The detection of plant disease using deep learning models has a gain a significant success in computer vision tasks, but even though most models obtain good performance, the application of these models on resource-limited edge devices is challenging. Having a trade-off between accuracy outputs and low computational costs becomes crucial . This research uses transfer learning on the PlantVillage dataset to evaluate six representative architectures. From well know CNNs to the last trending Vit. Beside accuracy, we measure the most important deployment metrics : model size, FLOPs, Parameters, and CPU inference latency. The finding shows us that Swin-Tiny has the highest accuracy ( 98.65%) yet needs huge resources (105 MB, 4.37 GFLOPs) whereas, the lightweight CNN (EfficentNet-B0) gives the best accuracy trade-off. Provides high level accuracy (97.10%) with low computational complexity (15.76 MB, 0.38 GFLOPs).
The Internet of Vehicles (IoV) enhances modern transportation but introduces significant cyber security vulnerabilities and privacy risks. Traditional centralized Intrusion Detection Systems (IDS) are inadequate for this environment, as they could compromise driver privacy. This study proposes a novel federated learning (FL) framework that enables collaborative cyber-attack detection across vehicles without sharing private data. We implemented and evaluated advanced FL algorithms (FedAvg, FedProx, SCAFFOLD, and FedNova) with a deep neural network (DNN) architecture. The framework is tested under a realistic non-identically and independently distributed (non-IID) data, where the federated clients are designed to detect specific in-vehicle Controller Area Network (CAN) attacks. Our experimental validation demonstrates that FedNova achieved an optimal F1-score of 81.67% for attack detection while maintaining complete data isolation. These findings establish that FL has the potential to be used as an IDS model in the IoV networks, while still preserving user privacy.
A physical layer addressing scheme that enables efficient multiple access is presented for wireless networks with multiple-input multiple-output (MIMO) communications tuned for maximizing beamforming gain. The method assigns unique, specially designed preamble sequences to each network node, which serve as physical addresses and enable this way selective training of the beamforming weights of each link. Communication is initiated using a previously proposed channel estimation procedure for SVD beamforming, where a source node transmits the preamble of the intended destination. For this scheme to be effective, the set of preambles must exhibit excellent correlation properties: low peak sidelobes in their autocorrelation to ensure accurate packet detection, and low peak cross-correlation to prevent a node from incorrectly detecting a preamble addressed for another. We propose an algorithm to find a set of binary sequences that jointly optimizes these properties. The performance of the addressing scheme is evaluated through extensive simulations. Receiver Operating Characteristic (ROC) curves are generated, demonstrating the trade-off between the probability of missed detection and the probability of false alarm with the proposed preambles. Results for a 4 × 4 MIMO system show that at a signal-to-noise ratio of 2 dB, a detection threshold can be chosen to achieve both missed detection and false alarm probabilities below 10−2, confirming the viability of the proposed method.
Let $\mathcal{T} = {\mathbb{F}_{{p^s}}} + {u_1}{\mathbb{F}_{{p^s}}} + {u_2}{\mathbb{F}_{{p^s}}} + {u_3}{\mathbb{F}_{{p^s}}} + u_1^2{\mathbb{F}_{{p^s}}}$ be a finite non-chain ring. In this paper, we propose a method of constructing quantum codes and LCD codes from constacyclic codes over $\mathcal{T}$. The first codes are obtained via the Gray map and the Calderbank-Shor-Steane construction from Euclidean dualcontaining constacyclic codes over $\mathcal{T}$. The second codes are obtained as the gray image of consta cyclic codes of LCD codes over $\mathcal{T}$.
Millimeter-wave phased array systems rely on periodic beam training intervals to maintain directional alignment, which interrupts data transmission and reduces effective throughput. To address this limitation, this paper introduces a dual-band architecture in which a narrow Steering Band performs continuous angular scanning using a Time-Varying Phased Array (TVPA), while a Primary Band sustains uninterrupted high-rate communication. The TVPA waveform produces a deterministic and range-invariant angular sweep, allowing receivers to identify their optimal sector without requiring suspension of the data link. The overhead comparison is carried out for a 2.16 GHz channel with Steering Band allocations from 25 to 400 MHz, and beam training durations from 100 to 104 µs across sweep intervals of 10 ms to 1 s. The analysis demonstrates a consistent trend in which the spectral cost of maintaining a continuous Steering Band remains fixed, while the cost of periodic sweep-based training increases as training duration becomes non-negligible. These results show that continuous steering can preserve directional alignment while avoiding the beam-training interruptions, offering a promising alternative to sector sweep procedures in directional wireless systems.
