This PDF file contains the front matter associated with SPIE Proceedings Volume 12970, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

Aiming at the problem that a large amount of water surface garbage causes serious pollution to the environment and is difficult to clean up in time, artificial intelligence equipment and machine vision target detection are combined. A deep learning-based lightweight floating object detection algorithm MobileNetV3-YOLOv4. MobileNetV3-YOLOv4 proposes to use a depth wise separable convolution method to replace the 3 × 3 conventional convolution in the network, replaces the original CSPDarknet53 backbone feature extraction network with MobileNetV3 network, and proposes to use a synthetic fusion pyramid network (SFPN) in the neck network of the model. A synthetic fusion Pyramid Network can reduce the gap between the detection accuracy and the original model while reducing the number of parameters and calculations. The number of parameters and calculations is only 20% of the original model, and the detection speed reaches 96 frames per second. Compared with the original YOLOv4 model, it is 69 frames per second, an increase of 39.1%.

Driver drowsiness detection is essential in the field of intelligent driving and should be solved timely. Although the application of convolutional neural networks has brought about great progress in this field, they do not perform well in complex driving scenarios due to their inability to extract comprehensive spatio-temporal information well. In this paper, a Hybrid model using Graph networks and Long short-term memory networks for Drowsiness Detection (HGLDD) is proposed for the first time to fuse the driver’s facial depth information and head posture information together with eye and mouth information from facial landmark sequence. The model extracts spatio-temporal depth features and determines whether or not the driver is in a drowsy state. On the Nthu-DDD dataset, the proposed model ultimately achieves an average accuracy of 98.01%, demonstrating its applicability to drowsiness detection tasks in real driving scenarios.

To improve the stability of high-proportion power electronic systems for power line communications and reducethenoise effect of power electronic equipment operation, this paper analyzes the principle and architecture of the deepconvolutional neural network (DCNN) algorithm. The noise modeling method based on the prediction parameters ofDCNN is proposed to reduce the reconstruction error and improve the noise immunity. Simulation results verifiedthat the proposed algorithm achieves superior performance in peak signal-to-noise ratio (PSNR) and reconstruction error.

Traditional mathematical statistics and machine learning models have limitations in soil moisture prediction. This study offers an algorithm based on XGBoost under correlation for soil moisture prediction at a depth of 10 cm, which can achieve an accuracy of 99.29 %. It has a greater advantage over several other machine learning models. The model is also optimized by Grid Search and Randomized Search for the model hyperparameter, and its model is proven to have good generalization ability and robustness.

Space-borne bistatic radar has stronger viability than monostatic radar. And larger radar cross section (RCS) makes anti-stealth capability of space-borne bistatic radar stronger. However, space-borne bistatic radar is also in down-looking working state, which will suffer serious clutter interference. Since the transmitter and receiver are not placed on the same platforms, bistatic configurations cause clutter characteristics more complicated. It is mainly manifested in resolution spatial variation and tanglesome space-time distribution. In this paper, we first establish the signal model of spaceborne bistatic radar based on satellite-earth relation. Then, we focus on arbitrary orbital plane configuration to study resolution spatial variation and space-time distribution characteristics against bistatic geometric relationships. Last, we adopt the full link evaluation model to evaluate the performance of space-time adaptive processing (STAP) under corresponding bistatic geometric relationships. This paper can provide significant references for system design of space-borne bistatic radar.

For the long-range and high-speed maneuvering space target detection, its performance is constrained by two challenges, the coherent integration loss due to Range migration and Doppler frequency spread, the contradiction between Range Ambiguity, and Doppler bandwidth ambiguity. This paper proposed a novel parameter estimation and integration method to solve the above two issues. Firstly, the waveform diversity technique is used to solve the range ambiguity problem. Then a matched filter bank for high-speed targets that can calculate the range ambiguity number is designed. Next, coherent integration is successfully achieved by estimating the target motion parameters and compensating for the echo signal. Finally, a two-step search strategy is proposed to improve the parameter estimation process, numerical experiments have validated that the complexity of the algorithm is reduced while ensuring the accuracy of the parameter estimation.

Reduced-dimension (RD) space-time adaptive processing (STAP) technique has achieved good clutter suppression performance in real data processing. However, the traditional F$A method suffers from performance degradation in presence of clutter fluctuation. Extended F$A method (E-F$A) is an effective approach to improve the performance in such a clutter environment, but it requires high computational complexity and sample support. In this paper, a novel reduced-dimension method for E-F$A clutter suppression is proposed. Firstly, we extract characteristics of clutter by tensor Tucker decomposition, which can preserve the structural characteristics of the clutter in the spatial and Doppler domains. Then, we select eigenvectors to construct the RD matrix based on principle components (PC) analysis. Finally, RD data is obtained by multiplying the RD matrix with the original data, and the weight vector for clutter suppression can be calculated. The experimental results based on real measured data validate the effectiveness of the proposed method.

Space-time adaptive processing (STAP) has been extended to the distributed space-borne multiple-input multiple-output (MIMO) radar system to improve the detection performance. Nevertheless, the transmitting waveforms are not completely orthogonal to each other, resulting in the STAP performance deterioration. In this paper, we established a novel STAP performance analysis model to reveal the influence mechanism of waveform properties. Firstly, the signal model of distributed space-borne radar with hybrid baseline and yaw angle was presented. Then, we derived the characteristic of the clutter covariance matrix for post-Doppler STAP, considering the effect of waveform periodic autocorrelation and cross-correlation sidelobe. Finally, the numerical simulation results, based on indicators of clutter eigenvalue spectrum and output signal-to-noise ratio loss, demonstrated the accuracy of the proposed analysis model. The aforementioned results provide important basis for the design of distributed space-borne MIMO radar system.

Aiming at the problems of high manual participation and long time-consuming sequencing in the process planning of typical spacecraft parts, a process route optimization method based on improved hybrid genetic algorithm is proposed. By analyzing the relationship between parts' machining features, the machining method of machining features is disassembled into steps and a step forward graph is established, and the initial process route is generated based on the constraints of state change rules. The optimization objective function considering the use cost and replacement cost of processing resources is constructed. Aiming at the problem of process route decision-making, the coding method of adding step sequence information and processing resource information is proposed. In order to ensure that the process route sequence meets the process rule constraints, the OBX crossover operator is used to construct the adaptive mutation operator, and the improved hybrid genetic algorithm is constructed by using the artificial bee colony algorithm for process route optimization. The feasibility of the method is verified by taking a spacecraft cabin part as an example.

Anylogic software is used for simulation and modeling to evaluate the influence of traffic organization and streamline design on the internal traffic organization of the port area. Taking Yulong Port in Yantai as an example, the average running time of the truck flow network was 1738.6s, with a minimum running time of 751.0s and a maximum running time of 2787.0s. By adjusting the simulation scenario, sensitivity analysis is conducted on factors such as flow changes, gate service efficiency, and loading and unloading efficiency. When the flow of office vehicles increases by 2.0times, the increase in the entry time of freight and office vehicles is more significant. When the efficiency of gate services is doubled, the entry time of office and freight traffic increases by 2.6 and 2.2 times, respectively, resulting in a significant increase in delay time. When the loading and unloading efficiency decreases by 1.5 times, the average operating time of freight vehicles increases by 22.2%, and the loading and unloading efficiency of trucks plays a key role in improving the efficiency of port collection and distribution. The research results can provide a reference basis for the planning and construction of port road traffic facilities.

High-precision load forecasting is a prerequisite for ensuring the safety and stability of power systems. In order to solve the problem of the insufficient generalization ability of a single model, this paper proposes a deep temporal neural network ensemble model based on multi-node short-term load forecasting. It can effectively reduce the prediction error, improve the prediction accuracy and shorten the prediction time to ensure the stable operation of the system. It can also improve the efficiency of clean energy utilization and optimize the level of resource allocation.

The one-bit analog-to-digital converters (ADC) can drastically reduce the system complexity and power consumption, which has attracted considerable research interest in the last decade. In this paper, the problem of one-bit quantized signal direction-of-arrival (DOA) estimation via sparse array is considered. The proposed method first gives an approximate reconstruction method to extend the aperture of sparse array, then a compressive sensing method is presented to obtain accurate DOA estimation. The simulation results demonstrate that the proposed method is well suited for scenarios of few antenna elements and small snapshots.

