aselin
e
DL
m
o
d
els
M
o
d
e
l
C
o
m
p
l
e
x
i
t
y
A
c
c
u
r
a
c
y
C
D
D
(
%)
D
F
D
(
%)
I
n
c
e
p
t
i
o
n
~
2
3
mi
l
l
i
o
n
89
82
D
e
n
seN
e
t
-
1
2
1
~
8
m
i
l
l
i
o
n
93
84
R
e
sN
e
t
-
50
~
2
5
mi
l
l
i
o
n
91
84
VGG
-
16
~
1
3
8
m
i
l
l
i
o
n
88
73
A
l
e
x
N
e
t
~
6
0
mi
l
l
i
o
n
84
72
P
r
o
p
o
se
d
(
L
EA
D
-
N
e
t
)
~
6
m
i
l
l
i
o
n
96
87
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
xp
la
in
a
b
le
s
o
ci
a
l m
ed
ia
d
is
a
s
ter ima
g
e
cla
s
s
ifica
tio
n
u
s
in
g
a
lig
h
tw
eig
h
t …
(
R
a
s
h
mi
K
a
n
g
o
ka
r
Ta
r
a
n
a
th
)
1471
4.
CO
NCLU
SI
O
N
T
h
is
p
a
p
e
r
p
r
o
p
o
s
e
d
L
E
AD
-
N
et,
a
DL
m
o
d
el
d
esi
g
n
e
d
f
o
r
e
f
f
e
cti
v
e
a
n
d
i
n
t
er
p
r
et
a
b
le
cl
ass
i
f
ic
ati
o
n
o
f
d
is
as
te
r
-
r
elat
e
d
i
m
a
g
es
f
r
o
m
s
o
c
ial
m
ed
ia
p
o
s
ts
.
T
h
e
p
r
o
p
o
s
ed
L
E
AD
-
Ne
t
is
d
ev
el
o
p
e
d
b
as
ed
o
n
th
e
in
t
eg
r
ati
o
n
o
f
E
f
f
ici
e
n
tN
etB
0
C
NN
a
r
c
h
it
ec
tu
r
e
a
n
d
m
o
d
i
f
i
ed
SE
att
e
n
ti
o
n
m
ec
h
an
is
m
to
ac
h
i
ev
e
a
f
o
cu
s
e
d
lea
r
n
i
n
g
o
f
c
o
m
p
le
x
f
ea
t
u
r
es
f
r
o
m
d
is
ast
er
i
m
a
g
es
.
T
o
m
a
k
e
m
o
d
el
tr
ai
n
i
n
g
m
o
r
e
r
o
b
u
s
t
,
it
i
s
h
y
p
er
p
a
r
a
m
e
te
r
s
ar
e
t
u
n
e
d
a
n
d
o
p
ti
m
i
ze
d
v
i
a
a
N
AS
-
lit
e
,
t
h
e
r
e
b
y
e
n
a
b
l
in
g
a
n
o
p
ti
m
a
l
b
a
la
n
c
e
b
et
wee
n
ac
c
u
r
ac
y
a
n
d
co
m
p
u
tati
o
n
a
l
ef
f
i
ci
en
cy
.
T
h
e
m
o
d
el
was
ev
al
u
at
e
d
o
n
t
wo
b
e
n
c
h
m
a
r
k
d
at
ase
ts
,
C
D
D
a
n
d
DM
D
,
u
s
i
n
g
ac
c
u
r
a
cy
,
F
1
-
s
c
o
r
e
,
R
OC
-
AU
C
,
a
n
d
m
o
d
el
co
m
p
l
ex
it
y
.
T
h
e
m
o
d
el
a
ch
ie
v
e
d
9
6
%
ac
c
u
r
ac
y
o
n
th
e
C
DD
d
at
ase
t
an
d
8
7
%
o
n
t
h
e
m
o
r
e
ch
all
en
g
i
n
g
DM
D
d
at
aset
,
a
n
d
o
u
t
p
e
r
f
o
r
m
s
s
e
v
e
r
al
b
as
eli
n
e
C
NNs
i
n
te
r
m
s
o
f
b
o
t
h
cl
ass
if
ic
ati
o
n
p
er
f
o
r
m
a
n
c
e
a
n
d
m
o
d
el
c
o
m
p
le
x
it
y
.
