Proceedings of the 2nd International Conference on Internet of Things, Communication and Intelligent Technology

The voltage transformer in the substation convert high voltage into low voltage, and its second circuit must be grounded at one point through the zero phase line N600. The long-term existence of N600 multi-point grounding increase the possibility of power grid accident. This situation is usually found through patrol inspection. However, the patrol inspection is conducted every six months, the circle is so long that the problem cannot be identified in time. This paper proposes a method to identify and find the N600 multiple grounding point in real time. It judges the situation of multi-point grounding through the changes of N600 grounding current value, and determines the position of redundant grounding point through working event in the substation when the event happens, to achieve real-time monitoring of multi-point grounding situation of N600. In addition, this paper also implemented an N600 grounding current monitoring device and an N600 multi-point grounding monitoring and decision-making system, and verified the effectiveness and real-time performance of this method and the system in a testing environment.

Aiming at the control methods of the streetlights, as well as the problems of high energy consumption caused by the continuous lighting of streetlights despite of few pedestrians at night in the park, this paper proposes a smart streetlight system for parks based on NB-IoT (Narrow Band Internet of things) technology, which uses STM32L431RCT6 single-chip microcomputer as the main control chip, collects the ambient illuminance information through a light sensor, and transmits it back to the HUAWEI CLOUD platform through the NB-IoT communication module, so as to realize automatic remote control of streetlights on the terminal side. When the Streetlight is in the lighting state, the brightness of the street lamp is set to 50%, and if the infrared sensor detects human movement, the brightness of the Streetlight is increased; and the brightness will be reduced after the human moves away. Experimental data show that the system can realize the automatic remote control of a single streetlight according to the ambient illuminance, and automatically adjust the lighting brightness of the streetlight at night by detecting human movement, so as to effectively save electricity and reduce energy consumption.

Disaster prediction can buy valuable time for power grid disaster prevention and reduction. In this study, we firstly propose a physical statistical method for substation inundation prediction, which can comprehensively consider geographic features, real-time water level height of substation, precipitation amount, substation water collection capacity and drainage capacity. Nine substations in Changsha City, Hunan Province, are then analyzed separately and inundation prediction models are established using the rainstorm and substation inundation process in May 2022. The accuracy of the models are finally verified by using the heavy rainfall process in July 2022. The results show that the model can reproduce the process of water level variation in the substation, with a height error of less than 5cm. Subsequently, those models can combine the refined precipitation prediction data to carry out real-time prediction of substation inundation risk and hopefully improve the substation rainstorm disaster resistance.

Substation patrol system is designed to realize the joint patrol of robots and video cameras based on image analysis technology. By the interconnection and interaction with the main/auxiliary equipment monitoring system of the substation or the one-key sequence control system or the patrol centralized monitoring system of the master station, intelligent linkages are realized, including forward linkage, reverse linkage and video double confirmation between the patrol system and the main/auxiliary equipment monitoring system of the substation. Taking advantage of the great flexibility of the robot and the high patrol efficiency of the video camera, the locations of the video cameras and the patrol path of the robot are planned properly to contain all the patrol points, in order to cover the patrol range of the substation, improve the patrol efficiency, and save project investment as far as possible. It can effectively support typical application scenarios of substation operation and maintenance, such as equipment intelligent patrol, daily operation monitoring, on-site patrol and abnormal fault disposal. Remote intelligent patrol is applied to replace manual patrol, and the original fault analysis based on experience judgment of operators is replaced with active forewarning, comprehensive judgment, intelligent diagnosis and intelligent decision-making.

The advent of the big data era has stimulated the collection, processing, and analysis of traffic data. This article proposes a deep learning-based models for vehicle speed prediction and improves it using three methods: BPNN (Backpropagation Neural Network), BPTT (BP Through Time), and LSTM (Long Short-Term Memory). The three methods are trained using different data volumes, including one week, one month, and two months. The effects of different data volumes on the different methods are compared with BPNN as the baseline. Additionally, it is worth noting that the experimental code involved in this paper does not utilize any existing libraries. Finally, the prediction results of the three methods are evaluated using RMSE (Root Mean Square Error) and MAE (Mean Absolute Error). The results show that when using one month of data volume, the difference between the three methods is more obvious, and BPTT outperforms BPNN and LSTM. Compared with BPNN, BPTT reduces RMSE by 2.68% and MAE by 0.86%.

In existing zero-trust grid security protection architectures, the intelligent trust platform incorporating static matching algorithms suffers from inefficiency and long adaptation cycles. To address this, the fuzzy analytic hierarchy process (FAHP) is introduced to improve and optimize the trust evaluation platform, ensuring the privacy of transmitted data and the security of the system through continuous evaluation and updating of trust values. During the data transmission process, the accessing entity is dynamically authorized based on the principle of least privilege. Finally, feasibility assessment is performed in the trusted access proxy to determine the reliability of the data. Simulation experiments validate the feasibility and effectiveness of the FAHP-based zero-trust grid security protection architecture.

In areas such as maritime rescue, fisheries management, traffic surveillance, and national defense, ship detection and recognition based on synthetic aperture radar pictures is crucial. Deep learning offers a fresh approach to high-performance detection and recognition for SAR ships in the past few years with the growth of artificial intelligence. This work examines the ship target recognition approach in SAR images and suggests a direction-aware inshore ship detection method in order to meet the challenges of multi-scale ship target detection in SAR images as well as the complicated background of ships stationed in ports. Multiscale features are observed by using the pyramid feature extraction module with attention method. Aiming at the phase ambiguity problem in OBB regression, the direction-aware classification regression head was designed to accurately determines the position and direction of ship targets. Finally, the experimental part verifies that the proposed method reduces the computational complexity of our method and ensuring the detection performance.

Traffic accident risk prediction is used to study the historical accidents, identify the relevant factors and predict accident risk in the future. The existing prediction methods mainly obtain predicted unit by regularly gridding the road areas, resulting in a decrease in accuracy and low practical value. To improve the prediction accuracy, this paper takes urban roads as the prediction unit, adopts graph convolution neural network and gated recursive unit, and proposes a spatiotemporal gated graph convolutional neural network model (STGG-CnovNet) fusing multi-source spatiotemporal data features. The model consists of spatial convolution, temporal convolution, and spatiotemporal convolution. In the spatial convolution module, the spatiotemporal map data is constructed, and the spatial correlation is captured. In the temporal convolution module, a gated cycle unit is used to model the time correlation of traffic accidents. In the spatiotemporal convolution module, constructing a road similarity map can capture the spatiotemporal correlation of nodes. On real datasets, experimental results demonstrate our method is better than other baselines.

