Al-Furat Journal of Innovations in Electronics and Computer Engineering

In this study, a method for skull stripping referred to as the 3D Enhanced
Residual U-Net is introduced. The approach combines the traditional U-Net
with an enhancement mechanism designed to improve both the effectiveness
and processing speed of the U-Net. An anisotropic diffusion filter (ADF)
reduces noise in MRI images while maintaining the edges of present objects.
This is accompanied by skull stripping to eliminate non-brain matter and
contrast enhancement to elevate the visual quality. The architectural
adaptations allow for rapid and stable training. As brain images vary
significantly from subject to subject, a deep learning approach will account for
these differences leading to consistent skull stripping results. Neurofeedback
Skull Stripping (NFBS) dataset was used for the proposed model formulation.
The results from the experiments show that the proposed approach is effective
and practical compared to previous methods. This method obtained an
impressive sensitivity rate of 0.9974, DSC of 1.0000, a specificity of 0.9983, an
IOU of 0.9831, an accuracy percentage of 0.9969, and a precision of 0.9961,
showing an actual ability to differentiate the distinct parts of the brain and
the skull.

Effective waste management, which includes processing garbage and
reducing the amount of uncontrolled refuse areas, is the first step
toward improving the environment. Waste classification and then
recycling them into alternatives as raw materials could improve the
environment including the quality of air and water. Using smart waste
bins is one of the proposed solutions to deal with waste management
processes. This work suggests an affordable design for smart waste
trash, which could be considered as an intelligent solution for waste
management. The system suggests using three colour-coded bins. Each
bin is customised for specific types of waste, which are food waste,
papers and metal cans or plastic waste. Mechanical sensors are used
for each bin in order to weigh it and notify the workers that the trash is
full. The other components of the suggested design are solar panel and
remote-control system. The solar panel is used for lighting and to
power surveillance cameras. The remote-control system is necessary
to monitor the whole operation of waste management. The proposed
system is recommended to be used in specific areas of society, such as
universities, schools, hospitals and malls as a first step and then
generalise it to include the whole city and even the public areas. Also,
in order to feature the need for such smart waste bins in different places
of the city, a questionnaire is proposed, and a survey is undertaken as
part of this work. Hence, including smart waste management systems
could ensure a significant advancement in urban planning. A
significant acceptance has been gained for the suggested design, and
the majority of participants have encouraged using those smart bins.

The primary objective of this research is to identify the structure of
modern communication networks, particularly 5G, and their efficiency in
transmitting and receiving data in a practical manner. The research
employed a network simulation method using the srsRAN Project 23.11,
the Samsung Galaxy S23, and LimeSDR Mini v2. The research concluded
that the new technology improved data transmission speed and reduced
latency by 74.6% (9.7 milliseconds). It also accelerated data downloads by
455%, reaching speeds of up to 512 Mbps. The network architecture
supports Massive MIMO, Beamforming, and an All-IP Core, making it a
suitable technology for use in many applications, such as remote surgery.

Holographic beamforming antennas help in increasing the capacity and
spectral efficiency of wireless networks while reducing interference by
employing advanced signal processing techniques for adaptive
beamforming patterns. Hence, these antennas can be effectively used in
5G and future communication systems. This review presents the state-ofthe-
art technology for holographic beamforming antennas and identify its
uniqueness compared with conventional beamforming technology. The
uniqueness of this technology is its capacity for controlling
electromagnetic waves. The review will also look into the main principles
and advantages of this technology and its future prospects. The study
reviews existing literature to compare holographic beamforming
technology with other technologies, such as conventional phased arrays
and massive MIMO to identify its specific advantages and disadvantages.
The advantages of high-frequency communications using holographic
beamforming can be attributed to its precise capabilities resulting from
its flexible design, simple hardware requirements, and beam steering
capabilities. Empirical studies on holographic beamforming antennas
identified specific advantages to include its flexible beam control system,
reduced power requirements, and accurate tracking capabilities to reduce
delay times for communications. The realization of holographic
beamforming systems depends on resolving three main obstacles, which
include high frequency propagation loss, fabrication constraints, and
energy efficiency concerns. Future investigators need to concentrate on
antenna optimization while developing improved fabrication methods and
smart control systems to improve performance and scalability for
practical 6G applications. The complete realization of holographic
beamforming potential in next generation wireless communications
depends on resolving the existing challenges.

