Membrane Systems (or P Systems) are a computational paradigm, inspired by the biological cell, that focuses on information processing through the use and formalization of distributed and parallel structures separated by membranes. Within these systems, a set of objects evolves through rules that modify and communicate them among regions. The work aims to formalize a Membrane System for the analysis of network traffic to detect volumetric anomalies within a packet-switched computer network, and classify them. We focus on two volumetric anomalies, i.e., Elephant Flow and DDoS , and on a high-cardinality anomaly, i.e., Super-Spreader , which is volumetric-by-effect. All these anomalies are typically related to well-known network attacks. After formalizing the model, an implementation on GPU using the CUDA platform is proposed, and compared to a CPU implementation. The reported results provide deep insights into the proposed model, highlighting its scalability and showing its ability in identifying and adapting to different scenarios of attacks, especially when they do not occur simultaneously, provided a correct tuning of the parameters. Experiments on a real trace show a very good detection performance for Elephant Flow and Super-Spreader anomalies (up to 0.95 F1 score), while the model struggles more in the detection of DDoS traffic (0.71 F1 score). The use of GPU for computation has proven to be well-suited for parallel processing tasks, such as those naturally enabled by Membrane Computing, strongly outperforming the use of CPU, with a time speedup of about 50 times versus the CPU. Understanding the characteristics and performance of this approach, as well as the alternatives proposed within the context of Membrane Computing, could open pathways to new strategies for networking system protection and guide future development of Network-based Intrusion Detection Systems based on P Systems.
Multi-agent UAV navigation in cluttered environments is challenged by deadlock, starvation-like persistent yielding, oscillation, and unsafe congestion, particularly in narrow passages and dynamically emerging bottlenecks. Existing approaches often rely on static planning assumptions, reciprocal compliance, or learning-based policies in which temporal ordering and fairness are not explicitly represented. This paper introduces T-CARE (Temporal Coordination-Aware Reinforcement Learning), a hybrid learning-heuristic multi-UAV coordination framework that integrates zero-shot constrained action selection with priority-aware temporal reservations. T-CARE extends the prior ZSE-CRL action-selection backbone by adding: (i) runtime spatiotemporal reservations, (ii) priority aging and wait-time accounting designed to reduce persistent yielding, (iii) temporary-goal reassignment for blocked or congested motion, (iv) stagnation detection with escape and recovery, and (v) geometry-agnostic bottleneck discovery with shared-corridor reuse. We evaluate T-CARE in simulation using three-swarm adversarial congestion stress tests, and ten-swarm reconstructed 3D urban/suburban environments, with each ten-swarm scenario containing one leader and three followers per swarm for a total of 40 UAVs, under strict zero-shot deployment. Across the tested scenarios, T-CARE achieved 100% success, 0% collision rate, and no observed persistent starvation or deadlock, while learning-only, reactive, and coordination-reduced baselines exhibited failures under the same evaluation protocol. These results support T-CARE as an empirically effective coordination architecture for simulated multi-UAV traffic, while formal convergence-time guarantees, communication-delay robustness, and real-world flight validation remain future work.
The proliferation of unmanned aerial vehicles (UAVs) in both civilian and military domains has intensified the need for autonomous counter-drone systems capable of operating without reliance on ground infrastructure. Existing ground-based and hybrid approaches suffer from high latency and complete failure under communication jamming or denial. This paper proposes a fully onboard (OB) architecture for autonomous drone-to-drone detection in both the visible (RGB) and thermal (IR) domains, where all perception and decision-making tasks are executed exclusively using the embedded computational resources of the unmanned aerial vehicle. In particular, this work analyzes the features of the physical components of the architecture (i.e., the Single-Board Computers or SBCs for short, the available sensors, and the UAV platforms) and their performances in the experimental settings. Various computational platforms are tested to assess their impact on the performance of the detection pipeline, evaluating specific parameters such as inference speed (fps), inference time (ms), power consumption (W) and operational autonomy. In order to enable a comprehensive evaluation, a ground-based (GB) counterpart was also implemented, where real-time video streams are transmitted from the drone to a ground station for processing and control commands are subsequently sent back. The onboard architecture offers significantly lower latency and complete independence from radio links and controllers, making it particularly suitable for applications requiring high robustness in communication-denied or contested environments. In particular, the findings highlight the advantages of the Jetson Orin Nano platform in achieving inference speeds up to 80.93 fps at 12.36 ms on YOLO v8n quantized models, overcoming state-of-the-art performances. According to our knowledge, this is the first fully onboard RGB-IR drone-to-drone visual detection architecture in the literature.
