Jurnal Teknologi Informasi dan Ilmu Komputer,
Год журнала:
2024,
Номер
11(6), С. 1399 - 1412
Опубликована: Дек. 10, 2024
Pengenalan
ekspresi
wajah
adalah
tantangan
penting
dalam
pengolahan
citra
dan
interaksi
manusia-komputer
karena
kompleksitas
variasi
yang
ada.
Penelitian
ini
mengusulkan
arsitektur
sederhana
Convolutional
Neural
Network
(CNN)
untuk
meningkatkan
efisiensi
klasifikasi
emosi
pada
dataset
kecil.
Dataset
digunakan
Jaffe,
terdiri
dari
213
berukuran
256x256
piksel
tujuh
kategori
ekspresi.
Citra-citra
tersebut
di-resize
menjadi
128x128
mempercepat
pemrosesan.
Data
diproses
menggunakan
CNN
3
lapisan
konvolusi,
2
subsampling,
dense.
Kami
mengevaluasi
model
dengan
5-fold
10-fold
cross-validation
estimasi
kinerja
robust,
serta
teknik
hold-out
(70:30,
80:20,
85:15,
90:10)
perbandingan
hasil
jelas.
Hasil
menunjukkan
akurasi
tertinggi
sebesar
90.6%
learning
rate
0.001
pembagian
85%
data
latih
15%
uji,
melebihi
lebih
kompleks.
Meskipun
tidak
transfer
atau
augmentasi
data,
tetap
unggul
dibandingkan
pendekatan
tradisional
seperti
Local
Binary
Pattern
(LBP)
Histogram
Oriented
Gradient
(HOG).
Dengan
demikian,
terbukti
efektif
pengenalan
Abstract
Facial
expression
recognition
is
a
significant
challenge
in
image
processing
and
human-computer
interaction
due
to
its
inherent
complexity
variability.
This
study
proposes
simple
architecture
enhance
the
efficiency
of
emotion
classification
on
small
datasets.
Jaffe's
consists
images
sized
pixels
across
seven
categories.
These
were
resized
accelerate
processing.
The
was
processed
using
comprising
convolutional
layers,
subsampling
dense
layers.
We
evaluated
with
5-fold-
for
robust
performance
estimation
techniques
clear
result
comparison.
results
indicated
highest
accuracy
training
testing
split,
surpassing
that
more
complex
models.
Although
does
not
employ
or
augmentation,
it
still
outperforms
traditional
approaches
such
as
Thus,
this
proves
effective
facial
Remote Sensing,
Год журнала:
2025,
Номер
17(3), С. 422 - 422
Опубликована: Янв. 26, 2025
The
main
objective
of
the
present
study
was
to
develop
an
integrated
approach
combining
remote
sensing
techniques
and
U-Net-based
deep
learning
models
for
lithology
mapping.
methodology
incorporates
Landsat
8
imagery,
ALOS
PALSAR
data,
field
surveys,
complemented
by
derived
products
such
as
False
Color
Composites
(FCCs),
Minimum
Noise
Fraction
(MNF),
Principal
Component
Analysis
(PCA).
Dissection
Index,
a
morphological
index,
calculated
characterize
geomorphological
variability
region.
Three
variations
U-Net
architecture,
Dense
U-Net,
Residual
Attention
were
implemented
evaluate
performance
in
lithological
classification.
Validation
conducted
using
metrics
accuracy,
precision,
recall,
F1-score,
mean
intersection
over
union
(mIoU).
results
highlight
effectiveness
model,
which
provided
highest
mapping
accuracy
superior
feature
extraction
delineating
flysch
formations
associated
units.
This
demonstrates
potential
integrating
data
with
advanced
machine
enhance
geological
challenging
terrains.
Journal of Geophysical Research Machine Learning and Computation,
Год журнала:
2025,
Номер
2(2)
Опубликована: Май 10, 2025
Abstract
Lithological
thin‐section
image
classification
is
crucial
in
geology.
Traditional
manual
methods
rely
on
expert
experience,
being
subjective
and
time‐consuming.
Convolutional
neural
network
(CNN)‐based
automated
has
potential
but
less
effective
with
more
rock
types
limited
training
data,
restricting
its
applications.
We
propose
a
lightweight
framework
that
integrates
the
multi‐head
self‐attention
(MSA)
mechanism
into
classical
convolutional
(CNN)
architectures,
hereinafter
denoted
as
MSA‐CNN.
Specifically,
we
employ
VGG16
AlexNet
backbone
networks
incorporate
MSA
to
enhance
feature
extraction
from
small‐scale
lithological
data
sets.
The
resultant
MSA‐VGG16
MSA‐AlexNet
models,
after
fine‐tuning,
can
capture
geological
features
effectively
continuously
improve
accuracy.
conducted
comprehensive
experiments
public
set,
which
be
partitioned
3,
34,
105
categories
respectively.
model
exhibits
strong
generalization
ability
across
all
tasks.
Notably,
most
challenging
scenario
categories,
outperforms
previously
reported
best‐performing
same
set
by
approximately
9.61%.
These
results
strongly
validate
effectiveness
of
integrating
CNNs
for
classification.
They
highlight
this
method
practical
applications
represent
significant
advancement
Purpose
Artificial
intelligence,
particularly
deep
learning
(DL),
has
increasingly
influenced
various
scientific
fields,
including
soil
mechanics.
This
paper
aims
to
present
a
novel
DL
application
of
long
short-term
memory
(LSTM)
networks
for
predicting
behaviour
during
constant
rate
strain
(CRS)
tests.
Design/methodology/approach
LSTMs
are
adept
at
capturing
long-term
dependencies
in
sequential
data,
making
them
suitable
the
complex,
nonlinear
stress–strain
soil.
evaluates
LSTM
configurations,
optimising
parameters
such
as
step
size,
batch
data
sampling
and
training
subset
size
balance
prediction
accuracy
computational
efficiency.
The
study
uses
comprehensive
set
from
numerical
finite
element
method
simulations
conducted
with
PLAXIS
2D
laboratory
CRS
Findings
proposed
model,
trained
on
lower
stress
levels,
accurately
forecasts
higher
levels.
optimal
setup
achieved
median
error
3.59%
5.10%
3.86%
presenting
setup’s
effectiveness.
Originality/value
approach
reduces
required
time
complete
extensive
testing,
aligning
sustainable
industrial
practices.
findings
suggest
that
can
enhance
geotechnical
engineering
applications
by
efficiently
behaviour.