Accurate and fully automated liver segmentation can significantly
improve disease detection and personalised treatment planning. Liver
cirrhosis is the final stage of liver disease and is characterised by fibrosis
and the remodeling of the liver, which increases the mortality rate.
Magnetic resonance imaging (MRI), particularly T1- and T2-weighted
sequences, provides a powerful non-invasive tool for assessing liver
structure and pathology. However, manual delineation of cirrhotic liver
regions is both time-consuming and prone to human error, with
performance heavily dependent on the radiologist’s expertise. Reliable
segmentation is further complicated by pronounced morphological
distortions and heterogeneous signal characteristics associated with
cirrhosis. In this study, we address these challenges using a fully
convolutional ResNet-UNet framework that employs cross-weighted
transfer learning, retrained on T1-weighted MRI data to capture general
anatomical representations and fine-tuned on T2-weighted images to
adapt to contrast and pathological variability, using the CirrMRI600+
dataset. The proposed model achieved the highest Dice coefficient of 0.89
and mIoU scores of 0.81 on the testing set, surpassing the performance of
training a single ResNet-UNet on T1 and T2 separately, without transfer
learning.
This cross-weighted training paradigm demonstrates the effectiveness of contrastdomain
pretraining for robust and efficient liver segmentation, paving the way for
multimodal MRI integration in hepatic disease assessment and advancing progress
toward automated cirrhosis staging and personalised therapeutic planning.

Accurate tracking of vehicles is a key problem in autonomous driving, especially
when the maneuvers are complex and/or under high noise sensing measurements.
Traditional model-based estimators, such as the EKF, can easily be affected by
linearization errors and are dependent on precise mathematical modelling and
carefully calibrated noise characteristics. In this paper, we propose an enhanced
learning-aided state estimator called KalmanNet GRU that maintains the recursive
property of Kalman filtering and introduces a neural dynamics learner based on a
GRU (gated recurrent unit) cell and a learnable Kalman gain in order to achieve
robustness against nonlinearities and measurement noises. The performance of the
proposed approach is demonstrated via the HDVT problem with nonlinear range–
bearing radar measurements at an inverse sensing noise level of 20 dB. The
experimental results reveal that KalmanNet significantly outperforms the EKF as it
can attain an average test MSE of −7.91 dB in contrast to 9.85 dB resulted by the
EKF, which means achieving nearly 17.76 dB improvement in performance. The
model proposed also has a stable estimation performance and yet holds an acceptable
processing time for real-time autonomous navigation. These results verify the utility
of recursive filtering by data-driven dynamics learning for nonlinear state estimation
in ITS.

Terrorist groups are resorting more to social networks to promote their
operations through Twitter. To attract new members, these groups
employ radical propaganda by posting content. Sentiment-based natural
language processing methods are widely used in practice to pre-process
such posts or tweets .A Semi-automatic annotation approach was used.
the first set of labels were created automatically with the help of a Zero-
Shot classifier (ZSc) , which text to category classification, after which
random tweets were manually verified. The labelled dataset was then
finally used to train the model. This paper gives a developed system of
terrorism-content classification utilizing the Cross-linguistic Language
Model RoBERTa Large (XLM-R Large) to categorize information in the
Arabic and English language in Twitter using structured and labelled
information. The accuracy of our system, which involved the
combination of automated labeling and deep learning, was 85% ,and it
was successful in identifying extremist and terrorist materials.

