Chest X-ray interpretation remains one of the most challenging tasks in medical imaging due to overlapping anatomical structures and the frequent coexistence of multiple thoracic pathologies. This study introduces a dual-scenario deep learning framework that integrates lung segmentation and multi-label classification with Grad-CAM-based interpretability. Two complementary preprocessing approaches are evaluated: (1) Scenario A, where the original image is concatenated with its corresponding lung mask (twochannel input), and (2) Scenario B, where only the segmented lung region is retained (masked input). Both configurations are tested using ResNet50, DenseNet161 and ConvNeXt-Tiny backbones on the NIH ChestX-ray14 dataset comprising 112,120 radiographs labelled with 14 diseases. Experimental results demonstrate that integrating anatomical priors improves model stability, interpretability, and classification accuracy. The best-performing setup-ConvNeXt-Tiny under the masked-lung scenario-achieves an average ROC-AUC exceeding 0.85, outperforming or matching recent state-of-the-art methods. Furthermore, Grad-CAM visualizations confirm that segmentation-driven masking enhances spatial coherence between model activations and true pathological regions. Overall, the proposed framework offers a reproducible, anatomy-aware, and interpretable approach for automated thoracic disease detection, with attention patterns that may align with clinically relevant regions.
The energy and hardware constraints of medical IoT devices limit the deployment of real-time medical imaging solutions in remote environments. This work presents FrugalVision, an integrated solution for the reliable transmission of medical images in degraded network environments. The system combines three innovations: a residual autoencoder with adjustable differentiable compression, a reinforcement learning agent (PPO)dynamically adapting compression parameters, and the MQTT protocol for data and metrics communication in a constrained network. Validated on three Decathlon medical segmentation datasets (hippocampus, heart, brain tumors), the solution maintains image quality with average PSNRs of 37.63 dB, 33.05 dB, and 35.02 dB, respectively, consistently exceeding the clinical threshold of 28 dB. Comparative analysis with state-of-art methods demonstrates a PSNR difference of up to +8.76 dB compared to conventional methods. Processing latency is 0.15 to 0.26 s per image, with adaptation decisions taking 0.6 to 2.3 ms. Modeled power consumption is 0.02 to 0.04 J on a standard architecture without dedicated hardware acceleration. The architecture enables real-time adaptation to network variations using a simulator based on real network traces from the 5GMeas dataset. This approach supports the deployment of medical imaging solutions in underserved areas.
This paper presents a novel fuzzy decision-making framework for decentralized task allocation in heterogeneous multi-robot systems, enhanced through particle swarm optimization. The proposed model, called Particle Swarm Optimization of Fuzzy Decision-Making for Task Selection (PSO-FDM-TS), automates the tuning of fuzzy membership function parameters to improve adaptability and robustness in uncertain environments with limited communication. Through extensive simulations in a warehouse transportation scenario, PSO-FDM-TS achieves a significant reduction in task completion error compared to both manually-tuned fuzzy systems and observation-based allocation approaches. Moreover, the results show that PSO-FDM-TS maintains superior performance across diverse operational conditions, including varying task heterogeneity, different robot team sizes, and large-scale deployments without the need for inter-robot communication. This work represents a key step toward scalable and efficient coordination of heterogeneous robot teams in real-world applications such as automated logistics and mission adaptation in hostile environments.
Blockchain's promise of trust and immutability has positioned it as a transformative technology in high-integrity domains. However, as adoption increases, challenges related to performance and node scalability become critical. This study presents the first empirical comparison of the Raft and SmartBFT consensus mechanisms within the Hyperledger Fabric framework. Key performance metrics, including latency, throughput, and CPU utilization, were evaluated. The results reveal a fundamental trade-off: while SmartBFT demonstrates higher resource consumption and limited scalability across increasing node counts, Raft provides greater efficiency and scalability, albeit with increased latency. Interestingly, the study also identifies unexpected performance behaviors of SmartBFT under specific conditions. These findings highlight the importance of selecting consensus algorithms based on application-specific requirements. All results are validated through statistical analysis. Overall, this work offers valuable insights into the design and optimization of Hyperledger Fabric-based data-sharing applications.
In this paper, a finite-time control synthesis method is presented for a class of nonlinear systems. The proposed method is based on a differential geometric approach, in which a virtual system that is finite-time stable with a strict-feedback form is constructed and is diffeomorphic to the system under consideration (i.e., the real system). The control law is derived via a transformation from the real system to the virtual system. This control law ensures that the original system has the same dynamic characteristics as the virtual system and is finite-time stable. The stability in finite-time control of the virtual system is analyzed based on the Lyapunov function. Theoretical analyses and numerical simulations are conducted on three systems and compared with existing methods to demonstrate the effectiveness of the proposed control law.
