Deep learning based ensemble model for accurate tomato leaf disease classification by leveraging ResNet50 and MobileNetV2 architectures
Scientific Reports,
Год журнала:
2025,
Номер
15(1)
Опубликована: Апрель 22, 2025
Global
food
security
depends
on
tomato
growing,
but
several
fungal,
bacterial,
and
viral
illnesses
seriously
reduce
productivity
quality,
therefore
causing
major
financial
losses.
Reducing
these
impacts
early,
exact
diagnosis
of
diseases.
This
work
provides
a
deep
learning-based
ensemble
model
for
leaf
disease
classification
combining
MobileNetV2
ResNet50.
To
improve
feature
extraction,
the
models
were
tweaked
by
changing
their
output
layers
with
GlobalAverage
Pooling2D,
Batch
Normalization,
Dropout,
Dense
layers.
take
use
complimentary
qualities,
maps
from
both
combined.
study
uses
publicly
available
dataset
Kaggle
classification.
Training
11,000
annotated
pictures
spanning
10
categories,
including
bacterial
spot,
early
blight,
late
mold,
septoria
spider
mites,
target
yellow
curl
virus,
mosaic
healthy
leaves.
Data
preprocessing
included
image
resizing
splitting,
along
an
80-10-10
split,
allocating
80%
training,
10%
testing,
validation
to
ensure
balanced
evaluation.
The
proposed
99.91%
test
accuracy,
suggested
was
quite
remarkable.
Furthermore,
guaranteeing
strong
performance
across
all
showed
great
precision
(99.92%),
recall
(99.90%),
F1-score
99.91%.
With
few
misclassifications,
confusion
matrix
verified
almost
flawless
even
further.
These
findings
show
how
well
learning
can
automate
diagnosis,
providing
scalable
accurate
solution
smart
agriculture.
By
means
intervention
agriculture
techniques,
strategy
has
potential
crop
health
monitoring,
economic
losses,
encourage
sustainable
farming
practices.
Язык: Английский
A Lightweight Framework for Protected Vegetable Disease Detection in Complex Scenes
Food Science & Nutrition,
Год журнала:
2025,
Номер
13(5)
Опубликована: Май 1, 2025
ABSTRACT
The
rapid
development
of
computer
vision
technology
has
provided
new
technical
support
for
smart
agriculture.
Vegetable
diseases
represent
a
significant
threat
to
agricultural
production,
with
severity
that
cannot
be
ignored.
However,
through
scientifically
effective
prevention
and
control
measures,
these
negative
impacts
can
significantly
mitigated.
Intelligent
disease
detection
systems,
as
advanced
methods
replacing
traditional
manual
inspection,
have
become
important
means
developing
agriculture
improving
the
efficiency
vegetable
production
management.
Nevertheless,
is
not
only
time‐consuming
labor‐intensive
but
also
faces
accuracy
limitations,
while
existing
still
encounter
series
challenges
when
confronting
complex
backgrounds,
diverse
manifestations,
varying
degrees
occlusion
in
real
cultivation
environments,
including
insufficient
anti‐interference
capabilities,
limited
precision,
suboptimal
real‐time
performance.
This
research
addresses
practical
data
acquisition
sample
scarcity
protected
by
proposing
an
innovative
strategy
implements
differentiated
augmentation
technique
combinations
different
categories
samples,
enhancing
model's
resistance
environmental
interference.
Based
on
integrated
concepts
machine
deep
learning,
we
developed
lightweight
network
named
VegetableDet.
innovatively
combines
Deformable
Attention
Transformer
(DAT)
YOLOv8n
backbone
architecture,
perception
capabilities
long‐range
feature
dependencies.
Simultaneously,
Channel‐Spatial
Adaptive
Mechanism
(CSAAM)
into
Neck
network,
achieving
precise
localization
enhancement
key
features.
To
address
issue
low
model
convergence
efficiency,
further
designed
hierarchical
progressive
transfer
learning
training
strategy,
effectively
accelerating
adaptation
process
accuracy.
Experimental
evaluation
demonstrates
our
custom
comprehensive
dataset,
VegetableDet
exhibits
excellent
performance
detecting
30
healthy
samples
across
5
types,
precision
(P),
recall
(R),
average
(AP)
all
exceeding
90%,
overall
mean
Average
Precision
(mAP)
reaching
94.31%.
powerful
adaptability
under
conditions,
providing
reliable
monitoring
diseases,
broad
application
prospects.
Язык: Английский
Deep Learning Method with Domain-Task Adaptation and Client-Specific Fine-Tuning YOLO11 Model for Counting Greenhouse Tomatoes
Applied System Innovation,
Год журнала:
2025,
Номер
8(3), С. 71 - 71
Опубликована: Май 27, 2025
This
article
discusses
the
tasks
involved
in
operational
assessment
of
volume
produced
goods,
such
as
tomatoes.
The
large-scale
implementation
computer
vision
systems
greenhouses
requires
approaches
that
reduce
costs,
time
and
complexity,
particularly
creating
training
data
preparing
neural
network
models.
Publicly
available
models
like
YOLO
often
lack
accuracy
needed
for
specific
tasks.
study
proposes
a
method
sequential
detection
models,
incorporating
Domain-Task
Adaptation
Client-Specific
Fine-Tuning.
model
is
initially
trained
on
large,
specialized
dataset
tomato
detection,
followed
by
fine-tuning
with
small
custom
reflecting
real
greenhouse
conditions.
results
light
YOLO11n
achieving
high
validation
(mAP50
>
0.83,
Precision
0.75,
Recall
0.73)
while
reducing
computational
resource
requirements.
Additionally,
was
developed
captures
unique
challenges
environments,
dense
vegetation
occlusions.
An
algorithm
counting
tomatoes
also
created,
which
processes
video
frames
to
accurately
count
only
visible
front
row
plants.
can
be
utilized
mobile
surveillance
systems,
enhancing
monitoring
efficiency
greenhouses.
Язык: Английский