2022 Innovations in Intelligent Systems and Applications Conference (ASYU),
Год журнала:
2023,
Номер
unknown, С. 1 - 6
Опубликована: Окт. 11, 2023
The
accurate
classification
of
news
and
commercial
jingles
is
essential
for
the
automated
generation
broadcast
flow.
Currently,
in
press
companies,
editors
manually
label
start
end
times
advertisements,
which
incurs
both
cost
time
loss.
Although
method
extracting
fingerprints
has
been
employed
to
detect
on
a
channel
basis
automatically
classify
music,
this
approach
falls
short
when
it
comes
classifying
new
produced
by
channels.
In
study,
we
created
dataset
segments
from
TV
channels
Turkey.
We
analyzed
most
effective
second
interval
or
commercials,
resulting
an
impressive
accuracy
score
98.18%.
By
leveraging
conducting
extensive
analysis,
have
made
significant
progress
accurately
jingles.
This
research
can
potentially
save
companies
costs
automating
process.
IEEE Sensors Journal,
Год журнала:
2023,
Номер
23(21), С. 26269 - 26278
Опубликована: Сен. 22, 2023
Industrial
equipment
failure
diagnosis
is
a
crucial
issue
that
impacts
the
national
industrial
manufacturing
level,
economic
cycle
development,
and
sustainable
technological
advancement.
A
multimodal
knowledge
graph
(MMKG)-based
intelligent
diagnostic
model
for
fault
proposed
to
address
issues
of
insufficient
inadequate
data
samples
encountered
when
using
single-mode
in
existing
equipment.
This
does
not
require
extensive
learning
complex
scenarios.
The
utilizes
an
improved
faster
region
with
CNN
(Faster
RCNN)
features
object
detection
module
extract
visual
information
feature
vectors
semiordered
main
nonmain
objects.
These
are
then
mapped
entity,
attribute,
relationship
cosine
similarity
correspondence
mapping.
semantic
matching
inference
performed
based
on
this
mapping,
resulting
set
triplets.
Finally,
bidirectional
autoregressive
transformers
(BARTs)
text
generation
processes
triplet
generate
texts.
Experimental
results
demonstrate
Faster
RCNN
achieves
1.2%
increase
confidence
trained
small
training
datasets.
accuracy
generated
description
texts
reaches
approximately
98%
compared
standard
presented
article
addresses
challenge
diagnosing
faults
equipment,
particularly
scenarios
limited
data,
such
as
substations.
It
enhances
target
effectively
even
scarce.
Additionally,
it
MMKG
enable
interpretable
decision-making.
Abstract
Class
imbalance
and
class
overlap
create
difficulties
in
the
training
phase
of
standard
machine
learning
algorithm.
Its
performance
is
not
well
minority
classes,
especially
when
there
a
high
significant
overlap.
Recently
it
has
been
observed
by
researchers
that,
joint
effects
are
more
harmful
as
compared
to
their
direct
impact.
To
handle
these
problems,
many
methods
have
proposed
past
years
that
can
be
broadly
categorized
data‐level,
algorithm‐level,
ensemble
learning,
hybrid
methods.
Existing
data‐level
often
suffer
from
problems
like
information
loss
overfitting.
overcome
we
introduce
novel
entropy‐based
sampling
(EHS)
method
highly
imbalanced
datasets.
The
EHS
eliminates
less
informative
majority
instances
region
during
undersampling
regenerates
synthetic
oversampling
near
borderline.
achieved
improvement
F1‐score,
G‐mean,
AUC
metrics
value
DT,
NB,
SVM
classifiers
well‐established
state‐of‐the‐art
Classifiers
performances
tested
on
28
datasets
with
extreme
ranges
Physica Scripta,
Год журнала:
2024,
Номер
99(9), С. 096001 - 096001
Опубликована: Июль 24, 2024
Abstract
The
chemical
industry
generates
a
broad
spectrum
of
hazardous
gases,
presenting
significant
challenges
for
conventional
detection
methods
due
to
their
diverse
properties
and
low
concentration
levels.
E-nose
systems,
employing
sensor
arrays,
offer
potential
the
determination
gas
mixtures.
This
study
presents
novel
algorithm,
CNN-ECA,
which
integrated
CNNs
attention
mechanisms
improve
recognition
accuracy
systems.
By
integrating
mechanism
module
into
CNN’s
convolutional
operations,
algorithm
emphasizes
critical
feature
information.
Three
gases
(ammonia,
methanol,
acetone)
mixtures
were
chosen
as
target
gases.
combined
with
various
networks
(SENet,
ECA,
CBAM)
construct
models,
then
employed
train
evaluate
data
collected
from
array.
results
compared
traditional
network
models
(KNN,
SVM,
CNN).
Experimental
findings
indicated
that
prediction
performance
CNN
surpassed
models.
Particularly,
CNN-ECA
model
demonstrated
highest
in
both
qualitative
quantitative
analyses.
promising
solution
mixed
by
synergizing
networks,
thereby
enhancing
reliability
measurements.
Moreover,
capitalizing
on
lightweight
architecture
model,
transfer
learning
techniques
adapt
it
deployment
Raspberry
Pi
hardware
platform.
facilitates
development
real-time
system
detection.
