Remote sensing change detection (RSCD) is crucial for ur- ban monitoring, environmental protection, and disaster as- sessment, but small-sample scenarios often lead to overfitting and inaccurate predictions on unseen data. To address this, we propose GSAG-CDGAN, an end-to-end framework integrat- ing Selective Noise Augmentation (SNA) to mitigate overfit- ting, an Attention-Guided Adversarial Network (AGAN) to enhance structural consistency, and a Perceptual Loss Mod- ule (PLM) to preserve semantic consistency. Experiments on CDData-50 show that GSAG-CDGAN improves F1-Score from 0.6954 to 0.8851, with notable gains in Recall and IoU, demonstrating enhanced robustness under small-sample con- ditions. Further evaluation on the WHU-CD dataset yields an F1-Score of 0.9502, confirming strong cross-dataset general- ization and the method’s effectiveness in diverse scenarios.

Existing federated clustering methods typically provide limited interpretability, which hinders users’ ability to understand and trust cluster assignments, particularly for ambiguous samples located near cluster boundaries. Moreover, the decentralized and unsupervised nature of federated learning makes it difficult to construct globally consistent clusters and achieve semantic alignment without sharing raw data. To address these limitations, we propose EFC (Explainable Federated Clustering) , a unified framework that integrates privacy-preserving global clustering with an explicit explainability module. EFC first maps client data into low-dimensional embeddings and then performs anchor-based federated alignment at the server to obtain a shared embedding space, enabling global clustering without exchanging original features. Importantly, EFC clarifies the source of explainability by coupling clustering with post-hoc, sample-level attribution: boundary or uncertain samples are identified in the shared space, and SHAP or LIME is applied to a surrogate decision model to quantify feature contributions to cluster assignments. Experiments on three heterogeneous modalities—structured (Wine), visual (CIFAR-10), and textual (AG News)—demonstrate that EFC improves clustering performance while also yielding more faithful and informative explanations relative to recent baselines. These results indicate that EFC provides a practical approach to interpretable unsupervised federated learning, supporting transparent and trustworthy deployment in privacy-sensitive settings.

In the Internet of Things (IoT) domain, end-to-end (E2E) planning tasks require distributed devices to collaboratively train deep models under highly dynamic environments. However, existing federated learning (FL) methods often assume homogeneous communication conditions and static node reliability, leading to suboptimal aggregation performance when confronted with heterogeneous uncertainty sources such as sensing noise, prediction bias, and communication instability. To address this challenge, we propose FedUAP (Federated Uncertainty-Aware End-to-End Planning), a novel framework that dynamically adjusts client contributions based on multi-source uncertainty and network topology information. Specifically, each IoV vehicle node within the broader IoT system estimates three uncertainty factors—prediction uncertainty, sensing uncertainty, and communication uncertainty—to represent its model reliability and transmission stability. A topology-aware weighting module further refines the aggregation by incorporating node connectivity and link quality. In addition, a temporal smoothing strategy is introduced to stabilize weight evolution over successive communication rounds. Extensive experiments on various E2E IoV-centric IoT planning scenarios demonstrate that FedUAP achieves superior convergence stability, communication efficiency, and planning accuracy compared with existing adaptive aggregation and uncertainty-based FL baselines. The proposed approach provides a promising direction toward uncertainty-robust and topology-adaptive federated optimization in large-scale IoT and IoV networks.

Large-scale deep learning models rely on wireless networks for distributed training approaches, which are essential to meet the immense computational and data demands. However, the stochastic nature of wireless environments introduces significant challenges such as variable delays, noise interference, and packet loss, which lead to degraded gradient synchronization and hinder model convergence. In this work, we propose a novel communication-aware distributed training (CADT) framework that integrates real-time channel state information (CSI) feedback into the gradient aggregation process. Unlike conventional methods that assume static or ideal communication conditions, CADT dynamically reweights gradients from each node based on instantaneous channel quality, enabling robust aggregation under adverse wireless conditions. By dynamically adjusting the contribution of each node based on instantaneous channel conditions, CADT effectively compensates for wireless impairments, thereby ensuring more reliable gradient aggregation and significantly improving both convergence speed and final model accuracy. Extensive experiments on CIFAR-10, CIFAR-100, ImageNet, and SVHN using Vision Transformer and ResNet-50 demonstrate that CADT outperforms baseline methods in terms of convergence, accuracy, and communication efficiency. In addition, we provide a rigorous theoretical analysis that establishes convergence guarantees under realistic wireless conditions, thereby advancing the theoretical foundation of distributed optimization in non-ideal communication environments.Our framework offers a practical solution for real-world scenarios such as edge computing, where communication constraints and environmental variability are dominant factors.

