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.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
,
p
p
.
4
9
3
3
~
494
5
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
5
.
p
p
4
9
3
3
-
494
5
4933
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Sa
tellit
e
ima
g
ery a
nd deep l
ea
rning
f
ra
mewo
rk
f
o
r hi
g
hwa
y
co
nstruction
with
minima
l veg
etati
o
n impa
ct
M
ee
na
k
s
hi,
Ra
j
esh
T
ha
lwa
g
a
l Ma
t
ha
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
r
i
n
g
,
D
a
y
a
n
a
n
d
a
S
a
g
a
r
U
n
i
v
e
r
si
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
Feb
25
,
2
0
2
5
R
ev
is
ed
J
u
l
13
,
2
0
2
6
Acc
ep
ted
J
u
l
21
,
2
0
2
6
Co
n
stru
c
ti
n
g
r
o
a
d
s
th
r
o
u
g
h
v
e
g
e
tate
d
terra
in
fo
rc
e
s
a
trad
e
-
o
ff
b
e
twe
e
n
m
in
imiz
in
g
e
c
o
l
o
g
ica
l
h
a
rm
a
n
d
p
re
se
rv
in
g
c
o
n
str
u
c
ti
o
n
a
l
e
a
se
.
Th
is
wo
r
k
p
re
se
n
ts
a
u
n
ifi
e
d
,
e
n
d
-
to
-
e
n
d
f
ra
m
e
wo
rk
th
a
t
re
c
o
n
c
il
e
s
b
o
th
o
b
jec
ti
v
e
s,
with
t
h
re
e
m
a
jo
r
c
o
n
tri
b
u
ti
o
n
s.
F
i
rst,
p
r
o
p
o
se
a
h
y
b
ri
d
p
e
rc
e
p
ti
o
n
m
o
d
e
l
t
h
a
t
fu
se
s
a
g
ra
p
h
c
o
n
v
o
lu
ti
o
n
a
l
n
e
two
rk
(G
CN)
—
c
a
p
tu
rin
g
sc
e
n
e
-
l
e
v
e
l
sp
a
ti
a
l
c
o
n
tex
t
—
wit
h
a
c
a
p
su
le
n
e
two
rk
(Ca
p
sN
e
t)
th
a
t
e
n
c
o
d
e
s
p
a
rt
-
wh
o
le
re
latio
n
sh
i
p
s
f
o
r
re
li
a
b
le
d
e
tec
ti
o
n
o
f
sm
a
ll
,
c
l
u
ste
re
d
v
e
g
e
tatio
n
;
th
is
c
o
n
fig
u
ra
ti
o
n
a
tt
a
i
n
s a
n
a
v
e
ra
g
e
p
re
c
isio
n
(AP)
o
f
0
.
9
2
5
,
a
v
e
ra
g
e
re
c
a
ll
(AR)
o
f
0
.
9
4
8
(i
n
ters
e
c
ti
o
n
o
v
e
r
u
n
io
n
(Io
U)
=
0
.
5
),
a
n
d
9
8
%
l
a
n
d
-
c
o
v
e
r
c
las
sifica
ti
o
n
a
c
c
u
ra
c
y
o
n
u
n
se
e
n
d
a
ta.
S
e
c
o
n
d
,
we
in
tro
d
u
c
e
a
n
e
c
o
lo
g
ica
l
-
se
n
siti
v
it
y
c
o
st
m
a
p
f
u
sin
g
v
e
g
e
tatio
n
n
o
rm
a
li
z
e
d
d
iffere
n
c
e
v
e
g
e
tatio
n
in
d
e
x
(
NDVI
)
with
terra
in
sl
o
p
e
,
o
v
e
r
wh
ich
we
e
v
a
lu
a
te
Dij
k
st
ra
,
A*
,
a
n
o
v
e
l
Ec
o
-
Aw
a
re
A*
e
m
b
e
d
d
in
g
a
v
e
g
e
tatio
n
-
a
v
o
i
d
a
n
c
e
p
e
n
a
lt
y
,
a
n
d
a
n
non
-
d
o
m
in
a
ted
so
rti
n
g
g
e
n
e
ti
c
a
lg
o
rit
h
m
II
(
N
S
G
A
-
II
)
m
u
lt
i
-
o
b
jec
ti
v
e
se
a
rc
h
th
a
t
jo
in
tl
y
m
in
imiz
e
s
p
a
th
len
g
th
a
n
d
v
e
g
e
tatio
n
d
a
m
a
g
e
.
T
h
ird
,
o
n
a
b
e
n
c
h
m
a
rk
m
o
d
e
led
o
n
t
h
e
Be
n
g
a
lu
ru
–
Ka
n
a
k
a
p
u
ra
r
o
a
d
(ND
VI
=
0
.
7
,
slo
p
e
=
0
.
3
),
t
h
e
e
c
o
-
se
n
siti
v
e
r
o
u
te
l
o
we
rs
se
g
m
e
n
t
ND
VI
b
y
ro
u
g
h
l
y
6
9
%
a
n
d
e
li
m
in
a
tes
a
ll
h
ig
h
-
c
a
n
o
p
y
e
x
p
o
su
re
wh
il
e
a
d
d
i
n
g
o
n
ly
a
b
o
u
t
8
%
to
p
a
th
len
g
t
h
v
e
rsu
s
a
n
o
p
ti
m
ize
d
n
o
n
-
v
e
g
e
tate
d
ro
u
te.
To
g
e
th
e
r,
t
h
e
se
c
o
m
p
o
n
e
n
ts
e
n
a
b
le au
to
m
a
ted
,
e
c
o
lo
g
ica
ll
y
a
wa
re
ro
a
d
p
lan
n
i
n
g
d
irec
tl
y
fr
o
m
ima
g
e
ry
.
K
ey
w
o
r
d
s
:
Ar
tific
ial
in
tellig
en
ce
C
o
m
p
u
ter
v
is
io
n
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Dee
p
lear
n
in
g
No
r
m
alize
d
d
if
f
e
r
en
ce
v
eg
etatio
n
in
d
ex
Satellite im
ag
e
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
:
Me
en
ak
s
h
i
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
Da
y
an
an
d
a
Sag
ar
Un
iv
er
s
ity
B
en
g
alu
r
u
,
I
n
d
ia
E
m
ail:
m
ee
n
ak
s
h
ib
4
3
7
@
g
m
ai
l.c
o
m
1.
I
NT
RO
D
UCT
I
O
N
Ar
tific
ial
in
tellig
en
ce
(
AI
)
h
as
b
ec
o
m
e
a
tr
a
n
s
f
o
r
m
ativ
e
f
o
r
c
e
in
in
f
r
astru
ctu
r
e
co
n
d
itio
n
m
o
n
ito
r
in
g
,
r
esh
ap
in
g
m
o
d
e
r
n
ass
et
m
an
a
g
em
en
t
p
r
ac
tices
ag
ain
s
t
th
e
b
ac
k
d
r
o
p
o
f
f
ast
-
p
ac
ed
u
r
b
an
i
za
tio
n
an
d
th
e
p
u
s
h
f
o
r
s
u
s
tain
ab
le
d
e
v
elo
p
m
e
n
t
.
C
o
n
v
en
tio
n
al
ap
p
r
o
ac
h
es
to
th
is
p
r
o
b
lem
h
a
v
e
lo
n
g
r
elied
o
n
m
an
u
al
in
s
p
ec
tio
n
,
g
r
o
u
n
d
-
b
ased
s
en
s
o
r
s
,
an
d
p
er
io
d
ic
v
is
u
al
s
u
r
v
e
y
s
to
g
au
g
e
th
e
s
tate
o
f
r
o
ad
s
,
b
r
id
g
es,
b
u
ild
in
g
s
,
an
d
s
im
ilar
ass
et
s
[
1
]
–
[
4
]
.
T
h
ese
m
eth
o
d
s
wo
r
k
u
p
to
a
p
o
in
t,
b
u
t
th
ey
ca
r
r
y
r
ea
l
lim
itatio
n
s
:
th
ey
d
em
an
d
s
u
b
s
tan
tial
lab
o
r
,
co
n
s
u
m
e
s
ig
n
if
ican
t
tim
e,
co
v
e
r
o
n
ly
lim
ited
ar
ea
s
,
an
d
ca
n
n
o
t
s
u
s
tain
c
o
n
tin
u
o
u
s
o
v
er
s
ig
h
t
o
f
ex
ten
s
iv
e
in
f
r
astru
ctu
r
e
n
etwo
r
k
s
.
C
o
m
p
o
u
n
d
in
g
th
ese
s
h
o
r
tco
m
in
g
s
,
tr
a
d
itio
n
al
m
o
n
ito
r
in
g
lar
g
ely
ig
n
o
r
es
th
e
en
v
i
r
o
n
m
e
n
tal
s
id
e
o
f
in
f
r
astru
ctu
r
e
d
ev
elo
p
m
e
n
t,
leav
in
g
v
e
g
etatio
n
lo
s
s
an
d
b
r
o
ad
er
la
n
d
-
c
o
v
er
ch
an
g
e
i
n
s
u
r
r
o
u
n
d
in
g
ar
ea
s
u
n
ac
co
u
n
ted
.
So
m
e
o
f
th
e
w
ay
s
in
wh
ich
th
e
en
v
ir
o
n
m
en
t
co
u
ld
b
e
af
f
ec
ted
n
eg
ativ
el
y
alo
n
g
th
e
r
o
u
tes m
a
y
n
o
t b
e
well
co
n
s
id
er
ed
in
r
o
ad
alig
n
m
en
t.
T
h
e
s
h
o
r
ta
g
e
o
f
i
n
f
r
astru
ctu
r
e
in
th
e
c
o
u
n
tr
y
m
ak
es
tr
a
n
s
p
o
r
t
r
o
u
tes
cr
u
cial
f
o
r
t
h
e
n
atio
n
’
s
d
ev
elo
p
m
e
n
t.
R
em
o
tely
s
en
s
e
d
d
ata
an
d
m
ac
h
i
n
e
lear
n
i
n
g
(
ML
)
alg
o
r
it
h
m
s
h
av
e
i
n
cr
ea
s
ed
th
e
ac
cu
r
ac
y
o
f
lan
d
-
co
v
e
r
class
if
icatio
n
,
wh
er
ea
s
m
u
lti
-
o
b
jectiv
e
o
p
tim
i
za
tio
n
m
ak
es
it
f
ea
s
ib
le
to
n
av
ig
ate
tr
ad
e
-
o
f
f
s
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.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
9
3
3
-
494
5
4934
am
o
n
g
m
u
ltip
le
o
b
jectiv
es
.
Yet,
ch
ain
p
lan
n
in
g
s
t
ill
en
tails
d
is
co
n
n
ec
ted
s
tag
es:
p
er
ce
p
tio
n
(
class
if
icatio
n
/s
eg
m
en
tatio
n
)
i
s
d
esig
n
ed
in
d
ep
e
n
d
en
tly
f
r
o
m
r
o
u
tin
g
,
o
r
o
p
tim
izatio
n
is
co
n
d
u
cted
o
v
e
r
a
cr
u
d
e,
h
eu
r
is
tic
co
s
t
g
r
id
.
E
c
o
R
o
u
te
I
‑
Net
r
em
e
d
ies
th
e
lack
th
r
o
u
g
h
t
h
e
f
o
llo
win
g
in
te
g
r
ated
an
d
m
o
d
u
lar
s
y
s
tem
r
an
g
in
g
f
r
o
m
p
er
ce
p
tio
n
to
d
ec
is
io
n
:
a
b
ase
m
o
d
el
au
g
m
e
n
tin
g
s
eg
m
en
tatio
n
/cla
s
s
if
icatio
n
p
ip
elin
e
(
ViT
-
L
ar
g
e
+
Swin
T
r
a
n
s
f
o
r
m
er
)
,
y
iel
d
in
g
h
ig
h
-
r
eso
lu
tio
n
t
h
em
atic
m
ap
s
.
E
m
p
ir
ically
,
th
e
in
teg
r
ated
s
y
s
tem
is
ac
cu
r
ate
an
d
y
ield
s
co
n
s
tr
u
ctiv
e
s
av
in
g
s
r
elev
an
t
to
th
e
ab
s
tr
ac
ted
p
r
o
ject
r
eq
u
ir
e
m
en
t.
C
o
n
tr
ib
u
tio
n
s
:
a
u
n
if
ied
p
e
r
ce
p
tio
n
-
to
-
d
ec
is
io
n
f
r
am
ewo
r
k
in
teg
r
atin
g
f
o
u
n
d
atio
n
m
o
d
el
-
b
ased
lan
d
c
o
v
er
m
a
p
p
in
g
with
m
u
lti
-
o
b
jectiv
e
r
o
u
tin
g
.
A
m
ath
em
atica
lly
b
ased
co
m
p
o
s
ite
co
s
t
s
u
r
f
ac
e
f
o
r
v
eg
etatio
n
,
h
y
d
r
o
l
o
g
y
,
ter
r
ain
,
an
d
c
o
n
s
tr
u
ctab
ilit
y
ter
m
s
.
I
n
r
ec
e
n
t
y
ea
r
s
,
AI
–
b
ased
a
p
p
r
o
ac
h
es,
p
ar
ticu
lar
l
y
th
o
s
e
t
h
at
m
ak
e
u
s
e
o
f
s
atellite
im
ag
er
y
,
h
av
e
em
er
g
ed
as
p
o
wer
f
u
l
an
d
s
ca
lab
le
to
o
ls
f
o
r
in
f
r
astru
ctu
r
e
m
o
n
ito
r
in
g
an
d
m
an
ag
e
m
e
n
t.
Hig
h
-
r
eso
lu
tio
n
s
atellite
im
ag
es
m
ak
e
it
p
o
s
s
ib
le
to
o
b
s
er
v
e
lar
g
e
g
eo
g
r
ap
h
ic
ar
ea
s
an
d
co
llect
u
p
d
a
ted
in
f
o
r
m
atio
n
at
f
r
eq
u
e
n
t in
ter
v
als,
en
ab
lin
g
n
e
ar
r
ea
l
-
tim
e
m
o
n
ito
r
in
g
o
f
cr
itical
in
f
r
astru
ctu
r
e
ass
ets
[
5
]
–
[
8
]
.
B
y
s
h
i
f
t
i
n
g
f
r
o
m
m
a
n
u
a
l
i
n
s
p
e
c
t
i
o
n
t
o
A
I
-
d
r
i
v
en
s
a
t
e
l
li
t
e
i
m
a
g
e
a
n
a
l
y
s
i
s
,
i
n
f
r
a
s
t
r
u
c
t
u
r
e
h
e
a
lt
h
c
a
n
b
e
e
f
f
i
c
i
e
n
t
l
y
a
s
s
ess
e
d
a
c
r
o
s
s
l
a
r
g
e
n
e
t
w
o
r
k
s
w
h
i
l
e
t
h
e
s
a
m
e
m
o
d
e
l
s
s
i
m
u
l
t
a
n
e
o
u
s
l
y
d
e
t
ec
t
a
n
d
q
u
a
n
t
i
f
y
v
e
g
e
t
a
t
i
o
n
l
o
s
s
f
r
o
m
e
x
p
a
n
s
i
o
n
,
m
a
i
n
t
e
n
a
n
c
e
,
a
n
d
u
r
b
a
n
d
ev
e
l
o
p
m
e
n
t
,
j
o
i
n
tl
y
s
u
p
p
o
r
t
i
n
g
s
u
s
t
a
i
n
a
b
l
e
,
r
es
i
l
ie
n
t
d
e
v
e
l
o
p
m
e
n
t
[
9
]
–
[
1
1
]
.
Su
b
s
tan
tial
o
b
s
tacle
s
s
til
l
s
tan
d
in
th
e
way
:
th
e
s
h
ee
r
s
ca
le
o
f
av
ailab
le
d
ata,
th
e
s
u
b
tlety
o
f
m
an
y
s
tr
u
ctu
r
al
d
ef
ec
ts
,
an
d
th
e
d
if
f
icu
lty
o
f
t
ellin
g
au
th
en
tic
v
eg
etatio
n
lo
s
s
ap
ar
t
f
r
o
m
ch
an
g
es
d
r
iv
en
b
y
s
ea
s
o
n
o
r
clim
ate.
