This PDF file contains the front matter associated with SPIE Proceedings Volume 13397, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

Satellites can complement terrestrial communication networks. The integration of 5G and NTN technology can provide full-area full-coverage services integrating space, space and ground. Based on this, this paper proposes a design solution for a 5G NTN communication terminal, introduces the system architecture of 5G NTN for power consumption information collection business, and describes the application of the terminal in this power business. 5G NTN communication terminal integrates 5G and satellite technology and makes full use of terrestrial and satellite networks to efficiently meet different deployment scenarios and diverse business needs of the power system.

With the rapid development of cloud computing technology, traditional network solutions have struggled to meet the demands of large-scale data centers for high performance, high availability, and flexibility. OVN, a subproject of OVS, offers a scalable and efficient virtual networking solution, significantly enhanced the performance and scalability of overlay network through the combination of SDN model and container technology. However, in large-scale SDN network clusters, the most used container orchestration engine Kubernetes is now facing performance bottlenecks. This paper introduces the relevant theories and key technologies of OVN and Kubernetes, analyzes the advantages and disadvantages of existing open-source community SDN scheme, and proposes a novel network architecture based on an improved Kube-OVN. By optimizing subnet partition strategies, flow table generation mechanisms, and alleviating the pressure on centralized controllers, the proposed architecture enhances the performance and reliability of large-scale Kubernetes networks. Finally, the paper validates the effectiveness of the proposed solution on large cloud computing platforms through experiments.

Research on Internet of Vessels composite communication network equipment provides an in-depth analysis of various maritime communication environment issues and proposes a design for composite communication network equipment consisting of two types of maritime mobile routing equipment. The first FPGA-based high-speed communication unit adopts Beamforming technology and uses the signal processing technology of sensor arrays to send and receive signals directionally to ensure the smooth connection between nodes and the connectivity of the network. This signal processing technique can be used in environments with open seas and sparse communication nodes. In order to improve the capacity of the communication system, the second one designs another communication unit by studying the allocation algorithm based on channel quality, which can be configured with three different frequency bands, thus guaranteeing the stability of the network transmission and the timeliness of the information transmission. This method is suitable for maritime communication of ships in close proximity.

The heating magnetic stirrer has always been an indispensable experimental instrument in chemical laboratories, with a high daily demand. However, current heating magnetic stirrers still rely on traditional hardware technology, only capable of operating in local mode, and lack the ability for remote temperature adjustment by users, resulting in low efficiency. This study proposes an innovative design that integrates the heating magnetic stirrer with remote IoT control, aiming to enhance laboratory safety and efficiency through the integration of IoT and Proportional-Integral-Derivative (PID) technology. The key component Arduino Nano board acquires sensor temperature and system operating status and collaborates with a wireless communication-enabled sensor system. The acquired data is transmitted through the ESP8266- 01 module to the Blinker cloud platform for storage and display, enabling users to monitor system status and adjust set temperatures via a mobile application. Additionally, we have incorporated fault display functionality into the WeChat mobile service. Through comprehensive comparison, our design ensures the safety of chemical experiments and enhances the efficiency of the heating magnetic stirrer.

Federated learning has resolved the issue of privacy-protected data sharing among vehicles in vehicular network environments. However, traditional federated learning schemes suffer from long convergence times and significant network resource consumption. To fulfill the future needs for real-time traffic condition data and to alleviate the bandwidth pressure on backbone networks, this paper introduces a privacy-preserving federated learning scheme for rapid sharing of real-time traffic conditions. This system begins by designing a method that involves selecting highquality networking partners and optimizing the federated network structure of vehicles, thereby enhancing the efficiency with which vehicles obtain real-time traffic condition data. The results demonstrate that the data update efficiency of this system exceeds that of conventional vehicular network federated learning mechanisms. It provides multiple benefits, such as reduced potential attack windows, improved data timeliness, and lower communication costs.

