Remote Sensing,
Journal Year:
2024,
Volume and Issue:
17(1), P. 90 - 90
Published: Dec. 29, 2024
In
post-flood
disaster
analysis,
accurate
flood
mapping
in
complex
riverine
urban
areas
is
critical
for
effective
risk
management.
Recent
studies
have
explored
the
use
of
water-related
spectral
indices
derived
from
satellite
imagery
combined
with
machine
learning
(ML)
models
to
achieve
this
purpose.
However,
relying
solely
on
can
lead
these
overlook
crucial
contextual
features,
making
it
difficult
distinguish
inundated
other
similar
features
like
shadows
or
wet
roads.
To
address
this,
our
research
explores
a
novel
approach
improve
segmentation
by
integrating
row-wise
cross
attention
(CA)
module
ML
ensemble
learning.
We
apply
method
analysis
Brisbane
Floods
2022,
utilizing
4-band
PlanetScope
and
indices.
Applied
as
pre-processing
step,
CA
fuses
band
index
into
each
peak-flood
image
using
operation.
This
process
amplifies
subtle
differences
between
floodwater
characteristics
while
preserving
complete
landscape
information.
The
CA-fused
datasets
are
then
fed
proposed
model,
which
constructed
four
classic
models.
A
soft
voting
strategy
averages
their
binary
predictions
determine
final
classification
pixel.
Our
demonstrates
that
enhance
sensitivity
individual
areas,
generally
improving
accuracy.
experimental
results
reveal
model
achieves
high
accuracy
(approaching
100%)
dataset.
may
be
affected
overfitting,
indicates
evaluating
additional
reduced
study
encourages
further
optimize
validate
its
generalizability
various
contexts.