The dimensionality reduction space-time adaptive processing (STAP) can effectively suppress clutter and jamming in both spatial and temporal dimensions simultaneously. The discrepancy between training data and detection data can make the performance of STAP degrade, we propose interrupted sampling repeater jamming( ISRJ) to counter the joint-domain localized space-time adaptive processing (JDL-STAP). Firstly, the theory of JDL-STAP model is presented. By analyzing the feasibility of ISRJ model, we find that ISRJ can generate numerous false targets with different distributions between sampling data and detection data by changing the retransmitted parameters. And these false targets can destroy the independent identical distribution (IID) of system sample data seriously. Both simulated and measured results corroborate the effectiveness of the ISRJ methods against JDL-STAP.

The DOA tracking problem of MIMO radar under various noises is studied, and a robust DOA tracking algorithm for monostatic nested array MIMO radar is proposed in this study. The nested array is used as the transmitting and receiving array of MIMO radar, which greatly increases the degree of freedom. In the aspect of DOA tracking, the signal subspace is tracked by the αFAPI algorithm, and then the DOA is calculated using the proposed reduced-dimensional root-MUISC algorithm. Simulation results show that the proposed algorithm can effectively track the DOA of targets under the background of additive white Gaussian noise, and its performance is better than that of traditional uniform linear array MIMO radar. Moreover, under the condition of low signal-to-noise ratio and small number of snapshots, the performance of the proposed algorithm is better than that of FAPI algorithm. Under the background of non-Gaussian noise, such as outliers or mixed noise, the algorithm is robust to a variety of complex noises, and shows better angle tracking performance than FAPI algorithm.

Recently proposed, the dual-function radar-communication (DFRC) system aims to tackle the problem of spectrum overcrowding by allowing the joint design of radar and wireless communications. This system employs the transmit beamforming technique and waveform diversity to embed information in radar transmissions. In this paper, a dual-function radar communication system based on directional modulation (DM) is proposed. A set of beam weight vectors are designed to satisfy the requirements of the DFRC system. Three improved beam weight design methods of DM, minimum beam weight difference method (MBWDM) , minimum waveform differential modulus sum method (MWDMSM) and minimum enemy beam gain method (MEBGM) , are proposed for DFRC systems to optimise the DFRC beam by increasing the phase distortion or reducing the beam gain in the non-communication direction, to enhance the security during DFRC communication. Compared to the unimproved DFRC system beam, all three DM methods design waveforms that effectively reduce the correct receive area near the receiver direction and reduce the likelihood of correctly demodulated information in that direction when eavesdropping direction is known. Waveforms generated using MWDMSM maintain high BER performance at all angles in the non-communication zone and therefore also have the ability to defend against enemies of unknown orientation.

This study focuses on the application of elevator door motion recognition, comparing the capability of 2D Convolutional Neural Networks (2D CNNs) and 3D Convolutional Neural Networks (3D CNNs) in extracting spatio-temporal information, further exploring the factors contributing to their differences. The inquiry stems from observations regarding the significant computational power demanded by 3D CNNs during embedded deployment, while the proposed improvements arise from multi-modal information fusion concepts. Through experimental validation, we establish the efficiency of 2D CNNs in motion recognition tasks, employing the computational simplicity of 2D CNNs to match the precision of 3D CNNs. This strategy leads to the introduction of a novel 2D CNN architecture, termed as the "Imitation C3D Network".

Radar radiation identification suffer from the problem that the existing identification method can not process radar radiation data streams which does not in the database. In this paper, in order to solve the problem, we first introduce Online Transfer Learning (OTL) into radar radiation identification field. OTL is a machine learning framework, which aims to attack an online learning task by transferring knowledge from some source domain to a target domain. By extending binary classification online transfer learning to multiple classification tasks, we then consider the problem of training multiple classification online transfer learning using radar radiation data streams. Finally, the simulation results validate the efficacy of radar radiation identification based on online transfer learning techniques.

To begin with, the inductance calculation of planar spiral coils for the wireless energy transmission system of implantable cardiac pacemakers was introduced, with a focus on the calculation of mutual inductance between equally-sized coaxial coils. Then, the effects of primary and secondary coil parameters, such as radial distance and number of turns, on mutual inductance were analyzed. The results show that the mutual inductance declines as the transmission distance increases, and it can reach a maximum value in case of a change in the number of turns.

Peking duck, an EU-China mutual geographical indication protected product, is vulnerable to counterfeiting due to its widespread popularity and high profit. Efficient and intelligent breed identification is becoming more and more important to protect interests of stakeholders along the Peking duck supply chain, as well as the reputation of Peking duck. 160 hyperspectral images from Peking duck and Cherry Valley duck breeds are collected and divided into training and verification datasets in a 3:1 ratio after extracting texture features by PCA and GLCM. SVM with polynomial and radial basis kernel functions, as well as BP neural network models are established to identify duck breeds. Results show the SVM model with radial basis kernel function performs the best with an identification accuracy of 94.17%, and further validation achieving a satisfactory identification accuracy of 95%.

In this paper, we proposed a novel visible and near-infrared image fusion (VIS-NIR fusion) method based on texture filter, aiming at the problems of artifact, color distortion and noise in traditional VIS-NIR fusion methods. Firstly, the structure information of the visible (VIS) image and the near infrared (NIR) image after texture removal is obtained by relative total variation (RTV) calculation as the base layer of the fused image; secondly, Secondly, the image noise weight is calculated by establishing a Bayesian classification model, and the noise information in the VIS information is adaptively filtered by joint bilateral filtering; finally, the fused image is acquired by color space conversion. The experimental results show that the proposed algorithm can preserve the spectral characteristics and the unique information of VIS and NIR images without artifacts and color distortion, and has good robustness as well as preserving the unique texture.

Based on the AIS data, this study proposes a method to improve the accuracy of ship emission model prediction by using interpolation preprocessing and LSTM-CNN model prediction. The method uses the third spline interpolation preprocessing method and LSTM-CNN time-series prediction model to predict pollutant emissions. Firstly, the cubic spline interpolation method is used to interpolate the data at different time intervals. The optimal interpolation interval is selected. Next, use the data interpolated with the optimal interval distance as the dataset, the trajectory and navigation status data of the ship are predicted using the LSTM-CNN model. Finally, emissions were estimated based on the Ship Traffic Emission Assessment Model (STEAM). The study shows that the error between the interpolated emissions and the real emissions is 1.4587%, and the prediction using the interpolated data is 1.2702% from the original emissions, with an error reduction of 0.1885%.

Under the condition of strong interference, sonar equipment faces the problem of ultra-wideband joint perception processing. The solution depends on the collection of multiple sonar equipment, which has the defects of high cost, low integration and low processing efficiency. This paper introduces the basic concept of binaural positioning and separation, and analyzes the research status of binaural positioning and separation for underwater scenes by referring to the mechanism of human binaural high gain and high resolution under the condition of small aperture dual elements. Based on the combination of sonar signal processing technology, it is proposed to carry out technical approaches such as underwater acoustic signal separation technology based on prior information and binaural differentiation, and homologous auditory clue fusion technology, which can be used for the development of binaural positioning and separation technology in underwater scenes.

With the development of artificial intelligence and robotics technology, the research and development of table tennis robots can help athletes to practice table tennis as partners, and can also serve as intelligent coaches for professional guidance. The key problem of the development of table tennis robot is the trajectory prediction of table tennis and the prediction of the ball point in the movement. Through the detection, analysis and modeling of the table tennis trajectory in the movement, the movement trajectory is predicted, and then helps the athletes to find out the counterattack strategy through data analysis, so as to carry out deeper tactical training. The main research content of this paper is to build the table tennis trajectory prediction model, first detect the table tennis detection and track the data set, and then establish the table tennis movement trajectory prediction network through LSTM. After the experimental data analysis, the method of this paper can get more accurate prediction results.

Intelligent Reflecting Surface (IRS) is capable of performing phase and amplitude transformations on incoming signals, thereby achieving precise signal transmission and improving coverage and transmission efficiency. However, these advantages can only be achieved when the precise Channel State Information (CSI) is known. Due to the passive nature of IRS elements, obtaining accurate CSI is challenging. Therefore, a combination of Compressive Sensing (CS) algorithm and Deep Learning (DL) is proposed to address this issue. Firstly, joint-link structure is used to optimize the traditional CS algorithm, enabling better Normalized Mean Square Error (NMSE) performance at lower pilot overhead and Signal-to-Noise Ratio (SNR). Secondly, the channel estimation problem is regarded as a denoising problem, where the output of the optimized CS algorithm is treated as noisy CSI. A multi-layered deep denoising block network is utilized to further denoise the CSI, resulting in more accurate estimation. Finally, experimental results demonstrate that the proposed algorithm outperforms the comparison algorithms under the same SNR, yielding better accuracy.