T
h
ese
f
ac
to
r
s
k
ee
p
t
h
e
f
u
s
io
n
o
f
h
ig
h
-
r
eso
lu
tio
n
s
atellite
d
ata
with
AI
-
b
ased
m
o
n
ito
r
in
g
an
ac
tiv
e
r
esear
ch
f
r
o
n
tier
,
o
n
e
th
at
ca
lls
f
o
r
d
ee
p
lear
n
in
g
m
o
d
els
ca
p
ab
le
o
f
p
ick
in
g
u
p
o
n
f
in
e
-
g
r
ain
ed
c
u
es
tied
to
s
tr
u
ctu
r
al
wea
r
,
er
o
s
io
n
,
a
n
d
s
h
if
tin
g
v
e
g
etatio
n
co
v
e
r
.
T
h
is
p
r
o
ject
r
e
s
p
o
n
d
s
with
a
u
n
if
ie
d
,
r
ea
l
-
ti
m
e
f
r
am
ewo
r
k
b
u
ilt
on
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etw
o
r
k
s
(
C
NNs
)
th
at
f
la
g
s
d
eg
r
a
d
atio
n
s
ig
n
als
s
u
ch
as
cr
ac
k
s
,
wea
r
,
a
n
d
er
o
s
io
n
wh
ile
s
im
u
ltan
eo
u
s
ly
m
ea
s
u
r
in
g
n
ea
r
b
y
v
e
g
etatio
n
lo
s
s
[
1
2
]
–
[
1
4
]
,
g
iv
i
n
g
p
la
n
n
er
s
a
s
ca
lab
le,
co
s
t
-
ef
f
ec
tiv
e
,
au
to
m
ated
s
u
b
s
titu
te
f
o
r
m
a
n
u
al
in
s
p
ec
tio
n
.
I
n
p
r
ac
tice,
t
h
is
en
ab
les
p
r
o
ac
tiv
e
m
ain
ten
a
n
ce
,
cu
ts
o
p
er
ati
n
g
co
s
ts
,
an
d
lim
its
h
ar
m
to
v
eg
etatio
n
[
1
5
]
,
[
1
6
]
,
in
k
ee
p
in
g
with
s
m
ar
t
city
o
b
j
ec
tiv
es
an
d
wid
er
s
u
s
tain
ab
ilit
y
tar
g
ets.
T
h
e
p
ap
er
m
ak
es
th
r
ee
co
n
tr
i
b
u
tio
n
s
.
First,
it
p
r
esen
ts
a
s
i
n
g
le
p
er
ce
p
ti
o
n
-
to
-
d
ec
is
io
n
f
r
am
ewo
r
k
th
at
jo
in
s
f
o
u
n
d
atio
n
-
s
ty
le
la
n
d
-
co
v
e
r
m
a
p
p
in
g
with
m
u
lti
-
o
b
jectiv
e
r
o
u
tin
g
.
Seco
n
d
,
it
d
ef
in
es
a
c
o
m
p
o
s
ite
co
s
t
s
u
r
f
ac
e,
i
n
clo
s
ed
m
ath
em
atica
l
f
o
r
m
,
th
at
co
m
b
i
n
e
s
v
eg
etatio
n
,
ter
r
ain
,
an
d
c
o
n
s
tr
u
ctab
ilit
y
ter
m
s
.
T
h
ir
d
,
it
i
n
tr
o
d
u
ce
s
an
E
c
o
-
A
war
e
A*
p
lan
n
er
a
n
d
a
n
o
n
-
d
o
m
in
ated
s
o
r
tin
g
g
e
n
etic
alg
o
r
ith
m
I
I
(
NSGA
-
II
)
Par
eto
s
ea
r
ch
th
at
m
a
k
e
t
h
e
v
eg
etatio
n
-
v
er
s
u
s
-
f
ea
s
ib
ili
ty
tr
ad
e
-
o
f
f
e
x
p
licit,
an
d
it
r
elea
s
es
a
f
u
lly
r
ep
r
o
d
u
cib
le
im
p
lem
e
n
tatio
n
.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
Mo
n
ito
r
in
g
i
n
f
r
astru
ctu
r
e
d
ev
elo
p
m
en
t
an
d
v
eg
etatio
n
co
v
e
r
d
y
n
a
m
ics
h
as
b
ec
o
m
e
a
v
er
y
im
p
o
r
tan
t
p
r
o
ce
s
s
with
r
eg
ar
d
s
to
m
o
d
e
r
n
u
r
b
a
n
izatio
n
,
clim
ate
tr
en
d
s
,
an
d
an
in
cr
ea
s
e
in
ec
o
s
y
s
te
m
d
eg
r
ad
atio
n
.
T
h
e
co
m
b
in
atio
n
o
f
s
atellite
im
ag
e
r
y
an
d
AI
,
esp
ec
ially
with
r
eg
ar
d
s
to
d
ee
p
lear
n
in
g
,
h
as
s
ig
n
if
ican
tly
en
h
a
n
ce
d
th
e
p
r
o
ce
s
s
o
f
o
b
s
er
v
in
g
s
tr
u
ctu
r
al
d
ev
elo
p
m
en
t
an
d
v
e
g
et
atio
n
d
estru
ctio
n
o
n
a
lar
g
e
s
p
atial
an
d
tem
p
o
r
al
s
ca
le.
T
h
e
in
itial
p
r
o
ce
s
s
was
m
ain
ly
d
o
m
in
ate
d
b
y
o
b
s
er
v
atio
n
s
an
d
r
u
le
-
b
ased
s
y
s
tem
s
,
wh
er
ea
s
m
o
d
er
n
s
y
s
tem
s
ar
e
in
cr
ea
s
in
g
ly
u
p
g
r
ad
in
g
to
d
ee
p
lear
n
in
g
a
n
d
m
u
lti
-
tem
p
o
r
al
s
atellite
im
ag
es.
E
ar
ly
ac
co
m
p
lis
h
m
e
n
ts
p
r
o
v
ed
th
e
ca
p
ab
ilit
y
o
f
f
in
e
-
s
c
ale
v
eg
etatio
n
an
aly
s
is
.
T
h
e
p
r
o
g
r
ess
ac
h
iev
ed
b
y
Ar
k
in
et
a
l.
[
1
]
p
av
ed
th
e
way
f
o
r
t
h
e
an
aly
s
is
o
f
v
e
g
etatio
n
lo
s
s
b
ased
o
n
v
er
y
h
ig
h
-
r
eso
lu
tio
n
im
ag
er
y
.
T
h
is
wo
r
k
f
o
cu
s
ed
o
n
th
e
ass
ess
m
en
t
o
f
th
e
s
ev
er
ity
o
f
f
ir
es
u
s
in
g
air
b
o
r
n
e
i
m
ag
er
y
.
Fin
e
-
s
ca
le
v
eg
etatio
n
an
aly
s
is
p
r
o
v
e
d
wo
r
k
ab
le
in
s
ev
er
al
ea
r
ly
s
tu
d
i
es.
W
o
r
k
in
g
f
r
o
m
v
er
y
-
h
ig
h
-
r
eso
lu
tio
n
air
b
o
r
n
e
im
ag
er
y
,
Ar
k
in
et
a
l.
[
1
]
ass
ess
ed
f
ir
e
s
ev
er
ity
an
d
,
in
d
o
in
g
s
o
,
ad
v
an
ce
d
th
e
m
ea
s
u
r
e
m
en
t
o
f
v
eg
etatio
n
lo
s
s
.
B
ai
et
a
l.
[
2
]
p
r
o
p
o
s
ed
a
n
im
p
r
o
v
ed
lu
m
in
a
n
ce
-
co
n
tr
a
s
t
s
alien
cy
m
ap
b
u
ilt
o
n
in
ter
f
er
o
m
etr
ic
s
y
n
t
h
etic
ap
er
tu
r
e
r
ad
a
r
(
I
n
SAR
)
co
h
er
en
ce
d
if
f
er
en
ce
s
an
d
u
s
ed
it
to
d
elin
ea
te
b
u
r
n
e
d
ar
ea
s
.
Plo
m
er
et
a
l.
[
3
]
to
o
k
a
C
NN
-
b
ased
r
o
u
te,
co
m
b
in
in
g
r
ad
ar
an
d
o
p
tical
im
ag
er
y
to
m
ap
b
u
r
n
e
d
ar
ea
s
an
d
ca
p
tu
r
e
f
ir
e
-
d
r
iv
e
n
v
eg
etatio
n
lo
s
s
.
B
u
lato
v
et
a
l.
[
4
]
,
[
5
]
m
o
v
e
d
p
ast
two
-
d
im
en
s
io
n
al
im
ag
er
y
alto
g
eth
er
,
u
s
in
g
th
r
ee
-
d
i
m
en
s
io
n
al
p
o
in
t
clo
u
d
s
to
s
ep
ar
ate
v
e
g
etatio
n
f
r
o
m
b
u
ilt
s
tr
u
ctu
r
es
an
d
ag
ain
r
ec
o
v
er
in
g
v
eg
et
atio
n
lo
s
s
r
eliab
ly
.
Salien
cy
m
ap
tech
n
iq
u
es
wer
e
f
u
r
th
e
r
in
tr
o
d
u
ce
d
in
t
h
e
an
al
y
s
is
b
y
[
2
]
u
s
in
g
th
e
I
n
SAR
co
h
er
en
ce
d
if
f
e
r
en
ce
in
d
eter
m
in
in
g
th
e
b
u
r
n
ed
ar
e
as.
T
h
is
p
r
o
ce
s
s
in
d
icate
d
t
h
e
ef
f
ec
tiv
en
ess
o
f
th
e
an
al
y
s
is
tech
n
iq
u
es
u
s
ed
i
n
th
e
s
tu
d
y
.
T
ec
h
n
iq
u
es
b
ased
o
n
th
e
u
s
e
o
f
C
NN
wer
e
f
u
r
th
er
ex
p
lo
r
e
d
b
y
[
3
]
.
T
h
e
w
o
r
k
f
o
c
u
s
ed
o
n
th
e
an
aly
s
is
b
ased
o
n
t
h
e
u
s
ag
e
o
f
r
ad
ar
an
d
o
p
tical
im
a
g
er
y
i
n
th
e
b
u
r
n
a
r
ea
d
eter
m
in
atio
n
.
T
h
is
in
d
icate
d
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
an
aly
s
is
in
ca
p
tu
r
i
n
g
lo
s
s
d
u
e
to
v
eg
et
atio
n
f
ir
es.
I
n
ad
d
itio
n
to
th
e
an
aly
s
is
tech
n
iq
u
es
b
ased
o
n
two
-
d
im
en
s
io
n
al
im
ag
er
y
,
tech
n
iq
u
es
b
ased
o
n
th
r
ee
-
d
im
en
s
io
n
al
p
o
in
t
clo
u
d
s
f
o
cu
s
ed
o
n
th
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
S
a
telite
ima
g
ery
a
n
d
d
ee
p
lea
r
n
in
g
fr
a
mewo
r
k
fo
r
h
ig
h
w
a
y
c
o
n
s
tr
u
ctio
n
w
ith
min
ima
l
…
(
Meen
a
ksh
i
)
4935
an
aly
s
is
b
etwe
en
v
eg
etatio
n
a
n
d
m
an
-
m
ad
e
s
tr
u
ct
u
r
es
wer
e
f
u
r
th
er
ex
p
lo
r
ed
b
y
[
4
]
.
T
h
e
p
r
o
ce
s
s
in
d
icate
d
th
e
ef
f
ec
tiv
en
ess
in
ca
p
tu
r
in
g
th
e
l
o
s
s
d
u
e
to
v
eg
etatio
n
in
th
e
s
elec
ted
r
eg
io
n
.
I
m
p
r
o
v
em
en
ts
in
d
ee
p
lear
n
in
g
tech
n
iq
u
es c
o
m
b
in
ed
with
t
h
e
ad
v
en
t
o
f
h
ig
h
-
r
eso
lu
tio
n
s
atellite
d
ata
h
av
e
f
u
r
th
e
r
in
cr
ea
s
ed
th
e
ac
c
u
r
ac
y
o
f
v
e
g
etatio
n
lo
s
s
d
etec
tio
n
.
C
h
o
et
a
l
.
[
6
]
u
s
ed
d
ee
p
l
ea
r
n
in
g
tech
n
i
q
u
es
in
co
m
b
i
n
atio
n
with
Plan
etSco
p
e
s
atellite
im
ag
er
y
f
o
r
b
u
r
n
ed
ar
ea
i
d
en
tific
atio
n
,
in
cr
ea
s
in
g
th
e
ac
cu
r
ac
y
o
f
u
n
item
p
o
r
al
v
eg
etatio
n
lo
s
s
d
etec
tio
n
.
Dix
o
n
et
a
l.
[
8
]
p
r
es
en
ted
th
e
co
n
ce
p
t
o
f
s
p
atio
te
m
p
o
r
al
d
ee
p
lear
n
in
g
tech
n
iq
u
es
f
o
r
r
ec
o
g
n
izin
g
c
an
o
p
y
-
s
ca
le
tr
ee
d
ea
th
in
r
esp
o
n
s
e
to
f
o
r
est
f
ir
es
f
o
r
p
r
ec
is
e
v
eg
etatio
n
lo
s
s
m
o
n
ito
r
in
g
o
v
er
tim
e.
J
av
e
d
e
t
a
l.
[
9
]
p
r
esen
ted
a
d
ee
p
lear
n
in
g
-
b
ased
m
eth
o
d
f
o
r
ch
a
n
g
e
d
etec
tio
n
o
f
u
r
b
a
n
f
o
r
est
co
v
er
u
s
in
g
v
er
y
h
i
g
h
r
eso
lu
tio
n
(
VHR
)
s
atellite
im
ag
er
y
.
As
d
ee
p
lear
n
in
g
m
atu
r
ed
a
n
d
h
ig
h
er
-
r
eso
lu
tio
n
d
ata
b
ec
a
m
e
av
ailab
le,
v
e
g
etatio
n
-
lo
s
s
d
etec
tio
n
im
p
r
o
v
ed
a
g
ain
.
I
n
s
tu
d
ies
[
6
]
,
[
7
]
ap
p
lied
d
ee
p
lear
n
in
g
to
u
n
item
p
o
r
al
Plan
etSco
p
e
im
ag
er
y
f
o
r
b
u
r
n
e
d
-
ar
ea
id
e
n
tific
atio
n
an
d
r
ep
o
r
ted
h
ig
h
er
ac
c
u
r
ac
y
.
Dix
o
n
et
a
l.
[
8
]
d
ev
elo
p
ed
s
p
atio
tem
p
o
r
al
tech
n
iq
u
es
th
at
r
ec
o
g
n
ize
ca
n
o
p
y
-
s
ca
le
tr
ee
m
o
r
tality
af
ter
f
ir
e,
wh
ich
lets
th
e
p
r
o
ce
s
s
b
e
f
o
llo
wed
p
r
ec
is
ely
th
r
o
u
g
h
tim
e,
wh
ile
J
av
ed
et
a
l.
[
9
]
d
etec
ted
ch
an
g
e
in
u
r
b
an
f
o
r
est
co
v
e
r
f
r
o
m
VHR
im
ag
er
y
.
Ve
g
etatio
n
r
ec
o
v
er
y
af
ter
d
is
tu
r
b
an
ce
d
r
ew
atten
tio
n
to
o
.
Priy
a
an
d
Van
i
[
1
0
]
b
u
ilt
an
u
n
s
u
p
er
v
is
ed
d
ee
p
m
o
d
el
f
o
r
p
o
s
t
-
f
ir
e
v
eg
etatio
n
ch
a
n
g
e
an
d
r
ec
o
v
er
y
;
Su
r
et
a
l.
[
1
1
]
tr
ac
k
ed
v
eg
etatio
n
d
eg
r
a
d
atio
n
ac
r
o
s
s
I
n
d
ia
f
r
o
m
m
u
lti
-
s
en
s
o
r
im
ag
er
y
;
an
d
An
a
n
d
et
a
l.
[
1
2
]
lin
k
e
d
lan
d
s
ca
p
e
ch
an
g
e
to
lan
d
-
s
u
r
f
ac
e
tem
p
er
atu
r
e
a
n
d
v
eg
etatio
n
in
d
ices.
R
eg
io
n
al
s
tu
d
ies
h
av
e
p
aid
p
ar
ticu
lar
atten
tio
n
to
clim
ate
-
d
r
iv
en
v
e
g
etatio
n
ch
an
g
e
.
Sh
u
k
la
et
a
l
.
