ABSTRACT This paper proposes a 6G‐driven federated edge intelligence framework for multi‐center stroke lesion segmentation. Lightweight MobileStroke‐U‐Net is deployed at the edge of each participating hospital for distributed MRI segmentation, enabling federated learning without data sharing. Gradient adaptive weighted aggregation (GradAdapt) is used at the Center Server layer to alleviate heterogeneous distribution offset across multiple centers. The global model is trained in the cloud and combined with homomorphic encryption and blockchain auditing to achieve end‐to‐end privacy protection. Experiments are conducted on two major multi‐center datasets, ATLASv2.0 and ISLES22: Compared with the best single model, FL‐MobileStroke‐U‐Net improves Dice by 1%–2% on both datasets. FL‐MobileStroke‐U‐Net improves Dice from 0.6458 to 0.6611 on ATLAS v2.0, and the 95% Hausdorff distance changes from 22.9573 to 21.2032; on ISLES22, Dice increases from 0.7527 to 0.7632, and the 95% Hausdorff distance changes from 11.8847 to 11.0762. The results show that the proposed multi‐center federated framework effectively balances privacy, security, and efficiency, significantly improves cross‐center segmentation performance, and provides a new approach for 6G‐enabled intelligent stroke management.

ABSTRACT Fuzzy clustering algorithms have been widely used for complex data patterns and image segmentation using fuzzy partitioning. However, their performance degrades in the presence of noise and higher‐dimensional spaces. To overcome these limitations, we propose a hybrid algorithm, DBKIFCM–PSO, which combines distance‐based kernelized intuitionistic fuzzy C means clustering with particle swarm optimization. This hybrid approach addresses challenges in noisy datasets and higher‐dimensional spaces, providing optimal solutions. We compare DBKIFCM–PSO with existing algorithms on various datasets and demonstrate its superior performance. By integrating DBKIFCM and PSO, our algorithm has shown improved performance on evaluation metrics like , , and over counterpart method by obtaining best values of 0.9595, 0.1036, and −61 120, respectively, on these metrics making it suitable for real‐world applications.

ABSTRACT Recently, wireless communication networks have played a vital role in environmental monitoring and other data‐driven applications. Even though these networks often struggle with limited energy and redundant data transmissions. Moreover, traditional routing protocols, such as the Cross‐layer Opportunistic Routing Protocol (CORP), rely heavily on static routing decisions with fixed‐cost functions, leading to a lack of adaptability. To address these issues, this study proposes a Mistral 7B‐based Cross‐layer Optimization (M7BCO), which integrates adaptive reasoning and prompt‐based telemetry compression for energy‐aware decisions. The proposed M7BCO model utilizes a Partially Informed Sparse Autoencoder (PISA) to select a minimal subset of informative nodes by learning spatial correlations while preserving data reconstructability. Then, the proposed M7BCO model generates a real‐time decision for next‐hop selection and transmits power adjustment as it replaces the static optimization with adaptive reasoning. Unlike pure sequential models, the proposed model introduced a lightweight training loop between PISA telemetry selection and Mistral 7B adaptive reasoning. From the results, the proposed M7BCO model achieved better results when compared to the existing CORP model in terms of Energy Efficiency (EE) of 22.5, 65.3, and 100.2 mJ for 150, 300, and 500 nodes respectively.

ABSTRACTReal‐time multi‐object detection on smartphones requires a careful balance of accuracy, latency, energy efficiency, and data privacy. We introduce FedEdgeDetect, a unified framework that combines federated learning with edge‐assisted inference to address these challenges holistically. The system incorporates a hardware‐aware YOLOv5s variant with lightweight attention modules for efficient on‐device execution. A capability‐clustered federated training protocol is designed to ensure privacy through differential noise injection and secure aggregation, while reducing communication overhead. At inference time, a dynamic controller adaptively partitions computation between the device and edge, optimizing for real‐time performance and energy consumption. Experiments across diverse datasets and devices demonstrate that FedEdgeDetect consistently improves detection accuracy, accelerates inference, enhances energy efficiency, and enforces strong privacy guarantees, outperforming existing mobile detection baselines.

ABSTRACT The internet of healthcare things (IoHT) has advanced considerably, improving healthcare operations and patient monitoring by continuously collecting data from health sensors attached to patients. Current congestion detection techniques are insufficient for early detection since senders often remain unaware of the size of the residual queue. The real‐time transmission of critical health data is essential, yet frequent congestion at intermediate nodes can lead to increased packet loss, delays, and diminished system reliability. To tackle these challenges, we propose a robust and low‐complexity QACA algorithm tailored specifically for patient‐centric IoHT networks, which dynamically adjusts the frequency of acknowledgments based on real‐time queue occupancy thresholds. By integrating interval‐based acknowledgments with a priority‐based queuing strategy, QACA ensures that high‐priority medical data is transmitted promptly, even in the face of heavy network loads. Simulation results indicate that QACA significantly improves performance over the analytical model and DCCA regarding packet loss and packet delay reduction. Moreover, the current framework may be enhanced in future work with the use of LMs to add predictive estimation of queue status, traffic classification, and an intelligent transmission scheduling, thus paving the way toward scalable and intelligent congestion management in next‐generation IoHT systems.

