Electronics,
Год журнала:
2025,
Номер
14(4), С. 731 - 731
Опубликована: Фев. 13, 2025
Infrared-visible
image
fusion
(IVIF)
is
an
important
part
of
multimodal
(MMF).
Our
goal
to
combine
useful
information
from
infrared
and
visible
sources
produce
strong,
detailed,
fused
images
that
help
people
understand
scenes
better.
However,
most
existing
methods
based
on
convolutional
neural
networks
extract
cross-modal
local
features
without
fully
utilizing
long-range
contextual
information.
This
limitation
reduces
performance,
especially
in
complex
scenarios.
To
address
this
issue,
we
propose
TCTFusion,
a
three-branch
transformer
for
visible–infrared
fusion.
The
model
includes
shallow
feature
module
(SFM),
frequency
decomposition
(FDM),
aggregation
(IAM).
three
branches
specifically
receive
input
infrared,
visible,
concatenated
images.
SFM
extracts
using
residual
connections
with
shared
weights.
FDM
then
captures
low-frequency
global
across
modalities
high-frequency
within
each
modality.
IAM
aggregates
complementary
features,
enabling
the
full
interaction
between
different
modalities.
Finally,
decoder
generates
image.
Additionally,
introduce
pixel
loss
structural
significantly
improve
model’s
overall
performance.
Extensive
experiments
mainstream
datasets
demonstrate
TCTFusion
outperforms
other
state-of-the-art
both
qualitative
quantitative
evaluations.
Remote Sensing,
Год журнала:
2024,
Номер
16(20), С. 3804 - 3804
Опубликована: Окт. 13, 2024
The
fusion
of
infrared
and
visible
images
together
can
fully
leverage
the
respective
advantages
each,
providing
a
more
comprehensive
richer
set
information.
This
is
applicable
in
various
fields
such
as
military
surveillance,
night
navigation,
environmental
monitoring,
etc.
In
this
paper,
novel
image
method
based
on
sparse
representation
guided
filtering
Laplacian
pyramid
(LP)
domain
introduced.
source
are
decomposed
into
low-
high-frequency
bands
by
LP,
respectively.
Sparse
has
achieved
significant
effectiveness
fusion,
it
used
to
process
low-frequency
band;
excellent
edge-preserving
effects
effectively
maintain
spatial
continuity
band.
Therefore,
combined
with
weighted
sum
eight-neighborhood-based
modified
(WSEML)
bands.
Finally,
inverse
LP
transform
reconstruct
fused
image.
We
conducted
simulation
experiments
publicly
available
TNO
dataset
validate
superiority
our
proposed
algorithm
fusing
images.
Our
preserves
both
thermal
radiation
characteristics
detailed
features
IEEE Transactions on Image Processing,
Год журнала:
2025,
Номер
34, С. 1340 - 1353
Опубликована: Янв. 1, 2025
In
this
paper,
we
introduce
MaeFuse,
a
novel
autoencoder
model
designed
for
Infrared
and
Visible
Image
Fusion
(IVIF).
The
existing
approaches
image
fusion
often
rely
on
training
combined
with
downstream
tasks
to
obtain
high-level
visual
information,
which
is
effective
in
emphasizing
target
objects
delivering
impressive
results
quality
task-specific
applications.
Instead
of
being
driven
by
tasks,
our
called
MaeFuse
utilizes
pretrained
encoder
from
Masked
Autoencoders
(MAE),
facilities
the
omni
features
extraction
low-level
reconstruction
vision
perception
friendly
low
cost.
order
eliminate
domain
gap
different
modal
block
effect
caused
MAE
encoder,
further
develop
guided
strategy.
This
strategy
meticulously
crafted
ensure
that
layer
seamlessly
adjusts
feature
space
gradually
enhancing
performance.
proposed
method
can
facilitate
comprehensive
integration
vectors
both
infrared
visible
modalities,
thus
preserving
rich
details
inherent
each
modal.
not
only
introduces
perspective
realm
techniques
but
also
stands
out
performance
across
various
public
datasets.
code
available
at
https://github.com/Henry-Lee-real/MaeFuse.
Briefings in Bioinformatics,
Год журнала:
2025,
Номер
26(2)
Опубликована: Март 1, 2025
Abstract
Parkinson’s
disease
(PD)
is
a
complex,
progressive
neurodegenerative
disorder
with
high
heterogeneity,
making
early
diagnosis
difficult.
Early
detection
and
intervention
are
crucial
for
slowing
PD
progression.
Understanding
PD’s
diverse
pathways
mechanisms
key
to
advancing
knowledge.
Recent
advances
in
noninvasive
imaging
multi-omics
technologies
have
provided
valuable
insights
into
underlying
causes
biological
processes.
However,
integrating
these
data
sources
remains
challenging,
especially
when
deriving
meaningful
low-level
features
that
can
serve
as
diagnostic
indicators.
This
study
developed
validated
novel
integrative,
multimodal
predictive
model
detecting
based
on
derived
from
data,
including
hematological
information,
proteomics,
RNA
sequencing,
metabolomics,
dopamine
transporter
scan
imaging,
sourced
the
Progression
Markers
Initiative.
Several
architectures
were
investigated
evaluated,
support
vector
machine,
eXtreme
Gradient
Boosting,
fully
connected
neural
networks
concatenation
joint
modeling
(FCNN_C
FCNN_JM),
encoder-based
multi-head
cross-attention
(MMT_CA).
The
MMT_CA
demonstrated
superior
performance,
achieving
balanced
classification
accuracy
of
97.7%,
thus
highlighting
its
ability
capture
leverage
cross-modality
inter-dependencies
aid
analytics.
Furthermore,
feature
importance
analysis
using
SHapley
Additive
exPlanations
not
only
identified
biomarkers
inform
models
this
but
also
holds
potential
future
research
aimed
at
integrated
functional
analyses
perspective,
ultimately
revealing
targets
required
precision
medicine
approaches
treatment
down
Plant Biotechnology Journal,
Год журнала:
2025,
Номер
unknown
Опубликована: Март 30, 2025
Summary
Increased
drought
frequency
and
severity
in
a
warming
climate
threaten
the
health
stability
of
forest
ecosystems,
influencing
structure
functioning
forests
while
having
far‐reaching
implications
for
global
carbon
storage
regulation.
To
effectively
address
challenges
posed
by
drought,
it
is
imperative
to
monitor
assess
degree
stress
trees
timely
accurate
manner.
In
this
study,
gradient
experiment
was
conducted
with
poplar
as
research
object,
multimodal
data
were
collected
subsequent
analysis.
A
machine
learning‐based
monitoring
model
constructed,
thereby
enabling
duration
trees.
Four
processing
methods,
namely
decomposition,
layer
fusion,
feature
fusion
decision
employed
comprehensively
evaluate
monitoring.
Additionally,
potential
new
phenotypic
features
obtained
different
methods
discussed.
The
results
demonstrate
that
optimal
learning
model,
constructed
under
exhibits
best
performance,
average
accuracy,
precision,
recall
F1
score
reaching
0.85,
0.86,
0.85
respectively.
Conversely,
novel
derived
through
decomposition
supplementary
did
not
further
augment
precision.
This
indicates
approach
has
clear
advantages
offers
robust
theoretical
foundation
practical
guidance
future
tree
response
assessment.