T
h
e
in
t
eg
r
ati
o
n
o
f
G
r
a
d
-
C
AM
f
u
r
t
h
e
r
p
r
o
v
i
d
e
d
v
is
u
al
ex
p
l
an
ati
o
n
s
o
f
th
e
m
o
d
el'
s
d
ec
is
i
o
n
s
,
c
o
n
f
ir
m
i
n
g
its
att
e
n
ti
o
n
t
o
d
is
ast
er
-
r
el
e
v
a
n
t
r
eg
io
n
s
i
n
t
h
e
i
m
a
g
es
an
d
en
h
a
n
ci
n
g
m
o
d
e
l
t
r
an
s
p
a
r
en
c
y
.
F
u
t
u
r
e
wo
r
k
m
a
y
e
x
p
l
o
r
e
t
h
e
i
n
c
o
r
p
o
r
at
io
n
o
f
m
u
l
tim
o
d
al
f
e
at
u
r
es,
s
u
c
h
as
tex
t
an
d
g
e
o
l
o
c
ati
o
n
m
et
a
d
at
a
to
e
n
h
a
n
c
e
t
h
e
m
o
d
el'
s
p
r
ac
t
ic
al
u
t
ilit
y
i
n
em
er
g
en
cy
r
es
p
o
n
s
e
a
p
p
li
ca
ti
o
n
s
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
R
ash
m
i K
an
g
o
k
ar
T
ar
an
ath
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Gee
ta
C
h
id
an
an
d
ap
p
a
Ma
r
a
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
av
aila
b
le
o
n
r
eq
u
est
f
r
o
m
th
e
co
r
r
esp
o
n
d
in
g
au
th
o
r
,
[
R
KT
]
.
RE
F
E
R
E
NC
E
S
[
1
]
M
.
K
r
i
c
h
e
n
,
M
.
S
.
A
b
d
a
l
z
a
h
e
r
,
M
.
El
w
e
k
e
i
l
,
a
n
d
M
.
M
.
F
o
u
d
a
,
“
M
a
n
a
g
i
n
g
n
a
t
u
r
a
l
d
i
sas
t
e
r
s:
A
n
a
n
a
l
y
s
i
s
o
f
t
e
c
h
n
o
l
o
g
i
c
a
l
a
d
v
a
n
c
e
me
n
t
s
,
o
p
p
o
r
t
u
n
i
t
i
e
s
,
a
n
d
c
h
a
l
l
e
n
g
e
s
,
”
I
n
t
e
r
n
e
t
o
f
T
h
i
n
g
s
a
n
d
C
y
b
e
r
-
Ph
y
si
c
a
l
S
y
st
e
m
s
,
v
o
l
.
4
,
p
p
.
9
9
–
1
0
9
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
o
t
c
p
s.
2
0
2
3
.
0
9
.
0
0
2
.
[
2
]
V
.
Jay
a
w
a
r
d
e
n
e
,
T.
J.
H
u
g
g
i
n
s,
R
.
P
r
a
san
n
a
,
a
n
d
B
.
F
a
k
h
r
u
d
d
i
n
,
“
T
h
e
r
o
l
e
o
f
d
a
t
a
a
n
d
i
n
f
o
r
m
a
t
i
o
n
q
u
a
l
i
t
y
d
u
r
i
n
g
d
i
s
a
st
e
r
r
e
sp
o
n
s
e
d
e
c
i
s
i
o
n
-
ma
k
i
n
g
,
”
Pr
o
g
r
e
ss
i
n
D
i
s
a
st
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
2
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
d
i
s
a
s.
2
0
2
1
.
1
0
0
2
0
2
.
[
3
]
Z.
E.
L.
K
h
a
l
e
d
a
n
d
H
.
M
c
h
e
i
c
k
,
“
C
a
se
st
u
d
i
e
s
o
f
c
o
mm
u
n
i
c
a
t
i
o
n
s
s
y
st
e
m
s
d
u
r
i
n
g
h
a
r
sh
e
n
v
i
r
o
n
me
n
t
s
:
A
r
e
v
i
e
w
o
f
a
p
p
r
o
a
c
h
e
s,
w
e
a
k
n
e
s
ses,
a
n
d
l
i
mi
t
a
t
i
o
n
s
t
o
i
m
p
r
o
v
e
q
u
a
l
i
t
y
o
f
ser
v
i
c
e
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
D
i
s
t
ri
b
u
t
e
d
S
e
n
s
o
r
N
e
t
w
o
r
k
s
,
v
o
l
.