ABSTRACT Industrial Internet of Things (IIoT) systems generate massive behavioral data, demanding efficient analytics frameworks for real‐time monitoring. This study proposes a multi‐dimensional feature fusion framework integrating XGBoost, tailored for IIoT‐driven behavioral pattern recognition. A four‐dimensional architecture is constructed to analyze critical attributes across contact degree, status, duration, and social relations, leveraging edge‐computed IIoT footprints (e.g., mobile signaling, network interaction data). The framework defines three behavioral modes and achieves 98.89% precision, 98.85% recall, and 98.85% F1‐score via XGBoost. Feature importance analysis identifies key indicators such as mobile number status and interaction frequency. This work demonstrates the potential of harmonizing AI with IIoT data fusion, providing a scalable solution for real‐time monitoring in Industrial Internet and future network architectures.

ABSTRACT In this article, we introduce the reconfigurable intelligent surfaces (RISs) restrict their general adoption, which employs digital pre‐distortion, deep learning‐based correction, and adaptive filtering to counteract real‐time RF impairments. The technique is highly applicable to future 6G networks because it enhances MIMO performance by reducing BER, improving phase noise resilience, and achieving the highest spectral efficiency. Thermal noise, phase noise, and nonlinearity loss are RF impairments that significantly reduce the effectiveness of MIMO communication in 6G networks. Signal distortion, phase instability, and spectrum inefficiencies are the consequences of these impairments, which further increase BER and reduce capacity. A dynamic distortion mitigation framework is required because conventional compensating strategies cannot respond to new scenarios in real time. These approaches come with extra latency and power usage, making them less suitable for real‐time use in 6G. However, there remains a challenge to using ML‐based adaptive filtering on high‐speed and low‐power hardware, even though it has been at the forefront regarding dynamically compensating RF impairments. The cost and complexity of deployment of hybrid beamforming and reconfigurable intelligent surfaces (RISs) restrict their general adoption, yet they enhance MIMO performance in RF impairment. The basic challenge for smooth operation in 6G‐enabled MIMO systems is to develop adaptive, low‐power, and computationally efficient solutions.

ABSTRACT With the rapid development of the Internet of Things (IoT), the demand for personalized fashion design in the IoT environment is growing, and fashion recommendation has gradually become a new research hotspot. However, existing fashion recommendation methods are often designed based on a single modality and contain a large number of parameters, making them unable to be effectively deployed on IoT edge devices with limited computing ability. Inspired by this, this paper proposes a novel personalized fashion color recommendation (FashionCR) framework based on a lightweight large vision‐language model for fashion design in the IoT environment. Specifically, this framework consists of an IoT‐based fashion color recommendation system and the FashionCR model. The recommendation system mainly introduces how to train the FashionCR model and deploy it to the edge devices. The FashionCR model leverages the visual branch of the CLIP model to accurately learn the physiological features of different individuals, such as skin color and face shape, and utilizes the text branch to efficiently process the text intentions input by users. Meanwhile, in order to meet the limited resources of the IoT environment, a lightweight modification has been implemented to the CLIP model. In addition, the 4‐season color theory is integrated into the FashionCR framework to achieve accurate color recommendation. Experimental results show that this framework performs excellently in various metrics, providing a new solution for the field of fashion design in the IoT environment and effectively improving the accuracy and personalization of color recommendation.

ABSTRACTTo address the challenges of low accuracy in volleyball individual action recognition caused by complex scenarios in volleyball sports, Faster‐ActionNet was proposed based on the backbone of YOLOv11. In this network, partial convolutions are adopted in both the backbone and neck modules to amplify critical feature representations while minimizing redundant computational and memory overhead. In the backbone network, the Feature Refinement and Fusion Network (FRFN) attention mechanism is integrated, which employs optimized and streamlined operations to reduce feature redundancy across channels. This enhancement significantly boosts the reconstruction quality of latent sharp images and alleviates the risk of critical feature degradation. Experiments evaluating the individual action recognition model on volleyball‐specific tasks have revealed superior performance, with the model of mAP attaining 88.2% accuracy and 75.6 frames per second (FPS) in individual action recognition. These results have surpassed state‐of‐the‐art benchmarks. This model demonstrates outstanding performance in real‐world applications, providing valuable technical insights for improving sports action recognition and advancing computer vision technologies.

ABSTRACT Improved dependability, reduced latency, and great spectrum efficiency, especially in MIMO systems, are essential for 6G networks to be achieved in wireless communication. However, RF impairments, such as thermal noise, phase noise, and nonlinearity losses, severely hinder multiple‐input multiple‐output (MIMO) communication systems. These deficiencies restrict performance in high‐frequency 6G settings due to signal distortion, phase instability, and a poor signal‐to‐noise ratio (SNR). An Adaptive RF Compensation and Distortion Mitigation Framework (ARC‐DMF) is proposed here to resolve these issues. Dynamic compensation for the distortions inflicted by radio frequency interference is accomplished through the framework's adaptive filtering and machine learning‐based correction system, ensuring good signal transmission and enhanced data integrity. The ARC‐DMF approach employs digital pre‐distortion (DPD) techniques, deep learning‐based compensation modeling, and real‐time distortion estimation to minimize the effects of nonlinear distortion effects, phase noise variations, and thermal noise fluctuations. Bit error rate (BER), capacity enhancement, error vector magnitude (EVM), and spectral efficiency are some of the important performance metrics tested in several 6G operational scenarios through substantial simulations. The simulation findings show that ARC‐DMF substantially improves the performance of MIMO communication over traditional compensation methods, with reduced BER, better resilience to phase noise, and more resilient transmission. Enabling efficient and reliable wireless communication in high‐frequency RF situations, this study's findings offer insights into RF impairment mitigation for next‐generation 6G MIMO systems.