UAV reconnaissance has become an important means of battlefield reconnaissance and has received much attention in the military development of various countries. Therefore, UAV target detection accuracy is an important factor that restricts the effectiveness of military reconnaissance. How to improve the efficiency of UAV target detection and design high-precision neural network algorithms is the main direction of research in this paper. In this paper, for the current UAV target detection accuracy is not high, the FPG-based Transformer algorithm is proposed, which is mainly based on the Transformer, the Transformer Block is improved, and at the same time, the FPG pyramid structure is introduced in the downstream task, and the results show that, after the improvement of the Transformer, it improves the target detection accuracy, which is conducive to improving the accuracy of UAV reconnaissance.

Quantum computing is a novel computational paradigm that harnesses the fundamental principles of quantum mechanics to perform calculations. In comparison to classical computers, quantum computers can exhibit significant acceleration and storage advantages for specific problems. Inspired by classical adjacency matrices, this paper proposes a quantum adjacency matrix that encodes distances between vertices as phase distances. The quantum adjacency matrix constructs unitary matrices associated with each vertex, which are then decomposed into various phase gates. Leveraging the phase kickback effect, the phase distances are stored in the quantum state's phases. The approach is implemented on a noisy superconducting quantum computer, and distance information is extracted through measurements of the constructed circuit. Due to the complexity of phase rotation operations within the circuit, substantial errors occur on current real quantum computers, preventing the attainment of anticipated results. This paper employs circuit swapping and mapping optimization techniques during the compilation stage. The circuit is compiled multiple times to generate a minimally deep compilation route. In the measurement phase, Qiskit's measurement error mitigation techniques are utilized, involving measurement fitters and filter calibration to mitigate errors in measurement outcomes. These measures facilitate the presentation of expected results on IBM's real superconducting quantum computer.

In recent years, with the development of inter satellite network technology and the flourishing of commercial micro satellites, the number of satellites in orbit has increased sharply, and the problem of frequency interference between satellites and constellations is particularly prominent. Due to the use of the same frequency band in similar satellite services as agreed by the International Telecommunication Union, the issue of co frequency interference between geostationary orbit satellites and non geostationary orbit satellites, as well as between different non geostationary orbit satellites, is becoming increasingly prominent due to real-time changes in relative positions. This article takes the Beidou Navigation Constellation as an example and proposes an interference evaluation method for calculating and evaluating inter satellite co frequency interference through modeling analysis.

With the development of satellite communication technology, the demand for satellite resources for various communication tasks is increasingly complex. In order to further improve the utilization efficiency of satellite resources and better deal with the resource competition between different types of services, this paper proposes a satellite communication task scheduling algorithm based on genetic algorithm. By matching the satellite task scheduling problem with the utility optimization problem, and designing the chromosome coding method and genetic operator matching with this problem, the convergence performance of genetic algorithm is effectively improved. Simulation results show that the algorithm is a fast convergence and efficient satellite communication task scheduling algorithm.

Linear array is one of the most widely investigated domains in array processing. Many techniques including beamforming, subspace methods, compressive sensing techniques, etc. have been developed for years. Uniform linear array (ULA) is well adopted in many scenarios. However, there are several bottleneck issues constraining ULA, which cannot be easily settle. For instance, the maximum detectable number of targets and mutual coupling. Sparse linear array (SLA) can surmount these two physical limitations simultaneously. Mutual coupling can be limited by separating antennas at sparser position, while degree of freedoms is enhanced using the spirit of difference coarray. In this review, the development from ULA to SLA is retraced to provide inspiring points for those who are interested in SLA exploration.

The JESD204B protocol serves as a high-speed serial interface protocol for analog-to-digital converters (ADC) and field-programmable gate arrays (FPGA), providing enhanced data throughput compared to traditional LVDS interfaces. This paper presents an overview of the high-speed serial interface technology based on the JESD204B protocol, offering a comprehensive explanation of the data link synchronization mechanism and data mapping process on existing hardware platforms. A meticulously designed code logic system is employed to facilitate the high-speed transmission of ADC sampling data to the FPGA. Ultimately, the functionality of the designed code logic is assessed through functional simulation and board-level verification, yielding results that validate its ability to meet the application requirements for high-speed sampling scenarios.

Distributed radar is a radar with multiple small aperture antennas scattered, which can greatly improve the physical aperture of radar antenna array. Large aperture antenna can realize high precision angle measurement. This paper studies the high precision angle measurement technology of distributed radar in different working scenarios, so as to deal with multiple measurement scenarios and detect multiple types of targets. When the SNR of target echo is very low, transmitting all coherent signal to realize focusing in space, and using full array antenna to improve the SNR of echo signal. When the SNR is enough, the distributed radar antenna can be coherent clustered, then achieve the measurement, and further improve the accuracy of angle measurement. Distributed radar can achieve maximum precision of angle measurement under different scenarios, which can give full play to its advantage and enhance its usefulness.

Users have various social influences due to their behavior characteristics in the Chinese Sina-Microblog. In order to quantify the impact of different user influences on the spread of public opinion such as public health emergencies, we establish a user-influence susceptible-forwarding-immune (UI-SFI) dynamic information propagation model, which is based on the forwarded quantities data of a real message about COVID-19 vaccine. By the method of binary tree classification, we use the median of followings, followers, and Weibo posts as the boundary to divide users into eight types who have different abilities for the propagation scope of the message. Our Model-based analyses show that the numerical results of the proposed eight types of average exposure rates are in accord with the actual situation. Describing important factors that affect the spread of information, our sensitivity analyses reveal two key types of users and support the strategies for regulating the development of public opinion.

This paper offers an experimental analysis of clutter degrees of freedom in airborne bistatic radar under diverse environmental conditions. It commences by establishing a clutter model tailored to airborne bistatic radar. Subsequently, through a series of experimental simulations, a comparative study is executed to scrutinize the influence of variables including aircraft flight velocity, flight altitude, baseline length, total baseline distance, and flight angles on clutter degrees of freedom. The research results are helpful to the understanding of the clutter property of airborne bistatic radar and the research of clutter suppression methods.

A novel spatial interference suppression algorithm has been proposed to address the challenges of conventional beamforming(CBF) methods, such as wide main lobes, high sidelobes, low resolution, and inadequate interference suppression capability. The algorithm combines the reconstruction of covariance matrices with functional beamforming method. Firstly, virtual interference sources are artificially introduced in the interference area to reconstruct the covariance matrix with added noise, leveraging the principle of automatic nulling in Minimum Variance Distortionless Response (MVDR), which suppresses strong interferences effectively. Subsequently, the reconstructed covariance matrix is subjected to eigenvalue decomposition and integrated with the functional beamforming approach. The eigenvalue matrix is raised to the power of v^-1 to obtain refined weight vectors, which are then employed for beamforming. The beamformed results are raised to the power of v to restore the information of source distribution, thereby enhancing beam resolution. The algorithm demonstrates its capability to effectively suppress sidelobe interference in intense noise environments and achieve high-resolution beamforming through simulation and experimental validation.

Road crack detection holds a crucial significance within the realm of transportation infrastructure management. Its role is instrumental in the preservation and upkeep of road networks, leading to a mitigation of potential accidents and minimizing vehicular wear and tear. The application of deep learning methods in this domain has yielded certain achievements. However, in complex background environments, existing models struggle to effectively extract crack pixels, and predictions regarding intricate details of road cracks lack precision. Addressing these challenges, we propose BSGAUNet(Bypass Supervision Global Attention U-Net) based on the U-Net framework. Global attention module notably combats the interference of background noise, facilitating the accurate extraction of crack pixels. Additionally, bypass auxiliary supervision module enhances the global perception capability of the encoder, enabling the model to more precisely identify crack pixels. To ascertain the accuracy and generalization capacity of the model, we conducted tests on publicly available datasets. The results demonstrated that our model's performance surpassed that of existing models. Furthermore, ablation experiments were employed to validate the effectiveness of the modules.

Vehicle line-pressing detection is of great significance in traffic safety, which is conducive to smooth traffic and avoiding traffic accidents. Currently, most vehicle line-pressing detection is based on 2D object detection, which provides inferior performance on accuracy and robustness. 2D object detection has inherent limitations, such as providing only positional information of vehicles in the form of bounding boxes without accurate vehicle shape and pose information. However, this information significantly impacts the effect of vehicle line-pressing detection. Therefore, this paper introduces 3D object detection technology into the area of vehicle line-pressing detection to enhance the precision of line-pressing detection. 3D object detection can provide more accurate and rich vehicle information such as vehicle shape and pose, and based on this information, a higher precision detection of line-pressing can be performed. In addition, this paper adopts an intuitive and efficient line-pressing judgment strategy. It involves obtaining the mask image of vehicles through 3D object detection and performing a logical "AND" operation with the lane mask image to determine whether the vehicle is over the line. The experimental results show that compared with the method based on 2D object detection, the vehicle line-pressing detection method based on 3D object detection has better accuracy and stronger robustness.

