Internet Technology Letters最新文献

In the English teaching environment driven by the Internet of Things, real-time assessment of students' emotional states based on video data is particularly crucial for understanding student engagement and improving teaching quality. Existing server-based deep networks are limited by long-distance video data transmission, which seriously restricts the real-time performance of sentiment analysis. Moreover, simple video data cannot guarantee robustness in complex scenes. To address these issues, this paper proposes a multimodal fusion emotion recognition framework based on the edge-cloud collaboration mechanism. Firstly, on the edge node, we exploit two complementary modalities of data: video sequences and facial landmark sequences, and design a lightweight dual-stream neural network based on the 3D MobileNetV3 and graph convolutional network to efficiently extract multimodal features. On the server, we adopt the Transformer-based cross fusion mechanism to implement multimodal fusion and emotion evaluation. The edge side is responsible for real-time preprocessing and primary feature extraction. In our proposed framework, the server is responsible for aggregating feature data from multiple edge nodes. The experimental results indicate that the proposed framework can achieve high-precision student engagement assessment with low latency.

Network transmission plays an important role in information system services of IoT, and network transmission quality is also an important indicator of the quality of information export in China. To measure the development of Internet quality in China's foreign trade, this paper calculates and analyzes the time series data of Internet export quality based on open source. This paper first analyzes the necessity and significance of the research. After comparing several types of time series methods, this paper chooses a statistics-based method to describe the Internet quality data, compares the statistical data before and after the epidemic, and draws a conclusion that the epidemic is irrelevant to the Internet quality. Further, this paper combined the cyclic neural network LSTM and principal component analysis methods to carry out calculations. LSTM method completed the analysis and prediction of the quality data of the Internet export, selected packet loss rate, RTT, and delay jitter as the indicators of Internet quality, and completed the preprocessing of the indicator data. To obtain comprehensive network quality indicators, this paper constructs a single index network quality evaluation model based on principal component analysis and completes the analysis and calculation of data analysis. The computation results demonstrate the great prediction accuracy of the LSTM referenced in this research for index data, and the only index result for measuring the quality of Internet exports can be obtained by using principal component analysis. In practical applications, the research in this paper can provide support for the subsequent network performance optimization and bandwidth utilization improvement of the IOT industry. Github link and data source of the paper is https://github.com/superpeace90/iotproject.

In today's world, most individuals are suffering from health complications due to sedentary lifestyles, food habits, and aging. This might have distracted them from their daily activities, and many individuals are not seeking medical help with their busy work schedules. To overcome these challenges, Wireless Body Area Network (WBAN) provides challenging solutions with continuous and remote monitoring systems. Tiny sensors are either embedded in the body or worn by the patients for tracking physiological factors, including heart pulse rate (bilirubin), pressures (alkphos), and levels of hemoglobin (albumin). The inserted or implanted sensors collect the data and transmit it to the central sink node for the aggregation of the readings, and reports are comprehensively generated and transmitted to the medical professionals for analysis. While transmitting the sensitive data, a security breach might occur with the wireless channels. To enhance security, Enhanced Juels and Sudan (EJS) encryption algorithm is deployed to safeguard both the data and sensor systems. This prevents unauthorized access to data. Whale Optimization Algorithm is deployed for improving optimized network performance and security. Simulations are conducted in the Cooja simulator and demonstrate enhanced convergence efficacy, optimal weight computation, and mitigated error rates, including False Rejection Rate (FRR) and False Acceptance Rate (FAR) for a robust health monitoring system.

Vehicular Ad Hoc Networks (VANETs) play a crucial role in intelligent transportation by enabling communication between vehicles and infrastructure. However, ensuring secure, reliable, and consistent data transfer remains challenging due to their dynamic nature. This paper proposes a Clustering-Based Ant Colony Optimization (CB-ACO) and Fuzzy Logic algorithm to address these issues. Clustering reduces network load and enhances stability, while ACO selects optimal cluster heads and routes, supported by Fuzzy Logic-based trust assessments for secure access control. Compared to existing algorithms, the proposed method improves packet delivery, reduces latency, and enhances security, offering a robust solution for next-generation VANETs.

To address overheating and maintenance challenges in wind turbine slip ring systems operating in harsh environments, this study develops an online temperature monitoring system integrating Industrial Internet of Things (IIoT) and edge computing. A deep learning-driven distributed resource allocation algorithm (DDLRA) is deployed on edge devices to achieve real-time slip ring temperature monitoring, intelligent fault prediction, and reduced latency. A multi-layer edge intelligence architecture is constructed, optimizing offloading strategies via distributed neural networks to balance energy efficiency and resource utilization. Experimental simulations using over 1000 temperature datasets from key components demonstrate that the model maintains prediction errors within 0.3°C, effectively identifying anomalies and adapting to variable conditions. The system also enables real-time line load scheduling and carbon brush temperature prediction. This study pioneers the application of online updatable AI models in slip ring monitoring, combining edge responsiveness and cloud collaboration, offering significant engineering value for improving wind turbine reliability.