This work studies the problem of dynamic coverage control of multiple Unmanned Aerial Vehicles (UAVs) in the 3 dimensional (3D) environment. In this work, an efficient multi-agent deep deterministic policy gradient-based dynamic surface coverage (MADDPG-DSC) algorithm is proposed. In MADDPG-DSC, a digital elevation-based surface area calculation method is introduced to effectively allocate the points of interest (PoIs). Next, a cooperative trajectory control policy with multi-agent deep deterministic policy gradient is developed to guide the UAVs. Comparing with existing works, MADDPG-DSC shows better performance in terms of larger coverage rate, higher connectivity and lower energy consumption.

The application of unmanned aerial vehicle (UAV) inspection is gradually popularized in photovoltaic power stations, but the existing UAV inspection planning methods are currently strongly limited by the application scenarios and consumes a large amount of manual work. To ameliorate this, an automatic zoning optimization path planning method for UAV inspection path in photovoltaic power station is proposed in this paper. For any application scenario or scale of the power station, the whole station inspection path can be generated systematically according to the actual layout information and inspection requirements of the photovoltaic power station, with no need to swap the batteries manually halfway. The method includes following steps, waypoint coordinates determination, hangar location and jurisdiction demarcation, flight zoning and path optimization. Moreover, dynamic planning and hybrid ecological symbiosis algorithm is carried out in hangar location selection and inspection path planning. The application case of an 80MW photovoltaic power plant in East China shows that the hybrid symbiotic organism search path planning algorithm performances greater stability at different power station scales. The proposed zoning optimization path planning method can be highly adapted to any power station distribution specificity and reduce the length of the inspection path in the whole station by 37.98%–68.4%, compared to the manual path planning scheme.

In the context of UAV trajectory planning for data collection, challenges such as the uncertainty of a large-scale dynamic unknown environment and the need for multi-UAV coordination are prevalent. To address these challenges, this paper proposes a UAV data collection trajectory planning algorithm based on the D3QN (Double Dueling Deep Q-Network) algorithm. The proposed algorithm enables multiple UAVs to dynamically plan their flight paths for data collection in unknown environments through centralized training and distributed application. The algorithm’s performance is improved by incorporating competition mechanisms, candidate node queues, and reward function reshaping techniques. Based on the simulation results, the proposed algorithm outperforms similar algorithms in terms of success rates and task durations.

Carrier synchronization is an important link in wireless communication systems. It ensures frequency and phase synchronization between the receiving end and the transmitting end, thereby achieving reliable data transmission. The accuracy of carrier synchronization is crucial to ensure the performance of communication systems, especially in the case of high-speed data transmission and multi-user access. This article first investigates the current mainstream carrier synchronization algorithm, analyzes its working principle, and analyzes the characteristics of its mathematical model. By comparing the mainstream synchronous carrier implementation methods, two of them were selected, and the simulation models of square ring and COSTAS ring carrier synchronization were built using MATLAB/Simulink simulation software and its code program. Simulink simulation showed that it can be well realized. Simulation design of square ring and COSTAS ring carrier synchronization, and analyzed and compared the impact of square ring and Costas ring carrier synchronization methods on the coherent demodulation effect. The feasibility of the two carrier synchronization technologies has been verified through simulation, and both can better coherently demodulate the signal and have a high degree of signal restoration.

At present, there is little research on dust concentration prediction of dust instrument verification device, and the research on dust concentration prediction of dust instrument verification device is still in its infancy. It is necessary to further explore and optimize relevant algorithms and applications in the future to meet the needs of dust concentration monitoring and control in the workplace. In the process of dust concentration monitoring, in order to meet the metrological verification requirements of dust sensors with different ranges, it is necessary to adjust the parameters to change the dust concentration in the detection area. In order to solve this problem, a dust concentration prediction method based on long-term and short-term memory network (LSTM) and gated cycle unit (GRU) is proposed to predict the dust concentration in the detection area. Under the condition that the dust mass flow rate and wind speed are constant, other parameters are appropriately adjusted to predict the dust concentration in the detection area, and the average percentage error and root mean square error of the three algorithms are compared. The measurement error of dust concentration prediction method is less than that of LSTM method and GRU method, which shows that this method is more suitable for the field of dust concentration prediction and has better applicability.

For the high concentration dust instrumentation calibration device, it is necessary to ensure the uniformity and stability of the dust concentration distribution in the detection area. In this paper, Fluent fluid simulation software is used to numerically simulate the dust particles in the flow field, and the DPM model is used to study the transportation law of the discrete phase (dust particles) and the continuous phase (wind) in the flow field, and to determine the optimal detection area. The optimal detection area is determined by setting up dust concentration detection surfaces at different locations from the dust source and comparing the uniformity of the dust concentration distribution; the optimal transport wind speed is determined by changing the wind speed and comparing the uniformity of the dust concentration distribution while controlling the constant mass flow rate of the dust.

After installing a wireless communication module and computing equipment on the UAV (Unmanned Aerial Vehicle), the UAV can provide better communication coverage and even edge computing services for ground nodes by taking advantage of its high maneuverability and altitude advantages. According to the characteristics of user mobility and energy sensitivity, the edge computing problem of an intensive intelligent terminal network is presented. In this paper path planning and task computing strategy optimization are studied. For multi-UAV scenarios, a multi-agent reinforcement learning algorithm combined with initial UAV deployment is proposed. Specifically, the GAK (Genetic Algorithm k-means) algorithm is used to optimize the initial deployment positions of all UAVs. Then, using the multi-agent reinforcement learning algorithm MADDPG solves the optimization problems of track planning and task calculation strategies in dynamic environments. The simulation results indicate that the GAK-MADDPG algorithm can effectively utilize the local computing power of UAV and intelligent terminal by reasonably planning the trajectory and task calculation strategy of UAV, thus saving energy consumption on the user side to the greatest extent.

The issue of elderly people’s travel safety has attracted widespread attention in society. To address this problem and in line with current research trends, this study proposes an analysis and prediction of elderly people’s fall behavior based on ZigBee signal strength features. Due to the significant changes in radial range caused by movements, this paper investigates how ZigBee signal attenuation features can be used to perceive different angles. Various phenomena such as refraction, diffraction, and scattering can cause different degrees of interference in the normal signal propagation when ZigBee signals encounter different situations. By analyzing the signs of signal reception and detecting changes in signal strength, the physical condition of individuals during signal transmission can be determined. Furthermore, to address the issue of low accuracy in fall detection estimation based on broader spectral indices, this paper proposes an improvement. It presents an algorithm for extracting fall features based on a wider range of spectral indices, namely the fall behavior recognition algorithm.