We investigate a mesh-free artificial neural network (ANN) solver for the variable-coefficient Helmholtz equation, a core model for time-harmonic wave phenomena in heterogeneous media. The proposed formulation represents the solution as a smooth neural function and trains it by minimizing a composite loss that enforces the PDE residual via automatic differentiation together with Dirichlet boundary data. We evaluate the method on a controlled two-dimensional (2-D) benchmark with spatially varying wavenumber, quantify accuracy using global error metrics (L2, MAE, RMSE), and analyze accuracy cost drivers such as collocation budgeting and boundary sampling. Results indicate that the ANN approach attains competitive accuracy relative to classical discretization pipelines while avoiding explicit meshing and offering flexibility for complex geometries. We discuss practical design choices (activation, loss weighting, sampling) that are important for oscillatory solutions and heterogeneous coefficients. The present study focuses on a specific 2-D variable-coefficient case with a known target function selected for reproducibility and ablations; therefore, the findings should not be interpreted as fully general. We outline extensions to higher-frequency regimes, non-smooth coefficients, mixed boundary conditions, and three-dimensional settings, where the same formulation applies with augmented inputs and scaled collocation/memory budgets. Overall, the study provides a reproducible benchmark and guidance for deploying ANN-based solvers on variable-coefficient Helmholtz problems.
Indoor multi-UAVs are increasingly being utilized for real-time surveillance, monitoring, inspection, entertainment, and inventory tracking in constrained environments such as warehouses, logistics, underground tunnels, and commercial facilities. However, the increased density of UAV deployments at results in WiFi network congestion and interference, which degrades Quality of Service (QoS). UAVs require reliable communication with a Ground Control Station and edge computing resources to execute their missions, and QoS, measured here by throughput, is a key factor. This paper presents a mobility-aware, application-layer resource management framework for indoor multi-UAV networks. The edge-based network controller periodically monitors key performance metrics (e.g., RSSI, jitter, throughput, delay) and dynamically adjusts UAV transmission rates and access point associations periodically. We introduce the UAV Mobility Aware Congestion Control (UMACC) algorithm and evaluate it using the ns-3 simulator under dynamic UAV mobility patterns, including linear, random, and predefined. Simulation results demonstrate that our mobility-integrated framework significantly improves overall network performance and maintains QoS under high-density UAV conditions when compared to typical behaviour that is observed using WiFi operation.
The rising incidence of cyberattacks underscores the need for effective and cooperative approaches to cybersecurity. This study presents a public-goods-based game-theoretic model to investigate cooperative investment decisions under cyberattack risks within Watts–Strogatz networks. Agents, representing interconnected entities, decide on contributions to a shared defense while facing uncertainty in attack likelihood and severity. The Watts–Strogatz topology allows exploration of network structures from regular lattices to small-world and fully random networks. Simulation results demonstrate that network topology strongly influences both the emergence of cooperation and the rate of successful cyberattacks. Moreover, risk perception interacts with network structure to shape investment behavior and overall system security. These insights inform the design of effective cybersecurity policies in interdependent networked systems.
The growing frequency and sophistication of cyberattacks underscore the need for adaptive and collaborative defense mechanisms. This study introduces a public-goods-based game-theoretic framework to model cooperative cybersecurity investments for protecting shared cybersystems. In the proposed setting, agents—representing interconnected entities—repeatedly decide whether to contribute to a collective defense while facing uncertainty in attack outcomes. The attacker is equipped with two distinct strategies: a mixed strategy and a reinforcement learning (RL)-based strategy. Simulation results reveal that the RL-based attacker outperforms the mixed strategy, dynamically adapting to defenders’ behaviors and substantially weakening cooperative investment as the defenders’ enhancement factor increases.
The increasing reliance of industrial infrastructures on the Industrial Internet of Things (IIoT) has made them more vulnerable to complex cyberattacks, especially Advanced Persistent Threats (APTs). To recognize and analyze these multi-stage incursions, we require computational models that can capture both structural and temporal connections in IIoT network architectures. This paper presents a framework based on Graph Neural Networks (GNNs) for detecting and classifying APTs in IIoT settings. We use the CICAPT-IIoT2024 dataset, which provides realistic multi-phase APT attack scenarios. The approach models system components, network communications, and process interactions as nodes and edges in a dynamic graph, allowing for relational learning and context-aware feature extraction. The GNN architecture leverages graph connectivity patterns and message-passing techniques to identify attack phases with greater accuracy and robustness. Experimental results show that this method outperforms traditional deep learning techniques and ensemble methods, particularly in early-stage anomaly detection. This paper highlights the potential of graph-based learning as an effective way to enhance IIoT infrastructure security against the changing behaviors of advanced persistent threats.