As the demand for power wheelchair (PWC) users increases, obstacle avoidance remains a significant challenge-often beyond the capabilities of manual control. A vision-only autonomous obstacle detection and avoidance system for PWCs is introduced, using just a single monocular camera. This low-cost solution integrates deep learning, computer vision, and mobile robotics without relying on expensive depth sensors. By adapting a deep learning model via transfer learning on a newly developed sidewalk dataset, the system effectively detects obstacles. Once detected, a novel image-space avoidance method dynamically adjusts the wheelchair's motor speeds to navigate safely around obstacles. Deployed on a standard PWC, the system demonstrates object detection accuracy comparable to conventional notebook-based methods while operating at 5 Frames Per Second (FPS) on a Raspberry Pi 4, with a total system cost under $300. Real-world tests on a university campus confirm successful detection and avoidance of nine common sidewalk obstacle classes, with the lightweight design using minimal computational resources. This shows that effective obstacle avoidance can be achieved with minimal hardware, avoiding bulky and costly sensor arrays. The proposed system offers a novel, affordable approach for retrofitting any PWC with minimal modifications using off-the-shelf components. This innovation enhances mobility for PWC users both indoors and outdoors, setting the stage for fully autonomous, vision-only PWCs in the future.
Artificial intelligence (AI) is increasingly embedded into optoelectronic (EO) and EO/IR defense systems to enable faster perception, autonomous or human-assisted decision-making, and resilient operation in contested environments. This paper presents an adapted PRISMA-like review and critical synthesis of AI-integrated EO systems for defense applications, with emphasis on system architectures and their performance implications across surveillance, recognition, safety-critical and mission-critical use cases. From an initial pool of approximately 3,700 records screened across major scientific databases, 32 primary studies were selected for in-depth analysis and categorized by sensing modality (EO, IR, hyperspectral, multi-sensor fusion), task type (detection, classification, tracking, anomaly detection), and deployment constraints (edge vs. centralized processing, latency, power, and reliability requirements). Rather than ranking methods by reported accuracy-which is often non-comparable due to heterogeneous datasets and evaluation protocols-this study highlights how architectural choices and operational conditions shape real-world performance, including common failure modes such as weather degradation, occlusion, sensor drift, domain shift, and adversarial interference. The synthesis identifies recurring design patterns (hybrid edge-cloud pipelines, safety-isolated compute domains, and human-autonomy teaming loops) and summarizes research gaps in standardization, certification/verification, robustness, and explainability needed for deployable defense systems.
Field-Programmable Gate Array (FPGA) technology has emerged as a costeffective solution for implementing predictive controllers. Among these, Finite-Control-Set Model Predictive Current Control (FCS-MPCC) has gained recognition as a powerful strategy for permanent magnet synchronous motor (PMSM) drives due to its fast dynamic response and capability to handle system constraints. However, practical implementation requires rigorous verification prior to hardware deployment.This paper presents a co-simulation-based verification framework that integrates Simulink with ModelSim to validate the FCS-MPCC algorithm for PMSM drives at the Register-Transfer Level (RTL). The proposed approach combines MATLAB/Simulink for system-level modeling and testbench generation with ModelSim for cycle-accurate HDL simulation of the predictive controller. The framework enables early-stage functional verification, facilitates the detection of implementation inconsistencies, and supports improved consistency between control design and FPGA realization. Simulation results demonstrate the feasibility and correctness of the proposed approach, highlighting its effectiveness as a verification stage prior to hardware implementation.
This paper addresses a transdisciplinary application area that uses systems thinking to concord disaster vulnerability, sustainability, and resilience. Through a System Dynamics framework, the study connects Sustainable Development Goals (SDGs) with the dynamic interactions among societal, economic, and environmental factors influencing preparedness and recovery. It illustrates how systems modeling can capture interdependencies and resource flows, supporting more effective strategies for sustainable and resilient disaster risk reduction, based on seven critical factors. A Vulnerability Index is introduced to explore the allocation of resources among these interconnected factors. From a practical point of view, this approach allows to connect SDGs targets with the analytical rigor of systems modeling, demonstrating the essential role of System Dynamics in understanding and managing complex, interdependent systems that underpin societal resilience, advancing complex decision support systems.