Convolutional Neural Networks (CNNs) are sole of the various forms of deep
learning that have become effective tools in computer vision. Machine learning has
been transformed by Deep Neural Networks (DNNs) with many parameters; their
impact is particularly evident in network architecture. CNNs are very good at image
classification problems because of their ability to concentrate on objects in images
and extract information using spatial relationships. In this work, two robust
architectures Xception and NASNetMobile are integrated into a unique CNN model
for image categorization based on transfer learning. Using Xception and
NASNetMobile, the model uses photographs with specified dimensions, often known
as "best windowing of images," as input to categorize the images into two groups.
To avert overfitting issues in CNN, a dropout layer is introduced after the outputs of
these designs are concatenated using a concatenate layer. This research evaluates the
proposed model for an exceptionally challenging dataset associated with diabetic
retinal disease. The "Diabetic Retinopathy 224*224 Grayscale images" dataset,
which is part of the "APTOS 2019 Blindness Detection" dataset on Kaggle, has 3662
images, of which 1875 show abnormal cases and the residual 1805 show normal
instances. With an accuracy of 97.50%, precision of 96.39%, recall of 98.64%, and
F1-score of 97.36%, the model fared extremely well in this test.

Due to their efficiency and torque density, they are commonly used in
electric vehicles and robotics. The catch is that except for mechanical
sensor input, torque and speed estimation in austere environments
remains the single most difficult obstacle. This paper presents an
exhaustive survey of observer-based torque estimation methods for
PMSM drives to analyze state-of-the-art contributions and highlight the
leading field research directions.
While observer-based techniques, such as EKF, UKF, MRAS, SMO
and Luenberger observers have been widely applied to handle this issue;
adaptive based nonlinear and AI based observers are arising to face such
problems. This was important for observers loops implementation in
transients, making load torque observers interesting from academic point
of view and they are trending around the recent research as to their better
disturbance rejection during dynamic conditions. Metaheuristic
optimization algorithms, including PSO, GA and GWO, have also been
employed to auto-tune the observer's and controller's parameters in order
to enhance estimation accuracy and dynamic performance.
These type of optimised based works have been used with Modern
control techniques such as Field Oriented Control (FOC) and Model
Predictive Control (MPC) to achieve better results. We subsequently offer
a critical perspective on the observer designs, state estimation
techniques, and control approaches, with the objective of furnishing some
useful guidance on the existing voids and future research directions.
Although, some progress has been achieved, most of these developments
are based on simulated studies and underline the need of real-time
implementation and hardware in the loop validation, high level of
uniqueness in drive system.

This comprehensive review systematically compares visionbased
and wearable sensor approaches for hand gesture recognition
(HGR), addressing a critical gap in existing literature that often treats
these modalities in isolation. While previous surveys have
predominantly focused on either vision-based methods or sensorbased
approaches separately, this review provides an integrated
analysis of both paradigms, highlighting their complementary
strengths and limitations across the entire HGR pipeline: from data
acquisition and preprocessing to feature extraction and classification.
We examine diverse datasets including UCI MYO Thalami and
RGB-based collections, analyzing preprocessing techniques (data
augmentation, noise reduction, normalization) and both traditional
machine learning (SVM, ANN, KNN) and deep learning methods
(CNN, RNN, LSTM). Our comparative analysis reveals that sensorbased
methods excel in controlled environments with precise motion
capture, while vision-based approaches offer greater usability at the
cost of environmental sensitivity. This review contributes a unified
framework for selecting appropriate HGR technologies based on
application requirements, computational constraints, and user needs,
particularly in healthcare rehabilitation, virtual reality, and assistive
technologies for hearing-impaired communities.

