Buletinul Institutului Politehnic din Iaşi. Secţia Electrotehnică. Energetică. Electronică,
Год журнала:
2023,
Номер
69(4), С. 9 - 29
Опубликована: Дек. 1, 2023
Abstract
Electroencephalogram
recordings
provide
insightful
information
concerning
the
diagnosis
and
prognosis
of
human
thinking
memory-related
processes,
aiding
researchers
physicians
during
Brain-Computer
Interface
systems
development.
In
electroencephalogram
memory
pattern
identification,
feature
extraction,
selection
are
determining
factors
for
an
impartial
data
description
accurate
classification.
The
signals
analyzed
in
this
study
collected
from
sixteen
electrodes
split
into
four
frequency
bands
specific
working
tasks
on
different
reasoning
scenarios.
Although
most
genetic
algorithm
based
optimization
procedures
tackle
minimization
a
classifier’s
error
rate
number
selected
features,
they
independent
how
configured,
either
single
or
multi-objective
manners,
major
problem
is
multidimensionality
quantity
redundant
noisy
recordings.
Since
objective
applied
separately
two
objectives:
misclassification
features
bias
final
results
to
direction,
all
these
limited
explorations
ground
use
better
sound
results.
Regarding
procedures,
compared
Pareto
ranking
schemes
meant
parents
survivors
evolutionary
optimization.
Usually,
methods
only
dominance
analysis
providing
partial
sorting
solutions
without
considering
strength
conflict
between
them.
paper
assign
ranks
by
combining
search
with
decisional
mechanism.
decision
implemented
through
adaptive
grouping
guide
towards
middle
first
fronts,
enabling
progressive
rejection
profitless
solutions.
population
several
groups
preserve
its
diversity,
supplementary
added
control
variety
valuable
information.
Finally,
layout
available
space
examined
clustering
individually
resulting
clusters
counteract
inherent
disadvantages
methods.
All
demonstrate
their
effectiveness
features.
Furthermore,
various
classifiers
distinctively
address
at
hand,
illustrating
mechanisms.
IEEE Journal of Biomedical and Health Informatics,
Год журнала:
2024,
Номер
28(9), С. 5227 - 5238
Опубликована: Июнь 17, 2024
Graph
neural
networks
(GNNs)
have
demonstrated
efficient
processing
of
graph-structured
data,
making
them
a
promising
method
for
electroencephalogram
(EEG)
emotion
recognition.
However,
due
to
dynamic
functional
connectivity
and
nonlinear
relationships
between
brain
regions,
representing
EEG
as
graph
data
remains
great
challenge.
To
solve
this
problem,
we
proposed
multi-domain
based
representation
learning
(MD
Electronics,
Год журнала:
2024,
Номер
13(13), С. 2530 - 2530
Опубликована: Июнь 27, 2024
Emotion
recognition
plays
a
crucial
role
in
affective
computing,
and
electroencephalography
(EEG)
signals
are
increasingly
applied
this
field
due
to
their
effectiveness
reflecting
brain
activity.
In
paper,
we
propose
novel
EEG
emotion
model
that
combines
the
ReliefF-based
Graph
Pooling
Convolutional
Network
BiGRU
Attention
Mechanisms
(RGPCN-BiGRUAM).
RGPCN-BiGRUAM
effectively
integrates
advantages
of
graph
convolutional
networks
recurrent
neural
networks.
By
incorporating
ReliefF
weights
an
attention
mechanism
into
pooling,
our
enhances
aggregation
high-quality
features
while
discarding
irrelevant
ones,
thereby
improving
efficiency
information
transmission.
The
implementation
multi-head
fusion
addresses
limitations
single-output
features,
achieving
optimal
selection
global
features.
Comparative
experiments
on
public
datasets
SEED
DEAP
demonstrate
proposed
significantly
improves
classification
performance
compared
classic
algorithms,
state-of-the-art
results.
Ablation
studies
further
validate
design
principles
model.
results
study
indicate
has
strong
potential
for
recognition,
offering
substantial
possibilities
future
applications.
Brain Sciences,
Год журнала:
2024,
Номер
14(5), С. 516 - 516
Опубликована: Май 20, 2024
Electroencephalography
(EEG)-based
emotion
recognition
is
increasingly
pivotal
in
the
realm
of
affective
brain–computer
interfaces.
In
this
paper,
we
propose
TSANN-TG
(temporal–spatial
attention
neural
network
with
a
task-specific
graph),
novel
architecture
tailored
for
enhancing
feature
extraction
and
effectively
integrating
temporal–spatial
features.
comprises
three
primary
components:
node-feature-encoding-and-adjacency-matrices-construction
block,
graph-aggregation
graph-feature-fusion-and-classification
block.
Leveraging
distinct
temporal
scales
features
from
EEG
signals,
incorporates
mechanisms
efficient
extraction.
By
constructing
adjacency
matrices,
graph
convolutional
an
mechanism
captures
dynamic
changes
dependency
information
between
channels.
Additionally,
emphasizes
integration
at
multiple
levels,
leading
to
improved
performance
emotion-recognition
tasks.
Our
proposed
applied
both
our
FTEHD
dataset
publicly
available
DEAP
dataset.
Comparative
experiments
ablation
studies
highlight
excellent
results
achieved.
Compared
baseline
algorithms,
demonstrates
significant
enhancements
accuracy
F1
score
on
two
benchmark
datasets
four
types
cognitive
These
underscore
potential
method
advance
EEG-based
recognition.
Buletinul Institutului Politehnic din Iaşi. Secţia Electrotehnică. Energetică. Electronică,
Год журнала:
2023,
Номер
69(4), С. 9 - 29
Опубликована: Дек. 1, 2023
Abstract
Electroencephalogram
recordings
provide
insightful
information
concerning
the
diagnosis
and
prognosis
of
human
thinking
memory-related
processes,
aiding
researchers
physicians
during
Brain-Computer
Interface
systems
development.
In
electroencephalogram
memory
pattern
identification,
feature
extraction,
selection
are
determining
factors
for
an
impartial
data
description
accurate
classification.
The
signals
analyzed
in
this
study
collected
from
sixteen
electrodes
split
into
four
frequency
bands
specific
working
tasks
on
different
reasoning
scenarios.
Although
most
genetic
algorithm
based
optimization
procedures
tackle
minimization
a
classifier’s
error
rate
number
selected
features,
they
independent
how
configured,
either
single
or
multi-objective
manners,
major
problem
is
multidimensionality
quantity
redundant
noisy
recordings.
Since
objective
applied
separately
two
objectives:
misclassification
features
bias
final
results
to
direction,
all
these
limited
explorations
ground
use
better
sound
results.
Regarding
procedures,
compared
Pareto
ranking
schemes
meant
parents
survivors
evolutionary
optimization.
Usually,
methods
only
dominance
analysis
providing
partial
sorting
solutions
without
considering
strength
conflict
between
them.
paper
assign
ranks
by
combining
search
with
decisional
mechanism.
decision
implemented
through
adaptive
grouping
guide
towards
middle
first
fronts,
enabling
progressive
rejection
profitless
solutions.
population
several
groups
preserve
its
diversity,
supplementary
added
control
variety
valuable
information.
Finally,
layout
available
space
examined
clustering
individually
resulting
clusters
counteract
inherent
disadvantages
methods.
All
demonstrate
their
effectiveness
features.
Furthermore,
various
classifiers
distinctively
address
at
hand,
illustrating
mechanisms.