Trigonometric functions such as sine and cosine are commonly used in signal analysis, but traditional methods for computing these functions can be too resource-intensive for systems with limited computing power. In this paper, algorithms for approximating sine and cosine are investigated that enable efficient signal analysis with lower computational complexity. The proposed algorithms use iterative approaches to achieve a precise approximation of trigonometric functions, which facilitates their real-time application in embedded systems. Instead of a linear iterative approach, the parallel computation of multiple iterations (PCMI) in a single operation is proposed. Time-domain signal analysis, which aims to detect frequency and phase changes, is a crucial component in many embedded system applications, such as vibration measurement, phase shift detection in communication, EEG signal analysis, data preprocessing for machine learning, and many others.
Accurate prediction of Type 2 Diabetes is essential for effective prevention and intervention strategies. This paper proposes a hybrid model that integrates an improved Louvain community detection algorithm with a domain-specific diabetes ontology to enhance prediction performance and interpretability. Three real-world datasets (Pima Indians, Diabetes 130-US Hospitals, and NHANES) were used for evaluation. Patients were grouped into clinically meaningful clusters using the enhanced Louvain method, followed by semantic reasoning over an ontology constructed from expert-defined medical rules and clinical features. Experimental results show that the proposed model outperforms conventional clustering and classification techniques across multiple metrics, achieving up to 97.56% accuracy in cross-validation. The use of ontological knowledge not only increases transparency in prediction outcomes but also supports semantic querying and dynamic model adaptation. This method demonstrates significant potential for real-world deployment in clinical decision support systems.
This paper presents a lightweight visual servoing framework based on Gaussian Process (GP) regression for real-time autonomous navigation of electric wheelchairs in corridor environments. The proposed system leverages Direct Visual Servoing (DVS), where global image features are mapped directly to velocity control signals using a GP model. Histogram of Oriented Gradients (HOG) features are extracted from camera images and used as inputs to the GP, which predicts angular velocity commands without requiring explicit geometric feature extraction or Jacobian computation. The GP model is trained using a dataset collected in varied indoor corridors, with control labels derived from a geometric-based control law. The system was deployed on a Raspberry Pi embedded platform and tested on a commercial wheelchair. Experimental results demonstrate that the proposed controller operates at 7 Hz, ensures robust trajectory correction under visual noise, and achieves over 90% accuracy in corridor-following scenarios, even under challenging lighting and occlusion conditions. This approach combines the adaptability of learning-based control with the simplicity and interpretability of classical visual servoing, offering a scalable and computationally efficient solution for assistive mobility applications.
Industry 4.0 and digital factories leverage key technologies such as virtual reality (VR) to design, simulate, optimise, and interact with physical production systems remotely or collaboratively. Recent advancements in industrial and aerospace applications demonstrate that integrating VR with versatile robotic control techniques can unlock new opportunities. VR offers an "extended arm" for physical environments by providing a user-friendly human-machine interface, enhancing sustainability, testability, and functionality in robotic control systems. This research focuses on a widely used and continuously evolving robotic arm, designed to operate in an integrated system that functions in both real and virtual environments. A 3-degree-of-freedom (3DOF) robotic arm was initially developed and later upgraded to a 5DOF system. The robotic arm is integrated with Virtual Reality (VR) interfaces based on the Robot Operating System (ROS) and Unity3D platforms. In the real-world scenario, inverse kinematics is applied for the robot's movement mechanism, while gradient descent algorithms are used for the virtual model. Forward kinematics using homogeneous transformation matrices has been employed to verify the accuracy of the robotic arm's movements. VR headsets and controllers are used to direct the virtual control of the robotic arm. The goal of this study is to develop integrated systems that enable control of real-world operations via remote observation, addressing challenges such as prototyping costs and safety risks. By offering an innovative approach to robotic arm control through VR integration, this work contributes to the advancement of Industry 4.0 technologies. The findings suggest potential applications across various industries, opening up new possibilities for remote operation and training in complex manufacturing environments.
In the context of complex industrial systems, effective maintenance strategies play a critical role in ensuring equipment availability,minimizing production downtime, and controlling operational costs. Although preventive and corrective maintenance approaches have been extensively studied, the dynamic selection of the most appropriate maintenance strategy remains a challenging issue, particularly under uncertain and disruptive operating conditions. This paper introduces PHADEM (Analytical Hierarchy Process for Maintenance Decision), an interactive multi-criteria decision-Making approach aimed at optimizing maintenance strategies. The proposed method simultaneously considers economic, technical, and environmental criteria, enabling real-time adaptation of maintenance policies to operational contexts. The model is implemented using a simulation framework that incorporates detailed representations of production stations, material flows, resources, and failure modes. The results demonstrate that the proposed hierarchical approach significantly enhances overall system performance while ensuring optimal resource allocation.