In the cloud-edge collaborative task scheduling framework, when the task volume on the edge node reaches a certain threshold, the edge node becomes unable to effectively migrate additional tasks. To solve the problem of ineffective optimization in task migration on the edge node, this paper proposes a task offloading strategy. Based on the Biogeography-based Optimization (BBO) algorithm, modifications are introduced to its initialization population strategy, migration method, and mutation method, enabling it to better address the multi-objective optimization problem of jointly minimizing delay and energy consumption. The leakage detection of pipelines is used to establish a simulation model, and experimentally verified that the algorithm can be more targeted to solve the scheduling in the unloading process of cloud-edge collaborative task offloading

With the development of intelligent and connected vehicles, automotive Cybersecurity has received widespread attention. The in-vehicle communication system, as an essential part of smart vehicles, poses significant security concerns. Currently, the CAN (Controller Area Network) bus transmits messages in plaintext without identity authentication and information encryption mechanisms, and a single CAN frame only has an 8-byte data field. Therefore, adding protective mechanisms for CAN messages is of great research significance. Based on this, we first propose a CAN message authentication mechanism. When the receiving party receives a message, it verifies the authenticity and integrity of the message. Secondly, we design a security log recording strategy for MCU controllers based on UDS (Unified Diagnostic Services) services. When a security event occurs, the event content can be recorded quickly. Finally, under a service-oriented architecture, we conduct a service-oriented analysis of automotive information security, designing an information security SOA (ServiceOriented Architecture) service composed of two foundational services and four atomic services, and detailing the standard interfaces for each service. Ultimately, we validate the secure communication strategy by deploying it on the NXP S32K146 chip. The validation results indicate that the proposed secure communication strategy effectively defends against spoofing attacks, tampering attacks, and replay attacks. It addresses the deficiency of lacking identity authentication in traditional CAN networks, ensuring the information security of intelligent connected vehicles.

This paper aims to enhance the security of the Advanced Encryption Standard (AES) by introducing two key improvements to the traditional fixed row shifting and key expansion methods. First, a dynamic row shifting technique is adopted, which determines the row shift amount based on the sum of each row's data modulo 4, thereby increasing randomness and making attacks more complex. Second, the Secure Hash Algorithm (SHA-256) is introduced into the key expansion process, selecting different bits of the hash value based on the expansion round to form the next round key, thus improving key unpredictability. Theoretical analysis and experimental validation demonstrate that the improved AES algorithm significantly enhances resistance to brute force attacks and increases encryption strength. Although there is a slight decrease in computational efficiency, the enhanced security is notable. This research provides a new approach for the field of cryptography, contributing to further improvements in the security and practicality of encryption algorithms

A sensor network is a novel type of distributed network comprising sensor nodes that perform data collection, information processing, and communication functions. The network is employed in a multitude of fields, including earthquake monitoring, agricultural production, smart homes, and battlefield reconnaissance. In this work, we propose a novel optimization model that employs edge computing nodes for data preprocessing and distribution in sensor networks. This approach markedly diminishes the computational load on the central server by conducting data processing and filtering at the network's edge, consequently reducing the volume of data that must be transmitted to the central server. In particular, we have developed a collaborative computing algorithm that allows multiple edge nodes to share computing tasks, thereby enhancing the speed and accuracy of data processing. Furthermore, the dynamic resource allocation algorithm is employed to adaptively adjust the resource allocation of edge nodes in response to real-time network conditions and task requirements, thereby enhancing the overall efficiency and reliability of the network. Our innovative approach to data preprocessing, including advanced filtering and compression at the edge nodes, and our dynamic resource allocation strategy, which adaptively adjusts resources based on real-time network conditions and mission needs, is significantly better than existing technologies. The experimental analysis indicates that the proposed method exhibits superior performance in terms of data processing speed and network latency compared to the traditional sensor network architecture.

This paper addresses the data association problem in multi-extended target tracking by proposing a variational Bayesianbased solution. The method models targets using random matrices. Within the variational Bayesian framework, association variables are used to establish data association between measurements and extended targets, and update equations are derived for estimating the motion and extended states of multiple extended targets. Simulation studies illustrate that the suggested data association technique efficiently allocates measurements, leading to efficient tracking of multiple extended targets.

5G Co-band signals have been widely used in wireless communication navigation. With the continuous development of 5G, people's requirements for positioning accuracy are constantly increasing. TOA and TDOA are commonly used measurement and positioning methods. TOA and TDOA positioning methods need to determine the arrival and reception time of signals, which puts higher requirements on the time synchronization accuracy in wireless networks. This paper studies the time synchronization of 5G Co-band signals in multi-hop networks. In view of the problem of reduced synchronization accuracy caused by the cumulative error of measurement noise in multi-hop grids, the child node uses logical clock drift and logical clock offset as the observation vector of Kalman filter to predict and correct the time phase estimation and frequency estimation of the parent node. The experiment shows that, the time synchronization accuracy of 5G Co-band signals in multi-hop network nodes has been dramatically improved.