1
5
,
n
o
.
2
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
7
7
/
1
5
5
0
1
4
7
7
1
9
8
2
9
9
6
0
.
[
4
]
K
.
S
e
n
e
v
i
r
a
t
n
e
,
M
.
N
a
d
e
e
sh
a
n
i
,
S
.
S
e
n
a
r
a
t
n
e
,
a
n
d
S
.
P
e
r
e
r
a
,
“
U
se
o
f
s
o
c
i
a
l
m
e
d
i
a
i
n
d
i
sas
t
e
r
ma
n
a
g
e
me
n
t
:
c
h
a
l
l
e
n
g
e
s
a
n
d
st
r
a
t
e
g
i
e
s
,
”
S
u
st
a
i
n
a
b
i
l
i
t
y
,
v
o
l
.
1
6
,
n
o
.
1
1
,
J
u
n
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
s
u
1
6
1
1
4
8
2
4
.
[
5
]
L.
B
e
l
c
a
st
r
o
e
t
a
l
.
,
“
U
si
n
g
s
o
c
i
a
l
me
d
i
a
f
o
r
su
b
-
e
v
e
n
t
d
e
t
e
c
t
i
o
n
d
u
r
i
n
g
d
i
s
a
st
e
r
s
,
”
J
o
u
rn
a
l
o
f
Bi
g
D
a
t
a
,
v
o
l
.
8
,
n
o
.
1
,
D
e
c
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
8
6
/
s
4
0
5
3
7
-
021
-
0
0
4
6
7
-
1.
[
6
]
A
.
S
.
A
l
b
a
h
r
i
e
t
a
l
.
,
“
A
s
y
s
t
e
m
a
t
i
c
r
e
v
i
e
w
o
f
t
r
u
s
t
w
o
r
t
h
y
a
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
a
p
p
l
i
c
a
t
i
o
n
s
i
n
n
a
t
u
r
a
l
d
i
sas
t
e
r
s,
”
C
o
m
p
u
t
e
rs
a
n
d
El
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
,
v
o
l
.
1
1
8
,
S
e
p
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
o
m
p
e
l
e
c
e
n
g
.
2
0
2
4
.
1
0
9
4
0
9
.
[
7
]
A
.
A
k
h
y
a
r
e
t
a
l
.
,
“
D
e
e
p
a
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
a
p
p
l
i
c
a
t
i
o
n
s
f
o
r
n
a
t
u
r
a
l
d
i
s
a
st
e
r
ma
n
a
g
e
me
n
t
s
y
st
e
ms:
a
met
h
o
d
o
l
o
g
i
c
a
l
r
e
v
i
e
w
,
”
Ec
o
l
o
g
i
c
a
l
I
n
d
i
c
a
t
o
rs
,
v
o
l
.
1
6
3
,
J
u
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
c
o
l
i
n
d
.
2
0
2
4
.
1
1
2
0
6
7
.
[
8
]
D
.
Y
u
a
n
d
Z.
H
e
,
“
D
i
g
i
t
a
l
t
w
i
n
-
d
r
i
v
e
n
i
n
t
e
l
l
i
g
e
n
c
e
d
i
s
a
st
e
r
p
r
e
v
e
n
t
i
o
n
a
n
d
mi
t
i
g
a
t
i
o
n
f
o
r
i
n
f
r
a
st
r
u
c
t
u
r
e
:
a
d
v
a
n
c
e
s,
c
h
a
l
l
e
n
g
e
s
,
a
n
d
o
p
p
o
r
t
u
n
i
t
i
e
s,
”
N
a
t
u
ra
l
H
a
za
r
d
s
,
v
o
l
.
1
1
2
,
n
o
.
1
,
p
p
.
1
–
3
6
,
M
a
y
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
6
9
-
0
2
1
-
0
5
1
9
0
-
x.
[
9
]
M
.
H
a
i
d
a
r
h
,
C
.
M
u
,
Y
.
Li
u
,
a
n
d
X
.
