Gas Sensor Drift Compensation Using Semi-Supervised Ensemble Classifiers with Multi-Level Features and Center Loss
ACS Sensors,
Год журнала:
2025,
Номер
unknown
Опубликована: Апрель 8, 2025
The
drift
compensation
of
gas
sensors
is
a
significant
and
challenging
issue
in
the
field
electronic
noses
(E-nose).
Compensating
sensor
has
great
benefit
improving
performance
E-nose
systems.
However,
conventional
methods
often
perform
poorly
due
to
complex
data
relationships
before
after
drifting,
or
require
label
information
for
both
nondrift
(source
data)
(target
enhance
performance,
which
hard
achieve
even
unrealistic.
In
this
study,
we
propose
semisupervised
domain
adaptive
convolutional
neural
network
(CNN)
based
on
ensemble
classifiers
multilevel
features,
pretraining,
center
loss
tackle
problem.
main
idea
make
full
use
features
extracted
from
apply
Hilbert
space's
maximum
mean
discrepancy
(MMD)
evaluate
similarity
at
different
levels.
Then
corresponding
MMD
used
as
weight
weighted
fusion
predictions
classifier
module,
so
obtain
more
reliable
result.
Furthermore,
optimize
training,
pretraining
help
feature
extractors
learn
robust
common
two
domains.
Center
also
applied
focused
learning
same
class.
results
sets
demonstrate
effectiveness
our
method.
average
classification
accuracies
under
settings
reach
76.06%
(long-drift)
82.07%
(short-drift),
respectively,
R2
score
reaches
0.804
regression
task,
improvements
compared
with
several
methods.
Our
work
provides
an
effective
method
algorithm
level
solve
problem
sensors.
Язык: Английский
CO Concentration prediction in E-nose based on MHA-MSCINet
Journal of the Taiwan Institute of Chemical Engineers,
Год журнала:
2025,
Номер
169, С. 105981 - 105981
Опубликована: Янв. 22, 2025
Язык: Английский
Research on Binary Mixed VOCs Gas Identification Method Based on Multi-Task Learning
Sensors,
Год журнала:
2025,
Номер
25(8), С. 2355 - 2355
Опубликована: Апрель 8, 2025
Traditional
volatile
organic
compounds
(VOCs)
detection
models
separate
component
identification
and
concentration
prediction,
leading
to
low
feature
utilization
limited
learning
in
small-sample
scenarios.
Here,
we
realize
a
Residual
Fusion
Network
based
on
multi-task
(MTL-RCANet)
implement
prediction
of
VOCs.
The
model
integrates
channel
attention
mechanisms
cross-fusion
modules
enhance
extraction
capabilities
task
synergy.
To
further
balance
the
tasks,
dynamic
weighted
loss
function
is
incorporated
adjust
weights
dynamically
according
training
progress
each
task,
thereby
enhancing
overall
performance
model.
proposed
network
achieves
an
accuracy
94.86%
R2
score
0.95.
Comparative
experiments
reveal
that
using
only
35%
total
data
length
as
input
yields
excellent
performance.
Moreover,
effectively
information
across
significantly
improving
efficiency
compared
single-task
learning.
Язык: Английский
Smart VOCs Recognition System Based on Single Gas Sensor and Multi-task Deep Learning Model
Sensors and Actuators B Chemical,
Год журнала:
2025,
Номер
unknown, С. 137853 - 137853
Опубликована: Апрель 1, 2025
Язык: Английский
Robust Odor Detection in Electronic Nose Using Transfer-Learning Powered Scentformer Model
ACS Sensors,
Год журнала:
2025,
Номер
unknown
Опубликована: Май 15, 2025
Mimicking
the
olfactory
system
of
humans,
use
electronic
noses
(E-noses)
for
detection
odors
in
nature
has
become
a
hot
research
topic.
This
study
presents
novel
E-nose
based
on
deep
learning
architecture
called
Scentformer,
which
addresses
limitations
current
like
narrow
range
and
limited
generalizability
across
different
scenarios.
Armed
with
self-adaptive
data
down-sampling
method,
is
capable
detecting
55
natural
classification
accuracy
99.94%,
model
embedded
analyzed
using
Shapley
Additive
exPlanations
analysis,
providing
quantitative
interpretation
performance.
Furthermore,
leveraging
Scentformer's
transfer
ability,
efficiently
adapts
to
new
gases.
Rather
than
retraining
all
layers
odor
set,
only
fully
connected
need
be
trained
pretrained
model.
Using
1‰
retrained
model,
model-based
can
also
achieve
accuracies
99.14%
various
gas
concentrations.
provides
robust
approach
diverse
direct
signals
real-world
applications.
Язык: Английский
Low-Power Chemiresistive Gas Sensors for Transformer Fault Diagnosis
Molecules,
Год журнала:
2024,
Номер
29(19), С. 4625 - 4625
Опубликована: Сен. 29, 2024
Dissolved
gas
analysis
(DGA)
is
considered
to
be
the
most
convenient
and
effective
approach
for
transformer
fault
diagnosis.
Due
their
excellent
performance
development
potential,
chemiresistive
sensors
are
anticipated
supersede
traditional
chromatography
in
dissolved
of
transformers.
However,
high
operating
temperature
power
consumption
restrict
deployment
battery-powered
devices.
This
review
examines
underlying
principles
sensors.
It
comprehensively
summarizes
recent
advances
low-power
detection
characteristic
gases
(H2,
C2H2,
CH4,
C2H6,
C2H4,
CO,
CO2).
Emphasis
placed
on
synthesis
methods
sensitive
materials
properties.
The
investigations
have
yielded
substantial
experimental
data,
indicating
that
adjusting
particle
size
morphology
structure
combining
them
with
noble
metal
doping
principal
enhancing
sensitivity
reducing
Additionally,
strategies
overcome
significant
challenge
cross-sensitivity
encountered
applications
provided.
Finally,
future
direction
DGA
envisioned,
offering
guidance
developing
applying
novel
gas-sensitive
Язык: Английский