In this paper, we present a comparative study of the multi-model approach and the nonlinear approach for predictive control of a Boost converter. The system is first Characterized by a collection of models. The optimal number of models is selected through the Elbow Method, Improved using Particle Swarm Optimization to improve partitioning efficiency. A genetic algorithm is used to refine the resulting models using the adequate validity. In contrast, predictive control combined with the Newton-Raphson method is applied to the Hammerstein model. Both control strategies are implemented on an STM32 microcontroller to enable real-time control of the Boost converter. The efficiency of the developed control strategies is analyzed and compared with respect to tracking precision, robustness, and execution time, demonstrating their suitability for power electronics applications.
The purpose of Video Anomaly Detection is to automatically identify abnormal spatiotemp oral patterns in surveillance systems. Traditional autoenco der frame reconstruction methods have made great progress in abnormal video detection. However, these methods often fail to adequately correlate global features with local emphasis feature and to harness feature information at various scales within the network. With the aim of addressing these issues, this study proposes a novel multi-scale Swin Transformer and memory-augmented attention feature fusion network for video anomaly detection(MSTMAF), which effectively performs multi-scale cross-fusion of global and fine-grained local features of the autoenco der. Specifically, a multi-scale Swin Transformer encoder is designed to achieve multi-layer feature fusion while extracting video frame features through a multi-branch skip connection operation. Furthermore, we design multi-attention fusion mechanism to refine attention, reducing the influence of background location in the feature map. This process heightens the high-level semantic representation of local features. Additionally, we introduce memory augmented module to better retain archetypal features of normal behavior from historical data. MSTMAF-Net achieved outstanding AUC performance on four publicly available datasets: UCSD Ped1(0.881), UCSD Ped2(0.982), CUHK Avenue (0.913), ShanghaiTech (0.764) and UCF-crime(0.831) demonstrating the effectiveness of our study.
In this paper, an interval observer-based fault diagnosis mechanism and an interval dynamic event-triggered robust fault-tolerant control strategy are proposed for uncertain nonlinear switched systems. Firstly, the interval observer design problem is transformed into a Sylvester equation solution problem by introducing the coordinate transformation method in the absence of actuator faults, and the corresponding interval observer design method is given. Secondly, based on the interval observer output, upper and lower bound estimates of the system outputs are constructed, and fault diagnosis is realized by monitoring whether the system output exceeds the preset bounds. In addition, an improved interval dynamic event triggered mechanism is proposed to optimize the utilization of communication and computational resources and reduce redundant information transmission. The mechanism incorporates interval estimation information and real-time residuals, which can effectively extend the triggering time interval while improving the fault detection performance. Based on the interval fault diagnosis results, a fault-tolerant control strategy based on the interval dynamic event triggered mechanism is proposed, and the conditions for asymptotic stabilization of the closed-loop system are given. Finally, a nonlinear model of a morphing aircraft system with variable wing curvature is used to verify the validity of the designed scheme.
Copolymerization reactors are crucial for polymer synthesis but remain difficult to regulate due to nonlinear kinetics, heat sensitivity, and disturbances such as inhibitor impurities. Traditional approaches like Proportional-Integral-Derivative (PID) controllers and linear Model Predictive Control (MPC) sometimes struggle with computational efficiency and accuracy trade-offs. This study presents a hybrid framework incorporating MPC, Neural Network (NN)-based gradient approximation, and adaptive Particle Swarm Optimization (PSO). The NN accelerates gradient computation to bypass iterative quadratic programming, while PSO enhances global search capabilities under actuator constraints. Validated on a nonlinear copolymerization reactor model, the hybrid method displays enhanced setpoint tracking and disturbance rejection compared to standard MPC and standalone PSO, with maintained computational feasibility. Robustness is emphasized by effective management of unmeasured inhibitor perturbations, achieving stable input adjustments and precise output regulation.
Efficient vehicle dynamics management is critical for high-performance driving and intelligent transportation systems. This paper focuses specifically on an informatics-driven, smartphone-based recommendation system that leverages sensor data and predictive modeling techniques to inform rider decisions without implementing explicit closed-loop control mechanisms. Using a discrete waypoint representation, the model derives road curvature and friction-limited corner speeds while considering acceleration and braking constraints. A forward-backward algorithm computes a feasible speed profile, ensuring adherence to physical and regulatory limits. Furthermore, we extend the model to multi-vehicle coordination, demonstrating its applicability to platooning scenarios. By synchronizing vehicle speeds and braking points, we can optimize traffic throughput in road bottlenecks, mitigating shockwaves and improving flow efficiency. This approach bridges the gap between performance driving techniques and intelligent transportation systems, providing a foundation for real-time speed advisory and autonomous driving applications.