H
e
,
“
Ex
p
l
o
r
i
n
g
t
r
a
d
i
t
i
o
n
a
l
,
d
e
e
p
l
e
a
r
n
i
n
g
a
n
d
h
y
b
r
i
d
m
e
t
h
o
d
s
f
o
r
h
y
p
e
r
s
p
e
c
t
r
a
l
i
ma
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
:
a
r
e
v
i
e
w
,
”
J
o
u
r
n
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
a
n
d
I
n
t
e
l
l
i
g
e
n
c
e
,
A
p
r
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
i
i
x
d
.
2
0
2
5
.
0
4
.
0
0
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
4
6
4
-
1
4
7
2
1472
[
1
0
]
K
.
A
b
r
a
h
a
m
,
M
.
A
b
d
e
l
w
a
h
a
b
,
a
n
d
M
.
A
.
-
Z
a
h
h
a
d
,
“
C
l
a
s
s
i
f
i
c
a
t
i
o
n
a
n
d
d
e
t
e
c
t
i
o
n
o
f
n
a
t
u
r
a
l
d
i
s
a
s
t
e
r
s
u
s
i
n
g
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
:
a
r
e
v
i
e
w
,
”
E
a
r
t
h
S
c
i
e
n
c
e
I
n
f
o
r
m
a
t
i
c
s
,
v
o
l
.
1
7
,
n
o
.
2
,
p
p
.
8
6
9
–
8
9
1
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
2
1
4
5
-
023
-
0
1
2
0
5
-
2.
[
1
1
]
L.
A
h
me
d
,
K
.
A
h
m
a
d
,
N
.
S
a
i
d
,
B
.
Q
o
l
o
m
a
n
y
,
J.
Q
a
d
i
r
,
a
n
d
A
.
A
.
-
F
u
q
a
h
a
,
“
A
c
t
i
v
e
l
e
a
r
n
i
n
g
b
a
s
e
d
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
f
o
r
w
a
s
t
e
a
n
d
n
a
t
u
r
a
l
d
i
s
a
st
e
r
i
ma
g
e
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
I
EEE
A
c
c
e
ss
,
v
o
l
.
8
,
p
p
.
2
0
8
5
1
8
–
2
0
8
5
3
1
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
0
.
3
0
3
8
6
7
6
.
[
1
2
]
Z.
M
a
,
W
.
Li
,
M
.
Z
h
a
n
g
,
W
.
M
e
n
g
,
S
.
X
u
,
a
n
d
X
.
Z
h
a
n
g
,
“
H
TC
V
i
T
:
a
n
e
f
f
e
c
t
i
v
e
n
e
t
w
o
r
k
f
o
r
i
ma
g
e
c
l
a
ssi
f
i
c
a
t
i
o
n
a
n
d
seg
m
e
n
t
a
t
i
o
n
b
a
s
e
d
o
n
n
a
t
u
r
a
l
d
i
s
a
st
e
r
d
a
t
a
se
t
s,”
T
h
e
Vi
s
u
a
l
C
o
m
p
u
t
e
r
,
v
o
l
.
3
9
,
n
o
.
8
,
p
p
.
3
2
8
5
–
3
2
9
7
,
A
u
g
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
3
7
1
-
023
-
0
2
9
5
4
-
3.
[
1
3
]
K
.
A
.
S
h
e
t
h
,
R
.
P
.
K
u
l
k
a
r
n
i
,
a
n
d
G
.
K
.
R
e
v
a
t
h
i
,
“
E
n
h
a
n
c
i
n
g
n
a
t
u
r
a
l
d
i
s
a
s
t
e
r
i
m
a
g
e
c
l
a
s
s
i
f
i
c
a
t
i
o
n
:
a
n
e
n
s
e
m
b
l
e
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
w
i
t
h
i
n
c
e
p
t
i
o
n
a
n
d
C
N
N
m
o
d
e
l
s
,
”
G
e
o
m
a
t
i
c
s
,
N
a
t
u
r
a
l
H
a
z
a
r
d
s
a
n
d
R
i
s
k
,
v
o
l
.
1
5
,
n
o
.
1
,
D
e
c
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
8
0
/
1
9
4
7
5
7
0
5
.
2
0
2
4
.
2
4
0
7
0
2
9
.
[
1
4
]
A
.
C
o
o
p
e
n
a
n
d
S
.
P
u
d
a
r
u
t
h
,
“
I
ma
g
e
c
l
a
ssi
f
i
c
a
t
i
o
n
b
a
se
d
o
n
d
i
sas
t
e
r
t
y
p
e
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
i
n
g
a
n
d
D
i
g
i
t
a
l
S
y
st
e
m
s
,
v
o
l
.
