Successes and limitations of pretrained YOLO detectors applied to unseen time-lapse images for automated pollinator monitoring
Abstract
Pollinating
insects
provide
essential
ecosystem
services,
and
using
time-lapse
photography
to
automate
their
observation
could
improve
monitoring
efficiency.
Computer
vision
models,
trained
on
clear
citizen
science
photos,
can
detect
in
similar
images
with
high
accuracy,
but
performance
taken
is
unknown.
We
evaluated
the
generalisation
of
three
lightweight
YOLO
detectors
(YOLOv5-nano,
YOLOv5-small,
YOLOv7-tiny),
previously
images,
for
detecting
~
1,300
flower-visiting
arthropod
individuals
nearly
24,000
captured
a
fixed
smartphone
setup.
These
field
featured
unseen
backgrounds
smaller
arthropods
than
training
data.
model
highest
number
trainable
parameters,
performed
best,
localising
91.21%
Hymenoptera
80.69%
Diptera
individuals.
However,
classification
recall
was
lower
(80.45%
66.90%,
respectively),
partly
due
Syrphidae
mimicking
challenge
smaller,
blurrier
flower
visitors.
This
study
reveals
both
potential
limitations
such
models
real-world
automated
monitoring,
suggesting
they
work
well
larger
sharply
visible
pollinators
need
improvement
less
sharp
cases.
Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown
Published: April 7, 2025
Language: Английский