In
automatic
driving,
its
detection
scene
is
different
from
the
simple
object
scene,
there
are
problems
such
as
complex
background,
large
change
in
size
of
similar
targets,
deformation
moving
targets
and
small
target
detection,
so
a
YOLO
V5-based
Convolutional
Block
Attention
Module
(CBAM)
Squeeze
Excitation
(SENet)
two
improved
algorithms
attention
mechanism,
through
to
enhance
ability
learn
specific
characteristics
target,
focus
on
obvious
details
road
scene.
Experiments
show
that
optimization
strategies
can
significantly
improve
accuracy,
also
effectively
model
V5
pay
global
information
detect
obstructed
autonomous
driving
scenarios.
Agriculture,
Год журнала:
2024,
Номер
14(4), С. 560 - 560
Опубликована: Апрель 1, 2024
Reasonably
formulating
the
strawberry
harvesting
sequence
can
improve
quality
of
harvested
strawberries
and
reduce
decay.
Growth
information
based
on
drone
image
processing
assist
harvesting,
however,
it
is
still
a
challenge
to
develop
reliable
method
for
object
identification
in
images.
This
study
proposed
deep
learning
method,
including
an
improved
YOLOv8
model
new
image-processing
framework,
which
could
accurately
comprehensively
identify
mature
strawberries,
immature
flowers
The
used
shuffle
attention
block
VoV–GSCSP
enhance
accuracy
detection
speed.
environmental
stability-based
region
segmentation
was
extract
plant
area
(including
fruits,
stems,
leaves).
Edge
extraction
peak
were
estimate
number
plants.
Based
plants
distribution
we
draw
growth
chart
(reflecting
urgency
picking
different
regions).
experiment
showed
that
demonstrated
average
82.50%
identifying
87.40%
ones,
82.90%
exhibited
speed
6.2
ms
size
20.1
MB.
technique
estimated
total
100
bias
error
images
captured
at
height
2
m
1.1200,
rmse
1.3565;
3
2.8400,
3.0199.
assessment
priorities
various
regions
field
this
yielded
80.53%,
those
provided
by
10
experts.
By
capturing
throughout
entire
cycle,
calculate
harvest
index
regions.
means
farmers
not
only
obtain
overall
ripeness
but
also
adjust
agricultural
strategies
both
quantity
fruit
set
plants,
as
well
plan
high-quality
yields.
IEEE Access,
Год журнала:
2024,
Номер
12, С. 14157 - 14167
Опубликована: Янв. 1, 2024
In
this
study,
we
propose
an
automated
system
for
measuring
the
size
of
strawberries
and
predicting
their
weight
using
AI
technology.
The
combines
computer
vision
techniques
with
LiDAR
sensor
data
to
accurately
estimate
dimensions
infer
weight.
By
integrating
deep
learning
models,
such
as
HRNet
keypoint
detection,
leveraging
capabilities
sensors,
minimize
human
intervention
achieve
precise
measurement.
relative
errors
width
height
are
3.71%
5.42%,
respectively,
exhibiting
a
lower
error
rate.
standard
deviation
0.19%
0.24%,
indicates
that
individual
had
very
low
rates
in
terms
measurements
height.
Weight
prediction
was
performed
through
regression
analysis
estimation.
Experimental
results
demonstrate
our
approach
enables
accurate
10.3%.
This
technology
holds
great
potential
strawberry
harvesting
classification
tasks,
facilitating
automation
these
processes.
Agriculture,
Год журнала:
2024,
Номер
14(3), С. 466 - 466
Опубликована: Март 13, 2024
The
origin
of
seeds
is
a
crucial
environmental
factor
that
significantly
impacts
crop
production.
Accurate
identification
seed
holds
immense
importance
for
ensuring
traceability
in
the
industry.
Currently,
traditional
methods
used
identifying
maize
involve
mineral
element
analysis
and
isotope
fingerprinting,
which
are
laborious,
destructive,
time-consuming,
suffer
from
various
limitations.
In
this
experiment,
near-infrared
spectroscopy
was
employed
to
collect
1360
belonging
12
different
varieties
8
distinct
origins.
Spectral
information
within
range
11,550–3950
cm−1
analyzed
while
eliminating
multiple
interferences
through
first-order
derivative
combined
with
standard
normal
transform
(SNV).
processed
one-dimensional
spectral
data
were
then
transformed
into
three-dimensional
maps
using
Gram’s
Angle
Field
(GAF)
be
as
input
values
along
VGG-19
network
model.
Additionally,
convolution
layer
step
size
1
×
padding
value
set
at
added,
pooling
layers
had
2
2.
A
batch
48
learning
rate
10−8
utilized
incorporating
Dropout
mechanism
prevent
model
overfitting.