As the core of modern industrial automation, Industrial Internet of Things (IIoT) can achieve data collection, control, and remote operations. However, the complex continuous connectivity exposes IIoT to increasing cyber intrusions. Existing intrusion detection methods often fail to adapt to diverse temporal attack patterns, lack effective feature modeling, and perform poorly under class imbalance. To this end, this article proposes a novel transformer-based IIoT intrusion detection method. First, an adaptive multi-scale window segmentation (AMSWS) strategy is proposed to dynamically segment traffic sequences via traffic variation. This can capture short-term and long-term intrusion behaviors. Second, an attention interaction transformer (AITrans) is proposed to enhance feature modeling by enabling inter-head and inter-layer attention interaction. Finally, an adaptive focal loss (AFL) is proposed to mitigate class imbalance by incorporating adaptive class weight and hardness-aware adjustment. This can improve detection performance on minority-class intrusions. Extensive experiments validate the effectiveness of the proposed method for IIoT intrusion detection.

The development of Internet-of-Things (IoT) services demands intelligent and adaptive mechanisms for online resource allocation under dynamic and uncertain environments. Intent-Based Networking (IBN) has emerged as a promising paradigm to align system behavior with high-level user intents. However, realizing intent-aware allocation in real time remains challenging due to uncertain resource availability and incomplete future information. This paper presents a modular framework that integrates semantic intent parsing, machine-learned resource prediction, and robust online decision-making. We propose IBN-ONMP, an IBN-based online resource allocation algorithm that leverages machine-learned predictions and adapts safety margins based on feedback to ensure feasibility and performance under uncertainty. We formally define the problem, establish theoretical guarantees including regret and competitive ratio bounds, and validate the approach on real-world and simulated datasets. Experimental results demonstrate that IBN-ONMP achieves high utility and robust performance across varying prediction error levels, which is consistent with theoretical analysis.

To diminish the substantial communication costs incurred by federated learning during the training of the global model and enhance the model update efficiency across both clients and server domains, we have integrated knowledge distillation into the federated learning framework. This integration has led to the development of a novel approach termed ClientsToServerKDFL, which streamlines the distillation process by directly transferring model insights from clients to the server for computational learning without the need for extensive computations across numerous clients. This iterative process ensures model accuracy and curtails communication expenses. Experimental data analysis has validated the efficacy of this algorithm.

In federated learning, frequent parameter transmission between clients and the server results in significant communication overhead, particularly due to redundancy within the parameters. To address this issue, we propose a Complementary Pruning for Device-to-Device Communication (FedCPD) method. This approach effectively reduces the amount of transmitted parameters by applying complementary pruning techniques on both the server and clients. Additionally, we decrease the communication frequency between clients and the server by employing chain updates among clients (i.e., device-to-device communication). We conducted experiments on the MNIST, FMNIST, CIFAR-10, and CIFAR-100 datasets, and the results demonstrate that our method significantly reduces communication costs while improving model accuracy.

Amidst the rapid integration of Medical Internet of Things (MIoT) into health monitoring ecosystems, electrocardiogram (ECG)-based biometric authentication has emerged as a pivotal component in securing smart healthcare architectures, leveraging its inherent biological uniqueness and real-time monitoring capabilities. Current ECG authentication methodologies face three MIoT-specific challenges: 1) Conventional feature extraction struggles with spatial heterogeneity in multi-device 12-lead signals; 2) Single-cycle analysis lacks generalizability across physiological states; 3) Environmental noise degrades edge computing robustness. To address these limitations, this study proposes ST-AuthNet, a lightweight ECG authentication framework that synergistically integrates spatiotemporal attention mechanisms with enhanced residual networks. First, we redesign the ResNet residual block architecture by replacing conventional 1× 1 convolutional downsampling with hybrid 2× 2 average pooling and 1× 1 convolutional operations, effectively mitigating low-amplitude morphological feature loss (e.g., P/T waves) during feature map compression. Next, a multi-head cross-attention mechanism is introduced to dynamically capture inter-lead spatial correlations and intra-PQRST temporal dependencies across ECG waveforms. Finally, an adaptive threshold decision module is developed to optimize model robustness against physiological variability and environmental perturbations through dynamic classification boundary adjustment. Evaluations demonstrate state-of-the-art performance with 99.77% (CYBHI), 88.60% (MIT), 76.33% (MIT2), and 92.44% (HeartID-V) accuracy, significantly outperforming existing methods in cross-scenario biometric verification.