With the rapid development of intelligent surveying and mapping industry, the application of Internet of Things and GNSS in deformation monitoring is becoming more and more extensive. Surveying and mapping engineering has high requirements for practical ability training, but traditional offline teaching methods cannot meet the needs of students, and the cost of offline education is relatively high. Therefore, based on WebGL technology and B/S architecture, with the background of high-rise deformation monitoring, this paper carries out a comprehensive chain integration of software and hardware operation functions in the production process such as the basic principle, data acquisition and data analysis of GNSS-based deformation monitoring technology, and builds a virtual simulation experiment system for high-rise deformation monitoring based on the Internet of Things and GNSS. The system simulates the basic system theory cognition in the process of collecting the Internet of Things and deformation data, the layout of the deformation monitoring network of the Internet of Things and the deformation data analysis and other functions, innovates the teaching method of modern surveying and mapping information technology practice courses, and uses virtual simulation technology to train innovative applied talents with professional competence and social adaptability. More accurately judge the possible disaster of the building, and take corresponding measures to avoid the occurrence of major accidents.

Due to the explosive growth of various applications, current wireless networks are confronted with heavy traffic burden, which makes the high data rate requirements of users can’t be satisfied. Millimeter wave (mmWave) spectrum with rich bandwidth can effectively cater for the suging traffic demand. Moreover, edge cache has been regarded as a promising approach for relieving the backhaul pressure. In this paper, we first construct a joint maximum distance separable (MDS) coded caching, power allocation and user association problem in mmWave enabled satellite Internet of Things (IoT), aims at maximizing the backhaul-aware network utility. The formulated mixed-integer non-linear programming problem (MINLP) is then solved through the decomposition approach. Specifically, the convex optimization technology is used to tackle the coded caching and power allocation subproblems, while the swap matching is implemented to solve the user association subproblem. Moreover, an iterative algorithm with low computational complexity, is designed to implement coded caching and resource allocation to two-sided exchange-stable (2ES) state. Finally, simulation results show the superiority of our proposed algorithm over benchmark schemes.

This paper relates to a security protection architecture for the Internet of Things, which aims to provide a defense in depth system with security genes to enhance the security and anti-attack ability of the Internet of things. The architecture is based on six security genes, including user, device, data, location, application and connection. By integrating trusted computing, identity mapping, mimic defense and other technologies, it realizes intelligent identity authentication, separation of access network and core network, and linkage of identity authentication and authorization. The architecture has the advantages of strong adaptability, strong security protection ability, high efficiency and reliability, and strong backward compatibility. Through this invention, the attack surface of the Internet of things can be effectively reduced, and the security and stability of the Internet of things can be improved.

Internet of Things (IoT) has demonstrated tremendous advantages in various industry and research fields. The IoT device number rapidly increased, followed by more serious safety hazard. These anomalies can influence the performance of system or even worse destroy the function of entire system. Anomaly detection methods are investigated to identify unusual states or malicious behaviors. This paper proposes a deep learning-based anomaly detection model to detect and classify anomalies in IoT. The proposed model is based on Residual Networks and Bi-directional GRU, which can fully utilize the spatial and temporal features of network traffic data. Moreover, attention mechanism is utilized to extract key features to improve the classification performance of the model. Experimental results show that the proposed model has better detection and classification performance.

Beidou space signal quality monitoring and assessment is one of the important means to ensure the normal operation of satellite navigation system. It not only provides space satellite status information support for ground operation and control system, but also provides important Beidou service quality information for the general users. In this paper, the Beidou space signal quality monitoring and assessment capability model and system mode, monitoring and assessing processes and architecture, performance evaluation method was proposed. The power spectrum deviation, the correlation loss less, S-curve deviation of BDS based on real data was assessed.

This paper proposes a simplified MIMO detection method for 5G reduced capability (Redcap) user equipment (UE) with high order modulation, which achieves a compromise between computational complexity and detection performance. With the freezing of the 5G Redcap standard for internet of thing (IoT) scenario applications, significant cost reduction is achieved by reducing UE capabilities. Redcap UE specifies that the maximum number of receiving antennas is 2, and the maximum modulation order is 256. Regarding this regulation, the MIMO detection method of two receiving antennas with 256-order quadrature amplitude modulation (256-QAM) is studied, which can improve detection performance and ensure low computational complexity. In existing works, QR decomposition (QRD)-M is a high-performance MIMO detection method, but its complexity is still high with 256-QAM. This paper proposes a further simplified MIMO detection scheme with 256-QAM based on QRD-M. The simulation results show that this method has better performance than linear detection, and has only a slight decrease in performance compared to QRD-M. This method also has reference value for MIMO detection of larger than two receiving antennas with high order modulation.

To satisfy the requirements of low phase noise, millimeter-wave broadband signals proposed by modern electronic systems, a frequency synthesis circuit design using cascade comb generators with a high-efficient mixing scheme is proposed. By the adoption of the highly integrating process, this frequency synthesizer is achieved in 110 mm × 80 mm × 20 mm, which covers 29 GHz–39.5 GHz. Within the operation frequency band, the phase noise is less than −100 dBc/Hz@1 kHz and less than −106 dBc/Hz@10 kHz. The spurious suppression is better than 60 dBc.

With the improvement of satellite sensor resolution, the data transmission rate also gradually increases. The data transmission system has adopted the high-frequency band to carry a higher data rate, and it needs to support high-order modulation modes such as 8PSK (8 Phase Shift Keying), 16APSK (16 Amplitude Phase Shift Keying), and an effective coding method with LDPC (Low Density Parity Check Code). The data transmission system in the satellite on the circular orbit is composed of baseband equipment, a power amplifier, and an antenna. The farthest distance of satellite to ground is considered under the low elevation of the ground station. The link allowance ensures that the whole arc meets the requirements. The link boundary conditions of the satellite on the circle orbit cannot be fully applicable on the elliptical orbit, which needs to adjust the system design method based on the data transmission system. We proposed a communication system on the elliptic orbit, in which the antenna’s widest beam angle is the coverage angle to the ground in the orbit perigee, and the EIRP (Effective Isotropic Radiated Power) in the orbit apogee is the maximum value. The simulation has ensured that a high-speed data transmission system can meet the elliptical orbit satellite data transmission in any access arc.

In power grids, since the synchrophasor measurement protocol lacks a built-in encryption mechanism, measurement data transmission and communication are vulnerable to false data injection attacks (FDIA) that can cause the state estimation (SE) results to deviate from the ground truth. The existing test environments are insufficient to validate the consequences of the vulnerability. Therefore, this paper proposes building a testbed based on a real-time digital simulator to reproduce SE process with synchrophasor data for the grid. The testbed encompasses multiple remote terminal units and phasor measurement units data transmission to the substation in the IEEE C37.118 or Modbus protocol. Based on these synchrophasors and other measurement information like voltage and power flow, nonlinear SE results are calculated satisfying physical laws including power flow balance and Kirchhoff laws. Honest Gauss Newton method is utilized to achieve more accurate state estimation outcomes, with the objective of minimizing the sum of squared error between measurement data and the estimated value. Finally, several stealthy FDIA experiments are conducted on this testbed considering the security issues of the data transmission protocol. The results show that these attacks can successfully bypass the residue-based bad data detection and falsify the grid states, leading to uneconomic and even insecure grid control and dispatch decisions.

