Geomatics,
Journal Year:
2023,
Volume and Issue:
3(4), P. 501 - 521
Published: Nov. 26, 2023
Detecting
changes
in
soil
micro-relief
farmland
helps
to
understand
degradation
processes
like
sheet
erosion.
Using
the
high-resolution
technique
of
terrestrial
laser
scanning
(TLS),
we
generated
point
clouds
three
2
×
3
m
plots
on
a
weekly
basis
from
May
mid-June
2022
cultivated
Germany.
Three
well-known
applications
for
eliminating
vegetation
points
cloud
were
tested:
Cloth
Simulation
Filter
(CSF)
as
filtering
method,
variants
CANUPO
machine
learning
and
ArcGIS
PointCNN
deep
sub-category
using
neural
networks.
We
assessed
methods
with
hard
criteria
such
F1
score,
balanced
accuracy,
height
differences,
their
standard
deviations
reference
surface,
resulting
data
gaps
robustness,
soft
time-saving
capacity,
accessibility,
user
knowledge.
All
algorithms
showed
low
performance
at
initial
measurement
epoch,
increasing
later
epochs.
While
most
results
demonstrate
better
PointCNN,
this
algorithm
revealed
an
exceptionally
plot
1,
which
is
describable
by
generalization
gap.
Although
created
highest
amount
gaps,
recommend
that
include
colour
values
combination
CSF.
Remote Sensing,
Journal Year:
2024,
Volume and Issue:
16(18), P. 3373 - 3373
Published: Sept. 11, 2024
Salt
marshes
provide
diverse
habitats
for
a
wide
range
of
creatures
and
play
key
defensive
buffering
role
in
resisting
extreme
marine
hazards
coastal
communities.
Accurately
obtaining
the
terrains
salt
is
crucial
comprehensive
management
conservation
resources
ecology.
However,
dense
vegetation
coverage,
periodic
tide
inundation,
pervasive
ditch
distribution
create
challenges
measuring
or
estimating
marsh
terrains.
These
environmental
factors
make
most
existing
techniques
methods
ineffective
terms
data
acquisition
resolution,
accuracy,
efficiency.
Drone
multi-line
light
detection
ranging
(LiDAR)
has
offered
fire-new
perspective
3D
point
cloud
potentially
exhibited
great
superiority
accurately
deriving
The
prerequisite
terrain
characterization
from
drone
LiDAR
filtering,
which
means
that
ground
points
must
be
discriminated
non-ground
points.
Existing
filtering
typically
rely
on
either
geometric
intensity
features.
may
not
perform
well
with
dense,
diverse,
complex
vegetation.
This
study
proposes
new
method
clouds
based
artificial
neural
network
(ANN)
machine
learning
model.
First,
series
spatial–spectral
features
at
individual
(e.g.,
elevation,
distance,
intensity)
neighborhood
eigenvalues,
linearity,
sphericity)
scales
are
derived
original
data.
Then,
selected
to
remove
related
redundant
ones
optimizing
performance
ANN
Finally,
reserved
integrated
as
input
variables
model
characterize
their
nonlinear
relationships
categories
(ground
non-ground)
different
perspectives.
A
case
two
typical
mouth
Yangtze
River,
using
6-line
LiDAR,
demonstrates
effectiveness
generalization
proposed
method.
average
G-mean
AUC
achieved
were
0.9441
0.9450,
respectively,
outperforming
traditional
information-based
other
advanced
methods,
deep
(RandLA-Net).
Additionally,
integration
individual–neighborhood
results
better
outcomes
than
single-type
single-scale
offers
an
innovative
strategy
derivation
under
novel
solution
deeply
integrating
radiometric
Remote Sensing,
Journal Year:
2023,
Volume and Issue:
15(5), P. 1400 - 1400
Published: March 2, 2023
Ground
filtering
is
one
of
the
essential
steps
for
processing
airborne
light
detection
and
ranging
data
in
forestry
applications.
However,
performance
existing
methods
still
limited
forested
areas
due
to
complex
terrain
dense
vegetation.
To
overcome
this
limitation,
we
proposed
an
improved
surface-based
filter
based
on
multi-directional
narrow
window
cloth
simulation.
The
innovations
mainly
involve
two
aspects
as
follows:
(1)
sufficient
uniformly
distributed
ground
seeds
are
identified
by
merging
lowest
points
line
segments
from
point
clouds
within
a
window;
(2)
complete
accurate
extracted
using
cyclic
scheme
that
includes
incorrect
elimination
internal
force
adjustment
simulation,
reconstruction
with
moving
least-squares
plane
fitting,
extraction
progressively
refined
terrain.
method
was
tested
five
sites
various
characteristics
vegetation
distributions.