In response to the shortage of time resources for traditional multifunctional phased array radar to perform multitasking, a multi-target tracking algorithm based on fusion measurement and a comprehensive strategy adaptive resource scheduling algorithm are proposed. The performance of these two algorithms in handling multitasking is verified through simulation experiments. This experimental platform integrates functions such as radar resource scheduling, signal processing, and data processing to evaluate the performance and resource scheduling efficiency of multifunctional phased array radar for target detection and tracking in multitasking environments. The results indicate that the proposed multi-target tracking algorithm based on fusion measurement can optimize the performance of the radar system and improve the accuracy of target signal detection and tracking within limited time resources. Compared to traditional resource scheduling algorithms, the comprehensive strategy adaptive resource scheduling algorithm can allocate system resources more reasonably in the context of multitasking scheduling, significantly improving the resource utilization efficiency of multifunctional phased array radar.

At present, UAV widely used in the dual-use field, the gimbal is the key component of the UAV hanging on the equipment and maintaining a stable posture, in order to strengthen the control efficiency of the power grid high-altitude workers, replace the traditional inspection mode of high-altitude workers, and build intelligent supervision and control methods relying on UAV high-altitude target recognition, dynamic tracking and face recognition, This paper proposes an application of non-inductive intelligent supervision technology based on the adaptive adjustment algorithm of UAV gimbal, which is used for the supervision of the qualification verification, identity verification, violation capture and other operations of aerial workers. Simulation experiment results show that the gimbal control system using fuzzy PID adaptive algorithm can reduce the overshoot by 20%, and the system stability time is about 0.2S. The pilot test verifies that compared with the traditional supervision method, the average comprehensive accuracy rate of target identification by the non-sensory supervision is as high as 97.8%, and the control efficiency has been greatly improved.

Traditional single-layer complex networks models show unsatisfactory practicality in modeling multi-functional, collaboratively operating aviation systems. This paper proposes a method for constructing a directed weighted super network model for aviation transport systems. Firstly, four functional sub-networks are constructed, including airline company sub-network, air route sub-network, airports-airline sub-network and the sector sub-network. On basis of the sub-networks, a super network model for aviation transport is introduced. Characteristics of the super network are then analyzed by using metrics such as functional-structural correlation coefficient, and node hyperdegree ratio. Results show that the super network structure is unstable, with low levels of association between most sub-networks, but stronger associations between air route-the Sector and air company networks-airports-airline. This work sets good bases for further vulnerability analysis of complex network.

In order to improve the ability of differential evolution (DE) to solve dynamic multi-objective optimization problems (DMOPs), in this paper, the information of historical population is used to designed a improved mutation operator, this operation to make up the deficiencies of the diversity for population. In addition, when the environment is changed, use center points and linear prediction model to design a prediction technique, this technique is used to track the changed optimal solution set. Merge the above strategies into NSGA-II and noted as (HPS-NSGA-II/DE). Finally, the performance of HPS-NSGA-II/DE is confirmed by some test functions.

When the shipboard satellite communication station captures and tracks the satellite, it needs to use the data source to provide support. In the current servo monitoring software, before capturing the satellite, it needs to manually select three data sources. After the onboard satellite communication station enters the self-tracking state, if the data source needs to be switched, it can only be switched manually in the servo monitoring software panel. If the data source is invalid or inaccurate, it is necessary to switch the data source in time to ensure the stable tracking of the antenna.

In this paper, we investigate the low complexity uplink transmission channel estimation (CE) algorithms for the problem of limited computational resources in massive multiple-input multiple-output (mMIMO) low Earth orbit (LEO) satellite Internet of Things (IoT) scenarios. We use angular-domain channel to characterize uplink transmission channel to exploit its sparse characteristics in the angular-domain and formulate the joint user activity detection (UAD) and CE problem as a sparse vector recovery problem by combining it with the sparse characteristics of the active user matrices in the satellite IoT scenario. Finally, we propose the enhanced simultaneous orthogonal matching pursuit (ESOMP) algorithm based on the theory of compressive sensing (CS), and verify it through the simulation experiment. The simulation results show that ESOMP algorithm can reduce the computational complexity while guaranteeing the algorithm's performance.