Diffusion strategies play a crucial role in distributed estimation, and with technological advancements, many effective diffusion algorithms resistant to impulsive noise have been proposed, with Diffusion maximum correntropy criterion being one of them. On the other hand, diffusion strategies require each node to share information with neighboring nodes, which introduces a certain communication overhead. To reduce communication costs, this paper leverages the interference caused by impulsive noise to design two communication mechanisms, where nodes execute corresponding communication strategies based on their individual states. Experimental results demonstrate that this algorithm effectively reduces communication overhead while maintaining performance as much as possible.

The old VLCC was taken as the study object, the PHAST software was used to simulate the oil film range of collision in the old VLCC, and predict the damage areas of the fire and explosion accidents caused by the collision. The simulation results had certain reference significance for the risk analysis in the old VLCC for offshore oil storage enterprise.

In response to the problem of security risks in the endorsement service phase of the consortium blockchain Hyperledger Fabric, the paper proposes an optimization scheme for the consensus mechanism based on the behaviour of endorsingnodes. By adding a behaviour monitoring module to the consortium blockchain architecture, the scheme records andanalyses the behaviour of each endorsing node in each transaction. In the monitoring module, a node capabilityvalueattribute is added to each node to reflect its operation status, and the normal operation of the endorsing node is judgedaccording to the capability value. The public key of the node and the hash value of the read-write set in the endorsement result are selected as the output information and passed to the monitoring module, so that the external module canobtainthe node behaviour information while ensuring that the business data on the blockchain is not leaked. Nodes withpoorendorsement performance will eventually be identified as malicious nodes due to their low capability value andbedeprived of endorsement rights. The experimental results show that this solution can timely and accurately detect andclean up the malicious nodes in the system that prevent transactions from reaching consensus, and ensure the normal execution of subsequent transactions.

Differential protection is an important technical guarantee for distribution network operation, which has strict communication delay requirements for data transmission. Short data packets carrying control information should be used for transmission. In this paper, we study a downlink short-packet communication model with RIS-assisted differential protection and a sum rate maximization problem is proposed to optimize active beamforming and passive beamforming using a successive convex approximate frame. Simulation results show that the proposed scheme can improve the sum achievable rate, especially under poor channel conditions.

Inertial measurement unit (IMU) evaluation, which plays an important role in the selection, storage scheme design, maintenance plan formulation, etc. of IMU, often uses the multi-attribute decision-making (MADM) methodandconsiders multidimensional information to produce a reliable result. To address the issues of bias and one-sidedness of asingle weighting method, a combination weighting method based on a rough set and information entropy is raisedinthispaper. This proposed method, integrating the statistical characteristics of data, information features, and decisionriskthat are necessary factors for data-driven weight assignment, constructs a feasible set of weight assignment methods. Based on the information entropy, the importance of each method in the constructed set can be calculated. Subsequently, the comprehensive weights under the feasible set can be yielded with the help of the rough set theory-based redundancyelimination method. Finally, the feasibility and effectiveness of the proposed methods are verified by a practical engineering case, i.e., the accurate evaluation of IMU.

Conversational machine reading comprehension (CMRC) requires models to effectively combine dialogue historyandanswer current questions. Previous works have shortcomings in handling historical information as they did not consider the role of historical questions in the learning process. Moreover, in the reasoning process, parallel input of multiplerounds of dialogue does not conform to human reasoning habits. Therefore, to address these limitations, this paperproposes the HistoryintoFlow model. In our model, we incorporate historical questions into the encoding layer, whichenables the model to extract complete historical information. In the reasoning layer of the model, we designa flowmodule that integrates intermediate representations generated from past conversations and performs reasoninginaccordance with the order of conversations. The final results show that the HistoryintoFlow model achieves an accuracyrate of 67.1% on the QuAC. Compared with some publicly available models, our model has improved in F1, HEQ-Q, and HEQ-D.

To realize fine production management and process control, a process reliability modeling method considering discrete state and continuous performance degradation is proposed in this paper. Process reliability is defined by the degree to which a process satisfies customer requirements, and the process elements covered by process reliability are analyzed. A process reliability modeling method based on a general generating function is proposed by integrating the discrete characteristics of stage capacity caused by failures and the continuous dynamic characteristics of long-term capacity caused by performance degradation. A case study is carried out for a part machining process, and the results show that this method can more accurately characterize the process state, and then provide a more accurate basis for production decision-making.

This study uses the important container rail-sea intermodal transport areas in China as the research area, and gets data through visiting research and network query to investigate the change in container rail-sea intermodal transport demand under the new development pattern. The generation and distribution of container freight in the source area are investigated using the 'four-stage' method. The market competitiveness of rail-sea intermodal transport is then investigated and assessed, and the Logit-based share rate model is developed to forecast the flow and direction of container rail-sea intermodal transport in the source area. Finally, the market share rate of container rail-sea intermodal transportation under various situations is investigated by varying the price and capacity level of container transportation.

In 2022, the rainfall and runoff in the Yangtze River basin were significantly low, which indicate a severe drought happened in the Yangtze River. Since August, the water level in the middle and lower reaches has continued to decline and broke the record of the lowest water level in the history of the same period. Taipingkou and Ouchikou, which belong to the three outlets of the Jingjiang River, have been cut off nearly three months in advance. In this paper, the low water propagation characteristics in the middle and lower reaches of the Yangtze River and the diversion characteristics of the three outlets of the Jingjiang River were studied. At the same time, MIKE 11 is used to quantitatively analyses the operation benefits of the Three Gorges Reservoir. The research results show that (1) the propagation time during the low flow period has been extended to a certain extent, compared to flood period, for example, the propagation time between Yichang and Shashi stations, Shashi and Lianhuatang stations, Lianhuatang and Hankou stations, Hankou and Jiujiang stations, Jiujiang and Datongjiang stations are respectively 2h, 5h, 9h, 4h, and 8h longer; (2) the attenuation coefficient between Yichang and Shashi, Shashi and Lianhuatang, Lianhuatang and Hankou, Hankou and Jiujiang, Jiujiang and Datong are 0.99, 0.76, 1.00, 0.64 and 0.74, respectively; (3) after the construction of the Three Gorges Reservoir, from 2003 to 2021, the average diversion ratio of the three outlets of Jingjiang River was 12%, which is 17% less than that from 1956 to 1966, and is 3% less than that from 1981 to 2002; (4) the operations of Three Gorges Reservoir have effectively raised the water level of the main hydrological stations in the middle and lower reaches of the Yangtze River.

Gait recognition is a biometric technology with important applications in identity verification and crime prevention. However, developing robust and accurate algorithms remains a challenge due to various factors such as environmental interference and clothing changes. Current gait recognition methods are categorized into template-based and sequence-based approaches, both of which have limitations. Template-based methods are simple but lose temporal information, while sequence-based methods preserve temporal information but are sensitive to unnecessary sequential constraints. Furthermore, these methods are not robust to various sources of noise and interference and often require a large number of gait frames for accurate recognition. Therefore, new approaches are needed to improve the accuracy and efficiency of gait recognition. We proposed the method, named Adaptive GaitSet, addresses these limitations by introducing an attention mechanism that allows the model to adaptively focus on important gait features. This method not only improves the accuracy and efficiency of gait recognition, but also enhances the robustness of the model to variations in walking conditions and sources of noise and interference. Experimental results show that the proposed method achieves the best performance on benchmark dataset.

With the development and widespread application of Model-Based Systems Engineering (MBSE) in the aerospace domain continue to grow, System Modeling Language (SysML) has become increasingly important as the most popular modeling language in MBSE. It is crucial to provide developers with an efficient and secure SysML editing tool to enhance their productivity. However, our research has revealed issues with the existing SysML editing tools, particularly in terms of data storage, resulting in a decline in efficiency as project data grows. Therefore, this paper proposes a GEF-based SysML development tool that supports developers in creating, editing, and exporting graphics. This tool addresses the data storage issues, improving development efficiency. Additionally, we present a model validation approach and extend the existing validation rules to ensure the correctness, reliability, and security of graphical elements and diagrams.