The increasing frequency and intensity of extreme meteorological events attributed to climate change have necessitated a paradigm shift in power grid resilience. Initiatives such as the RESISTO project address this challenge by bolstering critical infrastructure through the integration of advanced technologies, specifically artificial intelligence (AI) and thermal imaging. By prioritising proactive prediction, prevention, and detection, these interventions seek to ensure grid stability and service continuity under evolving environmental stressors. This study proposes a real-time novelty detection architecture for the thermal monitoring of power transformers, leveraging AI to provide early-warning signals for thermal anomalies. The experimental methodology involved the deployment of a distributed network of thermal cameras within Donana National Park (Spain), supported by a robust backend architecture for data ingestion, analysis, and automated alerting. The analytical system's long-term performance was validated using a thermodynamically simulated synthetic dataset spanning the full 2023 year, with performance evaluated through an event-based confusion matrix. Short-term efficacy was assessed using high-fidelity real thermal imaging. The root mean square error (RMSE) was used to quantify the predictive accuracy across both approaches, which produced values below 2.65% and 5.32% for synthetic data and real-world applications, respectively. These results confirm the system's robustness in anomaly detection and underscore its significant potential for enhancing the resilience of electrical distribution networks.
Accurate reconstruction of freeform engineering surfaces from measured point clouds is essential in modern CAD, CAM, CAE and industrial reverse-engineering pipelines. Conventional B-spline fitting often struggles to balance geometric fidelity, parameterization quality and computational efficiency, particularly when input data contain both smooth regions and localized irregularities. Although truncated hierarchical B-splines enable localized refinement, classical hierarchical fitting frequently applies refinement even when errors are predominantly caused by distorted parameterization rather than insufficient geometric resolution. This work presents a hybrid fitting framework that integrates local parametric optimization with adaptive truncated hierarchical refinement to achieve high accuracy while maintaining a compact surface representation. A global least-squares B-spline approximation establishes the baseline geometry, and an error-driven analysis identifies regions requiring improvement. Each region is first corrected through local optimization of parametric values to reduce mapping distortion, while refinement is introduced only when optimization alone provides insufficient error reduction. Truncation confines refinement to the selected region, limiting the recomputation to the associated control points. The framework is evaluated on several engineering geometries and consistently reduces refinement depth and degrees of freedom while achieving accuracy comparable to or better than classical hierarchical fitting. The results indicate that coupling parametric optimization with localized hierarchical refinement provides a computationally efficient and robust strategy for engineering-oriented freeform surface reconstruction.
As traffic cameras become prevalent, the automatic analysis of traffic scenes presents new opportunities and challenges. Advances in deep learning allow for automated characterization of traffic in such videos. This work aims to understand traffic flow without human supervision, focusing on the localization of road intersections. For this purpose, a three-stage method is proposed that uses a deep neural network for vehicle detection, an object tracker to recover vehicle trajectories, and unsupervised machine learning to detect potential incoming and outgoing traffic flows. The approach has been tested on a variety of real and synthetic videos, with satisfactory results across different camera positions, traffic patterns, and weather conditions. As a key part of the methodology, five options for clustering starting and ending track points were tested. These options included a basic strategy based on predefined spatially localized clusters, and the K-means algorithm with two methods to determine the optimal number of clusters: the Elbow method and the Silhouette score. Additionally, Mean Shift and the Density-Based Spatial Clustering of Applications with Noise were evaluated. An exhaustive analysis of the proposed clustering methods was conducted, including runtime at each stage, performance metrics, and the addition of noise to simulate tracker failures. The results demonstrated the feasibility of the proposed methodology and concluded that Mean Shift is the most suitable clustering method due to its balance of high performance, low runtime, and stable behavior against abnormal trajectory points.