Experimental
results
showed
could
accurately
separate
non-ground
different
environments,
average
kappa
coefficient
88.51%
total
error
4.22%.
Moreover,
comparative
experiments
proved
performed
better
than
classical
involving
slope-based,
mathematical
morphology-based
methods.
International Journal of Environmental Sciences & Natural Resources,
Journal Year:
2023,
Volume and Issue:
32(3)
Published: July 3, 2023
This
paper
adapts
the
deep
learning
pipeline
algorithm
based
on
Multi-Layer
Perceptron
(MLP)
Neural
Network
to
automatically
classify
forest
Light
Detection
And
Ranging
(LiDAR)
point
cloud.To
achieve
this,
Machine
Learning
(ML)
parameters
such
as
input
layer
elements,
number
of
hidden
layers,
activation
functions,
and
alpha
value
are
optimized
best
possible
performance.Regarding
important
role
geometric
features
in
layer,
most
suggested
literature
analyzed
employ
more
effective
ones
layer.As
a
result,
seven
features,
addition
3D
coordinates
cloud,
chosen
represent
first
layer.The
proposed
classifies
LiDAR
cloud
into
two
classes:
vegetation
terrain.The
approach
was
tested
using
points
clouds,
one
flat
area
other
mountain
area.The
results
provide
an
accuracy
score
greater
than
98%.The
obtained
result
confirms
high
efficiency
classification
regarding
envisaged
approaches
literature.Finally,
next
step
is
generalize
this
complicated
scenes
urban
areas.
PLoS ONE,
Journal Year:
2024,
Volume and Issue:
19(8), P. e0305118 - e0305118
Published: Aug. 29, 2024
In
order
to
solve
the
problem
of
image
quality
and
morphological
characteristics
primary
underglaze
brown
decorative
pattern
extraction,
this
paper
proposes
a
method
extraction
based
on
coupling
single
scale
gamma
correction
gray
sharpening.
The
single-scale
is
combined
with
sharpening
method.
improves
contrast
brightness
by
nonlinear
transformation,
but
may
lead
loss
detail.
Gray
can
enhance
high
frequency
component
improve
clarity
image,
it
will
introduce
noise.
Combining
these
two
technologies
compensate
for
their
shortcomings.
experimental
results
show
that
efficiency
last
element
enhancing
retention
detail
reducing
influence
showed
F1Score,
Miou(%),
Recall,
Precision
Accuracy(%)
were
0.92745,
0.82253,
0.97942,
0.92458
respectively.
Remote Sensing,
Journal Year:
2024,
Volume and Issue:
16(11), P. 1841 - 1841
Published: May 22, 2024
The
combination
of
Remote
Sensing
and
Deep
Learning
(DL)
has
brought
about
a
revolution
in
converting
digital
surface
models
(DSMs)
to
terrain
(DTMs).
DTMs
are
used
various
fields,
including
environmental
management,
where
they
provide
crucial
topographical
data
accurately
model
water
flow
identify
flood-prone
areas.
However,
current
DL-based
methods
require
intensive
processing,
limiting
their
efficiency
real-time
use.
To
address
these
challenges,
we
have
developed
an
innovative
method
that
incorporates
physically
informed
autoencoder,
embedding
physical
constraints
refine
the
extraction
process.
Our
approach
utilizes
normalized
DSM
(nDSM),
which
is
updated
by
autoencoder
enable
DTM
generation
defining
as
difference
between
input
nDSM.
This
reduces
sensitivity
variations,
improving
model’s
generalizability.
Furthermore,
our
framework
innovates
using
subtractive
skip
connections
instead
traditional
concatenative
ones,
network’s
flexibility
adapt
variations
significantly
enhancing
performance
across
diverse
environments.
novel
demonstrates
superior
adaptability
compared
other
versions
autoencoders
ten
datasets,
urban
areas,
mountainous
regions,
predominantly
vegetation-covered
landscapes,
Precision Agriculture,
Journal Year:
2024,
Volume and Issue:
26(1)
Published: Dec. 27, 2024
Abstract
Purpose
The
use
of
UAVs
(Unmanned
Aerial
Vehicles)
equipped
with
sensors
such
as
laser
scanners
offers
an
alternative
to
conventional,
labor-intensive
manual
measurements
in
agriculture,
they
enable
precise
and
non-destructive
field
surveys.
Methods
This
paper
evaluates
the
UAV-based
scanning
(RIEGL
miniVUX-SYS)
for
estimating
crop
height
plant
area
index
(PAI)
winter
wheat.
(Methods)
It
further
introduces
a
novel
ground
classification
method,
enhancing
early
growth
stage
through
sensor
attributes
like
intensity
pulse
shape
deviation.