Intelligent and flexible resource management systems will be required as Sixth-
Generation (6G) wireless networks advance in order to minimize energy
consumption and safeguard against security risks. To reduce power consumption in
active symbiotic secure communication networks, this study uses Reconfigurable
Intelligent Surfaces (RIS) to improve the performance of Successive Interference
Cancellation (SIC). Because it offers reliable, scalable, and nearly perfect solutions
to this complex, non-convex, multi-variable optimization problem, artificial
intelligence is essential. We use the Educational Competition Optimizer (ECO), a
unique metaheuristic algorithm based on human cognition, to determine the optimal
values for the beamforming vector and the active RIS reflection coefficient matrix.
The algorithm finds a good balance between exploration and exploitation by
simulating how competition changes over time in elementary, middle, and high
school. The simulation results show that the proposed AI-driven approach uses
89.0% less power than traditional passive RIS, 52.5% less power than active RIS
with random phases, 12.8% less power than Alternating Optimization with
Successive Convex Approximation (AO SCA), 10.1% less power than hybrid
Particle Swarm Optimization Grey Wolf Optimizer (PSO GWO), and 4.7% less
power than hybrid Deep Neural Network Genetic Algorithm Deep Reinforcement
Learning (DNN GA DRL). It also meets all security and reliability requirements with
100% feasibility. These results show that AI has the power to change how we
communicate in the future by developing systems that are safe, energy-efficient, and
long-lasting.

This review paper provides a glimpse of the recent developments of
Substrate Integrated Waveguide (SIW) technology integrated with
Frequency Selective Surfaces (FSS) antenna arrays to be implemented
over the fifth-generation (5G) system operating within the millimeterwave
(mmWave) regime. The review is primarily based on papers that
were generated within the timeframe between the years 2020 and 2025
and some foundational research works that embody the background
theoretical knowledge.
The document encapsulates the core ideas of the SIW and the FSS
technologies and the application of the antenna array to improve
radiation performance.
The latest works have additionally proven the effectiveness of
exponential-tapered FSS configurations and multilayer wideband
filters, which are used in increasing gain and interference reduction at
sub-6 GHz and at mmWave frequencies. Investigations carried out in
the area of microstrip–FSS hybrids and endfire SIW arrays [3] verify
that integration methods of this kind are essential for high-gain lowprofile
antennas applicable in terminal and base station 5G
applications.
Stacking technologies that make system design compact and efficient
are also introduced. The application of technologies to enhance gain,
improve the bandwidth, and design radiation patterns is analyzed
extensively.
An exhaustive comparison of open-loop designs that have been made
public is presented, focusing on engineering benefits, design
challenges, new research directions, and potential utilization within
terminal and base station equipment within 5G systems.

This review of literature outlines advances in antenna design based on
5G millimeter-wave (mm-wave) technology and establishes that it has the
potential to meet previously unrealized data rates, very low latency, and
spectral efficiency, and meet the growing needs of the current wireless
network. The current condition of the mm-wave technology and the
antenna array development, and what they can do in the 5G applications,
have been mentioned in the review. It also presents the ideas of more
advanced antenna design, including dielectric resonator antennas, with
multiple inputs and multiple outputs, and a single-layer MIMO antenna,
which can operate over a great variety of frequencies. Microstrip array
antennas in this regard are important, especially at 28 and 38 GHz, since
they have the ability to achieve high gain and provide high precision in
beamforming to reduce serious propagation losses at mm-wave
frequencies. Moreover, they are very small, inexpensive, and can be
combined with printed circuit technologies, and thus they suit well in
real-life 5G applications. These antenna arrays maximize spectral
efficiency and system capacity as they are able to support advanced
MIMO techniques and therefore can maximize communication
performance and reliability in congested cities.

Massive Multiple-Input Multiple-Output (MIMO) is a critical Technique
for next-generation communication systems, as it significantly improves
spectrum efficiency and system capacity, that must meet user
performance and quality of service requirements(QoS). However the
detection of signal in massive MIMO systems remains challenging due
to the limited performance of traditional algorithms and high
computational complexity. Over recent years deep learning has been
introduced as a method for improving signal detection algorithms in
massive MIMO systems .This study provides a systematic review of
Deep neural network-based detection techniques for massive MIMO
systems .A systematic literature review was conducted, drawing on
recent studies selected from major scientific databases and research
published between 2016 and 2025, with a focus on widely adopted
detection frameworks. A unified comparison was performed using key
criteria, including bit error rate (BER) performance, computational
complexity and practical feasibility. The reviewed methods were
categorized as data-driven, model-driven, and hybrid approaches,
allowing for the organization of analyses according to their design
principles. The results indicate that model-driven and hybrid techniques
enhanced trade-off between detection accuracy and computational
complexity, while purely data-driven methods require intensive training
and exhibit limited generalize ability under varying channel conditions.
Compared to existing surveys, this work provides a more standardized
classification and a more accurate comparative framework, highlighting
current research trends and identifying existing challenges, such as
scalability for large-scale systems ,effective training strategies and
robustness to realistic channel environments.

