Federated learning has emerged as a promising approach for privacy-preserving model training across decentralized UAV swarm systems. However, challenges such as data heterogeneity, communication constraints, and limited computational resources significantly hinder convergence efficiency in real-world scenarios. This work introduces a communication-aware federated learning framework that integrates multi-dimensional cost modeling with dynamic client aggregation. The proposed cost function jointly considers communication overhead, computation latency, and training contribution. A Shapley-inspired client evaluation mechanism is incorporated to guide aggregation by prioritizing high-impact participants. In addition, a two-phase training strategy is devised to balance learning accuracy and resource efficiency across different training stages. Experimental results on the MNIST and CIFAR-10 benchmark datasets under non-IID settings demonstrate that the proposed method achieves faster convergence, higher accuracy, and reduced communication-computation cost. These results highlight its suitability for deployment in bandwidth-constrained, resource-limited UAV edge environments.

Mobile data collectors (MDCs) play a very important role in Internet of Things (IoT) sensing networks. However, ensuring their trustworthiness against insider threats, such as on–off attacks and spatiotemporal fabrication, remains a critical challenge. Existing trust evaluation methods frequently struggle with these threats due to insufficient evidence dimensions and the inability to quantify behavioral stability. To address these limitations, this paper proposes an enhanced proactive trust evaluation system based on stability sequence extraction (E-PTES-S). E-PTES-S improves the evaluation accuracy by integrating five factors of evidence, stability-computation mechanisms, and an adaptive weight allocation scheme to maintain robustness even when proactive verification data is scarce. In addition to the usual interaction and proactive verification indicators, regional consistency (${\mathcal{T}}_{RC}$) and task timeliness (${\mathcal{T}}_{TT}$) are introduced to mitigate location falsification and transmit-time deviations more rigorously. Then, a sliding window technique is used to obtain an integrated evidence sequence, which includes a new continuous stability sequence (${\mathcal{F}}_{CSS}$) and traditional credible, untrustworthy, and uncertain sequences. This continuous stability sequence adds a variance-based incentive scheme to measure behavioral stability. Finally, the normalized trust value is derived from multiple indicators including multidimensional spatiotemporal evidence and stability metrics. Experimental results show that the proposed E-PTES-S achieves a normal node detection rate of 98.7% under complex dynamic conditions, outperforming the baseline PTES and Trust-SIoT algorithms by approximately 9% and 1%, respectively, while also improving the cumulative data collection profit by 4.8%. Furthermore, robustness analysis demonstrates that E-PTES-S exhibits excellent robustness against physical-layer uncertainties, successfully sustaining an 84.4% detection rate even under severe environmental shadowing.

The proliferation of sophisticated malware poses a persistent threat to cybersecurity. While visualizing malware as images enables the use of Convolutional Neural Networks, standard architectures are often inefficient and struggle with the high spatial and channel redundancy inherent in these representations. To address this challenge, we propose LR-MalConv, a new detection framework centered on a novel Low-Redundancy Convolution (LR-Conv) module. The LR-Conv module is uniquely designed to synergistically reduce both spatial redundancy, via a gating and reconstruction mechanism, and channel redundancy, through an efficient split–transform–fuse strategy. By integrating LR-Conv into a ResNet backbone, our framework enhances discriminative feature extraction while significantly reducing computational overhead. Extensive experiments on the Malimg benchmark dataset show our method achieves an accuracy of 99.52%, outperforming existing methods. LR-MalConv establishes a new benchmark for visualized malware detection by striking a superior balance between accuracy and computational efficiency, demonstrating the significant potential of redundancy reduction in this domain.

Due to the subjective nature of sensitive area radius, the effectiveness of current privacy protection strategies is often quite poor. To address this issue, a sensitive area radius design method of smart terminal is proposed by considering the total privacy budget and tolerance degree of adversary inference capabilities. With the increase in the tolerance degree of adversary’s inference capabilities, the sensitive area radius decreases. Based on the division results of the sensitive area, the privacy budgets of smart terminals are allocated separately. Then, a location data perturbation strategy is designed based on the allocated privacy budget and location encoding. Compared with the existing privacy protection strategies, the proposed method is more reasonable, which can improve the effectiveness of location privacy protection strategy. Finally, experimental results based on the real-world dataset show that the proposed algorithm can enhance data utility by 40%.