Sensors,
Journal Year:
2024,
Volume and Issue:
24(20), P. 6753 - 6753
Published: Oct. 21, 2024
The
accurate
detection
and
quantification
of
defects
is
vital
for
the
effectiveness
eddy
current
nondestructive
testing
(ECNDT)
carbon
fiber-reinforced
plastic
(CFRP)
materials.
This
study
investigates
identification
measurement
three
common
CFRP
defects-cracks,
delamination,
low-velocity
impact
damage-by
employing
You
Only
Look
Once
(YOLO)
model
an
improved
Eddy
Current
YOLO
(EDC-YOLO)
model.
YOLO's
limitations
in
detecting
multi-scale
features
are
addressed
through
integration
Transformer-based
self-attention
mechanisms
deformable
convolutional
sub-modules,
with
additional
global
feature
extraction
via
CBAM.
By
leveraging
Wise-IoU
loss
function,
performance
further
enhanced,
leading
to
a
4.4%
increase
mAP50
defect
detection.
EDC-YOLO
proves
be
effective
industrial
inspections,
providing
detailed
insights,
such
as
correlation
between
damage
size
energy
levels.
The international archives of the photogrammetry, remote sensing and spatial information sciences/International archives of the photogrammetry, remote sensing and spatial information sciences,
Journal Year:
2024,
Volume and Issue:
XLVIII-1-2024, P. 649 - 654
Published: May 10, 2024
Abstract.
With
the
progress
of
remote
sensing
sensors,
quality
optical
image
is
significantly
improved,
and
target
detection
on
it
can
extract
rich
feature
information.
However,
due
to
characteristics
with
various
sizes
a
large
proportion
number
small
targets,
increasing
difficulty
in
for
it.
In
response
this
challenge,
paper
proposes
an
improved
YOLOv8
algorithm
multi-scale
images.
First,
we
propose
PSPPF
module,
which
improves
model's
ability
adapt
different
data
distributions;
Second,
DSConv
introduced
into
Backbone
structure
reduce
complexity
network
while
maintaining
performance
model
detection;
Finally,
replace
original
loss
function
CIoU
MPDIoU
improve
localization
accuracy
prediction
box.
Applying
public
dataset
NWPU
VHR-10,
mAP
value
our
95.1%,
3.0%
higher
than
that
YOLOv8,
indicating
proposed
able
effectively
detect
targets
Making
sure
traffic
is
safe
and
well-managed
has
become
a
top
priority
in
the
world
of
contemporary
transportation.
Using
robust
YOLO
(You
Only
Look
Once)
v8
model
conjunction
with
Optical
Character
Recognition
(OCR)
technology,
this
research
explores
creation
deployment
state-of-the-art
detection
system.
The
main
goal
to
make
roads
safer
by
detecting
real-time
automatically
identifying
violations.
This
built
around
framework,
which
known
for
its
fast
accurate
object
detection.
With
OCR
technology
integrated,
system's
capabilities
are
greatly
enhanced.
Extracting
textual
information
from
licence
plates
allows
system
detect
violations
like
speeding.
In
order
it
work
wide
variety
international
contexts,
component
strong
can
handle
different
fonts,
sizes,
languages.
To
sum
up,
intelligent
management
systems
have
come
long
way
using
Framework
optical
character
recognition
violation
enhancement.
IEEE Access,
Journal Year:
2024,
Volume and Issue:
12, P. 98225 - 98238
Published: Jan. 1, 2024
In
recent
years,
speech
denoising
has
greatly
benefited
from
the
rapid
development
of
neural
networks.
However,
these
models
require
substantial
noisy-clean
pairs
for
supervised
training,
which
limits
their
widespread
use.
Although
there
have
been
attempts
to
train
networks
with
only
noisy
data,
existing
self-supervised
methods
often
suffer
a
lack
continuity,
low
noise
reduction
performance,
or
heavy
dependence
on
modeling.
this
work,
we
introduce
an
efficient
Dynamic
Multi-Focusing
Network
(DMFNet),
noise-only
trained
network
that
utilizes
multi-scale
connected
encoder-decoder
architecture
as
its
backbone.
Specifically,
designed
Spectral
Focusing
Unit
(SDFU)
enables
dynamically
adapt
shape
convolutional
kernels
while
learning
features,
thus
effectively
focusing
spectral
structure
human
voice.
Additionally,
Complex
Attention
Module
(CAM),
cross-space
specialized
feature
interaction
and
extraction.
Finally,
further
enhance
recovery
fine
details,
propose
Multi-Scale
Feature
Fusion
(CMFFU)
Scope
(CSFU)
adaptively
fuse
features
different
stages
in
encoding
process.
Extensive
evaluations
across
multiple
datasets
demonstrate
proposed
DMFNet
significantly
outperforms
other
state-of-the-art
methods.
Sensors,
Journal Year:
2024,
Volume and Issue:
24(20), P. 6753 - 6753
Published: Oct. 21, 2024
The
accurate
detection
and
quantification
of
defects
is
vital
for
the
effectiveness
eddy
current
nondestructive
testing
(ECNDT)
carbon
fiber-reinforced
plastic
(CFRP)
materials.
This
study
investigates
identification
measurement
three
common
CFRP
defects-cracks,
delamination,
low-velocity
impact
damage-by
employing
You
Only
Look
Once
(YOLO)
model
an
improved
Eddy
Current
YOLO
(EDC-YOLO)
model.
YOLO's
limitations
in
detecting
multi-scale
features
are
addressed
through
integration
Transformer-based
self-attention
mechanisms
deformable
convolutional
sub-modules,
with
additional
global
feature
extraction
via
CBAM.
By
leveraging
Wise-IoU
loss
function,
performance
further
enhanced,
leading
to
a
4.4%
increase
mAP50
defect
detection.
EDC-YOLO
proves
be
effective
industrial
inspections,
providing
detailed
insights,
such
as
correlation
between
damage
size
energy
levels.