[
1
3
]
ca
r
r
ied
o
u
t
a
s
p
atio
-
te
m
p
o
r
al
an
aly
s
is
o
f
v
eg
etatio
n
co
v
er
f
r
o
m
lin
ea
r
i
m
ag
in
g
s
elf
-
s
ca
n
n
i
n
g
s
en
s
o
r
(
L
I
SS
)
-
I
V
im
ag
er
y
with
AI
/ML
ch
an
g
e
d
etec
tio
n
.
Oth
er
wo
r
k
ap
p
lied
C
NNs
to
Plan
etSco
p
e
d
ata
f
o
r
a
cc
u
r
ate
lan
d
-
u
s
e
an
d
lan
d
-
c
o
v
er
class
if
icatio
n
.
Deep
-
lear
n
in
g
r
em
o
te
s
en
s
in
g
h
as
also
b
ee
n
u
s
ed
to
f
lag
n
o
r
m
alize
d
d
if
f
er
e
n
ce
v
eg
etatio
n
in
d
e
x
(
NDVI
)
-
r
elate
d
d
eg
r
ad
atio
n
i
n
m
in
in
g
an
d
h
ab
itat
zo
n
es
a
n
d
to
esti
m
ate
f
r
ac
tio
n
al
v
eg
etatio
n
co
v
e
r
f
r
o
m
Sen
tin
el
-
2
tim
e
s
er
ies,
an
d
h
y
b
r
id
r
ec
u
r
r
e
n
t
n
eu
r
al
n
etwo
r
k
(
R
NN
)
–
C
NN
m
o
d
els
h
av
e
b
ee
n
p
u
t
f
o
r
war
d
to
f
o
r
ec
ast NDV
I
f
o
r
lo
n
g
-
ter
m
v
eg
etatio
n
-
h
ea
lth
s
tu
d
ies [
1
3
]
–
[
1
6
]
.
R
ev
iew
ar
ticles
h
av
e
s
in
ce
d
r
awn
th
ese
th
r
ea
d
s
to
g
eth
er
,
p
o
in
tin
g
t
o
a
clea
r
m
o
v
e
to
w
ar
d
f
u
s
in
g
h
eter
o
g
en
e
o
u
s
d
ata
s
o
u
r
ce
s
a
n
d
to
th
e
g
r
o
win
g
d
o
m
in
an
ce
o
f
d
ee
p
lear
n
in
g
in
b
io
d
i
v
er
s
ity
m
o
n
ito
r
in
g
a
n
d
d
ef
o
r
estatio
n
m
ap
p
in
g
[
1
0
]
,
[
1
7
]
.
R
ec
en
t
s
tu
d
ies
h
av
e
wid
ely
em
p
lo
y
ed
d
ee
p
lear
n
in
g
an
d
r
em
o
te
s
en
s
in
g
f
o
r
v
eg
etatio
n
m
o
n
ito
r
in
g
,
ch
a
n
g
e
d
etec
tio
n
,
an
d
en
v
ir
o
n
m
e
n
tal
ass
es
s
m
en
t
u
s
in
g
m
u
lti
-
s
en
s
o
r
s
atellite
im
ag
er
y
.
R
eg
io
n
al
r
esear
ch
h
as
u
tili
ze
d
C
NNs,
NDVI
,
Sen
tin
el
-
2
,
Pla
n
etSco
p
e
,
an
d
h
y
b
r
i
d
R
NN
–
C
NN
m
o
d
els
to
m
ap
v
eg
etatio
n
d
y
n
a
m
ics,
clim
ate
-
in
d
u
ce
d
ch
a
n
g
es,
a
n
d
lo
n
g
-
t
er
m
v
e
g
etatio
n
h
ea
lth
.
C
o
m
p
r
eh
en
s
iv
e
r
ev
iews
em
p
h
asize
m
u
ltimo
d
al
d
ata
f
u
s
io
n
,
b
io
d
iv
er
s
ity
m
o
n
it
o
r
in
g
,
s
em
an
tic
s
eg
m
e
n
tatio
n
,
an
d
lar
g
e
-
s
ca
le
d
ef
o
r
estatio
n
d
etec
tio
n
.
I
n
f
r
a
s
tr
u
ctu
r
e
-
r
elate
d
s
tu
d
ies
h
av
e
co
n
tr
ib
u
te
d
th
r
o
u
g
h
la
n
d
-
co
v
er
s
eg
m
en
tatio
n
,
h
ig
h
way
p
ath
o
p
tim
izatio
n
i
n
co
r
p
o
r
atin
g
la
n
d
-
u
s
e
d
y
n
am
ics,
th
e
A*
alg
o
r
ith
m
f
o
r
o
p
tim
al
r
o
u
tin
g
,
an
d
Hu
ete'
s
SAVI
f
o
r
im
p
r
o
v
ed
v
eg
etatio
n
d
etec
tio
n
.
B
u
ild
in
g
o
n
th
ese
ad
v
an
ce
s
,
Priy
a
an
d
Van
i
[
1
0
]
ac
h
ie
v
e
d
o
v
er
9
6
%
ac
cu
r
ac
y
i
n
d
is
tin
g
u
is
h
in
g
g
en
u
i
n
e
v
e
g
etatio
n
lo
s
s
f
r
o
m
s
p
u
r
io
u
s
ch
a
n
g
es,
d
em
o
n
s
tr
atin
g
th
e
p
o
ten
tial o
f
d
ee
p
lear
n
in
g
f
o
r
ec
o
-
s
en
s
itiv
e
in
f
r
astru
ctu
r
e
p
l
an
n
in
g
.
Ma
ty
u
k
ir
a
an
d
Mh
an
g
ar
a
[
1
7
]
,
in
a
s
cien
to
m
etr
ic
r
ev
iew
o
f
t
h
e
liter
atu
r
e,
f
o
u
n
d
t
h
at
n
eu
r
al
n
etwo
r
k
s
d
o
m
in
ate
th
e
d
o
m
ain
an
d
th
a
t
C
NN
s
in
p
ar
ticu
lar
h
av
e
b
ec
o
m
e
s
tan
d
ar
d
b
ec
au
s
e
o
f
th
eir
ab
ilit
y
to
lear
n
h
ier
ar
ch
ical
f
ea
tu
r
es
d
ir
ec
tly
f
r
o
m
r
aw
in
p
u
t
d
ata.
T
h
is
m
o
tiv
ates
th
e
u
s
e
o
f
C
NN
an
d
g
r
ap
h
co
n
v
o
l
u
tio
n
al
n
etwo
r
k
s
(
GC
N)
b
ased
ap
p
r
o
a
ch
es
in
th
is
p
r
o
ject.
L
iu
et
a
l.
[
1
8
]
o
f
f
er
a
s
u
r
v
ey
o
f
r
o
a
d
ex
t
r
ac
tio
n
tech
n
iq
u
es
b
ased
o
n
d
ee
p
lea
r
n
in
g
th
at
in
clu
d
e
f
u
lly
s
u
p
e
r
v
is
ed
,
s
em
i
-
s
u
p
er
v
is
ed
an
d
u
n
s
u
p
e
r
v
i
s
ed
ap
p
r
o
ac
h
es
an
d
id
en
tify
r
elev
a
n
t
d
atasets
an
d
b
en
ch
m
ar
k
s
f
o
r
th
e
r
o
ad
s
e
g
m
en
tatio
n
m
o
d
u
le
d
e
v
elo
p
e
d
h
er
e.
Kr
ejca
r
a
n
d
Nam
az
i
[
1
9
]
e
x
p
lo
r
es
th
e
r
o
l
e
o
f
AI
in
r
em
o
te
s
en
s
in
g
m
o
r
e
g
e
n
er
ally
,
in
clu
d
in
g
la
n
d
c
o
v
er
class
if
icatio
n
,
o
b
ject
d
etec
tio
n
,
a
n
d
d
is
aster
r
esp
o
n
s
e,
b
u
t
also
i
d
en
tifi
es
ch
allen
g
es
th
at
th
e
c
o
m
b
in
ed
GC
N
-
ca
p
s
u
le
n
etwo
r
k
(
C
ap
s
Net
)
ap
p
r
o
ac
h
p
r
esen
ted
h
er
e
aim
s
t
o
ad
d
r
ess
,
s
p
ec
if
ically
r
elate
d
t
o
g
e
n
er
aliza
b
ilit
y
,
f
u
s
in
g
o
f
m
u
lti
-
s
o
u
r
ce
d
ata,
an
d
c
o
m
p
u
tatio
n
al
ef
f
icien
cy
.
T
ak
en
t
o
g
eth
er
,
th
is
b
o
d
y
o
f
liter
atu
r
e
i
n
clu
d
es
v
eg
etatio
n
lo
s
s
d
etec
tio
n
,
v
eg
etatio
n
clas
s
if
icatio
n
an
d
m
ap
p
i
n
g
m
eth
o
d
s
,
r
o
ad
ex
t
r
ac
tio
n
m
eth
o
d
s
,
a
n
d
m
o
r
e
b
r
o
a
d
ly
AI
ap
p
r
o
ac
h
es f
o
r
r
em
o
te
s
en
s
in
g
.
I
en
co
et
a
l.
[
2
0
]
p
r
o
p
o
s
e
tim
e
-
s
er
ies
wav
ef
o
r
m
an
d
im
ag
er
y
n
etwo
r
k
f
o
r
s
atellite
-
f
u
s
io
n
(
T
W
I
NS
)
,
a
d
u
al
b
r
a
n
ch
d
ee
p
lear
n
in
g
s
y
s
tem
wh
ich
co
m
b
i
n
es
Sen
tin
el
-
1
r
ad
a
r
an
d
Sen
tin
el
-
2
o
p
tic
al
tim
e
-
s
er
ies
d
ata
th
r
o
u
g
h
a
co
n
v
o
lu
tio
n
al
g
ated
r
ec
u
r
r
e
n
t
u
n
it
(
C
o
n
v
GR
U
)
ar
ch
itectu
r
e
co
u
p
led
with
atten
t
io
n
m
ec
h
an
is
m
s
in
o
r
d
er
t
o
tak
e
a
d
v
an
ta
g
e
o
f
co
m
p
lem
en
tar
y
s
tr
en
g
th
s
o
f
th
e
d
ata
in
s
p
ac
e
a
n
d
tim
e.
I
t
d
e
m
o
n
s
tr
ates
th
at
th
is
ap
p
r
o
ac
h
ca
n
en
h
a
n
ce
lan
d
-
co
v
er
class
if
icatio
n
ac
cu
r
ac
y
[
2
1
]
.
C
o
n
ti
n
e
n
t
al
-
s
ca
le
v
e
g
e
tati
o
n
p
att
er
n
m
o
d
e
lli
n
g
is
w
ell
e
x
p
l
o
r
e
d
th
r
o
u
g
h
m
u
lt
im
o
d
al
e
ar
th
o
b
s
e
r
v
ati
o
n
f
o
u
n
d
at
io
n
al
m
o
d
els
[
2
2
]
.
V
is
i
o
n
-
b
as
ed
g
r
a
p
h
n
e
u
r
al
n
et
wo
r
k
s
f
o
r
e
ar
th
o
b
s
e
r
v
at
io
n
h
a
v
e
d
e
m
o
n
s
tr
ate
d
a
d
v
a
n
t
a
g
es
i
n
l
a
n
d
co
v
er
class
if
i
ca
ti
o
n
.
F
ea
t
u
r
e
p
y
r
am
id
n
et
wo
r
k
s
a
n
d
f
u
ll
y
c
o
n
v
o
lu
ti
o
n
al
a
r
ch
ite
ct
u
r
es
h
av
e
b
ee
n
d
e
m
o
n
s
t
r
a
te
d
t
o
p
r
o
v
id
e
r
o
b
u
s
t
a
n
d
ac
cu
r
a
te
p
e
r
-
p
ix
el
c
h
an
g
e
d
ete
cti
o
n
i
n
r
em
o
te
s
e
n
s
i
n
g
.
E
f
f
ic
ie
n
t
an
d
r
o
b
u
s
t
tr
ee
b
as
e
d
m
o
d
els
li
k
e
L
i
g
h
tGB
M
s
e
r
v
e
as
s
tr
o
n
g
b
as
eli
n
es
f
o
r
l
ar
g
e
-
s
ca
l
e
class
if
ica
ti
o
n
t
as
k
s
[
2
3
]
–
[
2
4
]
.
Z
o
p
h
et
a
l
.
[
2
5
]
in
tr
o
d
u
ce
d
n
eu
r
al
ar
c
h
itectu
r
e
s
ea
r
ch
n
etwo
r
k
(
NASNet
)
,
u
s
in
g
n
e
u
r
al
ar
ch
itectu
r
e
s
ea
r
ch
to
au
t
o
m
atica
lly
d
is
co
v
er
a
tr
an
s
f
er
a
b
le
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.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
9
3
3
-
494
5
4936
c
o
n
v
o
l
u
t
i
o
n
a
l
"
c
e
l
l"
o
n
a
s
m
a
ll
p
r
o
x
y
d
a
t
a
s
e
t
(
C
I
F
AR
-
1
0
)
t
h
a
t
c
o
u
l
d
b
e
s
t
a
c
k
e
d
t
o
s
c
a
l
e
u
p
to
l
a
r
g
e
-
s
c
a
l
e
i
m
a
g
e
r
e
c
o
g
n
i
t
i
o
n
.
T
h
e
i
r
s
e
a
r
c
h
e
d
ar
c
h
i
t
e
c
t
u
r
es
s
u
r
p
as
s
e
d
h
a
n
d
-
d
e
s
i
g
n
e
d
n
e
t
w
o
r
k
s
o
n
I
m
a
g
e
Ne
t
w
h
i
le
l
o
w
e
r
i
n
g
c
o
m
p
u
t
a
t
i
o
n
a
l
c
o
s
t
,
d
e
m
o
n
s
t
r
ati
n
g
t
h
a
t
l
e
a
r
n
e
d
a
r
c
h
it
e
c
t
u
r
es
ca
n
t
r
a
n
s
f
e
r
e
f
f
e
c
ti
v
e
l
y
a
c
r
o
s
s
d
a
t
a
s
et
s
a
n
d
ta
s
k
s
.
T
h
e
a
b
o
v
e
d
i
s
c
u
s
s
i
o
n
o
n
t
h
e
s
t
a
t
e
o
f
t
h
e
a
r
t
d
e
m
o
n
s
t
r
at
es
a
c
l
e
a
r
a
r
e
a
w
h
e
r
e
s
i
g
n
i
f
i
ca
n
t
a
d
v
a
n
c
e
m
e
n
t
c
a
n
b
e
m
a
d
e
:
t
h
e
r
e
is
a
d
e
a
r
t
h
o
f
f
r
a
m
e
w
o
r
k
s
t
h
a
t
j
o
i
n
tl
y
a
d
d
r
e
s
s
i
n
f
r
a
s
t
r
u
c
t
u
r
e
c
h
a
n
g
e
a
n
d
as
s
o
ci
a
t
e
d
v
e
g
e
ta
t
i
o
n
l
o
s
s
u
s
i
n
g
a
s
i
n
g
le
,
i
n
t
e
g
r
at
e
d
p
i
p
e
l
i
n
e
.
U
t
i
li
z
i
n
g
m
u
lt
i
m
o
d
a
l
s
a
t
e
l
li
t
e
i
m
a
g
e
r
y
i
n
c
o
m
b
i
n
at
i
o
n
w
i
t
h
d
e
e
p
l
e
a
r
n
i
n
g
m
e
t
h
o
d
s
p
r
o
v
i
d
e
s
a
p
r
o
m
is
i
n
g
a
v
e
n
u
e
t
o
c
o
n
s
t
r
u
c
t
s
u
c
h
h
o
l
is
t
i
c
s
o
l
u
ti
o
n
s
.
M
o
r
e
b
r
o
a
d
l
y
,
th
e
c
o
n
v
e
r
g
e
n
c
e
o
f
ML
,
g
e
o
s
p
a
t
ia
l
a
n
a
l
y
s
is
,
a
n
d
o
p
t
i
m
i
z
a
ti
o
n
i
n
i
n
f
r
a
s
t
r
u
ct
u
r
e
d
e
v
e
l
o
p
m
e
n
t
h
a
s
b
e
c
o
m
e
a
n
a
c
t
i
v
e
r
es
e
a
r
c
h
d
i
r
e
c
t
i
o
n
,
o
n
e
t
h
a
t
t
h
is
p
r
o
j
ec
t'
s
s
e
g
m
e
n
t
at
i
o
n
,
c
o
s
t
m
a
p
p
in
g
,
a
n
d
r
o
a
d
-
p
l
a
n
n
i
n
g
o
p
t
i
m
i
z
a
t
i
o
n
b
u
i
l
d
s
u
p
o
n
.