Simultaneous Localization and Mapping (SLAM) systems typically rely on a prior map constructed during an initial deployment. In real-world environments, however, structural and semantic changes gradually invalidate this map, leading to degraded localization accuracy and, in severe cases, localization failure. This limitation poses a major obstacle to the long-term deployment of mobile robots in dynamic environments. This paper proposes a lifelong mapping framework with multi-view projection fusion (LLMF) that enables efficient map maintenance while preserving a consistent global coordinate system. The framework introduces two key design components. First, a multi-view point cloud projection alignment strategy based on Bird's-Eye View (BEV) and frontal view (FV) projections is employed to align point cloud maps acquired at different times without re-labeling previously defined operational points. Second, an image-based change detection and map update mechanism is developed, transforming computationally expensive 3D point cloud comparisons into efficient 2D image processing operations. The proposed framework is evaluated through qualitative experiments on the open-source MulRan dataset and quantitative long-term experiments conducted over more than nine months in a real farm environment. Experimental results demonstrate that LLMF maintains localization accuracy while significantly reducing the computational cost of change detection, lowering processing time from several hours to a few minutes. These results indicate that the proposed framework provides a practical and scalable engineering solution for long-term robot operation in changing environments.
With the proliferation of mobile smart devices, location data has become a critical asset in various applications. WiFi-based fingerprint positioning technology is one of the commonly used indoor positioning methods. However, the significant concern of privacy leakage has emerged as a crucial obstacle to its advancement. To address this issue, an enhanced privacy-preserving WiFi fingerprint localization scheme based on fingerprint recognition is proposed. The scheme initially employs the dummy fingerprint generation algorithm to create dummy location fingerprints that closely resemble the actual distribution, thereby obfuscating the user's true localization requests. The server then utilizes the Paillier homomorphic encryption algorithm for matrix multiplication selection to return encrypted query results corresponding to the real fingerprints, ensuring that user privacy data remains secure throughout the process. Furthermore, an enhanced dummy fingerprinting algorithm is proposed, aiming to optimize the movement entropy by leveraging location association information, hence improving location anonymity. Theoretical analysis and experimental results demonstrate the safety, effectiveness and practicality of the proposed scheme.
Crowd counting aims to estimate the number of individuals in images, and the use of multimodal data has been shown to significantly enhance counting accuracy. However, such approaches are highly sensitive to the loss or corruption of data from any single modality, leading to severe performance degradation. To address this limitation, a new problem setting-Modality-Reconfigurable Crowd Counting-is introduced, in which a model is required to maintain robust performance even when one of the input modalities (e.g., RGB or thermal) is perturbed or entirely unavailable. Modality reconfigurability is achieved through effective cross-modal information transfer, enabled by a Feature Patches Generator that leverages Margin Ranking Loss across multiple network layers to align and transfer discriminative features between modalities. Additionally, a Negative Knowledge Transfer Prevention module is incorporated to suppress misleading or detrimental cross-modal signals. State-of-the-art performance is demonstrated on RGB-T crowd counting benchmarks, with consistent accuracy maintained under both complete and degraded modality conditions.
Accurate segmentation of low-contrast images plays a crucial role in computer-aided diagnosis and treatment, particularly for early lesion detection and clinical decision support. To address the limitations of existing approaches in boundary localisation and multi-scale context modelling, we propose a lightweight and efficient hybrid segmentation framework, referred to as SwinFuseNet. The proposed architecture combines the strengths of detection-based and Transformer-based models. Specifically, the Global Pyramid Attention Backbone Network integrates a shifted window-based Transformer mechanism to enhance the global representation of blurred lesions in low-contrast images. In the feature aggregation stage, two dedicated modules-Dynamic Zoom Fusion and Spatial Interaction Fusion-are introduced to adaptively integrate information from multiple layers, effectively refining local boundary representations and fine-grained structural features. Additionally, a lightweight attention subnetwork is employed to highlight salient regions while suppressing background noise, thereby improving overall segmentation precision. Experiments conducted on four publicly available low-contrast image segmentation datasets (ISIC 2018, PH 2, LUNA16, Kvasir-SEG and Brisc2025) demonstrate that the proposed method significantly outperforms existing models, including variants of U-Net and a recent detection-based segmentation framework. On the ISIC 2018 dataset, the proposed network achieves a Dice coefficient of 0.9873 and an Intersection-over-Union score of 0.9566, representing improvements of 4.48% and 6.53% respectively over the current state-of-the-art, showing a remarkable 78% and 60% improvement, respectively, toward perfection from the best alternative algorithm. These results confirm the effectiveness and practical relevance of the proposed method in the domain of low-contrast medical image segmentation.