With the advancement of medical technology, electrocardiogram (ECG) monitoring plays a crucial role in the diagnosis of heart diseases. However, traditional ECG analysis methods rely on the subjective judgment of professional doctors, which has certain limitations. This paper proposes a technology for the detection and diagnosis of ECG abnormalities based on deep learning and signal processing, aiming to enhance the automation and accuracy of ECG signal analysis. Initially, this paper employs wavelet transforms for preprocessing of ECG signals, effectively removing noise while preserving key ECG features. Subsequently, Recurrent Neural Networks (RNNs), specifically Long Short-Term Memory networks (LSTM), are utilized for time-series analysis of ECG signals to differentiate between normal and abnormal ECG signals. Furthermore, a Convolutional Neural Network (CNN) is constructed to extract features from abnormal ECG signals for the identification of specific types of abnormalities. To augment the model's feature expression capability, this paper also introduces an attention mechanism, which enhances the weight of key signal segments, thereby improving the accuracy of abnormality detection. Experimental results demonstrate that this method achieves a high accuracy rate in the task of ECG abnormality detection, effectively assisting doctors in the diagnosis of electrocardiograms. This research not only has potential application value in intelligent monitoring systems and data mining technology but also provides a new technological approach for the automated diagnosis of electrocardiograms, with prospects for playing a significant role in the fields of industrial Internet of Things and intelligent manufacturing.

In recent years, the identification of flying objects in the air has been the focus of scholars' research, whether it is military or life, the application of UAV is more and more extensive. This paper presents a combined neural network method for classification and recognition of drones, birds and clutter. First, CNN is combined with recurrent neural network to improve network performance, and then it is combined with attention mechanism at different positions, which can play a role in regulating the features obtained by convolution and the output of recurrent network, etc. Finally, the results are input into the classifier optimized by optimization algorithm for classification. This method has better recognition effect than random forest, particle swarm BP algorithm, CNN combined with recurrent neural network algorithm, etc. The recognition rate of UAV reaches 0.99. The results of different algorithms are fused at decision level to achieve better results.

Orthogonal Frequency Division Multiplexing (OFDM) technology is a prevalent choice in the realm of wireless transmission. Despite its extensive use, the elevated Crest-to-Mean Power Ratio (CMPR) encountered in OFDM setups hinders their deployment in real-world scenarios. The employment of the Partial Transmit Sequence (PTS) strategy is a proven approach to mitigating CMPR, yielding substantial enhancements in its reduction. Yet, the intricacy of the PTS approach escalates with an increase in the count of sub-blocks, posing a difficulty in balancing computational intricacy with efficient CMPR control. This study introduces an optimized PTS algorithm utilizing Particle Swarm Optimization (PSO) to enhance the minimization of CMPR by fine-tuning the phase components. Our research includes simulation tests that affirm the efficacy of the novel method across varying sub-block counts and contrast it with standard OFDM configurations. The findings suggest that the PSO-enhanced PTS method not only effectively curtails CMPR but also achieves this with reduced computational overhead.

Big data is a data processing and application model based on cloud computing, with the ability to integrate and share data, cross reuse and form intellectual resources and knowledge services, etc. Big data has four characteristic quantities: massive (Volume), diverse (Variety), real (Value) and rapid (Velocity). With the traditional data mining using random analysis (sampling) is different, big data methods have not simply increase the amount of analytical data, but a systematic and structured analysis of massive data. Only through systematic analysis can we obtain enough intelligent, in-depth and valuable information. The basic process of analysing data using big data processing can be summarised in four steps, namely acquisition, pre-processing, statistics/analysis, and mining. Big data analysis of time series mainly involves five major aspects, namely visual analysis, data mining algorithm analysis, predictive analysis, semantic engine, data quality and data management. Among them, data mining algorithm analysis and predictive analytics are the main research focus of time series big data analytics, i.e., through the advantage of massive data, the use of algorithms to deeply analyse the attributes of the sequence, including the number, feature diversity, complexity, etc., and then according to the attribute characteristics of the sequence, mining analysis processing and other operations. Although big data is a newly emerging technology, its application in time series has achieved good results. The use of big data methods to analyse time series, on the one hand, solves the shortcomings brought by the traditional method of analysing only a small amount of data, on the other hand, big data technology goes deep into the data, and is able to dig out the useful information hidden behind the huge amount of data, so as to make the prediction and analysis more accurate and reliable.

