Geographies,
Год журнала:
2023,
Номер
3(3), С. 563 - 573
Опубликована: Авг. 30, 2023
The
creation
of
crop
type
maps
from
satellite
data
has
proven
challenging
and
is
often
impeded
by
a
lack
accurate
in
situ
data.
Street-level
imagery
represents
new
potential
source
that
may
aid
mapping,
but
it
requires
automated
algorithms
to
recognize
the
features
interest.
This
paper
aims
demonstrate
method
for
(i.e.,
maize,
wheat
others)
recognition
street-level
based
on
convolutional
neural
network
using
bottom-up
approach.
We
trained
model
with
highly
dataset
crowdsourced
labelled
Picture
Pile
application.
classification
results
achieved
an
AUC
0.87
wheat,
0.85
maize
0.73
others.
Given
are
two
most
common
food
crops
grown
globally,
combined
ever-increasing
amount
available
imagery,
this
approach
could
help
address
need
improved
global
monitoring.
Challenges
remain
addressing
noise
aspect
buildings,
hedgerows,
automobiles,
etc.)
uncertainties
due
differences
time
day
location.
Such
also
be
applied
developing
other
sets
e.g.,
land
use
mapping
or
socioeconomic
indicators.
Environmental Modelling & Software,
Год журнала:
2023,
Номер
172, С. 105931 - 105931
Опубликована: Дек. 16, 2023
Spatially
explicit
information
on
land
cover
(LC)
is
commonly
derived
using
remote
sensing,
but
the
lack
of
training
data
still
remains
a
major
challenge
for
producing
accurate
LC
products.
Here,
we
develop
computer
vision
methodology
to
extract
from
photos
Land
Use-Land
Cover
Area
Frame
Survey
(LUCAS).
Given
large
number
photographs
available
and
comprehensive
spatial
coverage,
objective
show
how
automatic
classification
could
be
used
reference
sets
validation
products
as
well
other
purposes.
We
first
selected
representative
sample
1120
covering
eight
types
across
European
Union.
then
applied
semantic
segmentation
these
neural
network
(Deeplabv3+)
trained
with
ADE20k
dataset.
For
each
photo,
extracted
original
identified
by
LUCAS
surveyor,
segmented
objects,
pixel
count
class.
Using
latter
input
features,
Random
Forest
model
classify
photo.
Examining
relationship
between
objects/features
Deeplabv3+
labels
provided
surveyors
demonstrated
classes
can
decomposed
into
multiple
highlighting
complexity
photographs.
The
results
mean
F1
Score
89%,
increasing
93%
when
Wetland
class
not
considered.
Based
results,
this
approach
holds
promise
automated
retrieval
rich
source
geo-referenced
now
becoming
through
social
media
sites
like
Mapillary
or
Google
Street
View.
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,
Год журнала:
2023,
Номер
16, С. 7662 - 7674
Опубликована: Янв. 1, 2023
The
local
climate
zone
(LCZ)
classification
scheme
is
effective
for
climatic
studies,
and
thus
timely
accurate
LCZ
mapping
becomes
critical
scientific
research.
Remote
sensing
images
can
efficiently
capture
the
information
of
large-scale
landscapes
overhead,
while
street-level
supplement
ground-level
information,
helping
improve
mapping.
Previous
study
has
proven
usefulness
in
enhancing
results,
however,
how
they
help
to
results
still
remains
unexplored.
To
unveil
underlying
mechanism
fill
gap,
this
study,
feature
importance
analysis
performed
on
experiments
using
different
data
sources
reveal
contributions
components,
correlation
adopted
find
relationship
between
street
view
key
indicators.
show
that
fusing
performance
considerably,
especially
compact
urban
types
such
as
highrise
midrise.
In
addition,
further
building
sky
embedded
contribute
most.
demonstrates
their
strong
correlations
with
indicators
which
define
scheme.
findings
us
better
understand
facilitate
future
studies.
The
accurate
classification
of
water
bodies
is
crucial
for
effective
resource
management,
particularly
when
leveraging
remote
sensing
and
deep
learning
techniques.
However,
achieving
precise
in
Very
HighResolution
Satellite
(VHRS)
images
poses
a
significant
challenge,
necessitating
identification
categorization
methodologies.
Existing
methodologies
lack
userfriendly
interfaces
struggle
with
real-time
integration
into
Geographic
Information
System
(GIS)
maps.
This
proposed
system
aims
to
develop
graphical
user
interface
(GUI)
that
facilitates
the
VHRS
images.
GUI
employs
preprocessing
techniques
such
as
Bilateral
filtering
false
color
composites.
Water
body
segmentation
performed
using
U-Net
model,
followed
by
Random
Forest
Classifier.
Additionally,
change
detection
analysis,
enabling
generate
suitable
vector
data
from
raster
temporal
variations
bodies.
detected
changes
are
seamlessly
integrated
GIS
maps,
ensuring
timely
updating
spatial
data.
approach
evaluated
an
urban
dataset
Kolkata,
West
Bengal,
India.