S
a
t
e
ll
i
te
-
b
a
s
e
d
l
a
n
d
c
l
ass
i
f
i
c
ati
o
n
h
a
s
a
d
v
a
n
c
e
d
c
o
n
s
i
d
e
r
a
b
ly
t
h
r
o
u
g
h
d
e
e
p
l
e
a
r
n
i
n
g
,
w
it
h
f
u
l
l
y
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
s
s
e
t
ti
n
g
a
n
e
a
r
l
y
b
e
n
c
h
m
a
r
k
f
o
r
s
e
m
a
n
t
i
c
s
e
g
m
en
t
a
t
i
o
n
a
n
d
U
-
N
e
t
R
o
n
n
e
b
e
r
g
e
r
e
t
a
l
.
,
[
2
6
]
,
f
ir
s
t
d
esig
n
ed
f
o
r
b
io
m
ed
ical
im
ag
es,
later
r
ep
u
r
p
o
s
ed
to
class
if
y
lan
d
co
v
er
in
to
f
o
r
ests
,
r
o
a
d
s
,
wate
r
b
o
d
ies,
an
d
p
lain
s
.
C
o
s
t
m
ap
p
in
g
,
wh
ich
ass
ig
n
s
ea
ch
lan
d
ca
teg
o
r
y
a
v
alu
e
r
e
f
lectin
g
its
ec
o
lo
g
i
ca
l
s
en
s
itiv
ity
an
d
co
n
s
tr
u
ctio
n
d
if
f
ic
u
lty
,
ex
ten
d
s
ea
r
lier
g
eo
g
r
ap
h
ic
in
f
o
r
m
a
tio
n
s
y
s
tem
(
GI
S
)
-
b
ased
ap
p
r
o
ac
h
es
th
at
ar
g
u
ed
f
o
r
em
b
ed
d
in
g
ec
o
l
o
g
ical
f
ac
to
r
s
d
ir
ec
tly
in
to
in
f
r
astru
ctu
r
e
p
lan
n
in
g
,
with
n
ewe
r
s
tu
d
i
es
p
air
in
g
GI
S
an
d
ML
to
g
e
n
er
ate
co
s
t
m
a
p
s
th
a
t
ad
ap
t
d
y
n
am
ically
t
o
ch
a
n
g
i
n
g
co
n
d
itio
n
s
.
L
o
n
g
-
s
tan
d
in
g
alg
o
r
ith
m
s
s
u
ch
as
Dijk
s
tr
a
[
2
7
]
an
d
A*
Har
t
et
a
l.
,
[2
8
]
co
n
tin
u
e
to
an
ch
o
r
s
h
o
r
test
-
p
ath
co
m
p
u
tatio
n
a
n
d
h
a
v
e
b
ee
n
ad
ap
ted
to
co
s
t
-
weig
h
ted
g
e
o
s
p
atial
s
ettin
g
s
,
with
r
ec
en
t
r
esear
ch
c
o
u
p
lin
g
th
em
to
m
ac
h
in
e
-
lear
n
in
g
-
d
er
iv
ed
co
s
t
m
a
p
s
s
o
th
at
co
m
p
u
te
d
r
o
u
tes m
in
im
ize
f
in
an
cial
an
d
en
v
ir
o
n
m
e
n
tal
co
s
t to
g
eth
er
.
Ad
v
an
ce
s
in
h
ig
h
-
r
eso
lu
tio
n
im
ag
er
y
a
n
d
d
ata
p
r
o
ce
s
s
in
g
n
o
w
s
u
p
p
o
r
t
f
in
e
-
g
r
ain
ed
ter
r
ain
an
aly
s
is
,
an
d
a
n
in
cr
ea
s
in
g
b
o
d
y
o
f
w
o
r
k
f
u
s
es
r
em
o
te
s
en
s
in
g
with
ML
to
l
o
ca
te
c
o
n
s
tr
u
ctio
n
p
ath
s
with
th
e
least
ec
o
lo
g
ical
d
is
r
u
p
tio
n
.
Su
s
tain
ab
ilit
y
co
n
s
id
er
atio
n
s
h
av
e
lik
ewise
m
o
v
ed
to
th
e
ce
n
te
r
o
f
in
f
r
astru
ctu
r
e
p
lan
n
in
g
,
with
p
r
io
r
s
tu
d
ies
d
o
cu
m
en
tin
g
th
e
en
v
ir
o
n
m
en
tal
to
ll
o
f
r
o
ad
b
u
ild
in
g
an
d
ca
llin
g
f
o
r
m
et
h
o
d
s
th
at
lim
it
h
ab
itat
f
r
ag
m
en
tatio
n
an
d
d
e
f
o
r
estatio
n
,
s
h
ap
in
g
f
r
am
ewo
r
k
s
th
at
weig
h
ec
o
n
o
m
ic
g
ain
s
a
g
ain
s
t
ec
o
lo
g
ical
co
s
ts
.
T
r
a
n
s
latin
g
an
aly
tical
o
u
tp
u
t
i
n
to
ac
tio
n
ab
le
g
u
id
an
ce
d
ep
e
n
d
s
o
n
ef
f
ec
tiv
e
g
e
o
s
p
atial
v
is
u
aliza
tio
n
,
an
d
p
r
o
g
r
ess
in
GI
S
an
d
ML
h
as
m
ad
e
in
ter
ac
tiv
e,
in
ter
p
r
eta
b
le
v
is
u
al
to
o
ls
av
ailab
le
to
s
tak
eh
o
ld
er
s
.
B
ey
o
n
d
r
o
a
d
s
,
co
m
p
ar
ab
le
s
eg
m
en
tatio
n
an
d
o
p
tim
izatio
n
tech
n
iq
u
es
h
av
e
f
o
u
n
d
u
s
e
in
u
r
b
an
p
lan
n
in
g
,
d
is
aster
r
esp
o
n
s
e,
r
eso
u
r
ce
allo
ca
tio
n
,
an
d
u
tili
ty
n
etwo
r
k
d
esig
n
,
p
o
i
n
tin
g
to
t
h
e
b
r
o
ad
e
r
ap
p
licab
ilit
y
o
f
th
is
m
eth
o
d
o
l
o
g
y
.
T
r
ac
k
in
g
h
o
w
in
f
r
astru
c
tu
r
e
g
r
o
ws
an
d
h
o
w
v
eg
etati
o
n
co
v
er
s
h
if
ts
h
as
tak
en
o
n
n
ew
im
p
o
r
tan
ce
a
m
id
m
o
d
er
n
u
r
b
an
izatio
n
,
a
c
h
a
n
g
in
g
clim
ate,
an
d
th
e
s
tead
y
lo
s
s
o
f
ec
o
s
y
s
tem
s
.
Pair
in
g
s
atellite
im
ag
er
y
with
AI
,
an
d
with
d
ee
p
lear
n
in
g
in
p
ar
ticu
lar
,
h
as
m
ad
e
it
f
a
r
ea
s
ier
to
o
b
s
er
v
e
b
o
th
co
n
s
tr
u
ctio
n
ac
tiv
ity
an
d
v
eg
etatio
n
lo
s
s
o
v
er
lar
g
e
ar
ea
s
an
d
lo
n
g
p
er
io
d
s
.
T
h
e
ea
r
lies
t
s
tu
d
ies
lean
ed
o
n
m
an
u
al
o
b
s
er
v
atio
n
an
d
r
u
le
-
b
ased
s
y
s
tem
s
;
cu
r
r
en
t
p
ip
eli
n
es
in
cr
ea
s
in
g
ly
r
ely
o
n
d
ee
p
lear
n
in
g
ap
p
lied
to
m
u
lti
-
tem
p
o
r
al
im
a
g
er
y
.
3.
M
E
T
H
O
D
Data
a
cq
u
is
itio
n
:
in
th
is
s
tu
d
y
,
th
e
s
atellite
im
ag
er
y
u
s
ed
w
as
p
r
o
v
i
d
ed
b
y
s
ev
er
al
r
en
o
w
n
ed
s
p
ac
e
ag
en
cies,
s
u
ch
as
th
e
I
n
d
ian
Sp
ac
e
R
e
s
ea
r
ch
Or
g
an
izatio
n
(
I
SR
O)
,
NASA,
th
e
E
u
r
o
p
ea
n
Sp
ac
e
Ag
en
cy
(
E
SA)
,
an
d
t
h
e
s
p
ac
e
s
tu
d
ies
in
s
titu
te
in
C
alif
o
r
n
ia
wh
ich
h
o
s
ts
s
atellite
d
ata
ac
r
o
s
s
v
ar
io
u
s
r
eso
lu
tio
n
s
in
s
p
atial,
s
p
ec
tr
al,
an
d
tem
p
o
r
a
l
ch
ar
ac
ter
is
tics
.
C
o
n
s
id
er
in
g
th
e
f
ac
t
th
at
o
u
r
a
r
ea
o
f
in
t
er
est
is
T
h
is
s
tu
d
y
p
r
io
r
itizes
s
atellite
d
ata
th
at
c
an
b
e
u
s
ed
to
s
tu
d
y
v
ar
io
u
s
ter
r
ain
an
d
clim
ate
co
n
d
itio
n
s
in
I
n
d
ia;
th
er
ef
o
r
e,
th
is
s
tu
d
y
s
elec
ted
im
ag
es
f
r
o
m
I
SR
O’
s
L
I
SS
-
I
I
I
to
d
ev
elo
p
an
d
test
o
u
r
m
o
d
el.
L
I
SS
-
I
I
I
h
as
4
s
p
ec
tr
al
b
an
d
s
–
b
o
th
v
is
ib
le
an
d
th
e
n
ea
r
-
in
f
r
ar
e
d
b
an
d
s
p
o
s
s
ess
a
s
p
atial
r
eso
lu
tio
n
o
f
2
3
.
5
m
o
v
e
r
a
s
wath
wid
th
o
f
ap
p
r
o
x
im
ately
1
4
1
k
m
wh
er
e
as
th
e
s
h
o
r
ter
wav
elen
g
t
h
in
f
r
ar
ed
b
a
n
d
p
o
s
s
ess
es
a
co
ar
s
e
s
p
atial
r
eso
lu
tio
n
o
f
7
0
.
5
m
a
n
d
s
wath
wid
t
h
o
f
a
p
p
r
o
x
im
ately
1
4
8
k
m
.
T
h
e
s
en
s
o
r
p
r
o
v
id
es
d
ec
en
t
tem
p
o
r
a
l
co
v
er
a
g
e
with
its
r
ev
is
it
tim
e
o
f
2
4
d
ay
s
,
f
r
ee
ly
av
ailab
le
th
r
o
u
g
h
B
h
u
v
an
,
th
e
n
atio
n
al
g
eo
-
p
o
r
tal,
o
v
er
a
s
p
an
o
f
m
o
r
e
th
an
a
d
ec
ad
e
co
m
p
r
is
in
g
h
is
to
r
ical
an
d
m
u
lti
-
s
ea
s
o
n
al
im
ag
er
y
with
r
elativ
ely
less
clo
u
d
co
v
er
.
I
m
ag
er
y
was
co
llected
f
r
o
m
s
ev
er
al
estab
l
is
h
ed
ag
en
cies,
am
o
n
g
th
e
m
th
e
I
SR
O,
NASA,
E
SA,
an
d
th
e
s
p
ac
e
s
tu
d
ies
in
s
titu
te
,
wh
o
s
e
ar
ch
iv
es
d
if
f
e
r
in
s
p
atial,
s
p
ec
tr
al,
an
d
tem
p
o
r
al
r
eso
lu
tio
n
.
Sin
ce
th
e
s
tu
d
y
f
o
cu
s
es
o
n
I
n
d
ian
r
eg
io
n
s
,
th
is
s
tu
d
y
f
av
o
r
e
d
im
ag
er
y
m
atch
e
d
to
I
n
d
ian
te
r
r
ain
an
d
clim
ate
an
d
s
ettled
o
n
I
SR
O's
L
I
SS
-
III
in
s
tr
u
m
en
t
f
o
r
b
u
ild
in
g
an
d
test
in
g
th
e
m
o
d
el.
L
I
SS
-
I
I
I
r
e
co
r
d
s
f
o
u
r
s
p
ec
tr
al
b
an
d
s
.
T
h
e
two
v
is
ib
le
b
an
d
s
an
d
th
e
n
ea
r
-
in
f
r
ar
ed
b
an
d
g
i
v
e
2
3
.
5
m
r
eso
l
u
tio
n
o
v
er
a
s
wath
n
ea
r
1
4
1
k
m
,
an
d
t
h
e
s
h
o
r
t
-
wav
e
in
f
r
ar
ed
b
an
d
wo
r
k
s
at
7
0
.
5
m
o
v
er
a
s
wath
o
f
ab
o
u
t
1
4
8
k
m
.
I
ts
2
4
-
d
ay
r
ev
is
it
cy
cle
k
ee
p
s
tem
p
o
r
al
co
v
er
ag
e
s
tead
y
,
an
d
b
ec
au
s
e
th
e
d
ata
ar
e
f
r
ee
th
r
o
u
g
h
B
h
u
v
an
,
I
n
d
ia'
s
n
atio
n
al
g
eo
-
p
o
r
tal,
m
o
r
e
th
an
a
d
ec
ad
e
o
f
m
u
lti
-
s
ea
s
o
n
al,
m
o
s
tly
lo
w
-
cl
o
u
d
s
ce
n
es
is
av
ailab
le,
wh
ic
h
m
ak
es
th
e
s
en
s
o
r
well
s
u
ite
d
to
tr
ain
in
g
r
o
b
u
s
t
s
eg
m
en
tatio
n
m
o
d
els.
T
h
is
s
tu
d
y
also
d
r
aws
o
n
th
e
p
u
b
lic
Sp
ac
eNe
t d
ataset,
wh
ich
o
f
f
er
s
h
ig
h
-
r
eso
lu
tio
n
im
ag
e
r
y
o
f
u
r
b
an
an
d
s
em
i
-
u
r
b
a
n
ar
ea
s
to
g
eth
e
r
with
d
etailed
g
r
o
u
n
d
-
t
r
u
th
l
ab
els
f
o
r
b
u
ild
in
g
s
,
r
o
ad
s
,
a
n
d
o
th
er
s
tr
u
ctu
r
es.
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
S
a
telite
ima
g
ery
a
n
d
d
ee
p
lea
r
n
in
g
fr
a
mewo
r
k
fo
r
h
ig
h
w
a
y
c
o
n
s
tr
u
ctio
n
w
ith
min
ima
l
…
(
Meen
a
ksh
i
)
4937
T
h
o
s
e
lab
els
m
a
k
e
s
u
p
er
v
is
ed
tr
ain
in
g
an
d
r
eliab
le
e
v
alu
atio
n
p
o
s
s
ib
le.
Sin
ce
Sp
ac
eNe
t
i
s
wid
ely
u
tili
ze
d
in
GI
S
ap
p
licatio
n
s
,
u
r
b
an
p
lan
n
in
g
,
an
d
d
is
aster
r
esp
o
n
s
e,
i
t
p
r
o
v
id
es
a
r
elev
an
t
b
en
c
h
m
ar
k
f
o
r
e
v
alu
atin
g
m
o
d
el
ac
cu
r
ac
y
in
s
eg
m
en
tin
g
v
ar
io
u
s
u
r
b
an
f
ea
tu
r
es
u
n
d
e
r
v
ar
y
in
g
c
o
n
d
itio
n
s
.
T
h
e
wo
r
k
ad
d
itio
n
ally
m
ak
es
u
s
e
o
f
t
h
e
p
u
b
licly
a
v
ailab
le
Sp
ac
eNe
t
d
ataset
th
at
s
o
u
r
c
ed
o
n
Kag
g
le.
T
h
e
Sp
ac
eNe
t
d
ataset
co
n
tain
s
u
ltra
-
h
ig
h
-
r
eso
lu
tio
n
p
ictu
r
es
o
f
city
s
ca
p
es/s
u
b
u
r
b
a
n
a
n
d
in
clu
d
es
co
r
r
esp
o
n
d
in
g
la
n
d
-
g
r
o
u
n
d
tr
u
th
s
o
f
b
u
ild
in
g
s
,
s
tr
ee
ts
an
d
o
th
er
s
;
t
h
ese
d
ata
allo
w
to
m
ak
e
a
p
r
o
p
er
tr
ain
in
g
o
f
an
y
s
u
p
er
v
is
ed
lear
n
in
g
m
o
d
el
as
well
as
an
u
n
b
iased
ev
alu
ati
o
n
o
f
t
h
e
m
o
d
el.
