Internet Technology Letters最新文献

Reliable crowd counting for IoT video analytics requires strong generalization across heterogeneous edge cameras. However, models trained on a labeled source domain often degrade on unseen cameras due to shifts in appearance, viewpoint, and density statistics. We propose MDANet, a deployment-oriented framework for cross-domain crowd counting that performs complementary alignment at three levels while keeping test-time inference identical to a lightweight backbone. At the data level, Fourier Amplitude Mix reduces camera-dependent style gaps by mixing low-frequency amplitudes. At the feature level, global–local High-Entropy Adversarial Regularization suppresses domain-discriminative cues under spatial heterogeneity. At the domain level, Density-Conditional Alignment modulates alignment strength according to predicted density to mitigate congestion-dependent errors. Extensive experiments show that MDANet achieves competitive or state-of-the-art accuracy with a favorable accuracy-efficiency trade-off, and additional evaluations under common stream degradations confirm its stability for edge deployment.

With the evolution of 6G wireless networks, wireless sensor networks are facing new challenges, particularly when it comes to energy efficiency and reliable data transmission. This paper proposes an energy-aware, intelligent routing framework that uses deep reinforcement learning to extend the lifetime of networks and increase data throughput in 6G networks. Through the implementation of a deep recurrent Q-learning mechanism, the framework enables dynamic routing with residual energy and node proximity as criteria for selecting the next hop, either in single-hop or multi-hop scenarios. As demonstrated by experimental results, the proposed model delivers higher packets, consumes less energy, and has a lower latency while achieving greater throughput than conventional PSO or clustering-based methods. WSNs of the future can take advantage of its robust routing capabilities.

Recent grammar error correction (GEC) systems have scaled rapidly in model size and architectural depth, creating a growing mismatch between algorithmic improvements and the latency and energy constraints of edge devices. The method reformulates English GEC as a task-constrained latent editing problem, where grammatical corrections are represented as low-rank perturbations in a compact linear subspace. A Tiny-LM-style weight re-parameterization aligns the latent editing vectors with a minimal set of re-parameterized weights, ensuring that English grammatical reasoning is concentrated in a hardware-friendly linear manifold. To improve correction fidelity under tight computational budgets, a two-stage progressive refinement strategy is employed: a fixed-window lookahead performs coarse structural edits, followed by a sparse consistency filter that selectively verifies candidate token corrections under INT8/INT4 quantization. The entire pipeline is static-shape and operator-regular, relying solely on linear, NPU-native operations for predictable latency and bounded memory footprint. Experiments on public datasets show that the proposed model outperforms large Transformer baselines in F0.5 score on typical edge NPUs while reducing latency by 3–7×, demonstrating that accurate, low-latency, on-device English GEC is achievable using generic NPU operators without heavyweight language models.

Recent grammar error correction (GEC) systems have scaled rapidly in model size and architectural depth, creating a growing mismatch between algorithmic improvements and the latency and energy constraints of edge devices. The method reformulates English GEC as a task-constrained latent editing problem, where grammatical corrections are represented as low-rank perturbations in a compact linear subspace. A Tiny-LM-style weight re-parameterization aligns the latent editing vectors with a minimal set of re-parameterized weights, ensuring that English grammatical reasoning is concentrated in a hardware-friendly linear manifold. To improve correction fidelity under tight computational budgets, a two-stage progressive refinement strategy is employed: a fixed-window lookahead performs coarse structural edits, followed by a sparse consistency filter that selectively verifies candidate token corrections under INT8/INT4 quantization. The entire pipeline is static-shape and operator-regular, relying solely on linear, NPU-native operations for predictable latency and bounded memory footprint. Experiments on public datasets show that the proposed model outperforms large Transformer baselines in F0.5 score on typical edge NPUs while reducing latency by 3–7×, demonstrating that accurate, low-latency, on-device English GEC is achievable using generic NPU operators without heavyweight language models.

