Remote Sensing,
Journal Year:
2024,
Volume and Issue:
16(2), P. 327 - 327
Published: Jan. 12, 2024
Given
the
large
volume
of
remote
sensing
images
collected
daily,
automatic
object
detection
and
segmentation
have
been
a
consistent
need
in
Earth
observation
(EO).
However,
objects
interest
vary
shape,
size,
appearance,
reflecting
properties.
This
is
not
only
reflected
by
fact
that
these
exhibit
differences
due
to
their
geographical
diversity
but
also
appear
differently
from
different
sensors
(optical
radar)
platforms
(satellite,
aerial,
unmanned
aerial
vehicles
(UAV)).
Although
there
exists
plethora
methods
area
sensing,
given
very
fast
development
prevalent
deep
learning
methods,
still
lack
recent
updates
for
methods.
In
this
paper,
we
aim
provide
an
update
informs
researchers
about
close
sibling
era,
instance
segmentation.
The
integration
will
cover
approaches
data
at
scales
modalities,
such
as
optical,
synthetic
aperture
radar
(SAR)
images,
digital
surface
models
(DSM).
Specific
emphasis
be
placed
on
addressing
label
limitations
era.
Further,
survey
examples
applications
benefited
discuss
future
trends
EO.
Earth system science data,
Journal Year:
2023,
Volume and Issue:
15(11), P. 4749 - 4780
Published: Oct. 30, 2023
Abstract.
In
China,
the
demand
for
a
more
precise
perception
of
national
land
surface
has
become
most
urgent
given
pace
development
and
urbanization.
Constructing
very-high-resolution
(VHR)
land-cover
dataset
China
with
coverage,
however,
is
nontrivial
task.
Thus,
this
an
active
area
research
that
impeded
by
challenges
image
acquisition,
manual
annotation,
computational
complexity.
To
fill
gap,
first
1
m
resolution
national-scale
map
SinoLC-1,
was
established
using
deep-learning-based
framework
open-access
data,
including
global
(GLC)
products,
OpenStreetMap
(OSM),
Google
Earth
imagery.
Reliable
training
labels
were
generated
combining
three
10
GLC
products
OSM
data.
These
images
derived
from
used
to
train
proposed
framework.
This
resolved
label
noise
stemming
mismatch
between
resolution-preserving
backbone,
weakly
supervised
module,
self-supervised
loss
function,
refine
VHR
results
automatically
without
any
annotation
requirement.
Based
on
large-storage
computing
servers,
processing
73.25
TB
obtain
SinoLC-1
covering
entirety
∼
9
600
000
km2,
took
about
months.
The
product
validated
visually
interpreted
validation
set
over
100
random
samples
statistical
collected
official
survey
report
provided
Chinese
government.
showed
achieved
overall
accuracy
73.61
%
κ
coefficient
0.6595.
Validations
every
provincial
region
further
indicated
across
whole
China.
Furthermore,
conformed
reports
misestimation
rate
6.4
%.
addition,
compared
five
other
widely
products.
had
highest
spatial
finest
landscape
details.
conclusion,
as
delivered
primal
support
related
applications
throughout
freely
accessible
at
https://doi.org/10.5281/zenodo.7707461
(Li
et
al.,
2023).
Remote Sensing,
Journal Year:
2024,
Volume and Issue:
16(2), P. 327 - 327
Published: Jan. 12, 2024
Given
the
large
volume
of
remote
sensing
images
collected
daily,
automatic
object
detection
and
segmentation
have
been
a
consistent
need
in
Earth
observation
(EO).
However,
objects
interest
vary
shape,
size,
appearance,
reflecting
properties.
This
is
not
only
reflected
by
fact
that
these
exhibit
differences
due
to
their
geographical
diversity
but
also
appear
differently
from
different
sensors
(optical
radar)
platforms
(satellite,
aerial,
unmanned
aerial
vehicles
(UAV)).
Although
there
exists
plethora
methods
area
sensing,
given
very
fast
development
prevalent
deep
learning
methods,
still
lack
recent
updates
for
methods.
In
this
paper,
we
aim
provide
an
update
informs
researchers
about
close
sibling
era,
instance
segmentation.
The
integration
will
cover
approaches
data
at
scales
modalities,
such
as
optical,
synthetic
aperture
radar
(SAR)
images,
digital
surface
models
(DSM).
Specific
emphasis
be
placed
on
addressing
label
limitations
era.
Further,
survey
examples
applications
benefited
discuss
future
trends
EO.