In federated learning , ensuring high accuracy while maintaining fairness across heterogeneous clients presents a significant challenge. Existing approaches often fail to adequately balance these objectives, especially in non-IID environments. To address this issue, we propose an adaptive regularization framework, F , which dynamically adjusts the balance between global accuracy and fairness during training. Our method introduces two fairness metrics—variance and mean absolute deviation (MAD)—to quantify performance disparities among clients. By incorporating these metrics into the loss function, we enable adaptive tuning of the regularization parameter to maintain global performance while minimizing client imbalances. Extensive experiments across diverse datasets and heterogeneous environments demonstrate that our approach significantly improves both accuracy and fairness, outperforming baseline methods such as FedAvg and FairFed. These results highlight the potential of F to achieve a more balanced and robust federated learning system .

To tackle the challenge of single-spectral object re-identification in complex and dynamic lighting scenarios, multi-spectral object re-identification, which integrates visible light and infrared information, is gradually taking the lead. Nevertheless, the significant heterogeneity across spectra causes formidable obstacles for this task. Most existing approaches alleviate inter-spectral disparities by amalgamating representations from different spectra, ignoring the selection of spectrum-specific crucial information. To address this issue, we propose a novel Representation Selective Coupling Network (RSCNet) for multi-spectral object re-identification. Specifically, we design an Attention-Fourier Token Sparsification (AFTS) module to adaptively sparse and join tokens from multi-spectral images in the attention domain and Fourier domain. This not only preserves spectrum-specific crucial information but also reduces inter-spectral gaps by selective coupling of multi-spectral representation. Meanwhile, to further align multi-spectral information and guide the model to learn more discriminative representation, we propose an Information Unification Constraint (IUC) learning strategy. Both feature-level information constraint and distribution-level information constraint are simultaneously deployed in IUC. Finally, we conduct extensive experiments on three multi-spectral object re-identification benchmarks, and the experimental results verify the effectiveness of our proposed method.

Federated learning significantly enhances intelligent transportation systems by enabling collaborative model training across multiple clients, thereby improving overall performance. However, the involvement of multiple participants, each using oversized models for local data processing, leads to substantial communication volumes and reduced communication efficiency. To address this issue, we propose the Federated Client and Global (FEDCG) pruning method, employing a two-stage pruning strategy at both the client and server levels. This approach uses mutual information to assess the importance of individual neurons or filters within a neural network, allowing for global pruning on the server. Specifically, federated learning connects multiple sub-models within intelligent transportation systems, such as autonomous vehicles and vehicle detection models, by reducing redundant parameters through pruning. When handling differences between various models, we use a parameter aggregation strategy to ensure the effectiveness of the global model. Our method first performs preliminary pruning at the client side to reduce local communication overhead, followed by further pruning at the server side to aggregate effective parameters from each client into a global model. This global model is constructed based on the pruned client models to ensure efficiency and accuracy. Extensive experiments demonstrate that FEDCG effectively reduces communication overheads during both the uploading and downloading phases while maintaining high accuracy and robustness across various datasets and neural network architectures. This method provides a valuable tool for practical federated learning in intelligent transportation systems.

Data heterogeneity is a prevalent challenge in intelligent transportation systems (ITS), often arising from variations in traffic patterns across different regions or time periods. For instance, certain traffic events, such as congestion, may be more frequent in urban areas during peak hours, while other events, like accidents, might occur more often in suburban regions, leading to slightly skewed label distributions. While federated learning provides an effective solution for distributed data, its performance can degrade when client datasets exhibit such label skew. To address this, we propose a strategy that combines Gaussian mixture clustering with oversampling. Gaussian mixture clustering can handle overlapping data points in model parameters, but insufficient client samples may limit local model training. To overcome this, we introduce a Gaussian mixture-based oversampling method to generate additional samples, enhancing the robustness of federated learning under slightly skewed label scenarios. Our experiments demonstrate that this method outperforms or matches existing approaches, ensuring more reliable and accurate ITS applications.