In this study, we recorded the electroencephalography (EEG) of human subjects when they watched a movie trailer. Based on the willingness preference from the subjects after finished the movie trailer watching, the EEG signals were analyzed to build regression models to predict the subjects' movie preferences. We proposed a new approach to rise the prediction accuracy. Specifically, the correlated component analysis (CCA) algorithm was used to calculate the inter-subject correlation (ISC) of each time segment in the movie trailer and the ISC of each channel in the EEG. These two states were selected to locate the moment when a specific area of the brain responds most strongly to the video stimulation, as an effective indicator for predicting individual movie preferences. The experimental results show that the method of selecting effective EEG signals at the level of ISC has achieved a better prediction effect on the GBDT regression model (R2 = 0.733, RMSE = 0.681). We also found that low-frequency EEG signals in the θ and δ bands of the parietal and occipital sites are more effective in predicting movie preferences than high-frequency EEG signals.

Due to the complexity of environmental factors, such as background and lighting, traditional vision-based gesturerecognition algorithms tend to be less robust. Similarly, gesture recognition algorithms based on neural networks havethe drawbacks of large computational requirements and high storage parameters, thereby hindering effective deployment. This paper proposes a novel gesture recognition algorithm based on infrared thermal imaging sensor, which effectivelyaddresses the above challenges. By collecting thermal imaging gesture data and performing preprocessing, an improvedbinary neural network is trained and deployed on a microprocessor device. Results show that the proposed binaryneural network achieves a recognition rate of 99% for three gestures, and 95% for five gestures. Compared with convolutional neural networks, which have higher computational requirements and more model storage parameters, the proposedalgorithm achieves a frame rate of 44FPS on STM32F4 microprocessor and 366FPS on H7, highlighting its potential asan effective gesture recognition technology using thermal imaging sensors.

The massive installation of smart meters on the customer side plays an important role in the collection and analysis of multi-user data. However, the current way of power grid collecting electricity data is consistent with the centralized grid control, and this way cannot cope with the new demand of multiparty data interaction and integration in the future. Blockchain technology integrates cryptography, distributed ledger technology and data sharing, which can enhance the security of power measurement data. To improve the performance of power data transmission and transaction, this paper combines blockchain smart contracts with the practicality in the field of power measurement, and proposes a smart contract based on the identification of hydrogen energy users, which analyzes the nodes containing hydrogen energy and identifies different nodes to complete the contract.

LiDAR (Light Detection and Ranging) is widely used in large-scale architecture measurement because of its high acquisition efficiency. But there are two problems that make it difficult to manage construction projects directly from measurement data: inefficient processing caused by large data size and inaccurate progress analysis caused by occlusion. For these reasons, this paper proposes a two-step approach that consists of data filtering and construction progress analysis. Firstly, to improve the efficiency of processing, a voxel filtering method with variable raster length by adding curvature information is proposed. Secondly, to overcome the occlusion. a construction progress analysis method with cumulative distribution from a projection perspective is performed. To verify this method's effectiveness and reliability, this paper carries out a set of data from a garage. The numerical results show that the proposed indicator can overcome the occlusion phenomenon and represent the distribution of the components efficiently. This paper also compares the degree of data fluctuation between 100% data and 25% data, the difference is less than 0.12 which proves the proposed approach can effectively preserve the key features of the original large-range building data.

With the deepening of the application scenarios of multiple CA digital certificates in the power industry, in order to solve the problem of mutual trust and mutual appointment of digital certificates issued by various Certificate Authority, this paper proposes a design method for digital certificate interconnection. By proposing and formulating a unified standard for digital certificates of the national power grid and designing a multi-Certificate Authority operation and management platform, it opens the channel of mutual trust and mutual appointment of digital certificates issued by different Certificate Authority, centralizes the full lifecycle management of digital certificates and brings a better experience to users of business systems.

Equipment entity recognition and attribute extraction are the basis of constructing equipment knowledge graph. In this paper, we firstly design a model framework for equipment entity recognition and its attribute extraction. Then, combining the advantages of BiLSTM and CRF, an equipment entity recognition method based on Dic+BiLSTM-CRF is proposed by constructing a domain-specific dictionary for equipment. Furthermore, the equipment entity attribute extraction method is designed based on HMM model and Viterbi algorithm. The experiment results show that compared with the traditional methods, the performance of equipment entity recognition based on Dic+BiLSTM-CRF is close to the general domain entity recognition level. The accuracy rate, recall rate and F1 value of equipment entity attribute extraction are higher than 80%.

To cope with the escalating malicious code variants, we propose a malicious code classification method (BiTCN-CA) based on bidirectional temporal convolution network (BiTCN) and channel attention (Channel Attention (CA) based malicious code classification method (BiTCN-CA). The method fuses malicious code opcode and bytecode features to show different details; and builds BiTCN models to take advantage of the backward and forward dependencies of the features; and introduces the channel attention mechanism to further explore the deep dependencies within the malicious code data. The model is validated on kaggle dataset, and the experimental results show that the classification accuracy of malicious code based on BiTCN-CA can reach 99.36% with fast convergence speed and low classification error, and finally the effectiveness of the model is proved by comparison experiment and ablation experiment.

In our paper, we propose a methodology that uses natural language processing technology, including the Latent Dirichlet Allocation model and Word2Vec word embedding model, along with various machine learning and deep learning algorithms. This approach allows us to analyze Computational Fluid Dynamics literature accurately and efficiently. By applying these techniques to extract topic and word relationships from the literature data, we can discover hotspots, research directions, and emerging trends in the field. Our results validate the effectiveness of this approach, providing valuable insights. This research also offers a framework for researchers to analyze scientific literature, helping them understand development trends and explore knowledge associations in their respective fields.

Corner reflector arrays are a crucial passive interference against anti-ship radar guides. High resolution range profile is a projected distribution of the electromagnetic scattering properties of a target in one dimension. Corner reflectors can be placed in various ways to form a variety of corner reflector arrays to approximate the range profile of a ship. Still, in practice, the physical dimensions of ships and corner reflector arrays vary greatly. However, the physical sizes of ships and angular reflector arrays differ considerably. In this paper, based on the principle of monopulse 3D imaging, using the three-dimensional scattering center model of ships and corner reflector array, we propose the minimum volume enclosing ellipsoid feature that can fully characterize the spatial morphological information of the target and discriminate the ships and angular counter arrays by SVM classifiers. The experimental results show that the proposed method in this paper has an identification accuracy of more than 93% for ships and corner reflector arrays at SCR10dB and more than 80% when the sea surface height is below 6m.

A significant area of research in the field of information retrieval is query expansion. To address the issue of inaccurate user query descriptions and mismatches with query terms during the information retrieval process, a semantic query expansion method based on Word2Vec is proposed. The method involves training low-dimensional word vectors using the Word2Vec distributed neural language probability model, selecting a candidate set of expansion words, and employing a query vector generation approach that focuses on the expansion words to filter the candidate set. This ensures that the selected expansion words effectively reflect the semantic and syntactic relevance of the entire query. Experimental results demonstrate that the semantic query expansion method based on Word2Vec improves both recall and precision, making it a valuable approach in the field of query expansion.

Nowadays, there is a general problem of communication difficulties in the deaf community for gesture expression of emotion, which leads to depression. This paper designed and implemented an emotional gesture recognition method for the recognition of gestures’ emotion. The recognition results of negative and positive emotions can improve the communication of the disadvantaged people. The method designed the recognition network based on convolutional neural network, spatial pooling, residual structure and then optimized the network with attention mechanism. The results of the experiments on real scene image dataset show that the network model proposed in this paper has a high recognition accuracy and good robustness.

The density of suspended particles is a crucial indicator for assessing the water quality, the study was based on high score 1 (GF-1) satellite remote sensing image, using a neural network model, choosing a double hidden layer set, building the Hangzhou bay turbid water suspended sediment concentration inversion model, to analyze the space-time characteristics of the suspended sediment density, and comparing with other conventional precision inversion method. The findings demonstrate that retrieving the suspended sediment concentration in Hangzhou Bay using the BP neural network model is viable, and compared with other conventional methods, the model has a higher correlation (R2 = 0.9398) for the inversion of turbidity water with higher suspended sediment concentration. In the spatial distribution of the Hangzhou bay area water suspended sediment concentration is high overall, Hangzhou bay water suspended sediment concentration to an average of 679 mg/L, the inversion results show that the water in Hangzhou Bay has a high overall concentration of suspended silt, internal differences obviously, among them the highest of the central region, and diminishing the coast gradually to bay mouth, respectively.

