IEEE Access,
Год журнала:
2023,
Номер
12, С. 245 - 257
Опубликована: Дек. 25, 2023
The
goal
of
music
genre
classification
is
to
identify
the
given
feature
vectors
representing
certain
characteristics
clips.
In
addition,
improve
accuracy
classification,
considerable
research
has
been
conducted
on
extracting
spectral
features,
which
contain
critical
information
for
from
clips
and
feeding
these
features
into
training
models.
particular,
recent
studies
argue
that
can
be
enhanced
by
employing
multiple
simultaneously.
Consequently,
fusing
a
consideration
in
designing
Hence,
this
paper
provides
short
survey
compares
performance
most
CNN-based
models
with
newly
devised
model
employs
late
fusion
strategy
features.
Our
empirical
study
12
public
datasets,
including
Ballroom,
ISMIR04,
GTZAN,
showed
CNN
outperforms
other
compared
methods.
Additionally,
we
performed
an
in-depth
analysis
validate
effectiveness
classification.
Scientific Reports,
Год журнала:
2024,
Номер
14(1)
Опубликована: Янв. 5, 2024
Abstract
Accurate,
fast,
and
intelligent
workpiece
identification
is
of
great
significance
to
industrial
production.
To
cope
with
the
limited
hardware
resources
factory
equipment,
we
have
made
lightweight
improvements
based
on
You
Only
Look
Once
v5
(YOLOv5)
proposed
a
YOLO
named
YOLO_Bolt.
First,
ghost
bottleneck
deep
convolution
added
backbone
module
neck
YOLOv5
detection
algorithm
reduce
model
volume.
Second,
asymptotic
feature
pyramid
network
enhance
utilization
ability,
suppress
interference
information,
improve
accuracy.
Finally,
relationship
between
loss
function
decoupling
head
structure
was
focused
on,
number
layers
redesigned
according
different
tasks
further
accuracy
model.
We
conducted
experimental
verification
MSCOCO
2017
dataset
homemade
bolt
dataset.
The
results
show
that
compared
YOLOv5s,
parameters
only
6.8
M,
which
half
original
On
dataset,
mAP
increased
by
2.4%.
FPS
104
frames/s.
0.5
4.2%,
our
method
1.2%
higher
than
latest
YOLOv8s.
improved
can
provide
effective
auxiliary
technical
support
for
detection.
IEEE Access,
Год журнала:
2024,
Номер
12, С. 9418 - 9439
Опубликована: Янв. 1, 2024
This
paper
proposes
a
Design
Space
Exploration
for
Edge
machine
learning
through
the
utilization
of
novel
MathWorks
FPGA
Deep
Learning
Processor
IP,
featured
in
HDL
toolbox.
With
ever-increasing
demand
real-time
applications,
there
is
critical
need
efficient
and
low-latency
hardware
solutions
that
can
operate
at
edge
network,
close
proximity
to
data
source.
The
toolbox
provides
flexible
customizable
platform
deploying
deep
models
on
FPGAs,
enabling
effective
inference
acceleration
embedded
IoT
applications.
In
this
study,
our
primary
focus
lies
investigating
impact
parallel
processing
elements
performance
resource
FPGA-based
processor.
By
analyzing
trade-offs
between
accuracy,
speed,
energy
efficiency,
utilization,
we
aim
gain
valuable
insights
into
making
optimal
design
choices
implementations.
Our
evaluation
conducted
AMD-Xilinx
ZC706
development
board,
which
serves
as
target
device
experiments.
We
consider
all
compatible
Convolutional
Neural
Networks
available
within
comprehensively
assess
performances.
With
the
urgent
global
demand
for
sustainable
development,
intelligent
education
driven
by
multi-source
physical
information
has
attracted
widespread
attention
as
an
innovative
educational
model.
However,
in
context
of
dual
carbon,
achieving
and
efficient
development
faces
many
difficulties,
one
important
challenges
is
how
to
effectively
evaluate
students.
The
application
deep
neural
networks
evaluation
direction
digitization.
Currently,
there
need
conduct
research
on
value
empowering
with
networks.