1
5
,
n
o
.
1
,
p
p
.
1
4
7
3
–
1
4
8
3
,
S
e
p
.
2
0
2
4
,
d
o
i
:
1
0
.
1
2
7
8
5
/
i
j
c
d
s
/
1
6
0
1
1
0
9
.
[
1
5
]
A
.
M
.
M
u
st
a
f
a
,
R
.
A
g
h
a
,
L
.
G
h
a
z
a
l
a
t
,
a
n
d
T
.
S
h
a
’
b
a
n
,
“
N
a
t
u
r
a
l
d
i
s
a
st
e
r
s
d
e
t
e
c
t
i
o
n
u
si
n
g
e
x
p
l
a
i
n
a
b
l
e
d
e
e
p
l
e
a
r
n
i
n
g
,
”
I
n
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
s
w
i
t
h
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
2
3
,
S
e
p
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
sw
a
.
2
0
2
4
.
2
0
0
4
3
0
.
[
1
6
]
M
.
H
.
B
a
sh
i
r
,
M
.
A
h
ma
d
,
D
.
R
.
R
i
z
v
i
,
a
n
d
A
.
A
.
A
.
E
.
-
La
t
i
f
,
“
Ef
f
i
c
i
e
n
t
C
N
N
-
b
a
se
d
d
i
sas
t
e
r
e
v
e
n
t
s
c
l
a
ss
i
f
i
c
a
t
i
o
n
u
s
i
n
g
U
A
V
-
a
i
d
e
d
i
m
a
g
e
s
f
o
r
e
m
e
r
g
e
n
c
y
r
e
s
p
o
n
se
a
p
p
l
i
c
a
t
i
o
n
,
”
N
e
u
r
a
l
C
o
m
p
u
t
i
n
g
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
6
,
n
o
.
1
8
,
p
p
.
1
0
5
9
9
–
1
0
6
1
2
,
Ju
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s0
0
5
2
1
-
0
2
4
-
0
9
6
1
0
-
4.
[
1
7
]
E.
Y
a
s
i
,
T.
U
.
S
h
a
k
i
b
,
N
.
S
h
a
r
mi
n
,
a
n
d
T.
H
.
R
i
z
u
,
“
F
l
o
o
d
a
n
d
n
o
n
-
f
l
o
o
d
i
ma
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
u
s
i
n
g
d
e
e
p
e
n
se
mb
l
e
l
e
a
r
n
i
n
g
,
”
Wa
t
e
r
Re
s
o
u
rce
s
M
a
n
a
g
e
m
e
n
t
,
v
o
l
.
3
8
,
n
o
.
1
3
,
p
p
.
5
1
6
1
–
5
1
7
8
,
O
c
t
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
2
6
9
-
024
-
0
3
9
0
6
-
9.
[
1
8
]
M
.
S
a
t
h
i
a
n
a
r
a
y
a
n
a
n
,
P
.
-
H
.
H
su
,
a
n
d
C
.
-
C
.
C
h
a
n
g
,
“
E
x
t
r
a
c
t
i
n
g
d
i
s
a
st
e
r
l
o
c
a
t
i
o
n
i
d
e
n
t
i
f
i
c
a
t
i
o
n
f
r
o
m
so
c
i
a
l
me
d
i
a
i
ma
g
e
s
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
D
i
s
a
st
e
r R
i
sk
Re
d
u
c
t
i
o
n
,
v
o
l
.
1
0
4
,
A
p
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
j
d
r
r
.
2
0
2
4
.
1
0
4
3
5
2
.
[
1
9
]
K
.
V
a
n
Ex
e
l
,
S
.
S
h
e
r
c
h
a
n
,
a
n
d
S
.
Li
u
,
“
o
p
t
i
m
i
z
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
mo
d
e
l
s
f
o
r
c
l
i
mat
e
-
r
e
l
a
t
e
d
n
a
t
u
r
a
l
d
i
sas
t
e
r
d
e
t
e
c
t
i
o
n
f
r
o
m
U
A
V
i
ma
g
e
s
a
n
d
r
e
m
o
t
e
se
n
si
n
g
d
a
t
a
,
”
J
o
u
rn
a
l
o
f
I
m
a
g
i
n
g
,
v
o
l
.
