E3S Web of Conferences,
Год журнала:
2025,
Номер
631, С. 01001 - 01001
Опубликована: Янв. 1, 2025
Ensuring
the
structural
safety
of
civil
infrastructure
is
vital
for
public
welfare
and
cost-effective
maintenance.
Crack
detection,
as
a
key
indicator
health,
has
transitioned
from
traditional
image
processing
to
advanced
deep
learning
methods.
This
paper
presents
systematic
review
crack
detection
technologies
organized
under
novel
“method-scenario”
framework
that
categorizes
techniques
based
on
their
underlying
algorithms
specific
application
contexts
(e.g.,
pavements,
bridges,
tunnels,
specialized
materials).
By
comparing
conventional
approaches
with
modern
models
multi-modal
fusion
techniques,
we
highlight
strengths
limitations
each
method
in
various
real-world
scenarios.
Our
analysis
reveals
critical
challenges—including
data
scarcity,
sensitivity
noise,
gap
between
theoretical
practical
deployment—which
must
be
addressed
enhance
reliability
generalizability.
We
conclude
by
proposing
future
research
directions
focused
integrating
physics-based
constraints
lightweight
computational
establishing
unified
evaluation
protocols
bridge
laboratory
precision
engineering
implementation.
Scientific Reports,
Год журнала:
2024,
Номер
14(1)
Опубликована: Апрель 1, 2024
Abstract
Crack
formation
is
a
common
phenomenon
in
engineering
structures,
which
can
cause
serious
damage
to
the
safety
and
health
of
these
structures.
An
important
method
ensuring
engineered
structures
prompt
detection
cracks.
Image
threshold
segmentation
based
on
machine
vision
crucial
technology
for
crack
detection.
Threshold
separate
area
from
background,
providing
convenience
more
accurate
measurement
evaluation
condition
location.
The
cracks
complex
scenes
challenging
task,
this
goal
be
achieved
by
means
multilevel
thresholding.
arithmetic-geometric
divergence
combines
advantages
arithmetic
mean
geometric
probability
measures,
enabling
precise
capture
local
features
an
image
processing.
In
paper,
thresholding
minimum
proposed.
To
address
issue
time
complexity
thresholding,
enhanced
particle
swarm
optimization
algorithm
with
stochastic
perturbation
detection,
criterion
function
adaptively
determine
thresholds
according
distribution
characteristics
pixel
values
image.
proposed
increase
diversity
candidate
solutions
enhance
global
convergence
performance
algorithm.
compared
seven
state-of-the-art
methods
several
metrics,
including
RMSE,
PSNR,
SSIM,
FSIM,
computation
time.
experimental
results
show
that
outperforms
competing
terms
metrics.
Materials Genome Engineering Advances,
Год журнала:
2024,
Номер
2(1)
Опубликована: Март 1, 2024
Abstract
Generative
adversarial
networks
(GANs),
as
a
powerful
tool
for
inverse
materials
discovery,
are
being
increasingly
applied
in
various
fields
of
science.
This
review
provides
systematic
investigations
on
the
applications
GANs
from
group
different
aspects.
The
basic
principles
first
introduced;
then
detailed
GANs‐based
studies
regarding
distinct
scenarios
across
composition
design,
processing
optimization,
crystal
structure
search,
microstructure
characterization
and
defect
detection
is
presented.
At
end,
several
challenges
possible
solutions
discussed
outlined.
overview
highlights
efficacy
science,
may
stimulate
further
use
more
intriguing
achievements.
Buildings,
Год журнала:
2024,
Номер
14(6), С. 1808 - 1808
Опубликована: Июнь 14, 2024
This
research
manuscript
presents
a
comprehensive
investigation
into
the
prediction
and
detection
of
reflective
cracking
in
pavement
infrastructure
through
combination
machine
learning
approaches
advanced
image
techniques.
Leveraging
algorithms,
models
were
developed
optimized
for
accuracy
efficiency.
Additionally,
efficacy
methods,
particularly
utilizing
Mask
R-CNN,
was
explored
robust
precise
identification
on
surfaces.
The
study
not
only
aims
to
enhance
predictive
capabilities
management
systems
(PMSs)
learning-based
but
also
seeks
integrate
technologies
support
real-time
monitoring
assessment
conditions.
By
providing
accurate
timely
cracking,
these
methodologies
contribute
optimization
maintenance
strategies
overall
improvement
practices.
Results
indicate
that
achieve
an
average
over
85%,
with
some
achieving
accuracies
exceeding
90%.
Moreover,
utilization
mask
region-based
convolutional
neural
network
(Mask
R-CNN)
demonstrates
exceptional
precision,
95%
across
different
types
weather
results
demonstrate
promising
performance
predicting
while
R-CNN
showcases
from
images.
underscores
importance
leveraging
cutting-edge
address
challenges
management,
ultimately
supporting
sustainability
longevity
transportation
networks.