White LED light source is used in RGB visible light communication (VLC) technology, composed of red, green and blue, the three basic colors for data transmission. The transmission rate of RGB VLC technology has been achieved more than 10Gbps, Through improved modulation technology and reception algorithm. However, the spectrum utilization efficiency of RGB visible light communication is low. Weak anti-multipath interference ability and serious nonlinear distortion are also problems. In this paper, Orthogonal frequency division multiplexing (OFDM) technology is used. Through this technology, the data transmission rate and communication quality are improved. And the theory of RGB visible light communication technology based on OFDM is analyzed, a communication model is established, and the frequency utilization efficiency and anti-multipath interference capability of OFDM are verified by MATLAB simulation.

With the increasing demand for applications such as marine resource development, marine environmental exploration, and ocean cross-medium operations, the demand for interconnectivity in various physical environments is becoming increasingly high, especially for efficient interconnection between air and water environments. However, due to the different characteristics of air and water media, as well as the impact of harsh environments, smooth passage through the air-water interface faces significant difficulties. Currently, the most extensively used method in engineering is to use relays to achieve air surface and water interconnectivity. In this article, we investigated the current research on air-sea cross-medium relay communication and discussed future research directions. We hope to use this article to promote the research of air-sea cross- medium relay communication and achieve an efficient, high-speed, and robust air-sea cross- medium relay communication system.

The improvement of service quality and resource utilization for users in edge areas is the research focus of D2D assisted relay technology. A high-performance SL-DRX scheme applied to 5G networks has been proposed. This scheme introduces PSCCH counters and traffic thresholds to improve the original network communication scheme, in order to solve the SLDRX Inactivity Timer timeout problem. The high-efficient SLDRX scheme proposed by the research institute has innovatively improved D2D relay technology.

Grid Cyber-Physical System (CPS) improves the intelligence of the grid by combining computing, communication and control technologies, but this new grid CPS system may also have some new security risks, such as new types of attacks on the connection between the physical and information networks. In this paper, we propose a deep reinforcement learning-based joint optimization scheme to improve the security and resource efficiency of multiple grid sensors by exploiting physical layer security (PLS) techniques in a scenario where a malicious eavesdropper can wiretap confidential grid information. We use Wyner’s wiretap coding scheme to prevent confidential information from being decoded and eavesdropped by malicious eavesdroppers. We minimize the system processing latency while securing the wireless communication process by jointly optimizing the transmit power of the grid sensor and the allocation of computing resource blocks. The optimization problem in this paper is formulated as a multi-agent cooperative optimal decision problem and is solved using a double deep Q-network algorithm. Simulation results demonstrate the robustness and effectiveness of the scheme in ensuring information security and reducing delay.

Benefit to the advantages of easy deployment and high flexibility, unmanned aerial vehicle (UAV) has been utilized to act as the aerial base station, providing communications service for target areas, such as remote regions and disaster areas. However, with the ever-increasing demand of high-speed and high-quality connections, the efficient multiple access constitutes the main challenge of the UAV communications network. Therefore, in this paper, we propose a hybrid multiple access strategy for UAV communications network, where both the non-orthogonal multiple access (NOMA) and the orthogonal multiple access (OMA) technology are invoked for the sake of efficiently handling the multi-users data-hungry connections. To expound, an adaptive Q-learning based trajectory optimization algorithm is conceived, which is capable of successively solving the problem of user clustering, power allocation and UAV’s trajectory, yielding the maximized achievable throughput. The numerical simulation results demonstrate that the proposed scheme has superiority in terms of average coverage and achievable rate, compared to that of the conventional OMA and NOMA schemes.

Free space quantum key distribution systems apply polarization coding or polarization multiplexing frequently to realize free space channel transmission. In order to ensure the correct detection and decoding of the signals, it is necessary for the optical system possessing a good polarization preservation property. Based on the polarized light transmission theory, a polarization signal transmission model of the optical system under non-approximate conditions is established, and the influence of polarization state of the incident light, the field-of-view angle of the incident light, and the refractive index of the membrane layer to the polarization characteristics of the optical system is analyzed. And the feasibility of improving the polarization maintaining capability of the system by optimizing the parameters of the optical system is discussed. The results show that for the line-polarized light that commonly used in quantum communication systems, the polarization maintaining of the system can be guaranteed over 98%.

In FANETs, failure of any critical node (cut vertex) separates the networks into disconnected components, resulting in a degradation of connectivity reliability. Therefore, it is crucial to detect the critical nodes to ensure connectivity maintenance in FANETs. Since the existing distributed approaches for detecting critical nodes still suffer from high overhead and low accuracy, this paper proposes an efficient geometric-partition-based distributed algorithm for detecting critical nodes using a novel partitioned framework and geometric theory. The proposed algorithm is divided into two phases, the first phase can detect most of the nodes under the partitioned framework using local neighbor information, the second phase further detects the remaining nodes by identifying geometric cycles formed between these nodes. The simulation results reveals that the proposed algorithm further improves the accuracy, can detect critical nodes in large scale networks more efficiently than existing distributed algorithms, with lower energy consumption and faster speed.

Advanced communication network technology is essential for real-time and reliable transmission of Smart Grid Cyber-Physical Systems (CPS) data to ensure the dependable operation of Smart Grid CPS. Combining satellite communication networks with Smart Grid CPS can effectively solve transmission congestion problems and high transmission delays. Low Earth Orbit (LEO) satellites are the primary research object in satellite communication due to their low cost, low transmission delay, and low propagation loss. However, the open space network environment where satellites operate makes it easy for illegal nodes to join the network and threaten effective operation. Rapidly changing network topology in LEO satellites makes it challenging for nodes to establish trust relationships, and traffic distribution in the network is uneven, making quality of service difficult to guarantee. We propose a blockchain based trustworthy load-balancing routing strategy for LEO satellites, introducing identity authentication technology based on blockchain to improve network security. We propose a multi-path transmission load balancing strategy based on deep reinforcement learning for intelligent routing decisions through a round-robin scheme. Experimental simulations show that our proposed strategy significantly improves the performance of average maximum link utilization, cumulative waiting delay, and total business transmission delay compared to the baseline strategy.

Automatic modulation recognition (AMR) is one of the essential techniques and a di cult challenge to crack in non-cooperative communication systems. Attention mechanisms have been widely applied to deep learning- based (DL) AMR, and its effectiveness has been proven. However, these methods still have problems of high complexity and low accuracy. This letter proposes a high-performance, lightweight framework combined with multi-scale grouped convolution (MGC) and second-order channel attention (SCA), named SCA-MGCLSTM. The MGC structure ensures that channel independent multi-scale depth features are extracted while considerably reducing the number of model parameters. Meanwhile, unlike general attention mechanisms that use first-order information, SCA leverages second-order information from feature maps to obtain more effective attention scores. Experiments on benchmark datasets show that our model outperforms existing deep learning methods regarding training speed and recognition accuracy.