This paper introduces a novel federated learning framework that integrates dynamic blockchain sharding and Byzantine fault tolerance mechanisms (FL-Sharding-BFT). The proposed framework dynamically adjusts shard configurations based on node capacity and network conditions, reducing communication overhead and enhancing model synchronization. The Byzantine fault tolerance mechanism further ensures robustness by identifying and isolating malicious nodes during model aggregation. It also ensures stronger robustness and faster convergence, highlighting its scalability and security for federated learning in 5G edge networks.

This paper introduces a novel federated learning framework that integrates dynamic blockchain sharding and Byzantine fault tolerance mechanisms (FL-Sharding-BFT). The proposed framework dynamically adjusts shard configurations based on node capacity and network conditions, reducing communication overhead and enhancing model synchronization. The Byzantine fault tolerance mechanism further ensures robustness by identifying and isolating malicious nodes during model aggregation. It also ensures stronger robustness and faster convergence, highlighting its scalability and security for federated learning in 5G edge networks.

In this work, a quad modal mode division multiplexing system utilizing linearly polarized (LP) modes, namely, LP[0,1], LP[2,2], LP[0,3], and LP[1,3], is designed and analyzed under the impact of free space optics (FSO) and multimode fiber integrated link impairments. Results depict that reliable transmission over 100 m fiber and 100 m FSO links can be achieved at 160 Gbps data rate considering 1–7.5 dB insertion loss, free-space weather, and turbulence. Also, the spatial shift of 12 μm and spatial tilt of 10° can be supported with low modal crosstalk. Compared to existing works, this work provides long-reach and high-speed communication for 5G based networks.

In this paper, propose an adaptive learning-based MEC framework for real-time roadside service provisioning in vehicular networks. This novel decision-supporting framework is used to optimize the task offloading, resource allocation and service scheduling in dynamic vehicular environment combined the machine learning technology. It is under the condition that it cooperates with RSUs to compute and send tasks. Referring to decomposition the offloading and resource management problem into sub-problems, addressing them via the deep reinforcement learning, this mixed-integer nonlinear optimization problem is established. The results in simulations demonstrate that the proposed model offers significantly higher success rates than baselines under various traffic loads, processing rates, number of MEC servers and computation resource requirements, which confirms its robustness and potential applications for ITS.

Industry 5.0 is catalyzing a paradigm shift toward human-centric intelligent manufacturing, where the operational, cognitive, and social attributes of individuals critically influence production quality and resilience. Traditional quality control methods, however, fall short in dynamically identifying and mitigating human-factor-induced risks within complex human-physical systems. To address this gap, this study proposes a proactive quality risk identification and control framework for ship propulsion system assembly, leveraging real-time Industrial Internet of Things (IIoT) data. The framework begins by establishing a comprehensive human factors risk point system, systematically mapping human roles such as operator, decision-maker, and social being to 24 specific process risk points. By integrating Reverse Failure Mode and Effects Analysis (FMEA) and Fault Tree Analysis (FTA), real-time fault logs are decomposed into meta-faults. Weighted Association Rule Mining (WARM) is then employed to construct a dual-layer correlation network between meta-faults and risk points, revealing hierarchical causal relationships and enabling precise, dynamic risk prioritization. Based on this, a human-machine collaborative risk management strategy is proposed. This research integrates human factors into digital quality management, providing an extensible methodology for proactive quality management aligned with Industry 5.0's human-centered objectives.

Aiming at the high-dimensional dynamic path optimization problem of BeiDou-ASEAN cross-border logistics, this paper proposes an industrial Internet of Things optimization method based on improved particle swarm algorithm. By introducing the Close distance function to reduce high-dimensional extreme dimensional interference, a dynamic inertial weight adjustment strategy and an adaptive variation operator are designed, and the path planning is optimized by combining BeiDou real-time data. Experiments show that the average relative error of the algorithm in the static road network is only 0.18%, and the dynamic event preoptimization efficiency is more than 85%, which is 41.9% higher than the convergence speed of the traditional algorithm, and the response level of 200 ms is achieved in the 1000-node road network. The study provides a high-precision real-time optimization scheme for cross-border logistics, which significantly improves regional logistics efficiency.