Buildings,
Год журнала:
2024,
Номер
14(12), С. 3832 - 3832
Опубликована: Ноя. 29, 2024
The
total
mileage
of
the
road
network
in
China
has
been
growing
rapidly
during
last
twenty
years.
With
development
deep
learning,
automatic
distr
ess
detection
method
is
more
accurate
and
effective
than
manual
detection.
However,
are
classified
into
five
grades
according
to
Chinese
standard
each
grade
its
own
characteristics.
A
single
model
cannot
effectively
identify
multi-grade
roads
with
different
materials
levels
distress.
This
study
proposes
a
YOLOv8-based
distress
strategy
adapted
for
multiple
grades.
improved
URetinex-Net
used
enhance
spatial
features
scenario
diversity
datasets.
Compared
base
YOLOv8
model,
enhancements
have
led
12%
increase
accuracy
cement
roads,
22.3%
improvement
speed,
5.5%
ordinary
asphalt
7.5%
recognition
highways,
9.3%
significant
effects.
refines
classification
based
on
their
matches
them
corresponding
artificial
intelligence
training
strategies,
providing
guidance
inspection
maintenance.
Buildings,
Год журнала:
2024,
Номер
14(5), С. 1442 - 1442
Опубликована: Май 16, 2024
Accurate
pavement
surface
crack
detection
is
crucial
for
analyzing
survey
data
and
the
development
of
maintenance
strategies.
On
basis
Swin-Unet,
this
study
develops
improved
Swin-Unet
(iSwin-Unet)
model
with
developed
skip
attention
module
residual
Swin
Transformer
block.
Based
on
channel
mechanism,
region
can
be
better
captured
while
feature
channels
assigned
more
weights.
Taking
advantage
block,
encoder
architecture
globally
feature.
Meanwhile,
information
efficiently
exchanged.
To
verify
performance
proposed
model,
we
compare
training
visualization
results
other
three
models,
which
are
Transformer,
Unet,
respectively.
Three
public
benchmarks
(CFD,
Crack500,
CrackSC)
have
been
adopted
purpose
training,
validation,
testing.
test
results,
it
found
that
iSwin-Unet
achieves
a
significant
increase
in
mF1
score,
mPrecision,
mRecall
compared
to
existing
thereby
establishing
its
efficacy
underlining
advancements
over
current
methodologies.
Computer-Aided Civil and Infrastructure Engineering,
Год журнала:
2025,
Номер
unknown
Опубликована: Март 3, 2025
Abstract
Pavement
crack
measurement
(PCM)
is
essential
for
automated,
precise
road
condition
assessment.
However,
balancing
speed
and
accuracy
on
edge
artificial
intelligence
(AI)
mobile
devices
remains
challenging.
This
paper
proposes
a
real‐time
PCM
framework
deployment,
incorporating
lightweight
distillation
network
surface
feature
algorithm.
Specifically,
the
proposed
instance‐aware
hybrid
module
combines
feature‐based
relation‐based
knowledge
distillation,
leveraging
instance‐related
information
efficient
transfer
from
teacher
to
student
networks,
which
results
in
more
accurate
segmentation
model.
Additionally,
algorithm,
based
distance
mapping
relationships
coordinate
extraction,
addresses
issues
with
branching
loss,
enhancing
efficiency.
Real‐time
was
performed
actual
roads
utilizing
robot
equipped
an
computing
unit.
The
precision
reached
84.37%,
frame
per
second
of
77.72.
Compared
ground
truth,
relative
error
average
width
ranged
6.42%
40.65%,
while
length
varied
between
1.48%
3.76%.
These
findings
highlight
feasibility
assessment
save
maintenance
costs.