Face video stylization aims to convert real face videos into specified reference styles. While one-shot methods perform well in single-image stylization, ensuring continuity between frames and retaining the original facial expressions present challenges in video stylization. To address these issues, our approach employs a personalized diffusion model with pixel-level control. We propose Local Consistency Guidance(LCG) strategy, composed of local-cross attention and local style transfer, to ensure temporal consistency. This framework enables the synthesis of high-quality stylized face videos with excellent temporal continuity.

Communication overhead remains a significant challenge in federated learning due to frequent global model updates. Essentially, the update of the global model can be viewed as knowledge transfer. We aim to transfer more knowledge through a compact model while reducing communication overhead. In our study, we introduce a federated learning framework where clients pre-train large models locally and the server initializes a compact model to communicate. This compact model should be light in size but still have enough knowledge to refine the global model effectively. We facilitate the knowledge transfer from local to global models based on pre-training outcomes. Our experiments show that our approach significantly reduce communication overhead without sacrificing accuracy.

Nowadays, sign language is becoming increasingly important in people’s daily life. Existing solutions are often based on wireless signals (e.g., acoustic, visible, and WiFi) or wearable sensors to recognize gestures, but they suffer from vulnerability to environmental influences, poor security, and high energy consumption, which prevent them from accurately capturing finger micromovements. In this article, we propose RF-Sign, which uses passive radio-frequency identification (RFID) tags to capture multiple finger micromovements simultaneously to enable sign language support. In particular, two main issues are studied. One is the problem of positional differences when users make the same gesture, and the other is the problem of segmenting consecutive gestures using only empirical thresholding methods and ignoring the existence of differences in thresholds for different gestures. For position differences, we propose position models to normalize the hand’s horizontal rotation angle and radial distance. For segmenting consecutive gestures, we use the received signal strength (RSS) trend of the reference tag to represent the finger micromovements state. The experimental results show that the average accuracy reaches 92.81% under different angles, distances, and other conditions.

More robust intelligent transportation systems including autonomous driving systems are in full flourish with the revolution of deep learning and the 6G wireless communication network. Vehicle Re-Identification, an indispensable branch of the intelligent transportation system, aims to retrieve specific vehicles captured from non-overlapping cameras. However, this is fundamentally challenging with the substantial inter-class similarity and substantial intra-class divergence. Embedding semantic information into vehicle re-identification task has gained ample interest, but the performance needs to be further improved. This work proposes a semantic-oriented feature coupling transformer (SOFCT) for vehicle re-identification as a solution. Specifically, the knowledge-based transformer is first embedded to model images with discriminative attributes. Second, original patches are divided into five semantic groups via semantics-patches coupling, and the feature extractions for different semantics are performed in the semantic feature extraction (SFE) transformer. Third, patch features are weighted via semantics-patches coupling in the patch feature weighting (PFW) transformer, the weighted feature is fed into subsequent encoders to excavate information. Finally, two groups of learnable semantics are embedded to automatically learn semantic features in the learnable semantic extraction (LSE) transformer. Experiments demonstrate that the proposed SOFCT method surpasses other state-of-the-arts with the mAP/Rank-1 of 80.7%/96.6%, 89.8%/84.5%, 86.4%/80.9%, and 84.3%/78.7% on VeRi776 and VehicleID.

Digital Twins (DTs) can generate digital replicas for mobile networks (MNs) that accurately reflect the state of MN. Machine learning (ML) models trained in DT for MN (DTMN) virtual environments can be more robustly implemented in MN. This can avoid the training difficulties and runtime errors caused by MN instability and multiple failures. However, when using data from various devices in the MN system, DTs must prioritize data privacy. Federated learning (FL) enables the construction of models without data leaving devices to protect DTMN data privacy. Nevertheless, FL’s privacy protection needs further improvement for it only guarantees device-level data ownership but ignores that models may retain private information from data. Therefore, this paper focuses on data forgetting in privacy protection, and proposes a novel FL-based unlearning framework (FedME2), which contains MEval and MErase modules. Guided by memory evaluation information from MEval and employing MErase’s multi-loss training approach, FedME2 gets accurate data forgetting in DTMN. In four DTMN virtual environments, FedME2 achieves an average data forgetting rate of approximately 75% for global models under FL and kept the influence on global models’ accuracy below 4%. FedME2 has better data forgetting and improves DTMN data privacy protection while guaranteeing model accuracy.