Online calibration of traditional magnetometers requires the assistance of inertial sensors, resulting in complex calculations and unstable accuracy. To this end, a six sigma constrained online calibration method for magnetometers is proposed, which reclassifies the errors of magnetometers and constructs an accurate magnetometer error model. Use Six Sigma (SS) to manage the calibration model of magnetometers and achieve error classification of magnetometers. A six sigma constrained extended Kalman filter (SEKF) method is proposed to calibrate the magnetometer error. The experimental results show that in the aspect of heading correction, the heading Root-mean-square deviation of the proposed magnetometer online calibration method is reduced by more than 32% and 12% respectively compared with the gyroscope assisted (GA) method and the extended Kalman filter calibration (EKF) method. A high-precision online self-calibration method for magnetometers without inertial sensor assistance has been implemented.

The traditional approach of obtaining railway freight car numbers is manually recognizing car numbers, which has the drawbacks of low efficiency and error-prone. In this paper, we propose a small-sample-based method for railway freight car number recognition. The original sample data consists of character samples cropped from the target region of railway freight car numbers, with an average of 15 samples per character. Each type of character sample data is augmented to 500 images and undergoes image processing. Subsequently, an image stitching algorithm is employed to combine the processed character samples into a training set image. Using this training set, a character detection model is trained on the YOLOv5 algorithm. By incorporating spatial positional relationships between car number characters, a car number extraction algorithm is designed to output the final result. We test the proposed method, and our method achieved a car number recognition accuracy of 89.58%, demonstrating credible recognition performance.

In order to solve the problem that the peak of the weak target in the azimuthal energy spectrum is lower than the background when using the passive sonar systems to detect targets, and it is unable to identify the target azimuth visually and quickly, a post-processing method that can identify the target azimuth and perform local enhancement is proposed in this paper. This method uses background equalization technology to compress the dynamic range of the energy spectrum. The background threshold of energy spectrum after normalization processing is estimated, and target azimuth is extracted. Combined with the background threshold, the correction coefficient is constructed in segments to achieve the local enhancement at the target azimuth. This method is able to distinguish targets and background visually and quickly, has the function of highlighting and identifying the azimuth of the target based on simulation and measured data.

Due to the complexity of the person re-identification scene, it is easy to be affected by visual changes such as illumination, occlusion, posture, etc, which makes the extracted person features will be insufficient and feature discrimination is poor, which in turn will lead to low retrieval accuracy. To address the above problems, a multi-scale attention pyramid network (SOAPNet) based on non-local block guidance of the spectral is proposed. First, the atlas non-local blocks are introduced into ResNet50 to capture remote dependencies more robustly and flexibly, so as to realize a wider range of information perception in the feature extraction stage; then, the multi-scale attention pyramid network is introduced into ResNet50, so that the model can focus more on the most critical and discriminative information in the person images, and this multi-scale attention strategy effectively solves the problem of The problem of insufficient features and poor discrimination in the person re-identification task, which enables the model to better distinguish person with different identities and extract more robust features. Experiments on Market1501, MSMT17, and DukeReID show that Rank1 is improved by 0.7%, 0.9%, and 1.2%, respectively, and mAP is improved by 0.5%, 1.0%, and 1.4%, respectively. The proposed method can significantly solve the above problems.

Under the current big data background, network crimes are happening frequently, and the network intelligence collection and analysis technology is particularly important for the governance of the crimes. To deal with the issue, we propose a new multi-level network intelligence collection and analysis practice approach, so as to improve the network intelligence collection and analysis ability of network security talents under the new situation, and adapt to the requirements of new network crime governance. First, we introduce the whole design of the network intelligence collection and analysis course, then elaborate on the details of the new multi-level mode, and finally analyze and summarize the preliminary exploration of the practice. The application results show that the multi-level practice approach is better than the common method, which is helpful to strengthen the participants' practical ability of network intelligence collection and analysis.

There are two types of binary Z-complementary pairs (ZCPs), Type-I ZCPs and Type-II ones. Existing known Z-optimal Type-II Odd-Length binary ZCPs (OB-ZCPs) have only for very limited lengths. In order to obtain Z-optimal non-periodic Type-II OB-ZCPs with more lengths, we present a new construction of Type-II OB-ZCPs of lengths 2^α 10^β 26^γ −3 (where α, β, γ are non-negative integers, α≥1 and α + β + γ ≥ 2 ) by using Turyn’s construction and deletion function. The proposed systematic construction can obtain not only Z-optimal Type-II nonperiodic OB-ZCPs but also optimal ones. The obtained type-II non-periodic OB-ZCPs provide more flexible choices for applications.

Charge-coupled device (CCD) spectrometers are commonly used for qualitative and quantitative analysis of substances. In order to reduce the influence of spectral signal noise on qualitative and quantitative analysis of substances, a hardware architecture of multilayer wavelet threshold denoising algorithm based on FPGA is designed. The folded multi-level architecture is adopted to realize the flexible selection of wavelet lifting scheme and wavelet decomposition layer number. In addition, the threshold processing hardware structure and wavelet decomposition structure are optimized. Experiments show that the proposed hardware architecture can realize the denoising of spectrum signals, and the hardware architecture is superior to the existing architecture in the utilization of DSP resources.

As a nascent framework in the field of signal processing, graph signal processing (GSP) presents a potent strategy for handling unstructured data. By infusing the least mean squares algorithm into graph signals, the GSP-LMS algorithm takes form. Nevertheless, the effectiveness of GSP-LMS wanes when confronted with non-Gaussian noise. To address the issue, the presentstudy introduces the maximum multi-kernel correntropy (MKC) criterion, presenting a robust distributed MKCbased graph signal estimation algorithm named GSP-DMKC. Benefiting from the adaptability of loss function, this algorithm demonstrates the capability to tackle intricate noise scenarios. Recognizing the constraints of communication resources in real-world scenarios, a partial transmission strategy is introduced to design the GSP-DMKC-RC algorithm, thereby mitigating communication overhead. GSP-DMKC-RC applies information compression during both adaptive and combinatorial steps. Furthermore, when opting for an appropriate compression rate, the GSP-DMKC-RC can attain estimation performance comparable to that of GSP-DMKC, concurrently easing the communication load. Ultimately, simulations are set up to affirm the excellence of algorithms.

In near-space fast-speed synthetic aperture radar (SAR), the azimuth multichannel technology is usually required to achieve high resolution wide swath (HRWS) imaging. Compared with traditional linear array SAR (LA-SAR), arc array SAR (AA-SAR) is with attractive omnidirectional and real-time dynamic imaging capacities. In the paper, the azimuth multichannel signal model of LA-SAR and AA-SAR are both established and compared, compared with the echo model of linear array SAR, the echo model of arc array SAR has an additional phase, which leads to the imbalance of multichannel. Directly using the linear array azimuth multi-channel reconstruction method, the imaging results will appear false targets. Finally, the inference is verified by point target simulation.

Firstly, the current cucumber pruning operation is carried out manually. Secondly, the existing target detection algorithm is not accurate enough to efficiently detect and identify the cucumber pruning target. In response to the above issues, this article proposes the LW-YOLO algorithm based on improved YOLOv5s-7.0. Initially, FCSP is constructed to replace the C3 module in the baseline model so as to reduce parameters, speed up detection, and improve detection accuracy. Then MCACB is designed to realize downsampling, reduce parameters, effectively aggregate feature information, and improve detection performance. Finally, the EW K-means clustering algorithm is proposed to estimate the prior boxes, which further improves the detection accuracy and robustness of the model. The experimental results show that the proposed algorithm has the smallest model size, only 8.39MB; the detection accuracy is the highest, with map@.5% reaching 97.1; less computing resources are needed, and GFLOPs is only 10.0 GB; and it runs faster, with 294.1 FPS. The various indicators of the LW-YOLO algorithm are not only better than the baseline model YOLOv5s-7.0 but also better than the latest models YOLOv6, YOLOv7, YOLOv7-tiny, and YOLOv8, which meet the detection and recognition requirements of cucumber pruning robots and can be deployed.

This paper addresses the issue of cross-reuse functionality among subsystems within a missile weapon system. The objective is to minimize development costs by formulating the missile weapon system's reliability assignment problem as a constrained combinatorial optimization challenge. To solve this problem, a parallel genetic algorithm is employed, utilizing a coarse-grained model and a multi-population evolution strategy. The presented example demonstrates the successful application of the parallel genetic algorithm in effectively solving the problem while achieving a high-quality solution.