1
1
,
n
o
.
2
,
Ja
n
.
2
0
2
5
,
d
o
i
:
1
0
.
3
3
9
0
/
j
i
m
a
g
i
n
g
1
1
0
2
0
0
3
2
.
[
2
0
]
V
.
D
u
b
e
y
a
n
d
R
.
K
a
t
a
r
y
a
,
“
F
l
o
o
d
-
F
i
r
e
N
e
t
:
a
n
o
v
e
l
f
l
o
o
d
i
ma
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
a
r
c
h
i
t
e
c
t
u
r
e
u
si
n
g
t
h
e
a
d
a
p
t
i
v
e
f
i
r
e
f
l
y
a
l
g
o
r
i
t
h
m,”
S
i
g
n
a
l
,
I
m
a
g
e
a
n
d
Vi
d
e
o
Pr
o
c
e
ss
i
n
g
,
v
o
l
.
1
9
,
n
o
.
8
,
A
u
g
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
7
6
0
-
0
2
5
-
0
4
2
3
5
-
z.
[
2
1
]
L.
W
e
n
,
Z.
X
i
a
o
,
X
.
X
u
,
a
n
d
B
.
L
i
u
,
“
D
i
sast
e
r
r
e
c
o
g
n
i
t
i
o
n
a
n
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
b
a
se
d
o
n
i
mp
r
o
v
e
d
R
e
sN
e
t
-
5
0
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
Ap
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
5
,
n
o
.
9
,
M
a
y
2
0
2
5
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
5
0
9
5
1
4
3
.
[
2
2
]
Y
.
S
h
a
o
a
n
d
L.
X
u
,
“
M
u
l
t
i
m
o
d
a
l
n
a
t
u
r
a
l
d
i
sas
t
e
r
sc
e
n
e
r
e
c
o
g
n
i
t
i
o
n
w
i
t
h
i
n
t
e
g
r
a
t
e
d
l
a
r
g
e
m
o
d
e
l
a
n
d
m
a
m
b
a
,
”
Ap
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
5
,
n
o
.
3
,
Ja
n
.
2
0
2
5
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
5
0
3
1
1
4
9
.
[
2
3
]
B
.
W
a
n
g
,
C
.
Z
h
a
o
,
J
.
Li
,
Q
.
S
h
e
n
g
,
a
n
d
X
.
L
i
n
g
,
“
U
2
D
D
S
‐
N
e
t
:
A
n
e
w
a
r
c
h
i
t
e
c
t
u
r
e
b
a
se
d
o
n
U
2
N
e
t
w
i
t
h
d
i
s
a
st
e
r
t
y
p
e
f
o
r
b
u
i
l
d
i
n
g
d
a
m
a
g
e
a
s
sessm
e
n
t
u
n
d
e
r
n
a
t
u
r
a
l
d
i
sa
st
e
r
s,
”
T
h
e
P
h
o
t
o
g
ra
m
m
e
t
ri
c
R
e
c
o
r
d
,
v
o
l
.
4
0
,
n
o
.
1
8
9
,
J
a
n
.
2
0
2
5
,
d
o
i
:
1
0
.
1
1
1
1
/
p
h
o
r
.
1
2
5
3
0
.
[
2
4
]
A
.
V
e
r
m
a
,
“
D
i
s
a
s
t
e
r
i
m
a
g
e
s
d
a
t
a
s
e
t
,”
K
a
g
g
l
e
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
k
a
g
g
l
e
.
c
o
m
/
d
a
t
a
s
e
t
s
/
v
a
r
p
i
t
9
4
/
d
i
s
a
s
t
e
r
-
i
m
a
g
e
s
-
d
a
t
a
s
e
t
[
2
5
]
F
.
A
l
a
m
,
F
.
O
f
l
i
,
a
n
d
M
.
I
mr
a
n
,
“
C
r
i
si
s
M
M
D
:
m
u
l
t
i
mo
d
a
l
t
w
i
t
t
e
r
d
a
t
a
s
e
t
s
f
r
o
m
n
a
t
u
r
a
l
d
i
s
a
st
e
r
s,”
i
n
1
2
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
AAA
I
C
o
n
f
e
re
n
c
e
o
n
W
e
b
a
n
d
S
o
c
i
a
l
Me
d
i
a
,
I
C
W
S
M
2
0
1
8
,
2
0
1
8
,
p
p
.