Cloud-assisted physical education teaching is an important direction for the development of smart education. However, the data processing and transmission of videos severely restrict the real-time performance of action analysis. To this end, this paper proposes an efficient edge cloud-assisted student movement recognition framework based on the graph convolutional network and human skeleton data. On the edge server, we use the YOLO-pose algorithm to generate robust human skeleton sequences and design an improved spatial–temporal dual stream graph convolutional neural network with an early fusion structure, which introduces the node weight module and the dynamic graph module to exploit long-distance dependency relationships of nodes. In the cloud server, we use a federated learning framework based on the density clustering mechanism to collaboratively train and aggregate parameters of models scattered across edge nodes. The experimental results show that our proposed model achieves excellent recognition accuracy on the self-built sports action dataset, providing an effective solution for intelligent and real-time feedback in outdoor sports teaching.

IIoT and digital twin-empowered micro level carbon governance is transforming EU-China carbon neutrality; this study defines the mechanism of DCCNAS (digital cockpits carbon neutral assimilation system) that facilitates real-time vehicle emission behavior interactions; however, it lacks theoretical support and remains unexplained. By employing a mixed-methods, first define the DCCNAS, subsequently describe the first-stage and second-stage structure of DCCNAS mechanism, demonstrate the exploitative and explorative green innovation with dual tacit knowledge transfer in through four cases. Second, develop a tripartite evolutionary model to quantitatively examine the micro-level asymptotic stability of exploratory versus exploitative green innovation strategies under dual tacit knowledge transfer. Automaker achieves dual tacit knowledge transfer through the diffusion of low-carbon technologies in driving behavior and community culture, empowers behavioral optimization and energy efficiency management within the DCCNAS framework to facilitate green transformation from individuals to organization. This study contributes in identifying DCCNAS and offers theoretical support and practical guidance for the green innovation practices of automakers, reveal the mechanism in achieving carbon neutrality goals in product design, technological upgrades, and system optimization.

With the rapid development of intelligent education, spoken English practice systems based on dialogue simulation have become essential tools for language learning. However, traditional cloud-based dialogue systems face significant challenges in real-time interaction due to high end-to-end latency caused by long-distance data transmission. To address this, this paper proposes a three-tier (device-edge-cloud) architecture for dynamic computational offloading and a lightweight multimodal edge computing framework (LMecf) that integrates speech, text, and prosodic features using a parameter-efficient transformer. Additionally, a latency optimization model is introduced to minimize delay while maintaining dialogue quality. Experimental results on the spoken English dialogue corpus (SEDC) show that LMecf reduces latency by $��42.3�\%�$ compared to cloud-only systems and $��28.7�\%�$ compared to edge-cloud hybrids, thus providing an effective solution for low-latency, high-quality spoken English dialogue simulation in intelligent education.

As microservices and containerized architectures gain prominence, Kubernetes has become the foundation for orchestrating distributed workloads. However, its default scheduling strategy is typically static and resource-centric, lacking responsiveness to changing workload patterns, inter-service communication needs, or energy efficiency goals. This paper introduces AICoLoc-K8s, an adaptive scheduling enhancement framework that augments Kubernetes with intelligent, real-time decision-making. AICoLoc-K8s leverages live system metrics collected from Prometheus and a lightweight AI model to inform pod placement decisions. Integrating a custom scheduler extender, the system dynamically evaluates candidate nodes based on multiple criteria, including CPU load, memory availability, and service tier affinity. The AI model is trained to optimize pod co-location, especially for workloads categorized by frontend, backend, and database roles. We implemented AICoLoc-K8s on an RKE2 cluster running inside KubeVirt virtual machines and validated its behavior through controlled experiments. The framework consistently demonstrated better placement alignment with workload characteristics than the default scheduler. Although quantifiable gains like latency and energy savings are reserved for future evaluation, initial results confirm the effectiveness of this real-time adaptive model.

As key smart grid edge nodes, power IoT terminal security directly impacts power network stability. However, traditional traceability techniques face sample imbalance, graph scale expansion, and computational delay. Thus, this paper proposes an attack traceability method for power IoT terminals based on dynamic causal graph. Firstly, residual generative adversarial networks generate synthetic attack data meeting timing and logic constraints to alleviate real attack sample scarcity; Secondly, a dynamic graph convolution causal reinforcement mechanism is designed—combining mutual information and attention weights to optimize attack traceability graph topology, reducing computational complexity while improving path inference accuracy; Finally, multi-level graph distillation transfers knowledge from complex graph attention networks to lightweight graph isomorphism networks, enabling efficient attack traceability in resource-constrained environments. Experiments show this method significantly boosts detection accuracy with few samples, outperforms traditional methods in cross-domain attack traceability accuracy, and cuts model computational overhead sharply, making it suitable for edge device deployment.