T
h
e
r
esu
lts
h
av
e
b
ee
n
q
u
a
n
titativ
ely
an
aly
ze
d
in
ter
m
s
o
f
in
ter
s
ec
tio
n
o
v
er
u
n
io
n
(
I
o
U)
,
th
e
m
o
s
t
co
m
m
o
n
an
d
ac
ce
p
ted
m
etr
ic
f
o
r
q
u
a
n
tify
in
g
s
eg
m
e
n
tatio
n
p
er
f
o
r
m
an
ce
.
I
o
U
is
d
ef
in
ed
as
th
e
r
atio
o
f
o
v
er
lap
to
u
n
io
n
ar
ea
.
W
ith
‘
A’
a
n
d
‘
B
’
th
e
s
et
o
f
p
ix
els
co
r
r
esp
o
n
d
in
g
to
g
r
o
u
n
d
-
t
r
u
th
an
d
esti
m
ated
s
eg
m
en
tatio
n
m
ap
s
,
r
esp
ec
tiv
ely
an
d
A
B
th
e
in
ter
s
ec
tio
n
,
I
o
U
is
ca
lcu
lated
as
in
(
1
)
.
(
,
)
=
|
∩
|
|
∪
|
=
|
∩
|
|
|
+
|
|
−
|
∩
|
(
1
)
I
o
U
also
k
n
o
wn
as
th
e
J
ac
ca
r
d
I
n
d
ex
,
m
ea
s
u
r
es
th
e
s
im
ilar
i
ty
b
etwe
en
th
e
p
r
e
d
icted
r
e
g
io
n
(
A)
an
d
th
e
g
r
o
u
n
d
-
tr
u
t
h
r
e
g
io
n
(
B
)
b
y
d
iv
id
in
g
t
h
eir
o
v
er
lap
p
in
g
ar
e
a
b
y
th
eir
co
m
b
in
ed
ar
ea
.
An
I
o
U
v
alu
e
clo
s
er
t
o
1
in
d
icate
s
h
ig
h
er
s
eg
m
en
tati
o
n
ac
cu
r
ac
y
,
wh
ile
0
in
d
icate
s
n
o
o
v
er
lap
.
T
h
e
r
esear
ch
f
r
a
m
ewo
r
k
d
escr
ib
es
a
wo
r
k
f
lo
w
th
at
u
s
es
a
p
air
o
f
i
m
ag
es
f
r
o
m
th
e
s
am
e
g
e
o
g
r
ap
h
ic
lo
ca
tio
n
tak
e
n
at
d
if
f
er
en
t
tim
es
(
f
o
r
ex
am
p
le,
s
atellite
o
r
ae
r
ial
im
ag
es),
as
s
h
o
wn
in
Fig
u
r
e
1
.
First,
th
e
in
p
u
t
im
ag
es
ar
e
p
r
ep
r
o
ce
s
s
ed
th
r
o
u
g
h
:
r
esize
to
a
u
n
if
o
r
m
s
ca
le,
n
o
r
m
aliza
tio
n
,
ex
tr
ac
tio
n
o
f
2
5
6
×
2
5
6
p
atch
es,
en
co
d
e
a
n
n
o
tatio
n
,
d
ata
au
g
m
en
tatio
n
(
r
o
tatio
n
,
f
lip
p
in
g
,
an
d
z
o
o
m
i
n
g
)
.
Nex
t,
t
h
e
p
air
s
o
f
p
r
ep
r
o
c
ess
ed
im
ag
es
ar
e
f
ee
d
in
to
th
e
U
-
Net
n
etwo
r
k
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
ch
an
g
e
d
etec
tio
n
,
o
u
tp
u
t f
ea
t
u
r
e
m
ap
s
th
at
h
ig
h
lig
h
t th
e
af
f
ec
te
d
ar
e
as.
Ar
ch
itectu
r
al
f
lo
wc
h
ar
t
a
n
d
p
r
o
ce
s
s
d
iag
r
am
o
f
th
e
f
r
am
ewo
r
k
r
esp
ec
tiv
el
y
:
f
ir
s
t
u
s
e
im
ag
e
to
i
n
p
u
t
,
s
ec
o
n
d
to
p
r
ep
r
o
ce
s
s
im
ag
es,
th
ir
d
u
s
e
d
ee
p
lear
n
in
g
to
d
etec
t
wh
eth
er
it
is
v
eg
etatio
n
,
i
f
n
o
t
th
e
n
d
etec
t
i
f
v
eg
etatio
n
l
o
s
s
,
an
d
th
en
to
d
ec
id
e
h
o
w
to
m
in
im
ize
v
eg
etatio
n
lo
s
s
as
s
h
o
wn
in
th
e
Fig
u
r
es
1
t
o
3
.
T
h
e
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
f
o
llo
ws
a
s
y
s
tem
atica
lly
p
r
o
ce
d
u
r
e
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
,
class
if
icatio
n
,
an
d
d
e
p
lo
y
m
e
n
t
u
s
in
g
C
ap
s
Net
s
an
d
GC
N
as
s
h
o
wn
i
n
th
e
Fig
u
r
e
2
.
T
h
e
p
r
o
ce
d
u
r
e
s
tar
ts
with
d
ata
c
o
llectio
n
f
r
o
m
av
ailab
le
s
o
u
r
ce
s
,
f
o
llo
wed
b
y
p
r
e
p
r
o
ce
s
s
in
g
o
f
th
e
d
ata
th
at
i
n
clu
d
es
clea
n
in
g
,
n
o
r
m
aliza
tio
n
,
d
en
o
is
in
g
,
an
d
en
h
an
ce
m
e
n
t
to
in
cr
ea
s
e
th
e
d
ata
q
u
ality
,
f
in
ally
s
p
litt
in
g
th
e
d
ata
in
to
tr
ain
in
g
,
v
alid
atio
n
,
an
d
test
s
ets
.
T
h
e
p
r
ep
r
o
ce
s
s
ed
d
ata
is
f
ir
s
t
p
a
s
s
ed
th
r
o
u
g
h
C
ap
s
Net
wh
er
e
lo
ca
l,
s
p
atial
f
ea
tu
r
es
ar
e
ex
tr
ac
te
d
an
d
en
co
d
ed
in
a
h
ier
ar
ch
ical
r
ep
r
esen
tatio
n
,
w
h
ich
p
r
eser
v
e
th
e
r
elatio
n
s
h
ip
s
b
etwe
en
o
b
jects.
Fig
u
r
e
1
.
Pro
p
o
s
ed
wo
r
k
f
r
am
ewo
r
k
Fig
u
r
e
2
.
Ar
c
h
itectu
r
al
f
lo
wch
ar
t a
n
d
m
eth
o
d
o
l
o
g
y
d
iag
r
am
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.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
9
3
3
-
494
5
4938
I
n
p
u
t
i
m
a
g
e
S
e
g
m
e
n
t
e
d
i
ma
g
e
Fig
u
r
e
3
.
Sam
p
le
i
n
p
u
t im
a
g
e
u
s
ed
f
o
r
tr
ai
n
in
g
a
n
d
s
eg
m
e
n
ted
im
ag
e
T
h
ese
h
ig
h
-
lev
el
f
ea
tu
r
e
v
ec
t
o
r
s
th
en
f
ee
d
in
to
a
GC
N,
wh
ich
lear
n
s
co
m
p
lex
d
e
p
en
d
e
n
cies
an
d
r
elatio
n
s
with
in
an
d
b
etwe
en
d
ata
s
am
p
les
.
A
r
ec
o
n
s
tr
u
ctio
n
lay
er
th
en
m
ap
s
th
e
GC
N
f
ea
tu
r
es
b
ac
k
to
th
e
lo
wer
-
lev
el
f
ea
tu
r
es,
p
r
eser
v
i
n
g
k
ey
in
f
o
r
m
atio
n
f
r
o
m
th
e
ca
p
s
u
les
an
d
ac
ts
as
a
r
eg
u
lar
izatio
n
ter
m
.
T
h
ese
r
ec
o
n
s
tr
u
cted
f
ea
tu
r
es
ar
e
in
p
u
t
in
to
a
class
if
icatio
n
la
y
er
t
o
g
en
e
r
ate
th
e
o
u
tp
u
t
p
r
ed
ictio
n
s
,
i.e
.
,
class
lab
els.
Mo
d
el
p
a
r
am
eter
s
ar
e
o
p
tim
ized
b
y
tr
ain
i
n
g
an
d
u
tili
zin
g
th
e
b
ac
k
p
r
o
p
ag
atio
n
alg
o
r
ith
m
an
d
a
n
a
p
p
r
o
p
r
iate
o
p
tim
izer
to
m
i
n
im
ize
class
if
icatio
n
er
r
o
r
,
th
e
n
ev
alu
ate
t
h
e
tr
ain
ed
m
o
d
el
u
s
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
a
n
d
a
n
aly
ze
th
e
r
esu
lts
.
C
o
m
p
ar
e
its
p
er
f
o
r
m
an
c
e
with
o
t
h
er
m
eth
o
d
s
to
test
i
f
th
er
e
is
s
ig
n
if
ican
t
im
p
r
o
v
em
e
n
t
in
p
e
r
f
o
r
m
an
ce
an
d
co
m
p
ar
e
ag
ai
n
s
t
th
em
.
Af
ter
v
alid
,
m
o
d
el
is
r
ea
d
y
f
o
r
d
ep
lo
y
m
en
t
t
h
at
ca
n
b
e
u
tili
ze
d
to
r
ea
l
wo
r
ld
p
r
o
b
lem
s
o
f
d
if
f
er
en
t
ar
ea
s
lik
e,
r
em
o
te
s
en
s
in
g
,
in
tellig
en
t
tr
an
s
p
o
r
tatio
n
,
a
n
d
cy
b
er
s
ec
u
r
ity
.
T
h
is
s
tu
d
y
d
esig
n
s
a
f
r
am
ew
o
r
k
b
ased
o
n
C
ap
s
Net
an
d
GC
N
wh
ich
tak
es
a
d
v
an
tag
e
o
f
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
lear
n
in
g
s
p
atial
co
r
r
elatio
n
s
an
d
g
lo
b
al
r
elati
o
n
s
h
ip
s
.
Seg
m
en
tatio
n
ac
c
u
r
a
cy
is
d
ef
in
ed
b
ased
o
n
th
e
o
v
e
r
lap
o
f
th
e
p
r
e
d
icted
r
eg
io
n
s
an
d
th
e
g
r
o
u
n
d
tr
u
t
h
r
eg
io
n
s
.
T
h
e
p
r
e
d
ictio
n
is
c
o
n
s
id
er
ed
c
o
r
r
ec
t
i
f
th
e
I
o
U
is
at
least
0
.
5
,
i.e
.
,
th
e
p
r
ed
icted
m
ask
a
n
d
th
e
g
r
o
u
n
d
tr
u
th
r
e
g
io
n
h
av
e
at
least
5
0
%
in
ter
s
ec
tio
n
o
v
er
th
e
u
n
io
n
.
Pre
cisi
o
n
I
o
U
≥
0
.
5
is
d
ef
in
ed
as
in
(
2
)
.
≥
0
.
5
=
≥
0
.
5
≥
0
.
5
+
≥
0
.
5
(
2
)
W
h
er
e
in
(
2
)
d
en
o
tes
th
e
p
r
ec
i
s
io
n
ca
lcu
lated
at
an
I
o
U
th
r
es
h
o
ld
o
f
0
.
5
o
r
g
r
ea
ter
.
He
r
e,
≥
0
.
5
r
ep
r
esen
ts
th
e
tr
u
e
p
o
s
itiv
e
d
etec
tio
n
s
t
h
at
ac
h
iev
e
a
n
I
o
U
≥
0
.
5
,
wh
ile
≥
0
.
5
d
en
o
tes
th
e
co
r
r
esp
o
n
d
in
g
f
alse
p
o
s
itiv
e
d
etec
tio
n
s
at
th
e
s
a
m
e
th
r
esh
o
l
d
.
R
ec
all:
I
o
U
≥
0
.
5
;
r
ec
all
m
ea
s
u
r
es
th
e
ab
ili
ty
o
f
th
e
m
o
d
el
to
co
r
r
ec
tly
id
e
n
tify
all
r
elev
a
n
t in
s
tan
ce
s
an
d
is
d
ef
in
ed
as
in
(
3
)
.
≥
0
.
5
=
≥
0
.
5
≥
0
.
5
+
≥
0
.
5
(
3
)
W
h
er
e
in
(
3
)
m
ea
n
s
f
alse
n
e
g
ativ
es
r
elate
d
to
TP
r
eg
i
o
n
s
with
I
o
U
v
alu
es
u
n
d
er
0
.
5
.
T
o
m
ea
s
u
r
e
m
o
d
el
p
er
f
o
r
m
an
ce
,
th
is
s
tu
d
y
u
s
e
s
two
ag
g
r
eg
ate
s
co
r
es:
wh
e
n
s
u
m
m
ed
o
v
er
all
g
r
o
u
n
d
t
r
u
th
o
b
jects
an
d
a
I
o
U
o
f
0
.
5
th
e
y
d
ef
in
e
two
s
u
m
m
ar
y
m
ea
s
u
r
es
(
av
er
a
g
e
p
r
e
cisi
o
n
(
AP
)
at
I
o
U
0
.
5
an
d
av
er
ag
e
r
ec
all
(
AR
)
at
I
o
U
0
.
5
)
t
h
at
ca
n
b
e
c
o
m
b
in
e
d
with
p
er
r
eg
io
n
m
et
r
ics
an
d
o
v
er
all
I
o
U
.
T
h
ese
r
e
p
r
esen
t
a
n
u
n
am
b
ig
u
o
u
s
way
to
ju
d
g
e
p
er
f
o
r
m
a
n
ce
o
f
an
in
s
tan
ce
s
eg
m
en
tatio
n
m
eth
o
d
f
o
r
s
atellite
im
ag
es.
Usi
n
g
th
is
s
tan
d
ar
d
as
in
s
p
ir
atio
n
th
is
s
tu
d
y
d
ev
elo
p
a
two
-
s
tag
e
in
s
tan
ce
s
eg
m
en
tatio
n
p
ip
elin
e.
A
g
r
ap
h
c
o
n
v
o
lu
ti
o
n
al
n
eu
r
al
n
etwo
r
k
(
GC
NN
)
,
f
ir
s
t
en
co
d
es
n
o
d
e
an
d
ed
g
e
f
ea
tu
r
es
d
er
iv
ed
f
r
o
m
th
e
g
r
id
th
at
r
e
p
r
esen
ts
th
e
im
ag
e
,
p
r
o
v
id
i
n
g
a
r
e
p
r
esen
tatio
n
o
f
g
lo
b
al
s
p
atial
d
ep
en
d
e
n
cies
an
d
lo
n
g
-
r
an
g
e
s
tr
u
ctu
r
al
p
atte
r
n
s
ac
r
o
s
s
th
e
en
tire
im
ag
e
[2
9
]
.
A
C
ap
s
Net
,
C
NN
(
ca
p
s
u
le)
p
r
o
ce
s
s
es
th
is
r
e
p
r
esen
tatio
n
a
n
d
r
ef
in
es
it
in
t
o
s
eg
m
en
tatio
n
s
b
y
m
o
d
ellin
g
r
elatio
n
s
h
ip
s
o
f
p
a
r
ts
o
f
th
e
s
eg
m
en
tatio
n
b
ein
g
with
in
s
eg
m
en
tatio
n
f
o
r
e
x
a
m
p
le.
A
f
u
ll
e
n
d
to
en
d
s
y
s
tem
tr
ain
ed
o
n
r
aw
im
ag
es
with
ac
co
m
p
a
n
y
in
g
m
eta
d
ata,
th
e
n
etwo
r
k
o
u
tp
u
ts
o
n
e
s
eg
m
en
tatio
n
m
ask
p
er
in
s
tan
ce
.