How to perform similarity search on high-dimensional data is a great challenge, especially for Euclidean distance, although the traditional E2LSH algorithm can cope with this problem well, the value of the parameter has a great influence on its effect. In this regard, a hash bucket parameter selection algorithm incorporating the Gaussian kernel function is proposed for its parameter selection problem, which improves the sensitivity to the data to obtain more accurate and efficient nearest neighbor search results in multi-distributed high dimensional data. Meanwhile, experimental comparisons with other similar algorithms are conducted on a common dataset, and their results show that the optimized E2LSH algorithm demonstrates improved performance in terms of speed and efficiency.

In the industrial domain, accurate detection and localization of abnormal images are crucial factors for ensuring production efficiency and quality. In recent years, methods in the field of unsupervised anomaly detection have predominantly centered around knowledge distillation models with a teacher-student structure. Although the application of knowledge distillation has significantly improved the precise identification of abnormal regions compared to traditional methods, the currently best-performing knowledge distillation models exhibit identical architectures for both the teacher and student networks. This limitation hampers the effective lightweighting of the student network and poses challenges in distinguishing between background and target objects in certain data with backgrounds, thereby affecting accuracy and wasting computational resources. To address this issue, we propose an innovative approach that integrates knowledge distillation with attention mechanisms for identifying abnormal regions in industrial anomaly images. Our method comprises two key features: firstly, transferring knowledge from a pretrained teacher network to a student network to enhance performance; secondly, incorporating attention mechanisms to direct the model’s focus towards potential abnormal regions, thereby enhancing detection accuracy. Our approach innovatively combines knowledge distillation with attention mechanisms, offering a novel solution for industrial anomaly image recognition. Through experimental validation, our method demonstrates superior performance compared to the original models, providing fresh insights for the field of industrial anomaly image analysis.

The analysis of time series data holds significant application value in many real-world scenarios, such as forecasting power generation and traffic flow. Recently, time series forecasting models based on representation learning have garnered widespread attention due to their ability to maximally retain meta-information. However, the current representation models for time series based on learning typically employ stacked convolutional neural networks (CNNs) as their primary encoder, constructing positive and negative samples at the timestamp level for various time series data. CNNs are primarily designed to capture local patterns. Yet, in time series forecasting tasks, long-term dependencies and global patterns are more crucial, and CNNs might not be adept at capturing these patterns. To address this issue, we introduce "Contrastive Learning Enhanced by Transformer Block for Time Series Forecasting." This method initially employs the Transformer to extract features characterized by global patterns. It then uses a contrastive loss function to encourage the model to discern similarities among positive samples and differences among negative ones. Ultimately, the features learned by the model are used for forecasting. Experimental results indicate that our approach outperforms traditional methods in time series forecasting tasks.

The glide path (GP) is an integral part of the Instrument Landing System (ILS), working in conjunction with the onboard receiver to provide glide guidance information for aircraft approaching and landing, operates in the 328.6 to 335.4MHz ultra-high frequency range, signal transmission mainly relies on the reflection of the protected area in front of the antenna. However, the mountain ahead has caused problems with the GP at multiple airports. The revised MH/T4003.1-2021 in 2021 is based on the original 2014 version, added obstruction requirements for obstacles ahead. Through computer modeling and simulation, this article conducts simulation analysis on the influence of the mountain in front of the GP on the glide slope structure and glide clearance under different obstruction angles, proposes a solution for the site selection and signal optimization after construction of the GP.

In the process of AI object recognition, objects need to be labeled in images, there are even multiple objects to be labeled in a single image, and the images needed for training reach thousands or even tens of thousands, which makes setting up sample integration a workload. Therefore, this paper proposes a Web-based automatic image object recognition system, which collects data based on the Web and allows multiple people to collaborate in labeling targets in images, and after finishing labeling, aggregates labeled images and builds a sample set; on this basis, the YOLO neural net model has trained automatically, and the study shows that this system can greatly improve the efficiency of sample set construction and quickly build a deep target. The study shows that this system can greatly improve the efficiency of sample set construction and quickly build a deep target recognition neural net, which has good application value.

To monitor the in-situ wear state of friction pairs, a friction noise signal processing method based on polymer interface wear performance prediction was introduced. Four types of friction noise signal datasets were collected from four different friction surfaces consisting of three constant load-speed points of six temperature points ranging from-120°Cto 25 °C and seven types of metal counterparts. Three wear states, including adhesive wear, wear, and a combination of both, were defined using field emission scanning electron microscopy for polymer wear surfaces. Six friction noise features were extracted from the time-domain signal of the sound using three machine learning algorithms, including random forest, deep forest, and XGBoost, to classify the three wear states. The results indicate that the deep forest algorithm exhibited the most effective classification performance. As a result, this method holds promising potential for on-site and real-time condition monitoring, as well as wear classification in the field of tribology applications.

The unmanned system is always employed in the uncontrolled radiation source accidents, tackling with geometric field mapping, obstacle avoidance exploration and radiation sources localization simultaneously. To balance the multiple subtasks, a novel source seeking and exploration strategy is proposed in this paper. Firstly, the geometric metric map and radioactive topology map are constructed respectively, while the evaluation functions are established to merge above both maps into the hybrid costmap. Additionally, the RRT* structure is introduced into the path planning module, leading to the capabilities of full-map expansion and multi-source exploitation. By combining the costmap with RRT* structure, the topological paths which satisfy source seeking and map exploration can be generated in the cluttered environment. Experimental results verify the effectiveness of the multi-task path planner, taking into account both multisource exploitation and environment exploration in an online manner.

Aiming at the huge and discrete storage of quality problem data in aerospace model quality management, a method for improving the efficiency of finding aerospace quality information by constructing a knowledge graph of aerospace quality and applying a question-answering system is proposed. The graph construction adopts a top-down construction method. First, the concept-attribute-relationship ontology model of aerospace quality knowledge is constructed. Then, based on combining expert knowledge to design the ontology concept of aerospace quality knowledge, the data to be processed is pre-processed by BERT. The training model is combined with deep learning technologies such as BiLSTM-CRF and self-attention mechanism to realize independent knowledge extraction, and then the Dice coefficient combined with the weighted average method is used to fuse the entity-relationship extraction results. Finally, based on the constructed knowledge graph, to improve the accuracy of semantic template matching, a natural language question semantic classification process model based on a weighted naive Bayesian classifier is established, which can further realize intelligent question answering. The results of the paper show that the intelligent application of knowledge graphs in the field of aerospace quality has a good prospect.

Existing CI/CD tools require users to define the entire process of software build and release, manage dependencies between projects, and determine the build sequence. The complexity and difficulty of project build and release are relatively high. This paper proposes a rapid build and release platform. Compared with the traditional CI/CD platform, it can automatically analyze project dependencies and generate project build and release processes. Users can complete software build and release without defining build and release pipelines, reducing the complexity and difficulty of software release.

This paper proposes a design approach for an encryption tool based on image processing and PDF manipulation to meet the encryption requirements of sensitive design drawings. The tool is developed using Python programming language and the OpenCV-Python library. Designers can choose encryption fields through a GUI, including modifying product name, removing stage markers, and changing approval date, among others. Additionally, there are two encryption modes available for selection: combined into an one encrypted PDF or each PDF encrypted separately. Batch encryption of all PDF drawings within a selected path can be completed very quickly, significantly reducing the time and workload required for manually encrypting multiple PDF drawings. This greatly improves encryption efficiency and eliminates tedious repetitive work. The program avoids human errors and ensures that sensitive information remains protected from unauthorized access.

The automatic assessment of X-ray weld image quality is an important basis for the automatic assessment of weld image defects. For ring weld X-ray welds, inspectors judge the image quality through human vision and physical blackness meter, which can easily lead to staff fatigue and thus misjudgment, to achieve an automatic assessment of X-ray weld image quality, this paper combines the subjective evaluation results with objective evaluation algorithms to propose a digital blackness meter model suitable for automatic assessment calculations, and gives the ring weld quality assessment algorithm The results show that the sensitivity of the proposed method can reach 98%. The results show that the proposed method has a sensitivity of 98. 5% and a specificity of 100%.