Results
estimation
shows
high
$$R^2$$
R2
score
$$99.69~\%$$
99.69%
but
systematically
lower
estimate
mean
absolute
error
7.4
cm
.
potential
PAI
derivation
is
analyzed
three
different
strategies
provides
overview
limitations
approach.
Additional
weighting
based
on
scan
angle
adaptation
extinction
coefficient
present
results
$$97.66~\%$$
97.66
0.25.
Conclusion
investigation
discusses
impact
calculated
gap
fraction,
which
describes
ratio
beams
penetrating
canopy
comparison
total
number
measurements.
Forests,
Journal Year:
2023,
Volume and Issue:
14(10), P. 1943 - 1943
Published: Sept. 24, 2023
Populus
euphratica
and
Tamarix
chinensis
hold
significant
importance
in
wind
prevention,
sand
fixation,
biodiversity
conservation.
The
precise
extraction
of
these
species
can
offer
technical
assistance
for
vegetation
studies.
This
paper
focuses
on
the
located
within
Daliyabuyi,
utilizing
PointCNN
as
primary
research
method.
After
decorrelating
stretching
images,
deep
learning
techniques
were
applied,
successfully
distinguishing
between
various
types,
thereby
enhancing
precision
information
extraction.
On
validation
dataset,
model
showcased
a
high
degree
accuracy,
with
respective
regular
accuracy
rates
being
92.106%
91.936%.
In
comparison
to
two-dimensional
models,
classification
is
superior.
Additionally,
this
study
extracted
individual
tree
euphratica,
such
height,
crown
width,
area,
volume.
A
comparative
analysis
data
attested
results.
Furthermore,
concluded
that
batch
size
block
training
could
influence
outcomes.
summary,
compared
2D
point
cloud
approach
exhibits
higher
reliability
classifying
extracting
poplars
tamarisks.
These
findings
valuable
references
insights
remote
sensing
image
processing
domains.
International Journal of Advanced Computer Science and Applications,
Journal Year:
2023,
Volume and Issue:
14(9)
Published: Jan. 1, 2023
The
extraction
of
buildings
from
multispectral
Light
Detection
and
Ranging
(LiDAR)
data
holds
significance
in
various
domains
such
as
urban
planning,
disaster
response,
environmental
monitoring.
State-of-the-art
deep
learning
models,
including
Point
Convolutional
Neural
Network
(Point
CNN)
Mask
Region-based
(Mask
R-CNN),
have
effectively
addressed
this
particular
task.
Data
application
characteristics
affect
model
performance.
This
research
compares
LiDAR
building
CNN
R-CNN.
Models
are
tested
for
accuracy,
efficiency,
capacity
to
handle
irregularly
spaced
point
clouds
using
data.
extracts
more
accurately
efficiently
than
CNN-based
cloud
feature
avoids
preprocessing
like
voxelization,
improving
accuracy
processing
speed
over
CNNs
can
with
variable
spacing.
R-CNN
outperforms
some
cases.
uses
image-like
instead
clouds,
making
it
better
at
detecting
categorizing
objects
different
angles.
study
emphasizes
selecting
the
right
or
accurate
depends
on
application.
For
data,
two
approaches
were
compared
utilizing
precision,
recall,
F1
score.
point-CNN
outperformed
had
93.40%
92.34%
92.72%
has
moderate
F1.
Geomatics,
Journal Year:
2023,
Volume and Issue:
3(4), P. 501 - 521
Published: Nov. 26, 2023
Detecting
changes
in
soil
micro-relief
farmland
helps
to
understand
degradation
processes
like
sheet
erosion.
Using
the
high-resolution
technique
of
terrestrial
laser
scanning
(TLS),
we
generated
point
clouds
three
2
×
3
m
plots
on
a
weekly
basis
from
May
mid-June
2022
cultivated
Germany.
Three
well-known
applications
for
eliminating
vegetation
points
cloud
were
tested:
Cloth
Simulation
Filter
(CSF)
as
filtering
method,
variants
CANUPO
machine
learning
and
ArcGIS
PointCNN
deep
sub-category
using
neural
networks.
We
assessed
methods
with
hard
criteria
such
F1
score,
balanced
accuracy,
height
differences,
their
standard
deviations
reference
surface,
resulting
data
gaps
robustness,
soft
time-saving
capacity,
accessibility,
user
knowledge.
All
algorithms
showed
low
performance
at
initial
measurement
epoch,
increasing
later
epochs.
While
most
results
demonstrate
better
PointCNN,
this
algorithm
revealed
an
exceptionally
plot
1,
which
is
describable
by
generalization
gap.
Although
created
highest
amount
gaps,
recommend
that
include
colour
values
combination
CSF.