To meet the design requirements of in orbit assembly spacecraft, this article proposes a method of reconstruction for the Spaceborne information network system for modular spacecraft in orbit assembly. The core of this method is to select a dedicated node on the SpaceWire routing chip as the information processing center for the link construction of functional module processing system. In order to handle the changes of system hardware structure, a series of dedicated link construction command set and specific execution logic are designed for system information network building. This method is also effective for in orbit module assembly and reconstruction of the modular satellites.

Massive MIMO is a key technology for next generation communication system, it can greatly increase the system sum-rate while the radiation power is significantly reduced. Different beam allocation, beam training, power allocation, and joint beam and power allocation algorithms have been employed with multiuser massive MIMO systems. The goal of this article is to present an overview of current research topics and future trends about beam and power allocation in massive MIMO. Specifically, beam selection, beam training, power allocation, and joint beam and power allocation algorithm have addressed. The discussed allocation schemes play a key role for allocating the beams and powers in such a fashion that system data rate has maximized and power consumption has decreased. Furthermore, the list of references is a good incentive for future researchers to work in this emerging field.

Traditional satellite networks should not guarantee Quality of Service (QoS) due to unbalanced resource utilization and high load. Therefore, a routing algorithm based on load balancing is proposed, and a genetic ant colony algorithm is used to guarantee multi-constrained QoS. Firstly, the potential traffic demand of the whole network is predicted and the heuristic function is optimized. Then, the path cost and pheromone update strategy are improved. Finally, the optimal path satisfying load balancing and QoS constraints is selected. Through simulation experiments, it is found that the proposed algorithm can effectively balance the service load of the satellite network, and significantly improve the performance in the aspects of end-to-end delay and packet loss rate.

To ensure the reliable communication in mobile ad hoc networks (MANETs) with highly dynamic environment, this paper investigates cross-layer routing problem to minimize the system average packet delivery delay. Firstly, we construct a multi-agent cross-layer routing framework, where each node learns its routing policy cooperatively based on local observations. Secondly, we construct a decentralized partially observable Markov Decision Process (Dec-POMDP) by modelling the cross-layer routing problem based upon cross-layer partially observable environmental information. Then, we utilize multi-agent framework and employ reinforcement learning (RL) method based on Deep Graph Neural network (DGN) to incorporate the observations of neighboring agents with the graph attention convolutional kernel, and use the method of experience replay (ER) and target network for network stabilization and model training. Simulation results show that our proposed algorithm can achieve a 46.6% improvement in cumulative reward compared to the baseline without utilizing DGN, and exhibit higher performance and enhanced stability than the baselines when the number of agents increases.

With the continuous development of 5G network scale, the traditional network planning method is no longer suitable for today’s needs. In order to meet the performance requirements of users on network coverage, service capacity, number of connections, mobility, network stability and other aspects, 5G network planning and construction must be more reasonable and more accurate. Therefore, this paper studies the method of 5G network planning based on a UMA urban macro station model. First, according to the parameters such as the market proportion of 5G operators, the penetration rate of 5G terminals, and the configuration of 5G frame structure, the capacity modeling is carried out to reasonably derive the capacity requirements of 5G networks. Secondly, according to the parameters such as transmission power, power loss and antenna gain, the coverage modeling is carried out, and the coverage requirements of 5G network are reasonably derived. Thirdly, according to the capacity and coverage planning theory, the upper computer application program is derived, modeled and designed, which can be used for reference by 5G network planners. This research method can accurately and scientifically evaluate and plan the capacity and coverage of 5G network, and has certain application and promotion value for 5G network construction.

The future applications of mobile communication require lower out-of-band (OOB) radiation and higher spectral efficiency. The widely used orthogonal frequency division multiplexing (OFDM) system suffers from severe OOB radiation, sensitivity to timing and frequency offsets, and the requirement for uniform waveform parameters across the entire frequency band. To adapt to the flexible and diverse nature of new application scenarios in mobile communication, a filtering multi-carrier modulation system based on time-frequency blocks is proposed. This system allows flexible adjustment of the occupied time-frequency block size according to the characteristics of the desired services, thereby controlling the peak-to-average power ratio (PAPR) and transmission delay of the system. Moreover, the introduced filter effectively suppresses sub-band spectrum leakage. Compared to the OFDM system, simulation results in additive white Gaussian noise channels and Rayleigh fading channels show that the proposed multi-carrier system exhibits better OOB radiation and PAPR, while achieving comparable bit error rate (BER) performance.

In this paper, a relay-aided non-orthogonal multiple access (NOMA) vehicle-to-vehicle (V2V) system model is proposed. The outdated channel state information (CSI) is derived based on the three-dimensional two-cylinder channel model. And the residual interference is derived using outdated CSI to characterize the degree of successive interference cancellation (SIC), i.e., perfect SIC and imperfect SIC. A power distribution scheme was presented to maximum the sum achievable rate of relay-aided NOMA-V2V system under consideration of residual interference. The effects of power allocation coefficient, imperfect SIC, transmission power ends on the achievable rates are analyzed for the relay-aided NOMA-V2V system. It provides theoretical guidance for V2V network on which multiple access method should be chosen.

Blast furnace is the core equipment of iron and steel smelting. Traditional inspection mainly relies on manual, the remaining problems include high labor intensity, low efficiency of inspection, inadequate inspection, and difficult digital display of inspection results. With the development of technologies such as UAV and online monitoring and diagnosis and their in-depth application in the field of inspection, firstly, an intelligent inspection business model of “UAV inspection + infrared scanning + data application and visual display” was introduced, then 5G and UAV were applied to temperature measurement of ironmaking blast furnace, blast furnace pipe, hot blast stove. Secondly, in order to realize the safety and stability of blast furnace production, StressWave analysis technology was applied to equipment predictive maintenance, especially for condition monitoring and fault diagnosis. In the specific application case of the gas-seal box and belt conveyor, StressWave on-line condition monitoring system was applied to listen for shock/friction raising events and quantify energy from shock and friction, through comprehensive analysis of on-line condition monitoring data to diagnose fault type and fault severity of gas-seal box and belt conveyor. The accuracy of diagnosis conclusion was verified in the application cases. Finally, this research content and application cases promote application of 5G+ condition monitoring technology in predictive maintenance of ironmaking blast furnace, through effectively improvement of inspection efficiency and quality to provide guarantee for the stable operation of ironmaking blast furnace.