T
h
is
s
tu
d
y
d
is
c
o
v
er
ed
in
ea
r
ly
ex
p
er
im
en
ts
th
at
tin
y
b
u
ild
in
g
s
,
o
f
ten
ap
p
e
ar
in
g
clu
s
ter
ed
,
ar
e
h
ar
d
to
d
is
tin
g
u
is
h
f
r
o
m
o
n
e
a
n
o
th
er
an
d
th
eir
im
p
ac
t
o
n
p
er
f
o
r
m
an
ce
is
h
id
d
e
n
with
in
AP
an
d
AR
m
ea
s
u
r
es,
wh
ich
g
iv
e
e
q
u
al
weig
h
t to
ev
er
y
o
b
ject
ir
r
esp
ec
tiv
e
o
f
p
h
y
s
ical
s
ca
le.
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
S
a
telite
ima
g
ery
a
n
d
d
ee
p
lea
r
n
in
g
fr
a
mewo
r
k
fo
r
h
ig
h
w
a
y
c
o
n
s
tr
u
ctio
n
w
ith
min
ima
l
…
(
Meen
a
ksh
i
)
4939
T
h
is
s
tu
d
y
co
m
b
at
th
is
th
r
o
u
g
h
d
esig
n
i
n
g
o
u
r
lo
s
s
f
u
n
ctio
n
with
two
c
o
m
p
lem
e
n
tar
y
weig
h
tin
g
s
ch
em
es,
th
e
f
ir
s
t
u
p
weig
h
ts
p
ix
els
n
ea
r
o
th
er
s
eg
m
en
ts
th
u
s
s
tr
en
g
th
en
in
g
th
e
b
o
r
d
er
s
o
f
ad
jace
n
t
o
b
jects,
wh
ile
th
e
s
ec
o
n
d
in
c
r
ea
s
es
th
e
weig
h
ts
o
f
s
m
all
s
eg
m
en
ts
s
o
th
ey
ar
e
n
o
t
u
n
d
e
r
r
ep
r
esen
t
ed
as
th
ey
co
n
tain
a
r
elativ
ely
s
m
all
p
r
o
p
o
r
tio
n
o
f
th
e
o
v
e
r
all
p
ix
els.
I
t
is
co
m
p
o
s
ed
o
f
two
lo
s
s
co
m
p
o
n
en
ts
:
a
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
ter
m
p
en
alizin
g
class
if
icatio
n
ac
cu
r
ac
y
an
d
a
Dice
lo
s
s
ter
m
wh
ich
p
en
alize
s
in
co
r
r
ec
t sp
at
ial
lo
ca
lizatio
n
an
d
wea
k
lo
ca
lizatio
n
,
th
is
s
tu
d
y
f
o
u
n
d
th
at
tr
ain
in
g
with
a
co
m
b
in
atio
n
o
u
t
p
er
f
o
r
m
e
d
ea
c
h
co
m
p
o
n
e
n
t
o
n
its
o
wn
.
I
m
ag
e
p
atch
es
wer
e
p
a
d
d
ed
(
r
ef
lectiv
ely
)
t
o
s
ize
3
2
0
×
3
2
0
th
is
e
n
s
u
r
ed
th
at
th
e
wh
o
le
o
b
ject
was
p
r
eser
v
ed
wh
en
s
eg
m
en
ts
wer
e
tiled
alo
n
g
b
o
r
d
er
s
an
d
it
is
an
ef
f
ec
tiv
e
way
to
s
to
p
o
b
jects
f
r
o
m
b
ein
g
cr
o
p
p
e
d
.
Gr
o
u
n
d
tr
u
th
m
ask
s
wer
e
g
en
er
ated
f
r
o
m
p
o
l
y
g
o
n
-
b
ase
d
an
n
o
tatio
n
s
s
o
th
at
we
ca
n
p
r
o
v
id
e
p
ix
el
-
b
ased
p
r
ed
ictio
n
s
f
o
r
c
o
m
p
ar
is
o
n
with
th
e
lo
s
s
f
u
n
cti
o
n
wh
ich
ca
n
also
ac
t
o
n
p
ix
els
d
ir
ec
tly
,
it
also
en
s
u
r
ed
th
at
we
h
a
d
th
e
s
am
e
ty
p
e
o
f
i
n
p
u
t
f
o
r
th
e
lo
s
s
f
u
n
ctio
n
an
d
o
u
r
n
etwo
r
k
'
s
p
r
ed
ictio
n
s
.
A
v
ar
iety
o
f
co
m
m
o
n
im
ag
e
au
g
m
en
tatio
n
s
wer
e
ap
p
lied
d
u
r
in
g
tr
ain
in
g
,
wh
ich
h
elp
s
th
e
n
etwo
r
k
to
g
en
er
alize
.
T
h
ese
wer
e
h
o
r
izo
n
tal
a
n
d
v
er
tical
f
lip
p
in
g
,
r
an
d
o
m
r
o
tatio
n
s
o
f
u
p
to
4
a
r
o
u
n
d
ev
er
y
9
0
,
a
s
m
all
v
ar
iatio
n
i
n
co
n
tr
ast,
an
d
a
s
m
all
v
ar
iatio
n
in
b
r
ig
h
t
n
ess
.
T
h
e
n
etwo
r
k
was
tr
ain
ed
with
th
e
1
-
cy
cle
lea
r
n
in
g
r
ate
s
ch
e
d
u
le
to
h
elp
s
p
ee
d
u
p
c
o
n
v
e
r
g
en
ce
an
d
im
p
r
o
v
e
g
en
er
aliza
tio
n
o
f
th
e
f
in
al
n
etwo
r
k
[
30
]
.
T
ab
le
1
s
h
o
ws 0
.
9
2
5
o
f
AP a
n
d
0
.
9
4
8
o
f
AR
at
I
o
U
=
0
.
5
th
r
esh
o
ld
s
f
o
r
h
eld
o
u
t te
s
t set d
em
o
n
s
tr
ate
m
o
d
el'
s
ca
p
ac
ity
f
o
r
th
is
co
m
p
licated
s
atellite
im
ag
es
s
eg
m
en
tatio
n
ap
p
licatio
n
.
T
h
r
o
u
g
h
th
at,
in
f
r
astru
ctu
r
e
ca
n
b
e
ev
alu
ated
c
o
n
s
id
er
in
g
th
eir
ef
f
ec
ts
o
n
v
eg
etatio
n
co
v
er
r
ed
u
cti
o
n
d
u
e
t
o
ex
p
a
n
s
io
n
o
f
in
f
r
astru
ctu
r
e,
g
r
o
wth
o
f
u
r
b
an
a
r
ea
,
a
n
d
lan
d
c
o
v
er
ch
an
g
es.
I
n
s
p
ir
ed
b
y
ef
f
icien
t
tr
ain
in
g
tech
n
iq
u
e
s
d
em
o
n
s
tr
ated
at
Stan
f
o
r
d
’
s
DAWNB
en
ch
co
m
p
etitio
n
-
a
co
n
test
d
esig
n
ed
to
r
ewa
r
d
r
a
p
id
,
l
o
w
-
co
s
t
tr
ain
i
n
g
o
n
b
e
n
ch
m
ar
k
s
s
u
ch
as
I
m
a
g
eNe
t
an
d
C
I
FAR
-
10
-
we
ad
o
p
ted
t
h
e
1
-
cy
cle
lear
n
in
g
r
ate
p
o
licy
d
ev
el
o
p
ed
i
n
ea
r
lier
wo
r
k
an
d
im
p
lem
en
ted
u
s
in
g
th
e
f
ast
AI
lib
r
ar
y
.
T
h
e
p
o
licy
cy
cles
t
h
e
lear
n
i
n
g
r
ate
th
r
o
u
g
h
o
u
t
a
s
in
g
le
tr
ain
in
g
r
u
n
.
T
h
is
ap
p
r
o
ac
h
all
o
ws
tr
ain
in
g
to
r
ap
id
ly
e
x
p
lo
r
e
t
h
e
lo
s
s
lan
d
s
ca
p
e
ef
f
icien
tly
wh
ile
av
o
id
in
g
th
e
s
h
allo
w,
non
-
g
en
er
alizin
g
s
h
ar
p
m
i
n
im
a.
T
h
e
Sto
ch
asti
c
g
r
a
d
ien
t
d
escen
t
(
SGD)
m
eth
o
d
with
m
o
m
en
tu
m
—
wh
ich
,
ac
co
r
d
in
g
to
th
is
s
tu
d
y
,
h
as
p
r
o
v
en
to
b
e
r
o
b
u
s
t
an
d
ca
p
a
b
le
o
f
p
r
o
d
u
cin
g
m
o
d
els
with
b
etter
g
en
er
aliza
tio
n
ab
ilit
y
—
was
u
s
ed
with
a
b
at
ch
s
ize
o
f
3
2
.
T
r
ain
in
g
was
p
er
f
o
r
m
e
d
in
s
ev
er
al
p
h
ase
s
:
I
n
itially
,
a
b
r
ief
war
m
-
u
p
p
h
ase
tr
ain
ed
o
n
ly
th
e
n
o
n
-
p
r
etr
ain
e
d
d
ec
o
d
er
lay
e
r
s
f
o
r
o
n
e
ep
o
c
h
with
a
lear
n
i
n
g
r
ate
o
f
6
.
T
h
is
allo
wed
th
e
r
an
d
o
m
ly
in
itialized
d
ec
o
d
e
r
lay
e
r
s
to
s
tab
ilize
with
o
u
t
d
is
r
u
p
tin
g
t
h
e
lear
n
ed
f
ea
tu
r
e
m
ap
s
o
f
th
e
p
r
etr
ai
n
ed
en
co
d
er
.
Af
ter
th
is
,
th
e
en
tire
n
etwo
r
k
was
u
n
f
r
o
ze
n
a
n
d
tr
a
in
ed
f
o
r
2
0
e
p
o
ch
s
ac
co
r
d
in
g
t
o
th
e
1
-
c
y
cle
p
o
lic
y
:
lear
n
in
g
r
ates
in
cr
ea
s
ed
lin
ea
r
ly
f
r
o
m
0
.
1
5
t
o
6
o
v
er
th
e
f
ir
s
t
4
5
%
o
f
th
e
r
u
n
to
allo
w
b
r
o
ad
e
x
p
lo
r
atio
n
an
d
escap
e
lo
ca
l m
in
im
a;
r
ates d
ec
r
ea
s
ed
lin
ea
r
ly
b
ac
k
to
0
.
1
5
b
y
th
e
9
0
% p
o
in
t to
s
ettle
in
to
a
s
tab
le
b
asin
;
an
d
f
in
ally
,
r
ates
wer
e
r
ed
u
ce
d
lin
ea
r
ly
to
0
.
0
0
1
5
o
v
er
th
e
f
in
al
1
0
%
f
o
r
f
in
e
co
n
v
er
g
en
ce
.
M
o
m
en
tu
m
was
ad
ju
s
ted
in
v
er
s
ely
t
o
lear
n
in
g
r
ate
d
u
r
i
n
g
all
1
-
cy
cle
tr
ain
in
g
p
h
ases
,
s
o
as
n
o
t
to
o
v
er
s
h
o
o
t
th
e
lo
s
s
at
h
ig
h
er
lear
n
in
g
r
ates
b
u
t
s
till
m
ain
tain
s
o
m
e
d
o
wn
war
d
m
o
tio
n
at
lo
w
o
n
es.
Fo
llo
win
g
th
e
f
u
ll
2
0
ep
o
ch
1
-
cy
cle
tr
ain
in
g
,
th
is
s
tu
d
y
p
e
r
f
o
r
m
ed
two
ad
d
itio
n
al
cy
cles
o
f
f
in
e
-
tu
n
in
g
with
lear
n
in
g
r
ates
o
s
cillatin
g
b
etwe
en
0
.
0
5
an
d
1
b
e
f
o
r
e
en
d
in
g
with
a
lear
n
in
g
r
ate
o
f
0
.
0
5
i
n
th
e
f
in
al
c
y
cle
to
r
ef
in
e
th
e
m
o
d
el
with
o
u
t r
is
k
in
g
ca
tast
r
o
p
h
ic
f
o
r
g
ettin
g
.
Tab
le
1
.
Train
i
n
g
m
a
tri
x
fo
r
th
e
d
a
tas
e
t
N
u
mb
e
r
o
f
e
p
o
c
h
s
Ti
me
f
o
r
t
r
a
i
n
i
n
g
A
c
c
u
r
a
c
c
y
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
C
y
c
l
e
1
10
8
4
ms/s
t
e
p
9
7
.
2
3
9
5
.
0
1
9
6
.
1
4
C
y
c
l
e
1
w
i
t
h
c
y
l
i
n
d
r
i
c
a
l
l
e
a
r
n
i
n
g
r
a
t
e
X
2
50
9
4
ms/s
t
e
p
9
8
.
3
4
9
6
.
4
5
9
7
.
5
6
Ov
e
r
a
ll
,
a
c
o
m
b
i
n
ati
o
n
o
f
t
h
i
s
m
u
l
ti
-
p
h
as
e
s
t
r
at
eg
y
–
w
ar
m
-
u
p
,
f
u
ll
1
-
c
y
cle
,
a
n
d
c
y
cl
ica
l
f
i
n
e
-
t
u
n
i
n
g
y
ie
ld
e
d
r
ap
id
co
n
v
e
r
g
e
n
c
e
w
h
ile
e
n
s
u
r
in
g
r
o
b
u
s
t
g
e
n
e
r
ali
za
t
io
n
.
T
h
e
m
o
d
e
l
was
t
r
a
in
e
d
f
o
r
b
o
t
h
6
-
l
a
b
el
an
d
3
-
la
b
e
l
v
ar
iat
io
n
s
o
f
th
e
s
e
m
a
n
tic
s
e
g
m
e
n
t
ati
o
n
o
f
a
er
ial
im
a
g
e
r
y
d
atas
et
u
s
i
n
g
e
x
a
m
p
le
i
m
ag
es,
w
it
h
t
h
e
l
att
er
p
r
o
d
u
ci
n
g
a
m
o
r
e
co
a
r
s
el
y
g
r
ain
e
d
o
u
tp
u
t
cl
ass
i
f
ic
ati
o
n
.
R
o
u
te
o
p
ti
m
i
za
t
io
n
al
g
o
r
i
th
m
s
:
o
n
t
h
is
s
u
r
f
ac
e
,
th
e
co
s
t
o
f
s
t
e
p
p
in
g
f
r
o
m
ce
l
l
u
t
o
a
n
8
-
c
o
n
n
e
cte
d
n
ei
g
h
b
o
r
v
is
th
e
av
er
a
g
e
t
r
a
v
e
r
s
a
l
co
s
t
o
f
t
h
e
tw
o
c
ells
,
s
ca
le
d
b
y
t
h
e
E
u
cli
d
e
a
n
s
t
e
p
le
n
g
th
(
,
)
,
wh
i
ch
is
1
f
o
r
o
r
t
h
o
g
o
n
al
m
o
v
es a
n
d
√
2
f
o
r
d
i
ag
o
n
a
l
o
n
es
as
in
(
4
)
.
(
,
)
=
1
2
(
[
]
+
[
]
)
⋅
(
,
)
(
4
)
E
d
g
e
weig
h
t
f
u
n
ctio
n
g
i
v
en
in
(
4
)
,
wh
ich
ex
p
lain
ab
o
u
t
th
e
tr
av
er
s
al
co
s
t
o
f
an
ed
g
e
jo
i
n
in
g
n
o
d
es
an
d
is
tak
e
n
as
t
h
e
m
e
an
o
f
th
e
en
d
p
o
in
t
co
s
ts
,
s
ca
led
b
y
th
e
g
e
o
m
etr
ic
d
is
tan
ce
b
et
wee
n
th
em
:
wh
er
e
,
(
,
)
is
weig
h
t
(
tr
av
e
r
s
al
co
s
t)
ass
ig
n
ed
to
t
h
e
ed
g
e
b
et
wee
n
n
o
d
es
an
d
,
[
]
,
[
]
is
ec
o
lo
g
ical/v
eg
etatio
n
co
s
t
v
alu
e
at
n
o
d
es
an
d
r
esp
ec
tiv
ely
,
(
,
)
is
E
u
clid
ea
n
(
o
r
g
eo
d
esic)
d
is
tan
c
e
b
etwe
en
n
o
d
es
an
d
in
m
etr
e
s
.
T
h
e
f
ac
t
o
r
½
a
v
er
ag
es
t
h
e
e
n
d
p
o
in
t
co
s
ts
,
s
o
t
h
at
an
ed
g
e
cr
o
s
s
in
g
f
r
o
m
a
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.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
9
3
3
-
494
5
4940
lo
w
-
co
s
t
ce
ll
in
to
a
h
ig
h
-
co
s
t
ce
ll
is
p
en
alize
d
p
r
o
p
o
r
tio
n
a
lly
to
th
e
m
ea
n
i
m
p
ac
t
alo
n
g
its
s
p
an
,
wh
ile
th
e
m
u
ltip
licatio
n
b
y
(
,
)
m
ak
es
th
e
weig
h
t
ex
ten
s
iv
e
in
p
ath
len
g
th
.