With the importance of communications for chipset networks. This paper studies the BER performance for communication considering impairments such as phase noise, dispersion and additive gaussian noise, the constructed system compensates for dispersion issues and then uses pilot tones to address and mitigate the effects of dispersion and phase noise. To further investigate the system's capability in handling dispersion and phase noise, comparative experiments were conducted to observe whether the system could still compensate for these issues under increased dispersion and phase noise conditions. The study found that the system performs well in addressing dispersion problems, however, its compensation effectiveness is limited when phase noise becomes excessively high.

This paper explores how to enhance the performance of deep learning image classification models by introducing attention mechanisms and optimizing feature extraction methods. Techniques such as CBAM (Convolutional Block Attention Module) and depthwise separable convolutions were employed, with experiments showing that the CBAM model achieved an accuracy of 89.2%, while the optimized model reduced computational cost to 1.20 GFLOPs. These results validate the effectiveness of the proposed methods in improving classification accuracy and reducing computational costs, offering new directions for future research.

Quality of Service (QoS), fairness, and resource sharing control in networks have not been fully addressed. Existing resource-sharing strategies depend on maintaining flow states within the network and executing different policies based on these states, which lack of scalability. These strategies require adjustments when traffic patterns shift and are tailored to specific network scenarios. In this paper, we presents PV-AIFO (Packet Value-Admission-In First-Out) Active Queue Management (AQM). PV-AIFO combines the packet scheduling and drop strategies of AIFO with a resource sharing mechanism based on the Per Packet Value (PPV) concept. This method utilizes the AIFO algorithm to maintain a sliding window that tracks packets and dynamically calculates their relative priorities. These priorities combined with the state of queue congestion determine packet admission control. Simulation experiments demonstrate PV-AIFO integrates the strengths of both PPV and AIFO concepts, maintaining target queue delays while implementing policy based resource sharing. The motivation behind this algorithm is simplicity and ease of deployment in networks, as the original PPV methodscheduler requires costly and hardware unsupported packet drops from the middle of the queue.

Federated learning allows users to locally train the raw data and pass the trained gradients to cloud server to get global model, protecting the data privacy. Most of the existing privacy preserving federated learning schemes use homomorphic encryption to protect gradient privacy, where all users hold the same private key. However, the gradient privacy of other users will be leaked, once any user’s key is compromised. Meanwhile, these schemes suffer from poor model accuracy and efficiency. Therefore, this paper proposes double verifiable privacy-preserving federated learning under multiple keys, in which Paillier homomorphic encryption is utilized to protect the user local data privacy. Furthermore, by combining verifiable secret sharing and linear homomorphic hash, our scheme enables users to verify the correctness of the random number shares and aggregation result, achieving double verifiability. The security analysis indicates that the proposed scheme meets the necessary security requirements, while the performance analysis demonstrates that our scheme reduces computation and communication costs in comparison with existing schemes.

The Internet of Things technology makes a large number of devices connected to the Internet, resulting in massive interactive data. However, data is susceptible to hacker attacks and malicious exploitation during the interaction process, resulting in serious data leakage and privacy issues. Penetration testing can simulate real attack behavior, help identify vulnerabilities and security vulnerabilities in the process of IoT information cloud interaction, and prevent malicious attackers from exploiting vulnerabilities to invade systems and obtain sensitive information. Therefore, vulnerability detection in the process of IoT information cloud interaction under penetration testing is proposed. By evaluating the defense mechanism status of the Internet of Things through bidirectional penetration, a machine learning model based on time Markov vector features is constructed. The model is trained using historical session flow data to identify session flows containing sensitive information. At the same time, hidden Markov models are used to classify normal information interaction behavior samples to detect security vulnerabilities during the information interaction process. The experimental results show that the success rate of using the proposed method for vulnerability detection in the process of IoT information cloud interaction remains above 98%. Relatively high, it can accurately detect vulnerabilities generated during the interaction process of IoT information cloud, ensuring the security of IoT.