Highly developed Man-in-the-Middle (MITM) attacks in Internet of
Things (IoT) networks present a significant security risk because they can
avoid protocol violations but introduce a small amount of behavioral
deviation such that more traditional intrusion detection systems cannot
detect it. This research will solve this issue by suggesting a hybrid neural
network that learns spatial, temporal and contextual features of IoT
traffic to classify multi-class MITM attacks. The model combines CNNLSTM,
Transformer-BiLSTM, and CNN-Transformer models to
collectively model local feature correlations, sequential dynamics, and
long-range dependencies. Large-scale experiments show that
performance has improved on all models following training, with the
CNN-Transformer having the highest overall performance, with Macro
F1-score of 97.72, accuracy of 98.00 and AUC of 98.41, and lower interclass
confusion and better stability through repetitions. The findings
attest to the idea that hybrid systems that consist of convolutional feature
generation and attention-based global modeling offer strong and balanced
detection of advanced variants of MITM attacks under heterogeneous IoT
conditions, which justifies their appropriateness in the context of
implementing effective IoT security in practice.

The speed of cyber-attacks and the increasing complexity of cyber theft
has made intrusion detection a focal point in today's cybersecurity
environment. The present paper is an overview of metaheuristic
optimization techniques to detect cyber-attacks using machine learning.
The purpose of this survey is to review the use of metaheuristic
algorithms in the improvement of intrusion detection systems (IDSs) in
terms of feature selection, parameter optimization, and multi-objective
optimization. This survey covers the main work that uses single-objective
and multi-objective metaheuristic algorithms such as evolutionary
algorithms, swarm intelligence algorithms, and hybrid metaheuristic
optimization approaches. The papers are reviewed in terms of their use
for IDS tasks, use of single-objective and/or multi-objective optimization,
and their performance using benchmarking datasets. The survey shows
some of the trends: metaheuristic algorithms are highly effective for
feature selection, classification performance, and a combination of
detection rate and false alarm rate. And hybrid solutions that use
metaheuristic algorithms and machine learning / deep learning
algorithms, such as Convolutional Neural Networks (CNN) and Gradient
Boosting, are more effective than single solutions. The advantages are
clear, yet the research also intersects some of the challenges related to
the methods, including scalability, efficiency, interpretability, and
adversarial attack resistance. Based on these insights, future research
will focus on developing adaptive and scalable optimisation approaches,
integrating optimisation approaches into real-time and distributed
systems, and designing interpretable and robust IDS. This survey offers
a practical perspective on the metaheuristic optimisation for cyber-attack
detection techniques, and the use of metaheuristic optimisation in
designing efficient and scalable IDS.

WSNs are currently actively implemented in Internet of Things (IoT)
settings where they are extremely susceptible to Denial-of-Service (DoS)
attacks because of their weak computational and energy capabilities. This
paper will provide experimental analysis of the machine learning and
deep learning models to detect multi-class DoS attacks using the WSNDS
dataset based on an optimization approach. The study will be aimed
at conducting systematic studies on the impact of feature selection, data
balancing, and hyperparameter optimization on the detection
performance under the resource constraints of the WSN settings. The
approach combines a dimensionality reduction technique (PCA), data
balancing technique (SMOTE), and hyperparameter optimization
technique (Ant Colony Optimization (ACO)). There are several
classification models which are experimentally tested both prior to and
following optimization to determine any enhancement in ability to detect
and classification balance. Findings indicate that optimization can
greatly lead to model stability, misclassification reduction, and
generalization improvement across types of attacks. The results offer
useful information on the choice and fine-tuning of the learning models to
achieve effective and dependable intrusion detection in WSN-based IoT
system.