T
h
is
s
tu
d
y
co
m
p
ar
es
f
o
u
r
p
lan
n
er
s
.
Dijk
s
tr
a
r
u
n
s
a
u
n
if
o
r
m
-
c
o
s
t
s
ea
r
ch
an
d
r
etu
r
n
s
a
p
r
o
v
a
b
ly
o
p
tim
al
least
-
co
s
t
p
ath
,
b
u
t
it
ex
p
an
d
s
n
o
d
es
in
ev
er
y
d
ir
ec
tio
n
[
2
5
]
.
A*
ad
d
s
an
ad
m
is
s
ib
le
E
u
clid
ea
n
h
eu
r
is
tic
ℎ
(
,
)
=
‖
−
‖
,
with
th
e
s
m
allest
ce
ll
co
s
t;
th
is
s
teer
s
t
h
e
s
ea
r
ch
to
war
d
th
e
tar
g
et
an
d
cu
ts
th
e
n
u
m
b
er
o
f
e
x
p
an
s
io
n
s
wh
ile
k
ee
p
in
g
th
e
p
ath
o
p
tim
al
o
n
t
h
e
co
m
p
o
s
ite
s
u
r
f
ac
e.
Pro
p
o
s
ed
E
co
-
Awa
r
e
A*
.
T
h
e
p
r
o
p
o
s
ed
p
lan
n
er
ad
d
s
an
e
x
p
licit
v
eg
etatio
n
-
av
o
id
an
ce
t
er
m
to
th
e
ed
g
e
m
o
d
el,
co
n
tr
o
lled
b
y
a
w
eig
h
t λ,
s
o
th
e
s
ea
r
ch
lean
s
to
war
d
lo
w
-
ca
n
o
p
y
c
o
r
r
id
o
r
s
as in
(
5
)
.
(
,
)
=
1
2
(
[
]
+
[
]
)
⋅
(
,
)
+
⋅
[
]
⋅
(
,
)
(
5
)
E
co
-
Awa
r
e
ed
g
e
weig
h
t
(
5
)
,
e
x
p
lain
th
e
h
e
u
r
is
tic
s
tay
s
ad
m
is
s
ib
le
with
r
esp
ec
t
to
th
e
b
ase
co
s
t,
s
o
th
e
s
ea
r
ch
r
em
ain
s
ef
f
icien
t,
an
d
th
e
ex
tr
a
ter
m
p
u
lls
th
e
r
o
u
te
awa
y
f
r
o
m
d
e
n
s
e
v
e
g
etatio
n
.
T
h
is
s
tu
d
y
s
et
λ
=
0
.
9
i
n
th
e
ex
p
er
im
en
ts
.
Sin
ce
o
n
e
f
ix
e
d
weig
h
t
ca
n
n
o
t
ex
p
r
ess
ev
e
r
y
p
la
n
n
er
'
s
p
r
io
r
ity
,
also
p
r
o
v
id
e
a
n
NSGA
-
II
m
u
lti
-
o
b
jectiv
e
s
ea
r
c
h
.
C
an
d
id
ate
co
r
r
id
o
r
s
p
r
o
d
u
ce
d
ac
r
o
s
s
a
r
an
g
e
o
f
ec
o
-
weig
h
ts
ar
e
ev
o
lv
e
d
with
f
ast
non
-
d
o
m
in
ated
s
o
r
tin
g
a
n
d
el
itis
t
s
elec
tio
n
,
m
in
im
izin
g
tw
o
o
b
jectiv
es
at
o
n
ce
,
c
u
m
u
lat
iv
e
len
g
th
-
c
o
s
t
an
d
cu
m
u
lativ
e
v
eg
etatio
n
ex
p
o
s
u
r
e,
an
d
th
e
s
ea
r
ch
r
etu
r
n
s
a
Par
eto
f
r
o
n
t.
A
k
n
ee
-
p
o
in
t
r
u
le
th
en
p
ick
s
th
e
s
o
lu
tio
n
n
ea
r
est th
e
id
ea
l
p
o
in
t
f
o
r
r
e
p
o
r
tin
g
.
T
h
e
r
o
u
te
-
o
p
tim
izatio
n
p
ip
eli
n
e
is
s
elf
-
co
n
tain
ed
a
n
d
r
e
p
r
o
d
u
cib
le.
So
th
at
th
e
wh
o
le
m
eth
o
d
ca
n
r
u
n
with
o
u
t
d
o
wn
lo
a
d
in
g
l
ar
g
e
im
a
g
er
y
tiles
,
it
g
e
n
e
r
ates
a
p
h
y
s
ically
p
lau
s
ib
le
b
en
c
h
m
ar
k
s
ce
n
e
d
eter
m
in
is
tically
:
a
s
m
o
o
th
el
ev
atio
n
f
ield
with
a
h
ig
h
-
s
lo
p
e
r
id
g
e,
a
d
en
s
e
f
o
r
est
b
elt
ly
i
n
g
ac
r
o
s
s
th
e
d
ir
ec
t
lin
e
b
etwe
en
th
e
en
d
p
o
in
ts
,
an
d
an
im
p
ass
ab
le
r
ip
ar
ian
b
ar
r
ier
.
T
h
is
s
etu
p
p
u
ts
th
e
v
eg
etatio
n
v
er
s
u
s
f
ea
s
ib
ilit
y
tr
ad
e
-
o
f
f
i
n
p
lain
v
iew
an
d
lets
all
p
la
n
n
er
s
b
e
co
m
p
a
r
ed
o
n
id
en
tical
in
p
u
t
s
.
Sam
p
le
s
atellite
im
ag
es
u
s
ed
f
o
r
tr
ain
i
n
g
th
e
p
r
o
p
o
s
ed
m
o
d
el
alo
n
g
with
th
e
ir
co
r
r
esp
o
n
d
in
g
s
eg
m
en
te
d
o
u
tp
u
ts
as
s
h
o
wn
in
Fig
u
r
e
3
.
T
h
e
s
eg
m
en
ted
im
ag
es
illu
s
tr
ate
th
e
s
u
cc
ess
f
u
l
ex
tr
ac
tio
n
o
f
v
e
g
etatio
n
a
n
d
o
th
e
r
lan
d
-
co
v
er
r
eg
io
n
s
,
d
e
m
o
n
s
tr
atin
g
t
h
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
s
eg
m
en
tatio
n
a
p
p
r
o
ac
h
in
ac
c
u
r
ately
id
en
tif
y
in
g
ter
r
ain
f
ea
tu
r
es.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
cr
em
en
tally
in
c
r
ea
s
in
g
th
e
n
u
m
b
er
o
f
GC
N
lay
er
s
an
d
b
lo
ck
s
with
in
C
ap
s
Net
m
o
d
u
le
ca
n
s
y
s
tem
atica
lly
r
ed
u
ce
m
o
d
el
m
is
class
if
icatio
n
s
o
f
s
m
aller
o
b
jects
a
n
d
f
u
r
th
e
r
s
ep
ar
at
e
d
en
s
ely
a
r
r
an
g
ed
in
s
tan
ce
s
.
Su
b
s
eq
u
en
tly
,
f
u
r
t
h
er
f
i
n
e
-
tu
n
i
n
g
t
o
s
p
ec
ialize
in
s
eg
m
en
tin
g
ac
cu
r
ate
in
s
ta
n
ce
b
o
u
n
d
ar
ies
an
d
b
etter
s
ep
ar
atin
g
clo
s
ely
-
p
a
ck
ed
s
m
all
o
b
jects
was
co
n
d
u
ct
ed
an
d
th
e
v
e
g
etatio
n
-
d
etec
tio
n
co
d
e
is
s
h
o
w
n
in
th
e
Fig
u
r
e
4
.
T
h
e
s
em
an
tic
s
u
b
d
iv
is
io
n
c
o
d
e
as
s
h
o
wn
in
Fig
u
r
e
5
an
d
r
esu
lts
o
f
th
e
p
r
o
p
o
s
ed
wo
r
k
as
s
h
o
w
n
in
Fig
u
r
e
6
.
W
h
en
th
is
s
tu
d
y
tr
ain
o
n
th
e
3
lab
els
d
ata
s
et,
th
e
s
am
e
m
o
d
el
p
r
o
d
u
c
ed
a
s
lig
h
tly
lo
wer
p
er
f
o
r
m
an
ce
m
etr
ic,
d
u
e
to
cl
ass
im
b
alan
ce
.
I
n
co
n
tr
ast,
th
e
m
o
d
el
tr
ain
ed
o
n
th
e
L
ab
el
d
atasets
ac
h
iev
ed
an
ac
cu
r
ac
y
s
co
r
e
8
5
%.
T
h
e
p
r
o
p
o
s
ed
GC
N
–
C
ap
s
Net
m
o
d
el
was
tr
ain
ed
u
s
in
g
t
h
e
s
atellite
im
ag
er
y
d
ataset
f
o
llo
win
g
th
e
p
r
o
ce
d
u
r
e
d
e
s
cr
ib
ed
in
m
eth
o
d
o
lo
g
y
s
e
ctio
n
.
T
o
id
e
n
tify
th
e
m
o
s
t
ef
f
ec
tiv
e
m
o
d
el
co
n
f
ig
u
r
atio
n
,
th
is
s
tu
d
y
s
y
s
tem
atica
lly
ex
p
er
im
en
ted
with
s
ev
er
al
p
a
r
am
eter
s
d
u
r
in
g
tr
ain
in
g
an
d
e
v
alu
atio
n
.
I
n
p
ar
ticu
lar
,
th
is
s
tu
d
y
v
a
r
i
ed
th
e
s
eg
m
en
tatio
n
co
n
f
id
e
n
ce
th
r
esh
o
ld
(
θ)
f
r
o
m
0
.
3
to
0
.
8
i
n
o
r
d
er
to
d
eter
m
in
e
th
e
s
ettin
g
th
at
y
iel
d
ed
th
e
b
est
in
s
tan
ce
-
lev
el
s
eg
m
en
tatio
n
p
er
f
o
r
m
an
ce
.
I
n
a
d
d
itio
n
t
o
th
r
esh
o
ld
tu
n
in
g
,
m
u
ltip
le
ar
ch
itectu
r
al
co
n
f
ig
u
r
atio
n
s
wer
e
ex
p
lo
r
ed
b
y
ch
an
g
in
g
th
e
d
ep
th
o
f
th
e
n
etwo
r
k
.
Var
io
u
s
co
n
f
ig
u
r
atio
n
s
o
f
th
e
GC
N
an
d
ca
p
s
u
le
b
lo
ck
s
wer
e
co
m
p
ar
ed
an
d
co
n
tr
asted
to
d
eter
m
in
e
th
eir
ef
f
ec
ts
o
n
s
eg
m
en
tatio
n
ac
cu
r
ac
y
.
T
h
e
o
v
er
all
p
ix
el
-
wis
e
ac
cu
r
ac
y
d
id
n
o
t
ch
an
g
e
s
u
b
s
tan
tially
-
p
r
im
ar
ily
s
in
ce
th
e
p
r
o
p
o
r
tio
n
o
f
b
ac
k
g
r
o
u
n
d
p
ix
e
ls
was
f
ar
g
r
ea
te
r
-
b
u
t
ev
e
n
a
f
ew
m
is
p
lace
d
f
o
r
eg
r
o
u
n
d
p
ix
e
ls
p
r
o
v
e
d
to
h
av
e
a
s
tr
o
n
g
im
p
ac
t
o
n
th
e
o
v
er
all
v
alu
es
f
o
r
p
r
ec
is
io
n
a
n
d
r
ec
all.
Fin
ally
,
ev
alu
atio
n
was
p
er
f
o
r
m
ed
o
n
th
e
o
f
f
icial
test
s
et
an
d
a
n
u
m
b
er
o
f
d
if
f
er
en
t k
ey
p
er
f
o
r
m
an
ce
m
et
r
ics,
n
am
ely
AP a
n
d
AR
wer
e
m
ea
s
u
r
ed
.
T
ab
le
2
s
u
m
m
ar
izes
th
e
f
in
al
r
esu
lts
ac
h
iev
ed
o
n
th
e
test
s
et.
T
h
e
d
ee
p
er
Net8
,
Net1
0
an
d
Net1
2
v
ar
ian
ts
,
r
u
n
at
=
0
.
5
co
n
s
is
ten
tly
in
cr
ea
s
e
th
e
d
etec
tio
n
p
er
f
o
r
m
an
ce
f
o
r
s
m
aller
o
b
je
cts,
an
d
also
allo
w
b
etter
s
ep
ar
atio
n
f
o
r
d
en
s
ely
s
p
ac
ed
in
s
tan
ce
s
,
wh
ile
r
ef
in
e
m
en
t
o
f
in
s
tan
ce
b
o
u
n
d
a
r
ies
co
n
tr
ib
u
te
with
s
lig
h
t
im
p
r
o
v
em
e
n
t,
th
e
f
in
al
r
esu
lt
o
n
th
e
h
o
ld
o
u
t
test
s
et
ac
h
iev
i
n
g
AP
=
0
.
9
2
5
an
d
AR
=
0
.
9
4
8
at
th
r
esh
o
ld
0
.
5
in
d
icate
s
th
at
th
e
h
y
b
r
id
ar
c
h
itectu
r
e
is
r
o
b
u
s
t
t
o
s
atellite
im
ag
e
ch
allen
g
es.
I
t
allo
ws
to
id
en
tify
a
n
d
est
im
ate
v
eg
etatio
n
lo
s
s
es
r
esu
ltin
g
f
r
o
m
in
f
r
astru
ctu
r
e
co
n
s
tr
u
ctio
n
,
u
r
b
an
s
p
r
awl
an
d
lan
d
u
s
e
ch
an
g
e,
s
u
ch
as
th
is
allo
ws
to
ev
alu
ate
a
co
n
s
tr
u
ctio
n
o
n
its
s
tr
u
ctu
r
al
h
ea
lth
as
well
as
ec
o
lo
g
ical
f
o
o
tp
r
in
t.
T
h
is
s
tu
d
y
esti
m
ates
o
v
er
all
lan
d
co
v
e
r
class
if
icatio
n
ac
cu
r
ac
y
at
9
8
%.
T
h
e
6
-
la
b
el
m
o
d
el
p
r
o
v
id
es
an
o
v
er
all
im
p
r
o
v
em
en
t
o
v
er
th
e
3
-
la
b
el
co
n
f
ig
u
r
atio
n
an
d
d
ep
th
c
o
n
tin
u
es
to
im
p
r
o
v
e
t
h
e
d
etec
tio
n
o
f
o
b
jects
o
n
an
in
s
tan
ce
-
lev
el
with
Net1
2
p
r
o
v
id
in
g
t
h
e
b
est
AP
an
d
AR
v
alu
es.
T
h
e
p
er
ce
p
tio
n
s
o
b
tain
ed
f
r
o
m
th
e
v
ar
io
u
s
n
etwo
r
k
s
f
o
r
m
p
ar
t
o
f
th
e
ec
o
-
s
en
s
itiv
ity
c
o
s
t
s
u
r
f
ac
e
f
o
r
th
e
r
o
u
te
o
p
tim
izatio
n
.
R
esu
lts
s
h
o
wn
in
T
ab
le
2
d
em
o
n
s
tr
ate
th
at
t
h
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
S
a
telite
ima
g
ery
a
n
d
d
ee
p
lea
r
n
in
g
fr
a
mewo
r
k
fo
r
h
ig
h
w
a
y
c
o
n
s
tr
u
ctio
n
w
ith
min
ima
l
…
(
Meen
a
ksh
i
)
4941
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
in
teg
r
ate
s
s
em
an
tic
im
ag
e
s
eg
m
e
n
tatio
n
,
GC
N
-
C
ap
s
Net
-
b
ased
v
eg
et
atio
n
d
etec
tio
n
an
d
en
v
ir
o
n
m
en
t
-
f
r
ien
d
ly
r
o
u
te
o
p
tim
izatio
n
ap
p
r
o
ac
h
es.