4
6
5
–
4
7
3
,
d
o
i
:
1
0
.
1
6
0
9
/
i
c
w
sm.
v
1
2
i
1
.
1
4
9
8
3
.
[
2
6
]
H
.
M
o
u
z
a
n
n
a
r
,
Y
.
R
i
z
k
,
a
n
d
M
.
A
w
a
d
,
“
D
a
m
a
g
e
i
d
e
n
t
i
f
i
c
a
t
i
o
n
i
n
s
o
c
i
a
l
me
d
i
a
p
o
st
s
u
s
i
n
g
m
u
l
t
i
mo
d
a
l
d
e
e
p
l
e
a
r
n
i
n
g
,
”
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
I
n
t
e
rn
a
t
i
o
n
a
l
I
S
C
RAM
C
o
n
f
e
r
e
n
c
e
,
v
o
l
.
2
0
1
8
-
M
a
y
,
p
p
.
5
2
9
–
5
4
3
,
2
0
1
8
.
[
2
7
]
V
.
N
i
k
o
l
a
o
u
,
S
.
M
a
ss
a
r
o
,
M
.
F
a
k
h
i
mi
,
L.
S
t
e
r
g
i
o
u
l
a
s,
a
n
d
W
.
G
a
r
n
,
“
C
O
V
I
D
-
1
9
d
i
a
g
n
o
s
i
s
f
r
o
m
c
h
e
st
x
-
r
a
y
s
:
d
e
v
e
l
o
p
i
n
g
a
si
mp
l
e
,
f
a
st
,
a
n
d
a
c
c
u
r
a
t
e
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
a
n
d
S
y
s
t
e
m
s
,
v
o
l
.
9
,
n
o
.
1
,
D
e
c
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
3
7
5
5
-
021
-
0
0
1
6
6
-
4.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Ra
shm
i
K
a
n
g
o
k
a
r
Ta
r
a
n
a
t
h
is
c
u
rre
n
tl
y
wo
r
k
in
g
a
s
a
ss
istan
t
p
ro
fe
ss
o
r
in
S
ri
Krish
n
a
I
n
stit
u
te
o
f
Tec
h
n
o
l
o
g
y
,
Ba
n
g
a
l
o
re
.
S
h
e
re
c
e
iv
e
d
h
e
r
B
.
E.
d
e
g
re
e
in
El
e
c
tro
n
ics
a
n
d
Co
m
m
u
n
ica
ti
o
n
E
n
g
in
e
e
rin
g
fr
o
m
Ja
wa
h
a
rlal
Ne
h
ru
Na
ti
o
n
a
l
Co
ll
e
g
e
o
f
E
n
g
i
n
e
e
rin
g
,
S
h
imo
g
a
.
S
h
e
re
c
e
iv
e
d
h
e
r
m
a
ste
r
d
e
g
re
e
in
S
o
ftwa
re
E
n
g
i
n
e
e
rin
g
fr
o
m
M
.
S
.
Ra
m
a
iah
In
stit
u
te
o
f
Tec
h
n
o
lo
g
y
,
Vis
v
e
sv
a
ra
y
a
Tec
h
n
o
l
o
g
ica
l
Un
i
v
e
rsity
.
S
h
e
is
p
e
ru
si
n
g
P
h
.
D
.
i
n
Co
m
p
u
ti
n
g
a
n
d
I
n
fo
rm
a
ti
o
n
Tec
h
n
o
l
o
g
y
,
Re
v
a
Un
i
v
e
rsity
.
S
h
e
h
a
s
e
fficie
n
tl
y
h
a
n
d
led
v
a
ri
o
u
s
ro
les
a
n
d
re
sp
o
n
sib
il
i
ti
e
s
in
De
p
a
rtme
n
t
o
f
C
o
m
p
u
ter
S
c
ien
c
e
a
n
d
En
g
i
n
e
e
rin
g
at
S
ri
Krish
n
a
In
stit
u
te
o
f
Tec
h
n
o
lo
g
y
.
He
r
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
a
rti
ficia
l
in
telli
g
e
n
c
e
a
n
d
m
a
c
h
in
e
lea
rn
in
g
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
ra
sh
m
ik
t1
9
@
g
m
a
il
.
c
o
m
.