In the era of intelligent manufacturing, the integration of IT and OT realizes collaborative manufacturing under the transparency of the entire data. In industrial manufacturing, the communication mode of Fieldbus has great limitations. It is difficult to meet the needs of the era of intelligent manufacturing. The complexity of Fieldbus has different physical interfaces, transmission mechanisms, and object dictionaries, which causes large delays and jitters in IT information collection and downlink of instructions. This paper designs a time-sensitive network data transmission system and transmission method, including an end device as a data transmission and reception point, a clock synchronization module that provides an accurate synchronous clock for the network, and a bridge device that acts as an exchange. A perfect data shaping and scheduling mechanism is adopted, so that strong real-time data can preempt weak real-time data. The preempted data can continue to occupy the channel for transmission after the strong real-time data transmission is completed, providing a set of fast transmission solutions for future deterministic data.

This paper discusses the design and application of smart energy monitoring platform. Facing the global challenges of energy security and environmental protection, a comprehensive monitoring solution integrating the Internet of Things, big data and artificial intelligence technology is proposed. The platform design focuses on real-time monitoring, data analysis and intelligent early warning, and solves problems such as data collection compatibility, big data migration and system stability. Through the edge iot agent terminal real-time acquisition of critical load data, combined with efficient data processing and analysis, to achieve accurate control of energy consumption. The communication architecture of the platform relies on ubiquitous iot and 4G/5G technology to ensure high-speed data transmission. The data architecture supports easy access and value mining of massive data, enhancing energy management decisions. The innovation of iot acquisition technology and data migration scheme ensures the practicability and flexibility of the platform. Practical applications show that the platform effectively improves the safety, efficiency and environmental performance of the energy system, and makes an important contribution to the digital transformation and sustainable development of the energy industry.

Hyperspectral imaging (HSI) provides a powerful guarantee for the accurate identification of various targets due to its rich spectral and spatial information. Anomaly detection, a significant application of HSI, is inevitably affected by noise during the imaging process. To achieve effective HSI anomaly detection in complex environments, this paper propose minimum noise fraction (MNF) and tensor low-rank anomaly detection by replacing principal component analysis (PCA) in PCA-tensor low-rank and sparse representation (TLRSR) with MNF as a preprocessing step. Experimental tests using real hyperspectral data demonstrate that the proposed method can effectively reduce noise in HSIs and significantly enhance the anomaly detection accuracy.

Internet of Things (IoT) devices are vulnerable to various cyber-attacks. Therefore, it is significantly crucial to design an effective intrusion detection system (IDS) for IoT security. However, IoT devices have limited resources such as computing resources needing to handle a great of data, which further increases the difficulty of precise intrusion detection. Moreover, most IDSs lack transparency. This study develops a feature selection-based hidden Markov model (HMM) for intrusion detection. We first establish a modified approach based on a fuzzy rough set for feature selection to select optimal features so that the computational burden can be reduced. Furthermore, an interpretable intrusion detection model is established based on the HMM. In addition, intrusion detection is achieved in the presence of partially missing data through the expectation-maximization algorithm. According to the simulation results, the proposed method is able to detect intrusion detection in IoT networks effectively.

Computed tomography (CT) imaging systems often produce images with lower resolution than ideal, particularly when using low-dose protocols to minimize radiation exposure. To address this limitation, we propose CTMTGAN, a novel super-resolution analysis method for enhancing CT image resolution using Generative Adversarial Networks (GANs). CTMTGAN integrates VMamba's TDSS module with Transformer-based attention blocks in its generator, trained with a PatchGAN discriminator and VGG-19 perceptual feature extractor. Experimental results demonstrate that CTMTGAN outperforms state-of-the-art methods in no-reference image quality metrics, producing clearer details, sharper edges, and fewer artifacts. Its ability to generate high-quality, high-resolution CT images suggests potential to improve diagnostic accuracy in clinical settings.

With the increasing application of remote sensing images, people have higher requirements for images’ quality. Therefore, improving remote sensing images’ resolution is of great significance. With the development of machine learning, neural networks (especially SRGAN) have significant advantage in single image super-resolution research. The SRGAN model was used to enhance remote sensing images’ resolution in the study. SRGAN’s generator used residual blocks, which improved SRGAN’s performance. SRGAN outperformed traditional algorithms in subjective visual perception and objective evaluation metrics such as PSNR and SSIM. Compared to SRCNN, SRGAN improved PSNR by 1.809 dB and SSIM by 0.015. In addition, the super-resolution image quality was significantly improved, which fully demonstrated the effectiveness of the proposed method.