T
h
e
s
eg
m
en
tatio
n
p
er
f
o
r
m
e
d
u
s
in
g
6
s
em
an
tic
class
es
p
r
o
v
id
es
ac
cu
r
ac
y
o
f
ab
o
v
e
8
5
%
with
m
ea
n
I
o
U
o
f
0
.
6
2
0
o
u
tp
er
f
o
r
m
in
g
th
e
3
-
class
ap
p
r
o
ac
h
.
T
h
e
GC
N
-
C
ap
s
Net
ar
ch
itectu
r
e
with
d
ep
th
s
1
2
r
esu
lted
in
th
e
b
est
p
er
f
o
r
m
a
n
ce
f
o
r
AP
(
0
.
9
2
5
)
an
d
AR
(
0
.
9
4
8
)
f
o
r
0
.
5
th
r
esh
o
l
d
.
Ov
er
all,
th
e
f
r
am
ewo
r
k
s
u
cc
ess
f
u
lly
att
ain
ed
a
lan
d
class
if
icatio
n
ac
cu
r
ac
y
o
f
9
8
%
an
d
d
eter
m
in
ed
e
f
f
icien
t r
o
a
d
alig
n
m
en
ts
th
at
m
in
im
ize
th
e
im
p
ac
ts
o
n
v
eg
etatio
n
.
Fig
u
r
e
4
.
Veg
etatio
n
-
d
etec
tio
n
r
o
u
tin
e
c
o
d
e
T
ab
le
2
.
T
r
ai
n
in
g
ac
c
u
r
ac
y
m
a
tr
ix
f
o
r
t
h
e
d
ataset
C
o
m
p
o
n
e
n
t
M
e
t
r
i
c
R
e
s
u
l
t
K
e
y
o
b
ser
v
a
t
i
o
n
S
e
g
m
e
n
t
a
t
i
o
n
m
o
d
e
l
(
3
-
l
a
b
e
l
)
A
c
c
u
r
a
c
y
~
8
3
%
S
l
i
g
h
t
l
y
l
o
w
e
r
d
u
e
t
o
c
l
a
ss i
mb
a
l
a
n
c
e
S
e
g
m
e
n
t
a
t
i
o
n
m
o
d
e
l
(
6
-
l
a
b
e
l
)
A
c
c
u
r
a
c
y
>
8
5
%
I
mp
r
o
v
e
d
b
y
b
e
t
t
e
r
l
a
n
d
-
c
o
v
e
r
r
e
p
r
e
se
n
t
a
t
i
o
n
S
e
g
m
e
n
t
a
t
i
o
n
m
o
d
e
l
(
6
-
l
a
b
e
l
)
M
e
a
n
I
o
U
0
.
6
2
0
Ef
f
e
c
t
i
v
e
p
i
x
e
l
-
l
e
v
e
l
s
e
g
me
n
t
a
t
i
o
n
G
C
N
–
C
a
p
sN
e
t
(
N
e
t
8
)
A
P
/
A
R
0
.
8
9
1
/
0
.
9
1
2
G
o
o
d
p
e
r
f
o
r
ma
n
c
e
a
t
m
o
d
e
r
a
t
e
d
e
p
t
h
G
C
N
–
C
a
p
sN
e
t
(
N
e
t
1
0
)
A
P
/
A
R
0
.
9
0
7
/
0
.
9
3
1
B
e
t
t
e
r
b
o
u
n
d
a
r
i
e
s
t
h
a
n
N
e
t
8
G
C
N
–
C
a
p
sN
e
t
(
N
e
t
1
2
)
A
P
/
A
R
0
.
9
2
5
/
0
.
9
4
8
B
e
st
-
p
e
r
f
o
r
mi
n
g
a
r
c
h
i
t
e
c
t
u
r
e
O
p
t
i
mal
t
h
r
e
s
h
o
l
d
Θ
=
0
.
5
B
e
st
b
a
l
a
n
c
e
H
i
g
h
e
s
t
p
r
e
c
i
si
o
n
–
r
e
c
a
l
l
b
a
l
a
n
c
e
P
r
o
p
o
se
d
f
r
a
m
e
w
o
r
k
C
l
a
s
si
f
i
c
a
t
i
o
n
a
c
c
u
r
a
c
y
9
8
%
R
e
l
i
a
b
l
e
v
e
g
e
t
a
t
i
o
n
-
l
o
s
s i
d
e
n
t
i
f
i
c
a
t
i
o
n
R
o
u
tin
g
p
er
f
o
r
m
an
ce
:
th
e
f
o
u
r
r
o
u
tin
g
s
y
s
tem
s
all
r
an
o
n
an
id
en
tical
b
en
ch
m
ar
k
s
ce
n
e
f
r
o
m
s
ec
tio
n
3
.
6
o
f
th
e
test
s
ec
tio
n
wh
er
e
s
tar
t
an
d
en
d
p
o
i
n
ts
wer
e
p
lace
d
in
o
p
p
o
s
in
g
co
r
n
e
r
s
o
f
th
e
m
ap
,
s
o
th
at
d
ir
ec
t
co
r
r
id
o
r
s
wer
e
f
o
r
ce
d
t
o
cr
o
s
s
th
e
ex
ten
s
iv
e
f
o
r
ested
s
ec
tio
n
.
E
ac
h
r
o
u
te
was
r
an
k
e
d
o
n
f
iv
e
q
u
a
n
titi
es:
p
ath
g
eo
m
etr
ic
len
g
th
(
in
s
tep
s
)
,
p
ath
cu
m
u
lativ
e
tr
av
er
s
al
co
s
t,
av
er
ag
e
s
am
p
le
v
alu
e
f
o
r
m
ea
n
NDVI
alo
n
g
th
e
r
o
u
te,
a
co
u
n
t f
o
r
th
e
am
o
u
n
t o
f
p
at
h
c
o
v
er
e
d
b
y
th
e
h
i
g
h
-
N
DVI
/can
o
p
y
ex
p
o
s
u
r
e
a
r
ea
s
(
t
h
e
p
r
o
p
o
r
tio
n
o
f
th
e
p
ath
s
eg
m
en
ts
h
av
i
n
g
NDVI
>
0
.
4
)
a
n
d
p
ath
av
e
r
ag
e
s
lo
p
e.
T
h
e
r
esu
lts
ar
e
s
h
o
wn
i
n
T
ab
le
3
,
it
s
u
m
m
ar
izes
th
e
co
s
ts
ar
e
lo
wer
f
o
r
th
e
ec
o
-
co
s
t
p
lan
n
er
s
b
ec
au
s
e
th
e
tr
ad
itio
n
al
r
o
u
te
ig
n
o
r
es
v
eg
etatio
n
an
d
is
s
co
r
ed
o
n
th
e
ec
o
s
u
r
f
ac
e
o
n
ly
f
o
r
c
o
m
p
ar
is
o
n
.
Dijk
s
tr
a
an
d
A*
r
etu
r
n
th
e
s
am
e
o
p
tim
al
p
ath
;
A*
r
e
ac
h
es
it
with
f
ewe
r
n
o
d
e
e
x
p
an
s
io
n
s
.
E
c
o
-
Awa
r
e
A*
an
d
th
e
NSGA
-
I
I
k
n
ee
s
o
lu
tio
n
tr
im
th
e
r
o
u
te
s
lig
h
tly
w
h
ile
h
o
ld
in
g
ca
n
o
p
y
ex
p
o
s
u
r
e
at
ze
r
o
.
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.
1
5
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
9
3
3
-
494
5
4942
T
ab
le
3
.
R
o
u
te
-
o
p
tim
izatio
n
c
o
m
p
ar
is
o
n
o
n
th
e
ec
o
-
s
en
s
itiv
ity
co
s
t su
r
f
ac
e
P
l
a
n
n
e
r
Le
n
g
t
h
(
p
x
)
To
t
a
l
c
o
st
M
e
a
n
N
D
V
I
C
a
n
o
p
y
e
x
p
.
(
%)
M
e
a
n
sl
o
p
e
(
°
)
Tr
a
d
i
t
i
o
n
a
l
(
l
e
n
g
t
h
+
s
l
o
p
e
)
1
7
8
.
3
6
3
.
6
0
.
4
1
8
4
9
.
0
6
.
7
8
D
i
j
k
st
r
a
(
e
c
o
c
o
s
t
)
1
9
5
.
3
3
1
.
2
0
.
1
3
1
0
.
0
7
.
2
7
A
*
(
e
c
o
c
o
st
)
1
9
5
.
3
3
1
.
2
0
.
1
3
1
0
.
0
7
.
2
7
Ec
o
-
A
w
a
r
e
A
*
(
p
r
o
p
o
se
d
)
1
9
1
.
9
3
1
.
4
0
.
1
2
9
0
.
0
7
.
6
8
N
S
G
A
-
I
I
(
k
n
e
e
so
l
u
t
i
o
n
)
1
9
1
.
9
3
1
.
4
0
.
1
2
9
0
.
0
7
.
6
8
T
h
e
co
m
p
ar
is
o
n
m
ak
es
th
e
v
eg
etatio
n
v
er
s
u
s
f
ea
s
ib
ilit
y
tr
ad
e
-
o
f
f
s
tar
k
.
T
h
e
co
n
v
en
tio
n
al
p
lan
n
er
,
wh
ich
m
in
im
izes
ju
s
t
len
g
th
an
d
s
lo
p
e,
s
en
d
s
th
e
co
r
r
id
o
r
s
tr
aig
h
t
th
r
o
u
g
h
th
e
f
o
r
est
b
el
t
-
with
m
ea
n
NDVI
0
.
4
1
8
,
r
o
u
g
h
l
y
h
alf
th
e
r
o
u
t
e
lies
u
n
d
er
ca
n
o
p
y
co
v
e
r
.
B
u
t
b
y
r
o
u
tin
g
o
n
th
e
co
m
p
o
s
ite
ec
o
-
s
en
s
itiv
ity
s
u
r
f
ac
e,
th
e
p
ath
b
en
d
s
awa
y
f
r
o
m
th
e
f
o
r
est
b
elt.
Me
an
N
DVI
f
o
r
th
e
p
r
o
p
o
s
ed
E
c
o
-
A
war
e
A*
p
ath
f
alls
to
0
.
1
2
9
(
a
6
9
% r
ed
u
ctio
n
)
,
an
d
t
h
e
f
r
ac
tio
n
u
n
d
er
ca
n
o
p
y
b
ec
o
m
es 0
% f
r
o
m
4
9
%,
m
ea
n
in
g
n
o
v
eg
etatio
n
co
v
er
o
v
er
th
e
co
r
r
id
o
r
.
C
u
m
u
lativ
e
v
eg
etatio
n
ex
p
o
s
u
r
e
is
6
7
%
less
th
an
in
th
e
o
th
er
s
ce
n
ar
io
.
Su
ch
an
ec
o
lo
g
ical
g
ain
ca
n
b
e
ac
h
ie
v
ed
f
o
r
ju
s
t
an
8
%
i
n
cr
ea
s
e
in
th
e
p
ath
len
g
th
an
d
a
s
lig
h
t
i
n
cr
ea
s
e
in
m
ea
n
s
lo
p
e.
Fig
u
r
e
5
o
v
er
lay
s
th
e
ca
lcu
lated
p
at
h
way
s
o
n
th
e
NDVI
d
ata.
T
h
e
r
e
ar
e
two
m
ain
tak
ea
way
s
.
First,
th
e
en
v
ir
o
n
m
en
tal
g
ain
co
m
es
f
r
o
m
th
e
co
s
t
s
u
r
f
ac
e
n
o
t
th
e
p
lan
n
e
r
-
in
f
ac
t
Dijk
s
tr
a
wo
u
ld
f
i
n
d
ex
ac
tly
th
e
s
am
e
p
ath
,
s
o
A*
is
ch
o
s
en
m
er
ely
f
o
r
ef
f
icien
c
y
.
T
h
e
s
ec
o
n
d
is
th
e
n
o
tio
n
th
at
E
co
-
Awa
r
e
A*
w
ith
its
s
in
g
le
tu
n
in
g
weig
h
t,
a
n
d
NSGA
-
I
I
th
r
o
u
g
h
p
r
o
v
id
i
n
g
th
e
f
u
ll
r
an
g
e
o
f
tr
ad
e
-
o
f
f
s
al
o
n
g
th
e
Par
eto
f
r
o
n
t
o
f
f
er
v
er
y
f
lex
ib
le
tr
ad
e
-
o
f
f
co
n
tr
o
l.
W
h
en
r
ea
d
alo
n
g
s
id
e
th
e
s
eg
m
en
tatio
n
s
tu
d
y
,
th
is
wo
r
k
also
s
u
g
g
ests
th
at
th
e
f
r
am
ewo
r
k
u
s
es
b
o
th
p
h
y
s
ical
an
d
ec
o
l
o
g
ical
f
ac
to
r
s
to
ev
alu
ate
in
f
r
astru
ctu
r
e
co
r
r
id
o
r
s
an
d
f
in
d
s
less
v
eg
etatio
n
d
is
tu
r
b
an
ce
i
n
th
e
p
r
o
p
o
s
ed
r
o
u
tes
th
an
ty
p
ical
r
o
u
tes.
I
n
Fig
u
r
e
5
,
t
h
e
tr
ad
itio
n
al
(
len
g
t
h
+
s
lo
p
e
)
r
o
u
te
cr
o
s
s
es
th
e
d
en
s
e
f
o
r
est
b
elt,
wh
ile
th
e
ec
o
-
c
o
s
t p
lan
n
e
r
s
d
eto
u
r
ar
o
u
n
d
it a
lo
n
g
a
lo
w
-
NDVI
co
r
r
id
o
r
.
Fig
u
r
e
5
.
Op
tim
ized
r
o
a
d
co
r
r
id
o
r
s
o
v
e
r
laid
o
n
th
e
NDVI
r
as
ter
T
h
e
co
m
p
a
r
is
o
n
clea
r
ly
illu
s
tr
ates
th
e
tr
ad
e
-
o
f
f
b
etwe
en
v
eg
etatio
n
an
d
f
ea
s
ib
ilit
y
.
T
h
e
d
ef
au
lt
p
lan
n
er
-
wh
ic
h
o
n
ly
s
ee
k
s
to
m
in
im
ize
len
g
t
h
an
d
s
lo
p
e
-
r
o
u
tes
th
e
co
r
r
id
o
r
s
tr
aig
h
t
ac
r
o
s
s
th
e
f
o
r
est
b
elt
;
th
e
r
o
u
te
h
as
an
a
v
er
ag
e
NDVI
o
f
0
.
4
1
8
an
d
alm
o
s
t
h
alf
th
e
r
o
u
te
r
u
n
s
b
en
ea
th
ca
n
o
p
y
co
v
e
r
.
T
h
e
r
o
u
te
wh
ic
h
u
s
es
th
e
co
m
p
o
s
ite
ec
o
-
s
en
s
itiv
ity
s
u
r
f
ac
e
as
its
s
ea
r
ch
p
r
io
r
ity
av
o
id
s
th
e
f
o
r
est
b
elt
b
y
h
u
g
g
in
g
t
h
e
b
o
u
n
d
ar
y
;
t
h
e
av
e
r
ag
e
NDVI
f
o
r
th
is
alter
n
ativ
e
E
co
-
Awa
r
e
A*
r
o
u
te
r
ed
u
ce
s
to
0
.
1
2
9
-
a
r
ed
u
ctio
n
o
f
alm
o
s
t
6
9
%,
an
d
th
e
ex
p
o
s
u
r
e
to
ca
n
o
p
y
co
v
e
r
r
ed
u
ce
d
f
r
o
m
4
9
·
0
to
0
%
-
th
e
r
o
u
te
d
o
es
n
o
t
g
o
b
en
ea
th
an
y
ca
n
o
p
y
.
C
u
m
u
lativ
e
v
eg
etatio
n
ex
p
o
s
u
r
e
d
ec
r
ea
s
es
b
y
6
7
·
2
%.
At
th
e
ex
p
en
s
e
o
f
ju
s
t
o
v
e
r
8
%
ad
d
itio
n
al
p
ath
len
g
t
h
an
d
a
s
m
all
in
cr
ea
s
e
in
av
er
a
g
e
s
lo
p
e,
s
ig
n
if
ican
t
g
ain
s
ar
e
ac
h
iev
ed
in
r
ed
u
ce
d
im
p
ac
t
o
n
v
eg
etatio
n
.
T
h
e
two
r
o
u
tes ca
n
b
e
s
ee
n
o
v
e
r
laid
o
n
th
e
NDVI
r
aster
in
Fig
u
r
e
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.