G
e
e
ta
Chi
d
a
n
a
n
d
a
p
p
a
M
a
r
a
is
c
u
rre
n
tl
y
wo
rk
i
n
g
a
s
p
ro
fe
ss
o
r,
in
S
c
h
o
o
l
o
f
Co
m
p
u
ti
n
g
a
n
d
In
f
o
rm
a
ti
o
n
Tec
h
n
o
lo
g
y
,
Re
v
a
Un
i
v
e
rsity
,
Be
n
g
a
lu
ru
.
S
h
e
h
a
s
c
o
m
p
lete
d
h
e
r
P
h
.
D
.
u
n
d
e
r
th
e
g
u
id
a
n
c
e
o
f
Dr
.
Ve
n
u
g
o
p
a
l
K
.
R
.
,
f
o
rm
e
r
v
ice
c
h
a
n
c
e
ll
o
r
,
Ba
n
g
a
lo
re
Un
iv
e
rsity
,
Be
n
g
a
l
u
ru
.
He
r
th
e
sis
ti
tl
e
is
“
Da
ta
a
u
d
it
in
g
a
n
d
se
c
u
rit
y
in
c
lo
u
d
c
o
m
p
u
ti
n
g
”
.
S
h
e
h
a
s
p
u
b
li
sh
e
d
3
0
+
p
a
p
e
rs
i
n
i
n
tern
a
ti
o
n
a
l
c
o
n
fe
re
n
c
e
s
a
n
d
j
o
u
rn
a
ls
.
S
h
e
h
a
s
p
u
b
li
sh
e
d
6
p
a
ten
ts
a
n
d
2
p
a
ten
ts
h
a
v
e
b
e
e
n
g
ra
n
te
d
.
S
h
e
h
a
s
a
u
th
o
re
d
a
b
o
o
k
e
n
ti
tl
e
d
,
“
S
e
c
u
re
re
tri
e
v
a
l
a
n
d
a
u
d
it
in
g
tec
h
n
iq
u
e
s
in
c
lo
u
d
”
,
p
u
b
li
sh
e
d
b
y
I.
K.
I
n
tern
a
ti
o
n
a
l
P
v
t.
L
td
.
S
h
e
re
c
e
iv
e
d
b
e
st
p
a
p
e
r
a
wa
rd
f
o
r
t
h
e
p
a
p
e
r
ti
tl
e
d
“
EAOD
BT:
e
fficie
n
t
a
u
d
it
in
g
f
o
r
o
u
tso
u
rc
e
d
d
a
tab
a
se
wit
h
to
k
e
n
e
n
fo
rc
e
d
c
lo
u
d
st
o
ra
g
e
”
,
in
th
e
p
r
o
c
e
e
d
in
g
s
o
f
IEE
E
C
o
n
fe
r
e
n
c
e
WI
ECON
-
ECE
2
0
1
9
.
S
h
e
is
re
v
iew
e
r
o
f
S
p
rin
g
e
r
j
o
u
r
n
a
l
a
n
d
h
a
s
c
h
a
ired
se
ss
io
n
s
in
v
a
rio
u
s
i
n
tern
a
ti
o
n
a
l
c
o
n
fe
re
n
c
e
s
.
S
h
e
is
c
u
rre
n
tl
y
g
u
i
d
in
g
4
sc
h
o
lars
a
t
REVA
Un
iv
e
rsity
.
As
a
p
ro
fe
ss
o
r,
sh
e
h
a
s
e
fficie
n
tl
y
h
a
n
d
led
v
a
rio
u
s
r
o
les
a
n
d
re
sp
o
n
si
b
il
it
ies
in
S
c
h
o
o
l
o
f
Co
m
p
u
ti
n
g
a
n
d
I
n
fo
rm
a
ti
o
n
Tec
h
n
o
l
o
g
y
,
REVA
U
n
iv
e
rsit
y
.
He
r
re
se
a
rc
h
a
re
a
s
in
c
lu
d
e
c
l
o
u
d
c
o
m
p
u
ti
n
g
,
c
o
m
p
u
ter
n
e
two
rk
s,
a
rti
ficia
l
in
telli
g
e
n
c
e
,
m
a
c
h
in
e
lea
rn
in
g
,
a
n
d
in
tern
e
t
o
f
th
in
g
s
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
g
e
e
tac
.
m
a
ra
@r
e
v
a
.
e
d
u
.
i
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.