We
first
studied
principles
characteristics
network
technology
evaluation;
second,
three
major
advantages
were
pointed
out:
objectivity
evaluating
diversified
data,
accuracy
perception
information,
mining
data
finally,
key
faced
clarified
from
perspectives
environment,
theoretical
knowledge,
interpretability.
This
provides
new
ideas
methods
lays
foundation
breaking
through
traditional
era
carbon
development.
Proceedings of the Institution of Mechanical Engineers Part O Journal of Risk and Reliability,
Год журнала:
2024,
Номер
unknown
Опубликована: Янв. 17, 2024
To
address
the
problem
of
reliance
on
a
priori
knowledge
and
difficult
hyperparameter
selection
in
feature
fusion.
The
effect
different
convolutional
kernel
sizes
filters
fusion
is
investigated
firstly,
based
which
an
artificial
Gorilla
Troops
Optimizer
(GTO)
enhanced
Convolutional
Long-Short
Term
Memory
Neural
Network
(CNN-LSTM)
method
for
bearing
lifetime
prediction
suggested.
GTO
algorithm
was
used
to
optimize
hyperparameters
such
as
size
CNN-LSTM,
pooling
layer
size,
batch
number
hidden
neurons,
rate
learning
with
goal
minimizing
mean
squared
error
remaining
useful
life
(RUL)
prediction.
From
optimized
CNN-LSTM
network
analyze
monitored
performance
degradation
data,
construct
health
indicators
(HI)
reflecting
degradation,
build
model.
Typical
cycle
data
has
been
validation
proposed
method.
results
indicate
that
have
better
trending
robustness,
leading
smaller
errors
outcomes.
IEEE Access,
Год журнала:
2024,
Номер
12, С. 24071 - 24078
Опубликована: Янв. 1, 2024
In
this
paper,
we
propose
an
innovative
image
enhancement
algorithm
called
"Dual-Enhancing-Dense-UNet
(DEDUNet)"
that
simultaneously
performs
brightness
and
reduces
noise.
This
model
is
based
on
Convolutional
Neural
Network
(CNN)
algorithms
incorporates
techniques
such
as
Decoupled
Fully
Connection
(DFC)
attention,
skip
connections,
shortcut,
Cross-Stage-Partial
(CSP)
dense
blocks
to
address
the
noise
removal
aspects
of
enhancement.
The
dual
approach
offers
a
new
solution
for
restoring
improving
high-quality
images,
presenting
opportunities
in
fields
computer
vision
processing.
Our
experimental
results
substantiate
superior
performance
proposed
algorithm,
showcasing
significant
improvements
key
indicators.
Specifically,
achieves
Peak
Signal-to-Noise
Ratio
(PSNR)
19.17,
Structural
Similarity
Index
(SSIM)
0.71,
Learned
Perceptual
Image
Patch
(LPIPS)
0.30,
Mean
Absolute
Error
(MAE)
0.09,
Multiply-Accumulate
(MAC)
0.696G.
These
highlight
algorithm's
remarkable
quality
capabilities,
demonstrating
considerable
advantage
over
existing
methods.
Experimental
demonstrate
efficiency
terms
improvement
compared
Geomatics Natural Hazards and Risk,
Год журнала:
2024,
Номер
15(1)
Опубликована: Март 25, 2024
Application
of
drone
technology
combined
with
LiDAR
and
Virtual
Reality/Augmented
Reality
in
surface
mining
operational
optimization
is
growing
on
a
high
trajectory.
The
sector
has
demonstrated
increased
interest
using
drones
for
everyday
tasks
such
as
bench
face
mapping,
dump
planning,
dragline
disposal
per
balance
diagram
mining.
One
the
key
requirements
application
3-dimensional
mapping
area
space
management
safe
efficient
manner.
can
assist
judicious
near
keeping
mind
available
dump/pit
slope
stability
requirements.
This
research
article
presents
review
how
Technology
based
cloud
computing
architecture
accelerate
mine
planning
activity
simple
side
cast
method.
Furthermore,
it
discusses
current
applications
AI-driven
3D
computer
vision
techniques
automating
data
analytics
point
clouds
extraction
terrain
parameters
plan
strategies
by
proper
positioning
large
mine.