Detecting small objects reliably is particularly difficult in modern neural architectures, where scale imbalance, background clutter, and high object density frequently degrade feature quality and prediction accuracy. To address these challenges, we propose YOLO-Super Resolution and Attention (YOLO-SRA), a multi-scale neural architecture enhanced with attention and super-resolution. The architecture introduces High-Resolution Feature Enhancement (HRFE) to better represent small objects without incurring high computational cost, a Grouped Multi-Scale Split Attention (GMSA) mechanism to efficiently extract features from densely distributed objects, and Weighted Fine-Grained Cross-Scale Fusion (WFCF) network for adaptive multi-scale feature integration with Unmanned Aerial Vehicle (UAV)-specific adjustments. The Spatial-Attentive Non-Maximum Suppression (SA-NMS) strategy is further employed to reduce missed detections in overlapping regions. Extensive experiments on the VisDrone dataset demonstrate that YOLO-SRA outperforms the baseline YOLOv11, achieving 12.5% and 11.3% increase in mAP 50 and mAP 50 : 95 , respectively, while reducing the parameter count by 6.0M. These results highlight the effectiveness of the proposed approach as a methodological contribution to neural systems for small object detection in aerial image.
Recent works on pose-agnostic anomaly detection (PAD) have addressed the challenge of identifying visual defects when the test object's pose is unknown, that is, when test images may depict the same object but in arbitrary orientations not seen in the reference anomaly-free dataset. In this unsupervised setting, models rely only on the knowledge of non-defective samples and their task is to detect anomalies appearing anywhere on the object surface. Current state-of-the-art approaches, such as OmniPoseAD, SplatPose, and SplatPose+, have advanced the field by introducing dedicated algorithms and frameworks for pose-agnostic anomaly detection. The present work consists of an engineering-oriented integration effort aimed at adapting existing PAD approaches to realistic industrial scenarios in which background clutter must be addressed for practical deployment. Two main contributions are provided: first, a simulated dataset for pose-agnostic anomaly detection with realistic industrial scenes; second, a complete pipeline that handles the introduced scenarios. Experimental results, carried out in comparison with the state-of-the-art SplatPose+ and measured in terms of pixel-level AUROC, AUPRO, image-level AUROC, and F-1 -score, demonstrate good performance on the proposed dataset. Code is available at: https://github.com/enmarchi/3dpad_background.
Industrial processes are becoming increasingly complex, requiring advanced modelling techniques to understand their behaviour and improve their performance. In this context, deep learning algorithms have proven to be effective tools for modelling dynamic systems, with Recurrent Neural Networks (RNNs) being particularly suitable for time-series data. However, the computational complexity of deep learning models can be a limitation in industrial environments, where real-time responses are required.This work proposes the use of Deep Echo State Networks to model an industrial system. The aim is to evaluate its performance in real-time industrial applications when running on embedded devices. The approach is validated on a process composed of four interconnected water tanks, which exhibits typical nonlinear industrial dynamics. Among several candidate architectures (including vanilla RNNs or LSTMs), Deep ESNs were selected for their balance of accuracy and computational efficiency. Different input-output setups and number of Deep ESN layers are tested, and results are compared with LSTMs in terms of accuracy and execution time. Finally, the best Deep ESN models are implemented on industrial embedded devices to evaluate the possibility of running these models in real time.The proposed approach achieved up to a 33% reduction in RMSE and a 14% improvement in R 2 compared to traditional reservoir computing, highlighting its superior predictive performance. The results show that Deep ESN models can effectively model the industrial system, with the best configurations achieving high accuracy and low execution times, demonstrating the feasibility of running these models in real time in industrial environments.
