Electronics,
Год журнала:
2024,
Номер
13(9), С. 1654 - 1654
Опубликована: Апрель 25, 2024
Traditional
object
detection
methods
using
static
cameras
are
constrained
by
their
limited
perspectives,
hampering
the
effective
of
low-confidence
targets.
To
address
this
challenge,
study
introduces
a
deep
reinforcement
learning-based
visual
perception
enhancement
technique.
This
approach
leverages
pan–tilt–zoom
(PTZ)
to
achieve
active
vision,
enabling
them
autonomously
make
decisions
and
actions
tailored
current
scene
outcomes.
optimization
enhances
both
process
information
acquisition,
significantly
boosting
intelligent
capabilities
PTZ
cameras.
Experimental
findings
demonstrate
robust
generalization
method
across
various
algorithms,
resulting
in
an
average
confidence
level
improvement
23.80%.
Journal of Marine Science and Engineering,
Год журнала:
2024,
Номер
12(9), С. 1617 - 1617
Опубликована: Сен. 11, 2024
Cold-water
coral
(CWC)
reefs,
such
as
those
formed
by
Desmophyllum
pertusum
and
Madrepora
oculata,
are
vital
yet
vulnerable
marine
ecosystems
(VMEs).
The
need
for
accurate
efficient
monitoring
of
these
habitats
has
driven
the
exploration
innovative
approaches.
This
study
presents
a
novel
application
YOLOv8l-seg
deep
learning
model
automated
detection
segmentation
key
CWC
species
in
underwater
imagery.
was
trained
validated
on
images
collected
at
two
Natura
2000
sites
Cantabrian
Sea:
Avilés
Canyon
System
(ACS)
El
Cachucho
Seamount
(CSM).
Results
demonstrate
model’s
high
accuracy
identifying
delineating
individual
colonies,
enabling
assessment
cover
spatial
distribution.
revealed
significant
variability
between
within
areas,
highlighting
patchy
nature
habitats.
Three
distinct
community
groups
were
identified
based
percentage
coverage
composition
abundance,
with
highest
group
being
located
exclusively
La
Gaviera
canyon
head
ACS.
research
underscores
potential
models
VMEs,
facilitating
acquisition
high-resolution
data
essential
understanding
distribution,
structure,
ultimately
contributing
to
development
effective
conservation
strategies.
Technologies,
Год журнала:
2024,
Номер
12(11), С. 220 - 220
Опубликована: Ноя. 5, 2024
The
adversarial
robustness
of
image
quality
assessment
(IQA)
models
to
attacks
is
emerging
as
a
critical
issue.
Adversarial
training
has
been
widely
used
improve
the
neural
networks
attacks,
but
little
in-depth
research
examined
way
IQA
model
robustness.
This
study
introduces
an
enhanced
approach
tailored
models;
it
adjusts
perceptual
scores
images
during
enhance
correlation
between
model’s
and
subjective
scores.
We
also
propose
new
method
for
comparing
by
measuring
Integral
Robustness
Score;
this
evaluates
resistance
set
perturbations
with
different
magnitudes.
our
increase
five
models.
Additionally,
we
tested
adversarially
trained
16
conducted
empirical
probabilistic
estimation
feature.
Electronics,
Год журнала:
2024,
Номер
13(9), С. 1654 - 1654
Опубликована: Апрель 25, 2024
Traditional
object
detection
methods
using
static
cameras
are
constrained
by
their
limited
perspectives,
hampering
the
effective
of
low-confidence
targets.
To
address
this
challenge,
study
introduces
a
deep
reinforcement
learning-based
visual
perception
enhancement
technique.
This
approach
leverages
pan–tilt–zoom
(PTZ)
to
achieve
active
vision,
enabling
them
autonomously
make
decisions
and
actions
tailored
current
scene
outcomes.
optimization
enhances
both
process
information
acquisition,
significantly
boosting
intelligent
capabilities
PTZ
cameras.
Experimental
findings
demonstrate
robust
generalization
method
across
various
algorithms,
resulting
in
an
average
confidence
level
improvement
23.80%.