Reconfigurable Intelligent Surfaces (RIS) have gained significant attention as a key enabler for next-generation wireless networks, offering the potential to intelligently shape the radio environment. Recently, hybrid RIS architectures have been proposed to enhance spectral and energy efficiency, as well as to provide greater flexibility in system design. However, existing works on hybrid RIS largely overlook the nonlinear distortions caused by the power amplifiers (PAs) employed by the RIS active elements. This paper investigates the impact of these nonlinearities on the performance of Orthogonal Frequency Division Multiplexing (OFDM) systems in a hybrid RIS setup. We derive analytical expressions for spectral efficiency (SE) under both linear and nonlinear conditions. In order to reduce the impact of nonlinear distortions on system performance, a channel-aware dynamic selection of active elements is proposed. Simulation results show that the proposed dynamic selection strategy significantly improves SE and effectively reduces the impact of nonlinearities in hybrid RIS systems.
Electrical power systems are complex and dynamic networks of interconnected components, inherently susceptible to disturbances and faults. Timely and reliable fault detection is crucial to maintain system stability and prevent large-scale outages, especially as power grids become increasingly distributed and incorporate high-capacity generation units. While traditional machine learning models have been applied for fault classification, they often lack the transparency and interpretability required for deployment in safety-critical environments. To address this, we design Kolmogorov–Arnold Networks (KANs) for explainable fault detection in power transmission systems. KANs decompose complex relationships into simpler, interpretable components, allowing the contributions of individual input features to classification decisions to be traced. Our experimental evaluation demonstrates that KANs accurately detect and classify diverse fault types while remaining computationally efficient. The framework not only achieves high predictive performance but also provides inherent interpretability.
The development of 6G mobile communications, expected around 2030, aims to deliver significant advancements over 5G, including data rates up to 1 Tbps. The THz band presents a promising opportunity for achieving these high data rates due to its large contiguous bandwidth. However, its adoption comes with substantial challenges, including high path loss, low penetration power, and the presence of incumbent services such as the Earth Exploration Satellite Service (EESS). EESS relies on highly sensitive passive sensors, necessitating strict interference regulations for any new services operating within the THz spectrum. This paper investigates the interference impact of 5G and 6G networks on EESS operations through extensive Monte Carlo simulations. Various interference mitigation techniques, including equivalent isotropically radiated power (EIRP) reduction, tapering algorithms and Uplink (UL) and Downlink (DL) throughput limitations were applied to assess their effectiveness in minimizing interference. The results indicate that, even with these mitigation techniques, the received interference levels exceed the established thresholds for both 5G and 6G scenarios. These findings highlight the need for further research into other technologies and techniques for interference management. It is likely that a combination of interference reduction techniques will be required to enable coexistence between 6G networks and incumbent THz-band services.
In a context of increasing digitalisation of administrative processes, cybersecurity has become a strategic issue for states, particularly Burkina Faso. Unfortunately, there is a lack of research into cybersecurity in Burkina Faso. In this article, we present an approach for identifying vulnerabilities in applications and websites from Burkina Faso’s cyberspace according to the OWASP Top 10 2021. Implementing this approach enabled us to collect 241 websites and web applications from various fields. Analysing the security risks of a sample of 20 websites and web applications identified 18,521 web vulnerabilities, forming the basis of a dataset called "BF-WeakWeb-2025". Six of the CWE identifiers found are listed among the 2024 CWE Top 25 most dangerous Software Weaknesses. To the best of our knowledge, this is the first study of web vulnerability analysis based on the OWASP Top 10 in Burkina Faso’s cyberspace. This dataset addresses the inadequacy and obsolescence of existing datasets in the field of cybersecurity. The dataset was used to fine-tune three Large Language Models (LLMs) — BERT, Llama, and Flan-T5 — to detect and classify the six CWE identifiers: CWE-693, CWE-79, CWE-1021, CWE-352, CWE-264, and CWE-89. Analysis of the results shows that the fine-tuned models correctly classify CWE identifiers with an accuracy rate of 98%, and are also robust against unbalanced data. This demonstrates the quality of the dataset.
Recently, cyber threat detection has become an integral part of cybersecurity teams' operations. Additionally, more threat-related information is now being shared on social media platforms than through any other source. Due to the vast amount of data on these platforms, manual supervision is no longer feasible. Therefore, machine learning (ML) models play a crucial role in identifying threat intelligence in social media posts. This paper aims to compare the capabilities of various ML models in detecting relevant and timely intelligence on X (formerly known as Twitter)—a prominent platform that brings together cybersecurity professionals, hackers, and political influencers. We trained several models, including ML algorithms (Logistic Regression, K-Nearest Neighbors, Random Forest, Support Vector Machine, XGBoost) and deep learning (DL) models (BiLSTM and CNN), using the multitask-cyberthreat-detection dataset and compared our findings with previous research. Our analysis shows that Logistic Regression consistently delivered the best performance among traditional ML algorithms in related studies, while CNN achieved the highest accuracy among DL models.