Traditional multi-robot path planning methods have limitations, such as difficulties in handling complex environments and long planning paths. To address these problems, a path planning algorithm based on conflict search and safety corridor as constraints is proposed for cooperative control of multiple robots, and MPC is used to solve the nonlinear optimization problem. The method overcomes the limitations of traditional methods and can complete the path planning task faster and achieve good results in complex environments. Specifically, a safe driving path is first planned between the start and end points of the environment map, and then a safe corridor is constructed on the safe path and the path planning parameters are optimized based on it. By viewing the robot’s status and motion trajectory in Rviz, it is verified that the system can accomplish tasks such as multi-robot path planning and cooperative obstacle avoidance. Compared with the conflict search approach, the average reduction of CBS path length after MPC optimization is 2.472% and the average reduction of ECBS path length is 2.581% after the introduction of the safety corridor constraint. The experimental results show that the proposed method can effectively reduce the total length of multi-robot path planning.

In this paper, an improved lightweight YOLOv5 based on a face target detection algorithm is proposed. First, build a small data set on top of the public data set. Next, the anchor frame is reset and the data annotation is standardized they improve the quality of the data set and improve the accuracy and speed of the model. The simulation results showed that the accuracy and mAP value of the improved YOLOv5 network reached 95.52% and 95.15%, respectively. They both have a 15% increase, and the computing speed has also increased by 19.00%. At last, this paper realizes faster and more accurate detection than the traditional YOLOv5 network.

Aiming at the problem of fast and accurate curve fitting required for aircraft direction trajectory data points, this paper uses the cubic spline function instead of the weights in the MLP neural network, and proposes a method for fitting aircraft trajectory data points using the spline weight function MLP neural network. The paper firstly introduces the MLP neural network as well as the spline weight function. Next, the cubic spline weight function equation is derived. Finally, the method of this paper is compared with the neural network method by performing curve fitting experiments on discrete data points of an airplane trajectory. The results show that the method in this paper reduces the error by 6.9% compared with the neural network method, which is closer to the actual airplane flight trajectory.

Automatic guided vehicle multi-target point navigation plays an extremely important role in logistics and transportation, industrial automation, warehouse management and other industries. Multiple navigation points are abstracted as points in a two-dimensional raster map, and each navigation point is assumed to be a city point, thus establishing a mathematical model for travelers. This paper proposes a custom distance calculation algorithm to calculate the global path length between navigation points. An ant colony algorithm is used as the multi-objective navigation optimization algorithm, and the global search capability is enhanced by using Levy flight in the construction of the solution part, and a loga-rithmic function is introduced to make the overall step length of Levy flight change dynamically. For the problem that the ant colony algorithm is prone to fall into local optimum, two local search algorithms are used to search in turn, incor-porating the advantages of different local search algorithms. The improved ant colony algorithm reduces the error by 0.05%-1.46% compared with the basic ant colony algorithm plus one local search. Finally, the feasibility of the whole sys-tem design is verified by using Turtlebot3 in a joint ROS-Gazebo simulation.

This paper studies the small sample target detection based on SAR images. Aiming at the problem of target semantic information loss in small sample targets, the residual network structure is optimized and improved based on YOLOv3 algorithm, and the data enhancement method is used to increase the number of small sample ships in the dataset. In this paper, three different ship detection data sets are processed, and the public datasets of multi-source SAR satellite ground object types are classified according to the different shooting satellites and polarization methods. A total of ten standard VOC datasets suitable for different algorithms are produced. In this paper, the constructed algorithm is compared with three comparison algorithms. The results show that although the detection performance of YOLOv3 is better than that of RCNN series, the detection accuracy of the two for small sample targets needs to be improved. Aiming at the small sample ship target with complex background in the data set, the detection effect of our algorithm is better. The mAP is used to verify the detection accuracy. The results show that the improved algorithm’s mAP is 2% higher than others.

In this paper, we give two methods to estimate common remainder from its residue set with errors. Two types of redundancy schemes are considered simultaneously, i.e., redundant residue number system (RRNS)and remainder redundancy. The proposed algorithms have two main steps. First, the proper cluster is obtained by deleting half of the redundancy remainders. Based on the clustered residual set, two algorithms are proposed by searching based method and mean based method, respectively. Simulations show that the two proposed algorithms are better than no-cluster based method. In addition, the proposed mean based algorithm has more advantages, especially in terms of computational complexity.

In today’s information age, with the rapid development of network technology and a large number of applications of sensors in daily life. The number of various types of data has shown a great growth. These data not only bring us convenience and benefits, but also pose a great challenge to us, that is, how to explore the information of the data and effectively analyze the specific significance contained in the data has become a hot research direction. In the large amount of data obtained, most of the data are in the normal state and only represent the basic “value” information. Compared with the data in the normal state, the information carried by a small part of the data is more worthy of attention. The emergence of these data is the so-called “exception”. The occurrence of exceptions indicates that the system that should be running normally has changed, and these changes often have a negative impact on the system. Looking at these applications, it is not difficult to find that the actual frequency of these anomalies accounts for a very small proportion compared with a large number of normal time, but its value is very huge. Therefore, real-time online anomaly detection of convective data has very important value and significance for scientific research and industrial application. In the process of this research, I have a more comprehensive and detailed understanding of the research field of “anomaly detection of stream data” through theoretical learning, apply the deep learning model to anomaly detection of stream data, and test the performance of the model through experiments. By comparing the performance of traditional anomaly detection algorithms, unsupervised and semi supervised machine learning models and neural networks such as LSTM/HTM in stream data anomaly detection, we try to find an algorithm that can detect stream data information in real time and provide anomaly alarm in time.

SQL injection attacks are a commonly used network attack method. To effectively detect and prevent such attacks, this paper proposes a SQL injection detection method based on a knowledge base of error codes associated with SQL injection. The proposed method is comprised of three main components: a data preprocessing module, an automatic detection feature extraction module, and a design module for the error code knowledge base. Using the error code knowledge base, the input SQL statements are matched in real time. As soon as a successful match is detected, the system promptly identifies it as a SQL injection attack and initiates the necessary response measures. By accumulating new error codes, the detection model can be further trained on new samples, thereby enhancing its recognition ability and expanding the detection range of the model. Based on experimental results, the error code knowledge base method achieves an accuracy of 97.34%. Furthermore, it maintains an accuracy of over 96% when tested on a new data set. When compared to traditional feature detection methods, it shows higher accuracy, precision, and recall rates.

The rapid increase in satellites deployment numbers and schedules for low-earth orbit mega-constellations have brought to the fore the issue of limited orbital resource in recent years. But the exploitation of the intrinsic capacity estimation of orbit-based resources has consistently posed a challenge in terms of reaching a clear and unified conclusion. We are attempting to research this. Our analysis covers two key perspectives, static distribution optimization and dynamic safe rendezvous, offering a comprehensive understanding of the issue. We then employ the spherical gradient descent method to conduct theoretical derivation, followed by experiments and numerical analysis, including dynamic simulations of rendezvous distances. Through these, we can make the estimation of maximum number of satellites within a single shell while ensuring a safe distance of at least 50 km at any given time. Simulation results show that the optimization can increase the single-shell capacity while also increasing the safety distance. Finally, based on the experimental results, recommendations for the constraints on the low-earth orbit capacity and the total number of deployable satellites are proposed.