This
resulted
construction
GAF-VGG
successfully
decoded
output
accurate
place-of-origin
labels
detection.
findings
suggest
exhibits
superior
performance
compared
both
original
PCA-based
terms
accuracy,
recall,
specificity,
precision
(96.81%,
97.23%,
95.35%,
95.12%,
respectively).
GAF-VGGNet
effectively
captures
NIR
features
origins
without
requiring
feature
wavelength
extraction,
thereby
reducing
training
time
enhancing
accuracy
origin.
Moreover,
it
simplifies
(NIR)
modeling
complexity
presents
novel
approach
analysis.
Research Square (Research Square),
Год журнала:
2024,
Номер
unknown
Опубликована: Янв. 4, 2024
Abstract
This
research
paper
presents
a
structured
approach
to
address
the
critical
concerns
associated
with
water
quality
assessment
and
underwater
waste
detection,
employing
advanced
machine
learning
techniques.
It
commences
an
exposition
on
significance
of
pollution's
impact
aquatic
ecosystems.
Subsequently,
methodology
employed
in
this
study
encompasses
utilization
YOLOv8
model
for
identification
waste,
rule-based
classifier
evaluation
quality,
application
XGBoost
algorithm
predicting
potability.
The
ensuing
sections
delve
into
practical
implementation
these
components,
offering
in-depth
insights
their
technical
intricacies
seamless
integration.
A
thorough
follows,
substantiating
system's
effectiveness
reliability
three
key
dimensions:
assessment,
potability
prediction.
As
indicated
by
lower
map50-95
score,
Yolov8
showed
impressive
precision
recall
recognising
positive
cases;
however,
improvements
are
required
complex
object
detection
scenarios.
Analysing
confusion
matrix
revealed
particular
categories
that
needed
be
improved.
On
other
hand,
produced
encouraging
outcomes,
demonstrating
excellent
accuracy,
f1
precision,
variety
categories,
underscoring
its
efficacy
precise
sample
class
concludes
transformative
potential
multifaceted
bolstering
environmental
conservation
safeguarding
ecosystems
against
pernicious
effects
pollution.
Applied Sciences,
Год журнала:
2024,
Номер
14(10), С. 4213 - 4213
Опубликована: Май 16, 2024
The
recognition
and
localization
of
strawberries
are
crucial
for
automated
harvesting
yield
prediction.
This
article
proposes
a
novel
RTF-YOLO
(RepVgg-Triplet-FocalLoss-YOLO)
network
model
real-time
strawberry
detection.
First,
an
efficient
convolution
module
based
on
structural
reparameterization
is
proposed.
was
integrated
into
the
backbone
neck
networks
to
improve
detection
speed.
Then,
triplet
attention
mechanism
embedded
last
two
heads
enhance
network’s
feature
extraction
accuracy.
Lastly,
focal
loss
function
utilized
model’s
capability
challenging
targets,
which
thereby
improves
recall
rate.
experimental
results
demonstrated
that
achieved
speed
145
FPS
(frames
per
second),
precision
91.92%,
rate
81.43%,
mAP
(mean
average
precision)
90.24%
test
dataset.
Relative
baseline
YOLOv5s,
it
showed
improvements
19%,
2.3%,
4.2%,
3.6%,
respectively.
performed
better
than
other
mainstream
models
addressed
problems
false
positives
negatives
in
caused
by
variations
illumination
occlusion.
Furthermore,
significantly
enhanced
proposed
can
offer
technical
assistance
estimation
harvesting.
Research Square (Research Square),
Год журнала:
2025,
Номер
unknown
Опубликована: Март 31, 2025
Abstract
In
order
to
achieve
high-precision
and
fast
detection
of
apple
targets
in
complex
orchard
environments,
this
study
proposed
a
lightweight
target
recognition
method
YOLOv10s-Star.
First,
based
on
the
YOLOv10s
model,
StarNet
is
used
as
backbone
network
reduce
number
parameters
calculations,
SCSA
attention
mechanism
added
PSA
module.
By
co-focusing
spatial
channel
mechanisms,
feature
extraction
ability
model
enhanced;
improved
BiFPN
module
structure
neck
full
fusion
utilization
deep
map
semantic
information
shallow
position
information,
thereby
improving
accuracy;
finally,
DyHead
head
designed
replace
original
scale
perception,
task
accuracy
efficiency
task.
Experimental
results
show
that
mAP
value
YOLOv10s-Star
92.4%,
5.06M,
amount
calculation
12.9G,
average
inference
speed
126.3
fps.
It
maintains
high
while
being
improves
speed.
suitable
for
deployment
embedded
devices
picking
robots,
laying
foundation
realization
intelligent
picking.