This paper introduces the Time-Varying Phased Array (TVPA), a waveform-level model based on the Frequency Diverse Array (FDA) that enables continuous angular scanning while preserving spatial coherence with range invariance. The TVPA is derived by imposing formal constraints on frequency offset, range drift, and temporal phase variation, resulting in a time-evolving array factor that closely approximates a conventional phased array at each resolvable instant. The model maintains consistent beam structure across time, including stable beamwidth, sidelobe symmetry, and angular resolution. Simulations confirm that the beam remains coherent over both short and long timescales, validating the TVPA as a potential solution for deterministic angular scanning in radar and communication systems.
The growing complexity and energy demands of Blockchain systems have intensified the need for intelligent and sustainable optimization strategies. This paper conducts a PRISMA-based systematic review of modern Artificial Intelligence approaches designed to enhance Blockchain security and energy efficiency. By analyzing recent advancements such as multi-agent reinforcement learning and graph neural networks, the study identifies how AI contributes to securing consensus mechanisms, detecting anomalies, and reducing energy consumption. Unlike prior reviews, this work provides an integrated perspective that connects AI-driven optimization with both security resilience and environmental sustainability. The findings highlight current research gaps, particularly the absence of large-scale experimental validation, and outline future directions toward intelligent, adaptive, and energy-aware Blockchain systems.
The explosive growth in wireless data traffic and connected devices calls for sophisticated approaches to radio spectrum and energy utilization. A key challenge lies in jointly managing user association and resource allocation, which directly determine how users connect to base stations and how power and bandwidth are distributed among them. While prior work has introduced various optimization and heuristic methods, many formulations simplify the problem by neglecting intercell interference, often by artificially limiting bandwidth or assuming fixed user rates. In this paper, we propose a novel interference-aware framework that explicitly incorporates intercell interference into the joint optimization of user association and resource allocation. Unlike prior approaches that avoid interference by restricting resources, we jointly optimize base station association, power control, and bandwidth allocation while accounting for inter-cell interference. Our method combines a heuristic association strategy with a barrier-method-based iterative solver for resource allocation. Simulation results show that explicitly modeling interference improves the achievable network throughput by up to 50% compared to interference-agnostic baselines, while maintaining rate feasibility under higher load. However, these gains come with a 10-15% increase in total power expenditure and a noticeable reduction in fairness (Jain’s index dropping from about 1.0 to 0.57), highlighting the trade-offs between throughput, energy efficiency, and user fairness in interference-aware resource management.
This study, in a real testbed environment, compares the performance of 802.11ax (Wi-Fi 6) and 802.11ac (Wi-Fi 5) Wireless Local Area Networks (WLAN) at 5 GHz band using the highest channel width of 160 MHz. We focus on TCP/UDP protocols with IPv4/IPv6 producing new results for throughput, Round Trip Time (RTT), and CPU Usage. Experimental results show that, with wireless security implemented, using UDP over IPv4 at a packet size of 1408 bytes, 802.11ax achieved a maximum throughput of 1133 Mbps, while 802.11ac achieved a throughput of 958.4 Mbps. We observe that the lowest RTT and CPU Usage was for 802.11ax. This was achieved with UDP over IPv4, For both 802.11ax and 802.11ac, we could not reach the theoretical bandwidth, even when using the highest channel width, and only one client.
The sixth generation (6G) of wireless networks demands high-performance communication solutions to achieve massive connectivity and ultra-low latency services. To achieve high throughput, rate-splitting multiple access (RSMA) offers interference management compared to conventional schemes. On the other hand, unmanned aerial vehicle (UAV) in heterogeneous network (HetNet) provides high coverage, improves throughput, and achieves massive connectivity. In this paper, we integrate UAV assisted HetNet with RSMA, and user clustering to maximize system sum rate in the presence of eavesdroppers while guaranteeing power constraints. The optimization problem jointly optimizes the common and private rates along with beamforming. The ptoblem is inherently non convex and solved using successive convex approximation (SCA) and first order Taylor expansion. The preformance of the RSMA-based scheme is measured in terms of the sum rate and is compared with NOMA based system. Numerical results show that RSMA-based system outperform conventional NOMA systems across various configurations. In particular, the results show higher achievable sum rates with increased number of antenna, and robustness to varying UAV transmit power and cluster separation distances.
