Neural Computing and Applications,
Journal Year:
2024,
Volume and Issue:
36(20), P. 12483 - 12503
Published: April 22, 2024
Abstract
Visual
inspection
of
defective
tires
post-production
is
vital
for
human
safety,
as
faulty
can
lead
to
explosions,
accidents,
and
loss
life.
With
the
advancement
technology,
transfer
learning
(TL)
plays
an
influential
role
in
many
computer
vision
applications,
including
tire
defect
detection
problem.
However,
automatic
difficult
two
reasons.
The
first
presence
complex
anisotropic
multi-textured
rubber
layers.
Second,
there
no
standard
X-ray
image
dataset
use
detection.
In
this
study,
a
TL-based
model
proposed
using
new
from
global
company.
First,
we
collected
labeled
consisting
3366
images
20,000
qualified
tires.
Although
covers
15
types
defects
arising
different
design
patterns,
our
primary
focus
on
binary
classification
detect
or
absence
defects.
This
challenging
was
split
into
70,
15,
15%
training,
validation,
testing,
respectively.
Then,
nine
common
pre-trained
models
were
fine-tuned,
trained,
tested
dataset.
These
are
Xception,
InceptionV3,
VGG16,
VGG19,
ResNet50,
ResNet152V2,
DenseNet121,
InceptionResNetV2,
MobileNetV2.
results
show
that
fine-tuned
DenseNet21
InceptionNet
achieve
compatible
with
literature.
Moreover,
Xception
outperformed
compared
TL
literature
methods
terms
recall,
precision,
accuracy,
F1
score.
it
achieved
testing
73.7,
88,
80.2,
94.75%
score,
respectively,
validation
73.3,
90.24,
80.9,
95%
IEEE Access,
Journal Year:
2024,
Volume and Issue:
12, P. 50949 - 50965
Published: Jan. 1, 2024
The
objective
of
this
paper
is
to
develop
a
novel
emotion
recognition
system
from
electroencephalogram
(EEG)
signals
using
effective
connectivity
and
deep
learning
methods.
Emotion
an
important
task
for
various
applications
such
as
human-computer
interaction
and,
mental
health
diagnosis.
aims
improve
the
accuracy
robustness
by
combining
different
(EC)
methods
pre-trained
convolutional
neural
networks
(CNNs),
well
long
short-term
memory
(LSTM).
EC
measure
information
flow
in
brain
during
emotional
states
EEG
signals.
We
used
three
methods:
transfer
entropy
(TE),
partial
directed
coherence
(PDC),
direct
function
(dDTF).
estimated
fused
image
these
each
five-second
window
32-channel
Then,
we
applied
six
CNNs
classify
images
into
four
classes
based
on
two-dimensional
valence-arousal
model.
leave-one-subject-out
cross-validation
strategy
evaluate
classification
results.
also
ensemble
model
select
best
results
majority
voting
approach.
Moreover,
combined
with
LSTM
performance.
achieved
average
F-score
98.76%,
98.86%,
98.66
98.88%
classifying
emotions
DEAP
MAHNOB-HCI
datasets,
respectively.
Our
show
that
can
increase
combination
achieve
high
automated
recognition.
outperformed
other
state-of-the-art
systems
same
datasets
four-class
classification.
Symmetry,
Journal Year:
2022,
Volume and Issue:
14(10), P. 1976 - 1976
Published: Sept. 21, 2022
In
the
present
work,
we
propose
a
novel
method
utilizing
only
decoder
for
generation
of
pseudo-examples,
which
has
shown
great
success
in
image
classification
tasks.
The
proposed
is
particularly
constructive
when
data
are
limited
quantity
used
semi-supervised
learning
(SSL)
or
few-shot
(FSL).
While
most
previous
works
have
an
autoencoder
to
improve
performance
SSL,
using
single
may
generate
confusing
pseudo-examples
that
could
degrade
classifier’s
performance.
On
other
hand,
various
models
utilize
encoder–decoder
architecture
sample
can
significantly
increase
computational
overhead.
To
address
issues
mentioned
above,
efficient
means
generating
by
generator
(decoder)
network
separately
each
class
be
effective
both
SSL
and
FSL.
our
approach,
trained
random
noise,
multiple
samples
generated
decoder.
Our
generator-based
approach
outperforms
state-of-the-art
FSL
approaches.
addition,
released
Urdu
digits
dataset
consisting
10,000
images,
including
8000
training
2000
test
images
collected
through
three
different
methods
purposes
diversity.
Furthermore,
explored
effectiveness
on
FSL,
demonstrated
improvement
3.04%
1.50%
terms
average
accuracy,
respectively,
illustrating
superiority
compared
current
models.
Engineering Applications of Artificial Intelligence,
Journal Year:
2023,
Volume and Issue:
123, P. 106205 - 106205
Published: March 31, 2023
In
the
Machine
Learning
(ML)
literature,
a
well-known
problem
is
Dataset
Shift
where,
differently
from
ML
standard
hypothesis,
data
in
training
and
test
sets
can
follow
different
probability
distributions,
leading
systems
toward
poor
generalisation
performances.
This
intensely
felt
Brain-Computer
Interface
(BCI)
context,
where
bio-signals
as
Electroencephalographic
(EEG)
are
often
used.
fact,
EEG
signals
highly
non-stationary
both
over
time
between
subjects.
To
overcome
this
problem,
several
proposed
solutions
based
on
recent
transfer
learning
approaches
such
Domain
Adaption
(DA).
cases,
however,
actual
causes
of
improvements
remain
ambiguous.
paper
focuses
impact
normalisation,
or
standardisation
strategies
applied
together
with
DA
methods.
particular,
using
\textit{SEED},
\textit{DEAP},
\textit{BCI
Competition
IV
2a}
datasets,
we
experimentally
evaluated
normalization
without
methods,
comparing
obtained
It
results
that
choice
normalisation
strategy
plays
key
role
classifier
performances
scenarios,
interestingly,
use
only
an
appropriate
schema
outperforms
technique.