This study aims to address challenges in sentiment text classification, such as emotional shifts, contextual dual meanings, and the limitations of traditional models in robustness and feature extraction. We propose a neural network model that integrates Bert, BiLSTM, TextCNN, and Attention mechanisms, along with adversarial training and contrastive learning methods. Bert is used for word embedding, BiLSTM captures sentence dependencies, TextCNN extracts local features, and Attention mechanisms finely adjust semantic relationships. Fast Gradient Method (FGM) enhances robustness, and a Dual Contrastive Learning (DualCL) strategy improves feature representation. Experimental results show that the model significantly outperforms traditional methods and other deep learning models in accuracy and robustness.

Usually, the energy carried by IoT perception nodes is limited and cannot be replenished, which is one of the bottleneck issues for the practical application of IoT. This article analyzes the energy consumption model for IoT perception layer network. Firstly, the energy consumption characteristics of the perception layer are analyzed. Subsequently, based on the analysis of the node energy consumption model, the linear equidistant energy consumption model and the linear optimized distance energy consumption model was analyzed and studied. Through MATLAB simulation, it has been proved that as the number of hops increases, single hop transmission consumes significantly more energy than multi hop transmission, and the overall energy consumption under optimized distance model is lower.

With the continuous expansion of network applications, traditional network structures are no longer suitable for modern network applications. SDN, as a new type of network architecture, decouples the control plane from the data plane, simplifies network configuration work based on logic centralized controllers and open programming interfaces, thereby achieving a flat management mode and flexible data forwarding function. However, the centralized control and programmability of SDN networks make them face more complex network security issues, among which DDoS attacks are one of the biggest security threats faced by SDN networks. DDoS attacks are a distributed coordinated attack that utilizes multiple network sources, with strong deceptive and destructive characteristics. This article studies the effective detection of network attacks using weighted KNN algorithm based on different datasets in SDN environment, providing a basis for defending against DDoS attacks.

With the rapid development of autonomous driving technology, the autonomous driving ability of vehicles has become an important research direction in intelligent transportation systems. Convolutional Neural Network (CNN) has become one of the core technologies in autonomous driving vision systems due to its excellent performance in image recognition and classification. Embedded systems also play an important role in the realization of autonomous vehicle driving due to their high efficiency and real-time nature. In this study, a vehicle autonomous driving scheme based on Convolutional Neural Network (CNN) and embedded system is proposed. Firstly, the convolutional neural network was used to perform realtime image processing and feature extraction on the road scene. Specifically, through the superposition of multi-layer convolutional layers and pooling layers, features such as edges, textures, and objects in the image are extracted layer by layer, so as to achieve efficient recognition of complex road environments. In this process, the convolutional layer is used to extract local features, while the pooling layer is used to reduce dimensionality and prevent overfitting to ensure the robustness and efficiency of the model. Secondly, an embedded system was designed and optimized, and the trained CNN model was deployed on the system to ensure real-time processing power and efficient energy consumption management. The design of the embedded system focuses on the optimal allocation of hardware resources and the effective control of energy consumption to meet the real-time operation needs of vehicles under different road conditions. Specifically, the embedded system uses high-performance processors and low-power hardware modules to ensure fast inference and realtime decision-making capabilities of CNN models. In addition, the overall performance and reliability of the system are further improved through the co-design of software and hardware. Combining the above two technical advantages, the scheme can effectively reduce energy consumption and improve the stability and response speed of the system while realizing autonomous driving.

This study aims to improve the consumption efficiency of clean electricity in coastal areas through Bayesian network models. This article utilizes the probabilistic reasoning ability of Bayesian networks to analyze various factors that affect power consumption, construct a power consumption efficiency prediction model, and verify the effectiveness of the model through simulation. The research results indicate that Bayesian networks can effectively identify key influencing factors, improve the scientific nature of power consumption decisions, and thus enhance the overall consumption efficiency of clean electricity in coastal areas. In terms of the impact on power dispatch optimization, it was found that the average prediction error was -0.24 MW, with a standard deviation of 3.12 MW, and most of the errors were concentrated in the range of -5 to 5 MW. The experimental results have verified the effectiveness and reliability of Bayesian networks in optimizing power systems.