Nowadays, aerial images of intelligent aircraft are widely used in all aspects of production and life. However, due to the limitations of airborne equipment, aerial images often have problems of low precision and high noise. Although, super resolution (SR) based on convolution neural network (CNN)can solve the above problems to some extent, huge number of parameters and computational overhead make these algorithms difficult to deploy on onboard computers. To address such limitations, a multi-scale residual attention and distillation network (MRADN) is designed. Firstly, with the aim to maintain the network lightweight enough, a multi-scale distillation network using depth-wise separable convolution (DSC) is proposed. Then, a multi-scale residual channel attention block (MS-RCAB) is designed, which leads to the network pays more attention to high-frequency details. What’s more, for the purpose of using feature information from different scales, a multi-scale attentional feature distillation block (MS-AFDB) is constructed. Compared with the networks in recent years, MRADN has advantages in parameters, computational complexity, processing speed and accuracy which has been proved by large number of experiments. Furthermore, experiments on self-built dataset has determined that this network is suitable for aerial images.

With its flexibility and active defense capabilities, the power grid buffer network has attracted widespread attention as a novel means of power grid defense. This article proposes an intelligent active defense method specifically designed to mitigate penetration attacks on power grid buffer networks. In this method, attackers typically employ intelligent penetration attacks based on reinforcement learning, which model the penetration process as a Markov decision process. Attackers continuously train themselves through trial and error to optimize their penetration paths, thus enhancing their attack capabilities. To prevent malicious exploitation of intelligent penetration attacks, the power grid buffer network introduces a deceptive defense method aimed at countering attack strategies based on reinforcement learning. This method first gathers necessary information (state, action, reward) during the construction of the attack model by attackers. It then generates deceptive actions through state dimension inversion and confuses attackers by flipping reward value signs, thereby implementing deceptive defense at the early, middle, and late stages of penetration attacks on the power grid buffer network. Finally, this article conducts simulation experiments to compare the defensive effectiveness of the proposed method in three stages of the power grid buffer network’s defense against intelligent penetration attacks. The experimental results demonstrate that the proposed method reduces the success rate of intelligent penetration attacks based on reinforcement learning.

Multi-modal expressions in social media contain richer information and have a wider spreading effect of fake news. Therefore, how to effectively use multi-modal information to accurately extract the representation features of fake news and detect them timely has become an urgent research task to be solved. Compared with single-text content, the difficulties faced by multi-modal fake news detection tasks mainly include: (1) extracting pertinent features to accurately represent fake news across various modalities poses a challenge; (2) there is a lack of unified feature representation methods to correlate multi-modal features such as text and image. To address these challenges, we propose a Multi-modal Pretrained Model (MPM) for detecting fake news by incorporating multimodal information. Extensive compared experiments were conducted on a multimedia dataset from Twitter named MediaEval2015. The experimental results demonstrate that the detection accuracy of MPM reached 91.8%, which is 21.8% better than the unimodal detection method and 41.7% better than the multimodal baseline model. The results verified the feasibility of incorporating multimodal information and the effectiveness of MPM.

In the face of ever-evolving challenges in agricultural disease management, particularly for the vital potato crop, this research endeavors to harness the power of Bayesian deep learning techniques for accurate and robust disease diagnosis. Drawing upon the Plant Village dataset, we developed and juxtaposed multiple models to discern the efficacy of uncertainty quantification using Monte Carlo Dropout (MC Dropout). Our comparative analysis across models emphasizes the profound impact of MC Dropout, underlining its superiority in enhancing model performance and reliability. The models enriched with MC Dropout not only demonstrated high diagnostic accuracy but also provided invaluable insights into prediction uncertainties, thereby bolstering the trustworthiness of the diagnosis. This study substantiates the promise of Bayesian methodologies in agricultural deep learning applications, laying the groundwork for future research that seeks to seamlessly merge precision with reliability in crop disease detection.

This paper presents a load transfer strategy generation method for intelligent distribution network based on prediction.Obtain section data of system operation from power distribution monitoring system. According to the primary model information and equipment operation status of the system, the topology structure of the system operation is calculated through the topology service to provide a basis for load transfer. Obtain the system prediction results from the prediction system, input the topology and prediction results of the system operation into the load transfer expert system, and calculate and generate the transfer strategy according to the weight configuration of the transfer. The transfer strategy is stored in the database or displayed on the interface to provide decision-making basis for operators.

In order to solve the problems faced by specialized vehicle charging stations such as public transportation and logistics in the construction and operation process, such as relatively cumbersome charging facility access, complex charging operation process, single settlement type, and lack of vehicle assessment methods, etc. This article proposes the idea of building a dedicated vehicle charging service platform based on the Internet of Things. The demonstration application results indicate that the technical solution of the dedicated charging service platform effectively solves the problems faced by dedicated vehicle charging, and the service platform provides important theoretical basis and application exploration for the development of specialized vehicle charging service business, and has important research and application value.

Objective: A random forest algorithm was used to analyze non-intellectual factors that directly affect student achievement at the K-12 level and provide targeted strategies for addressing these factors. Methodology: Student learning data from the Kalboard 360 Learning Management System were selected. Non-intellectual influences on student performance were assessed using a single-factor analysis and and random forest models to rank the importance of independent variables and the scores were categorized into three levels: high, medium, and low, for independent analysis. Results: The single-factor analysis revealed 11 non-intellectual factors that were statistically significant (P < 0.05). In the ranking of importance, the three predominant variables influencing academic performance are the frequency of course access, the number of hand-raising instances in class, and the grade level of absenteeism. The frequency of course access dominates the high score bracket, the number of instances of hand-raising in class dominates the medium score bracket, and the grade level of absenteeism dominates the low score bracket. Conclusion: School teachers should focus on non-intellectual factors besides traditional teaching techniques and adopt strategies such as providing rich online resources and motivating students to learn. Through this, educators can improve academic performance from a fresh perspective.

Amidst the ongoing advancements in modern industrial automation and intelligent manufacturing, the inspection operations of underground substations have become a pivotal research domain. The traditional manual inspections, hampered by inefficiencies and potential misjudgements, are gradually falling short of contemporary demands. To address this, our research focuses on harnessing the object detection and classification capabilities of deep learning. Specifically, we centered our study on YOLOv8, analyzing and experimenting with its application in image recognition for underground substations. The experimental data suggests that YOLOv8 exhibits exceptional proficiency in tasks like target detection, image segmentation, and classification. For instance, it achieved commendable results in switch status detection, readings of pointer instruments, and binary classification of knife switch statuses. Overall, deep learning introduces an innovative, accurate, and efficient detection method for underground substations, laying a robust technological foundation for future intelligent manufacturing and automation.