Earthwork Allocation and Transportation (EAT) significantly impacts construction cost in Railway Alignment (RA) optimization, but slight research has been devoted to this problem. To this end, an RA-EAT bi-level optimization framework is developed. A concurrent RA-EAT design model is formulated at the upper-level. At the lower level, the EAT system is modelled by incorporating earthwork allocation sections partition, soil waste/borrow pits selection and access roads generation. To solve this bi-level model, a tailored solution strategy is proposed. Specifically, a candidate pool for borrow/waste pits is established via a moving window method. Then, RA alternatives are generated using a Particle Swarm Optimization (PSO). Afterward, the EAT model is solved with a hierarchical approach, including determining EAT sections along an RA with a divide-and-conquer method and configuring access roads with a modified Dijkstra's algorithm. Thus, the total RA-EAT solution can be iteratively evolved through the PSO rationale. Finally, the proposed method is verified and analyzed in a real-world case via a sensitivity analysis and a comparative experiment.
Left Bundle Branch Block (LBBB) diagnosis is crucial for patient stratification and the selection of individuals who are likely to respond to Cardiac Resynchronization Therapy (CRT). The pathophysiological distinction between LBBB and strict LBBB (sLBBB) is investigated in this research with a view to optimizing diagnostic criteria and therapy. ECG signals were transformed into the vectorcardiographic (VCG) domain, where QRS loops were divided into two halves at the time of the velocity peak computed over the discrete derivates of the x, y, and z leads. From each half, angles and norms were extracted in all VCG planes, along with ratios between VCG peak velocity and VCG fidutial points. These were used to train machine learning models for classification into Healthy, LBBB, and sLBBB categories. The analysis identified four most significant features for the discrimination task: (1,2) peak velocity time relative to QRS onset/offset, (3) maximum norm of the early QRS loop in the frontal plane, and (4) QRS angle in the horizontal plane. These features preserved essential differences in conduction dynamics and electrical disturbances among the three groups. In particular, the time from velocity peak to QRS offset was the most discriminative feature, with progressive prolongation from Healthy to LBBB to sLBBB classes. This reduced 4-feature set achieved an accuracy of 0.85 and an F1-score of 0.83, which was on par with 15-feature-based models. Finally, the integration of explainable artificial intelligence (xAI) into these simplified models enabled the derivation of transparent diagnostic rules for LBBB, improving clinical interpretability on more reliable diagnostic decisions.
The uninterrupted growth of transactions carried out over the Internet (e.g., adoption of digital payments) can lead to potential exposure to serious security problems. Regarding the implementation of innovative technological solutions, the scientific community strives to develop approaches which can effectively protect the entities from adversarial cyber-threats. This research focuses on improving network security, developing the Hybrid Ensemble Deep Learning Intrusion Detection System ( HEDL infinity ). It is the optimized version of the previous two older releases, namely HEDL-IDS and HEDL - IDS 2 . The Ensemble consists of two Deep Neural Networks (DNN), four Convolutional Neural Networks (CNN) and five Recurrent Neural Networks (RNN) with parallel LSTM layers. The classifier of each Ensemble employs an improved Custom Vote process, following the Weighted Vote and the Majority Vote principles. The HEDL infinity was successfully validated on the UNSW-NB15 dataset, with overall accuracy equal to 99.56% during the training phase and 99.24% during the testing phase. The high values of the performance indices during testing confirm that the updated version is a robust tool that can be used in real-world, to significantly reduce the exposure of the network's users, paving the way for further research efforts.
In autonomous systems and robotics, acoustic signals provide valuable information for tasks such as acoustic source localization and recognition (LR), particularly in environments where visual sensing is limited. This paper investigates two unmanned aerial vehicles (UAVs)-based real-world scenarios that leverage acoustic scene awareness: (1) localization and recognition of human speech for search-and-rescue missions, and (2) detection and classification of other UAVs for counter-drone applications. To address these tasks, we design two deep learning models based on convolutional neural networks (CNNs) and a feature-based approach. These models process acoustic signals captured by two types of microphone arrays mounted on UAVs: a 4-microphone linear array and a 19-microphone spherical array. Each model performs direction of arrival (DOA) estimation and source classification under challenging ego-noise conditions using real-world datasets recorded in controlled experimental setups. We evaluate the models across different signal-to-ego-noise ratios and training configurations. Results show robust performance in both localization and recognition tasks, with approximately 6 degrees mean error and 7 degrees root mean square error (RMSE) for DOA estimations in the human speech scenario with multi-speaker classification accuracy till 0.95, and 3-5 degrees mean error and 7-11 degrees RMSE for DOA estimations in the UAV sound scenario with multi-UAV classification accuracy till 0.98. This demonstrates the potential of deep acoustic learning for UAV-based scene understanding in complex operational environments.