The Border Gateway Protocol (BGP) is the established inter-domain routing protocol utilized on the Internet. However, its original design placed greater emphasis on protocol efficiency and robustness rather than addressing security and trust concerns of network participants, consequently leaving the inter-domain routing system susceptible to numerous security risks. This thesis examines the inherent security vulnerabilities in the BGP inter-domain routing protocol, including hijacking, tampering, and leakage resulting from its internal flaws. It also addresses the significant security risks associated with centralized trust verification mechanisms that contribute to management centralization and trust monopolization. Furthermore, it integrates blockchain technology to propose a decentralized trusted BGP security mechanism, which ensures the authenticity and traceability of inter-domain protocol message sources and provides protection against interdomain routing hijacking.

In today’s fast-paced, high-pressure society, sleep disorders are increasingly prevalent. Long-term, daily sleep monitoring is essential for maintaining human health. To achieve multidimensional, convenient, and reliable sleep monitoring, this study introduces a novel IoT-based system designed for long-term tracking of individual sleep patterns. The system captures data across three dimensions—sleep environment, objective sleep data, and subjective sleep experiences—using environmental sensors, wearable devices, and a smartphone app for self-reporting. Data is processed, stored, managed, and visualized via a cloud platform. Field experiments with volunteers were conducted to validate the system’s effectiveness, demonstrating its precision and stability. The system enables comprehensive and long-term sleep monitoring, tracks nightly changes in user sleep experiences, and identifies environmental factors strongly associated with sleep depth. This provides critical insights for improving sleep quality and highlights the system’s broad application potential.

Cracks are one of the main defects in concrete structures, and timely detection and management of cracks are crucial for maintaining the stability and safety of these structures. Addressing the issues of low accuracy and efficiency in existing deep learning object detection models applied to concrete crack identification, this paper proposes a method based on an improved YOLOv8s for concrete crack detection. Initially, the C3STR is used to replace part of C2f in the backbone network of YOLOv8s to enhance the extraction of complex global features; secondly, a CA attention mechanism is introduced to improve the feature extraction across different layers; finally, the SPPFCSPC replaces the SPPF spatial pyramid pooling module to enhance feature fusion across various spatial hierarchies. The experimental results demonstrate that compared to conventional YOLOv8s, this method enhances detection precision by 3.09% to 94.34%, with a detection speed of 224 frames per second. Additionally, it allows for the quantitative calculation of crack length and average width.

This study provides an algorithm of a self-adaptive preset performance controller is designed for robot trajectory tracking problems with modeling errors and uncertain disturbances, and the stability of the system is proved using Lyapunov stability theorem. This controller can ensure that the tracking hin the preset boundary in the presence of uncertainty and external disturbances in the system, and has the characteristics of fast convergence and small overshoot. The effectiveness and performance of the controller were verified through Matlab simulation.

In the realm of Chinese long-text categorization, the advent of the big data era has led to an exponential surge in information volume, thereby imposing stricter demands on the accuracy and efficiency of automatic classification systems. This study introduces an enhanced model, hereby referred to as "FastText-BT-Att" based on the FastText framework. This model further deepens text representation through the integration of Bigram and Trigram features and incorporates a Multi-Head Attention mechanism to finely identify and weigh key information within the text. Comparative experiments demonstrate that the FastText-BT-Att model achieves higher accuracy rates in long-text classification tasks. The model is significantly outperforming traditional models, underscoring its enhanced effectiveness, highlighting its effectiveness and superiority in the task of categorizing Chinese long texts.

This study centers on enhancing the energy efficiency diagnostics of an integrated smart energy platform. It seeks to boost the platform's accuracy and practicality in predicting energy consumption and analyzing energy efficiency by incorporating a PSO-BP neural network model. The research demonstrates that this model effectively manages the nonlinear relationships within complex energy systems, enabling precise energy consumption forecasts. Verification of the model indicates a strong alignment between predicted and actual energy use, with minimal error, underscoring its high level of accuracy and dependability. Furthermore, the platform's architecture has been refined, featuring the integration of edge computing and enhancements to the communication framework, which greatly improve real-time monitoring and data processing. In summary, this work introduces an innovative approach to the energy efficiency assessment of integrated smart energy platforms, aiding businesses in their efforts to conserve energy and reduce emissions while advancing intelligent and precise energy management practices.