With the rapid development of information technology, image and text data have increased dramatically. Image and text matching techniques enable computers to understand information from both visual and text modalities and match them based on semantic content. Existing methods focus on visual and textual object co-occurrence statistics and learning coarse-level associations. However, the lack of intramodal semantic inference leads to the failure of fine-level association between modalities. Scene graphs can capture the interactions between visual and textual objects and model intramodal semantic associations, which are crucial for the understanding of scenes contained in images and text. In this article, we propose a novel scene graph semantic inference network (SGSIN) for image and text matching that effectively learns fine-level semantic information in vision and text to facilitate bridging cross-modal discrepancies. Specifically, we design two matching modules and construct scene graphs within each matching module for aggregating neighborhood information to refine the semantic representation of each object and achieve fine-level alignment of visual and textual modalities. We perform extended experiments in Flickr30K and MSCOCO and achieve state-of-the-art results, which validate the advantages of our proposed approach.

Future worldwide 6G research will drive the evolution of emerging intelligent control technologies, such as intelligent speed advisory systems (ISA), to a more advanced generation. As a special type of ISA, consensus-based speed advisory systems (CSAS) can be widely used to recommend a consensus speed for a vehicle platoon, enabling minimizing energy consumption or emissions over a planned route. Recently, speed recommendation services that protect data privacy (i.e., how to obtain an optimal speed in a privacy-preserving way) have drawn tremendous attention. However, current approaches could still encounter service trust issues with central servers and the malicious behavior of vehicles. Furthermore, existing research lacks considering road safety constraints (i.e., safe distance between adjacent vehicles and road speed limits) that are essential for the practical deployment of CSAS. To address the above issues, this paper proposes Eco-CSAS, a safe and eco-friendly consensus speed advisory system using blockchain. We formulate an optimization problem subject to the minimum following distance and maximum road speed limit to minimize the energy consumption of the automatic vehicle platoon. In addition, we introduce a consortium blockchain and cryptographic primitives to ensure service trust and data privacy. We implement the system on the Hyperledger platform, and experimental results show that the system can achieve speed recommendations in a trustworthy and privacy-preserving manner while ensuring a secure platoon.

Now public transportation, included bus and subway occupies a greater role in city transport, and the layout of traffic stations is the most important part of planning and design. However, some unpredicted factors for construction of traffic stations results in a low utilization rate of public transportation resources, for example, the layout of bus stops is chaotic, there is no clear layout scope, and there is a lack of integration with residents’ travel hotspots. Towards these challenges modern transport faces, we firstly analyze the distribution of bus stops and subway stations to determine the area range needs to be optimized in the traffic net from the perspective of time and space. And then, we propose an optimization method, called ’partial area clustering’ (PAC), to improve the utilization by changing and renewing the original distribution. The novel method was based on the K-means algorithm in the field of machine learning. PAC worked to search the suitable bus platforms as the center and modified the original one to the subway. Experiment has shown that the use of public transport resources has increased by 20%. The study uses a similar cluster algorithm to solve transport networks’ problems in a novel but practical term. As a result, the PAC is expected to be used extensively in the transportation system construction process.

Vehicle to Grid (V2G) refers to the optimal management of the charging and discharging behavior of electric vehicles through reasonable strategies and advanced communication. In the process of interaction, there are three stakeholders: the power grid, operators (charging stations), and EV users. In real life, the impact of peak‐valley difference caused a lot of power loss when charging. At the same time, the loss of current is also a loss for power grid companies and EV users. In this paper, we propose a multiobjective optimization method to reduce the current loss and determine the relationship between the parameters and the objective function and constraints. This optimization method uses a genetic algorithm for multiobjective optimization. Through the analysis of the number of vehicles and load curve of AC class I and AC class II electric vehicles before and after optimization in each period, we found that the charging load of electric vehicles played a role of valley filling in the low valley price stage and played a peak‐cutting role in a peak price period.