The train timetabling problem in liberalized railway markets represents a challenge to the coordination between infrastructure managers and railway undertakings. Efficient scheduling is critical to maximizing infrastructure capacity and utilization while adhering as closely as possible to the requests of railway undertakings. These objectives ultimately contribute to maximizing the infrastructure manager's revenues. This paper sets out a modular simulation framework to reproduce the dynamics of deregulated railway systems. Ten metaheuristic algorithms using the MEALPY Python library are then evaluated in order to optimize train schedules in the liberalized Spanish railway market. In addition, an analysis of the scalability of the problem has been carried out by comparing the results with those obtained with a classical mathematical model such as SCIP in Pyomo. The results show that the Genetic Algorithm outperforms others in revenue optimization, convergence speed, and schedule adherence. Alternatives, such as Particle Swarm Optimization and Ant Colony Optimization Continuous, show slower convergence and higher variability. The results emphasize the trade-off between scheduling more trains and adhering to requested times, providing insights into solving complex scheduling problems in deregulated railway systems.
Future 6G networks are expected to meet extreme Quality-of-Service (QoS) requirements. From delays below one hundred microseconds to bitrates above ten gigabit per second. Besides, these requirements must be guaranteed in massive scenarios, with densities above ten million devices per square kilometer, and be compatible with intense mobility where speed can reach up to one thousand kilometers per hour. In this context, network resource management cannot be static. Any fixed network configuration meeting such extreme requirements should be designed for the worst case, and then it would be oversized and non-profitable. To facilitate a tailored and efficient network resource distribution, dynamic management techniques are needed; so 6G nodes can envision the upcoming needs and get adapted to provide the expected QoS. However, previously reported predictive models for network resource management are either exclusively focused on software instances and service availability, or they consider network resources are a static pool to be optimally distributed among devices and/or verticals. New dynamic management schemes are required, centered on devices and their specific characteristics (high density, mobility, privacy restrictions, etc.), as well as they can predict the number of needed resources to serve the upcoming demand. This paper fills this gap. We propose a predictive management algorithm which can calculate the probability of network congestion for a given amount of resources, using the traffic theory and Gaussian models. Those models are generated through a federated scheme, where base stations periodically monitor the devices' resource consumption and produce a partial model with additive noise to preserve the devices' privacy. The network core collects all partial models and uses clustering technologies to produce a time-variant global model describing the dynamic resource demand. Scheduling policies are implemented to ensure the management algorithms do not have a relevant impact on network operations. An experimental validation based on simulation tools is also provided. Results show the achieved prediction precision is close to 93%, and the network resource consumption reduces up to 26% compared to a static configuration.
Accurate differentiation between Left Bundle Branch Block (LBBB) and its strict subtype (sLBBB) is essential for optimizing patient selection for Cardiac Resynchronization Therapy (CRT), yet remains clinically challenging. This study proposes and compares two graph-theory-based pipelines for automated classification of 12-lead electrocardiograms (ECGs) into Healthy, LBBB, and sLBBB categories. Functional connectivity graphs were constructed from inter-lead measures, including Pearson correlation, cross-correlation, and phase difference. The first approach combines Graph Signal Processing (GSP) with machine learning. Graph filtering was performed via spectral decomposition of the Laplacian matrix, selecting dominant eigenmodes and reconstructing signals through the inverse Graph Fourier Transform-integrating spatial and temporal features. The second approach converted connectivity matrices into grayscale images, classified using a Convolutional Neural Network (CNN), and incorporated Explainable AI (XAI) via Grad-CAM to visualize inter-lead interactions and enhance model transparency. The GSP-based method using phase difference and a Support Vector Machine achieved the highest performance (mean balanced accuracy = 0.8317 ), while the CNN-based approach with cross-correlation images reached 0.7646 , offering improved interpretability. Both methods distinguished pathological from healthy cases, but precise classification between LBBB and sLBBB remains challenging. These results highlight the complementary value of graph-based ECG analysis and support future hybrid models for CRT stratification.