As satellite remote sensing and aerial photography technologies continue to advance in recent years, there has been a noticeable increase in both the resolution and image quality of remote sensing images. Furthermore, an abundance of data sources has emerged, intensifying the challenges associated with detection. To address the challenges posed by small object size and dense distribution in remote sensing images, an innovative solution has been introduced. This solution entails an enhanced rotating object detection algorithm which leverages the power of vision Transformer technology. By utilizing this approach, the aim is to overcome the limitations of poor robustness and low detection accuracy commonly encountered in such scenarios.The enhancement of the feature extraction capability of the detection algorithm in YOLOv4’s feature fusion part is achieved through the introduction of the MS-Transformer module. This module, known for its self-attention mechanism, facilitates the acquisition of pertinent information among targets, thereby bolstering the algorithm’s ability to detect densely distributed targets. Moreover, the advancement of the five-coordinate YOLOv4 object detection framework enables the realization of multi-angle remote sensing object detection. To mitigate the issue of overlapping prediction frames on dense targets, the model incorporates the soft-NMS suppression method, ultimately refining the detection performance. The efficacy of the proposed algorithm in improving the model’s detection capability is substantiated through experimentation using the DOTA dataset.

In various tasks of self-supervised monocular depth estimation, the prevalent norm has been the adoption of a U-shaped network structure, utilizing the convolution module as the foundational operator, leading to notable accomplishments. However, the inherent limitation of convolution operations, characterized by a restricted receptive field, often poses challenges in explicitly capturing long-range dependencies. The Transformer, originally designed for sequence-to-sequence prediction, presents a global self-attention mechanism capable of capturing long-range dependencies but may compromise localization abilities, lacking in low-level details. This paper introduces TransIndoor, a robust alternative for self-supervised monocular depth estimation that combines the strengths of the Transformer and convolution. TransIndoor effectively extracts both global context information and local spatial details simultaneously. Furthermore, a novel local multi-scale fusion block is introduced to enhance fine-grained details by processing skipped connections within the encoder through the primary CNN stem. The comprehensive validation of TransIndoor using the NYU Depth V2 dataset and ScanNet demonstrates its capability to generate satisfactory depth maps, addressing the limitations of existing methods.

Citation networks, as an important type of graph-structured data, have been widely applied in fields such as academic research, scientific collaboration, and patent analysis. In this study, we construct a large-scale citation network dataset with rich citation relationships based on public datasets such as CNKI (China National Knowledge Infrastructure). We utilize Graph Convolutional Network (GCN) [1] for efficient classification and analysis in the domains of machine learning, deep learning, and neural networks. The experimental results demonstrate that our approach not only enhances the accuracy of citation network classification but also effectively captures complex relationships and local features among nodes, offering practicality and application value.

To excel in image super-resolution, deep neural network-based models often employ a stacking strategy of network modules. However, this approach leads to parameter explosion and information redundancy, thereby restricting the feasibility of deploying these models on mobile devices. To tackle this challenge, this study introduces a lightweight Dual-path Large Kernel Learning approach, namely DLKL, and applies it to image super-resolution. In DLKL, the first step involves the utilization of multi-scale large kernel decomposition to effectively establish long-range dependencies between pixels and successfully preserve local information. Subsequently, DLKL enhances the feature expression ability and achieves effective feature fusion through its dual-path network architecture. Furthermore, DLKL reduces the number of parameters while maintaining performance, thus striking a balance between network performance and efficiency. The remarkable performance of DLKL has been validated through a multitude of experiments by quantitative metric tests and visual evaluations. Comparative analysis, conducted against other prevailing algorithms, provides evidence of the DLKL method’s consistent proficiency in generating images featuring enhanced texture clarity and more faithful representation of natural structures. These findings serve to reinforce the method’s inherent superiority and robustness.

In this paper, a real-world denoising method for surveillance videos based on the trilateral filter is proposed. The algorithm is faced with challenges, including the absence of “ground truth” images and the complexity of the spatio-temporal distribution of the noise signal. However, experimental results have demonstrated that noise on stationary objects in such situations can be easily eliminated by averaging neighboring frames. Consequently, effective noise removal throughout the entire video can be achieved by accurately tracking and filtering moving objects along their trajectories. The model can be broadly divided into four steps. Firstly, coarse motion vectors are obtained through bilateral motion estimation. Secondly, the error vectors are judged and corrected using an amplitude-phase filter. Thirdly, these vectors are refined by performing a full searching in a small area. Finally, the noise is removed by applying a trilateral filter along the trajectory. The effectiveness of our model has been confirmed through numerous experiments, showcasing superior performance in both visualization and quantitative testing.

In recent years, the development of mobile internet and big data has propelled the digital transformation of the finance industry, enabling inclusive finance across society. Meanwhile, financial fraud detection technologies need continuous updates and improvements to combat new fraud tactics. Currently, the development of anomaly detection techniques in the traditional finance industry lags behind the pace of financial digitalization. In addition, emerging technologies lack practical application in financial scenarios. Therefore, we test the anomaly detection performance of two graph neural networks, GCN and GAT, on the DGraph dataset and compare with the MLP model. Experiments demonstrate that graph neural networks outperform the fully connected network and achieve good performance on financial anomaly detection.

Estrogen-related metabolic enzyme gene polymorphisms have been demonstrated to be linked to breast cancer, and in this paper, a novel noninvasive breast cancer prediction model was developed utilizing machine learning algorithms incorporating estrogen metabolic enzyme gene single nucleotide polymorphisms (SNPs). To precisely forecast the susceptibility to breast cancer,, the coded data of 14 SNPs from enrolled breast patients and normal women were randomly shuffled, with 80% of the data designated as training data, the remaining 20% reserved as the test group to be validated. Single factor analysis was performed to screen independent risk factors, and subsequent application of Breast Cancer with Single Nucleotide Polymorphisms - Machine Learning model (BCSNP-ML) prediction model was completed using Light Gradient Boosting Machine (LightGBM) algorithm. A total of 14 SNPs variables from 280 subjects were utilized in this study. Single factor analysis indicated that a meaningful association between SULT1A1 rs1042028, CYP1A1 rs1048943, CYP1B1 rs1056827, CYP1A1 rs1056836 and the incidence of breast cancer, with 14 variables demonstrates a notable area under the receiver operating characteristic curve (AUROC) of 0.809. The AUROC of the BCSNP-ML model constructed by four variables was 0.831. Additionally, BCSNP-ML is visualized and interpretated in the paper using SHapley Additive exPlanations analysis to further validate that the model exhibits great potential as a robust tool for clinical forecasting of breast cancer.