Convolutional neural networks (CNNs) applied in motor imagery brain-computer interface (MI-BCI) can employ large network sizes to learn rich information contained in electroencephalography (EEG) signals for achieving high classification performance. However, the large-size networks lead to high computational resources and long training time so that hindering CNNs in the practical application of BCIs. To address this issue, we propose a multi-scale convolutional neural network with attention mechanism (MS-CNN) in this study. The proposed model has the following properties: 1) it optimizes a multi-scale architecture for learning the spatial-spectral information related to MIs directly, avoiding the extra step of feature extraction. 2) it utilizes attention mechanism for efficient information enhancement while keeping a relatively small network size. The MS-CNN achieves the maximum classification accuracy value of 88.46% (±6.54) compared with existing baseline methods of the BCI competition IV 2a dataset.

Visual place recognition is an essential function for robot system, which can effectively reduce position error via traversing same place. However, a major challenge in this field is to be robust against viewpoint and illumination changes caused by environmental variations. To solve this problem, we propose a novel visual place recognition method in coarse-to-fine manner, which incorporating both high-level semantic feature and bag context information. Firstly, the semantic features of an image are obtained by adopting deep learning network, which converts the image data to the instance representation in more simple way. In coarse stage, we utilize the Bag of word (BoW) model to quantify semantic features by calculating their frequency distribution. In fine stage, we enhance the spatial correlations of semantic objects to further distinguish similar scenes. Through the above two stage, we achieve a robust pipeline of visual place recognition to tackle viewpoint and condition variations. Experimental results on several datasets, including both indoor scenes and outdoor scenes, show that our method achieve good performance, especially for extreme conditions.

Among the 3D target reconstruction methods, the neural radiation field is a micro-renderable process proposed on the basis of classical body rendering, which realizes the effective combination of neural field and graphics components and has shown its potential in the field of new viewpoint generation. Although the neural radiation field can generate high-quality reconstruction results, it requires rendering the scene under each viewpoint, leading to slow and time-consuming modeling. By contrast, the 3D point cloud, which is generated using the attention mechanism, presents both the spatial geometry of the object and preserves the color-texture features of the points around the object. Combining the neural radiation field with the attention point cloud avoids the rendering of missed points, resulting in improved reconstruction speed. Furthermore, the surface texture information of the object is preserved in the ray-tracing-based rendering pipeline. According to experimental results, the neural radiation field based on the attention point cloud method maintains the visual consistency of geometric contours with the original rendering method, thus achieving fast 3D target reconstruction. Quantified results based on SSIM, PSNR, and LPIPS further confirm the feasibility of this algorithmic model.

Digital processing is the core module of 3D printing technology. In response to the problems of complex system operation, limited adjustable parameters and high configuration requirements of the current traditional 3D printing pre-processing software, this paper studied about 3D printing digital processing. In order to efficiently obtain model geometry contour display and layered slicing data, the research and development of 3D printing pre-processing software is carried out in Windows 10 environment using Python language and VSCode editor, with the visualisation tool VTK as the graphics drawing interface, to realise the fast reading display and slicing operation of STL model files. Based on the analysis of the software requirements and functional requirements, the core algorithms were optimised to perform slicing and tool filling path planning operations on the model, and the final output of G code showed stable and efficient program performance, laying the foundation for the development of fused deposition modeling (FDM) 3D printing technology.

In response to the problem of decreased robustness and accuracy in visual inertial systems due to the inability to extract sufficient point features in weak-textured environments, a dynamic line feature algorithm is proposed, adding the number of co-visible point features as a dynamic threshold to adjust the line feature extraction process. Line features are easier to track in weak-textured environments, providing more geometric information and improving system robustness. The backend utilizes a sliding window algorithm to establish an objective function and jointly optimize the visual-inertial information to improve system accuracy. Experimental results using publicly available datasets demonstrate that compared to the VINS-Mono system, the algorithm exhibits better accuracy and robustness under weak-textured environments.

Drilled shafts, widely applied in structural engineering foundations such as highways and railway bridges, are favored due to their mature construction techniques, high bearing capacity, and wide applicability. However, inaccurate positioning of the concrete pouring interface within the drilled shaft can easily lead to problems such as overpouring, resulting in material waste and increased costs. The currently used rope and weight method for measurement is time-consuming, labor-intensive, and unable to accurately determine the concrete surface. To dynamically monitor the concrete pouring interface, this paper proposes a recognition technology based on a multi-sensor data acquisition cloud platform and the PSO-GA-BP neural network method. By monitoring data such as turbidity, conductivity, and pH value, the pouring interface is identified using the PSO-GA-BP neural network model, and the location and height are calculated to achieve intelligent monitoring of the concrete pouring interface. This approach effectively solves the issues of overpouring or underpouring.

The cold chain logistics of oysters starts from the moment they are harvested and ends when they reach the consumers’ homes. The process includes various stages: pre-processing, packaging, pre-cooling, processing, cold storage, loading, transportation, and sales. Due to inadequate quality control in the logistics process, fresh oysters often experience a significant decline in quality during transportation and storage, resulting in substantial losses. To address this issue, this article presents a real-time, traceable, and visual quality monitoring system for oyster cold chain logistics. It combines RFID technology with the development of a unique binding technique for preservation boxes and utilizes a hybrid technology incorporating “RFID+4G+DTU+MQTT+cloud storage + Web” to enable real-time collection of oyster transportation environmental parameters and distributed, non-volatile, and cost-effective data storage. A supporting B-S mode system management platform was developed and deployed on a cloud server. The system, with the flow of information as its main focus, provides functions including personnel management, oyster identification and traceability, data visualization monitoring, and quality prediction. The online operation of this system has proven to be highly real-time and stable, effectively enhancing the quality monitoring capability throughout the entire oyster transportation process, and it has demonstrated significant economic and social benefits.

This paper introduces FSD-YOLOv8, an advanced YOLOv8 algorithm specifically adapted for the precise detection of flames and smoke in various complex environments. By integrating a preliminary image dehazing phase, our algorithm significantly enhances the accuracy of smoke detection by reducing atmospheric confusion. Furthermore, we introduce the application of dilated convolutions within the YOLOv8 framework, which aids in detecting intricate features associated with fire and smoke. We present comprehensive evaluations of our approach using a large, dedicated flame and smoke dataset. Our findings demonstrate that FSD-YOLOv8 exhibits superior performance in detecting flame and smoke compared to conventional methods, paving the way for further advancements in machine learning-based fire detection systems.

In the process of optimizing the big data management system, the previous intelligent methods cannot be effectively calculated. This paper first introduces the research status of parallel optimization for Convolutional neural network model training, and explains some shortcomings. Aiming at the problems still existing in the parallel optimization of existing Convolutional neural network model training, an optimization processing method oriented to Convolutional neural network is proposed. The convolutional neural network model is optimized based on the distributed parallel processing framework. The results show that the comprehensiveness and rationality of the deep learning algorithm are greater than 83%. By using a multi parameter server, the gradient calculation parallelism is improved, the communication delay loss in parallel parameter update is reduced, and data preloading is adopted to reduce the data reading time, Improve the efficiency of network model training.

FPGA (Field Programmable Gate Array) is a gate array that can be programmed in real time and is widely used in the electronics industry. In this paper, a moving pattern display system based on VGA (Video Graphics Array) display is implemented by using VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) language "top-down" design and FPGA opening board and display. The system was developed with the aim of developing a portable vision protection device whose basic function is to display dynamic ICONS on the screen. Taking this paper as an example, its function is to make a 50×50 pixels white square display and move in 1024×768×60 display mode, the white square will recognize the display border and automatically bounce at the border. In the future, the display content can be modified according to needs, and the transformation of dynamic pinball games, low-power fun signs and other achievements can be completed.

Environmental conditions with smoke severely degrade the image quality, obscuring numerous features and fine details in polarimetric images. Hence, the process of smoke removal and the recovery of polarized information in polarimetric images are of paramount importance. To address this issue, the paper proposes a novel smoke removal algorithm for polarimetric images based on Generative Adversarial Networks. It successfully restores the polarized intensity responses of four distinct polarization angles, captured by a focused plane polarimetric camera in a smoke-free environment, making the recovered images indistinguishable from those captured without smoke. Experimental validation demonstrates the superiority of the proposed method over existing classical smoke removal algorithms in terms of both subjective visual quality evaluation and objective performance metrics. The method achieves remarkable results on a natural visible light polarimetric smoke image dataset.

This article presents the establishment of a machine vision system intended for the retrieval of table tennis balls within a training hall. Capturing images of table tennis balls of various colors in the RGB color format was achieved through the utilization of a high-resolution camera affixed to the pick-up robot. To counteract the influence of environmental lighting conditions in the training hall, the acquired RGB images were converted into the HSI color format. Subsequently, image segmentation was executed based on predefined hue and saturation thresholds, enabling the precise identification and categorization of table tennis balls. The utilization of Canny edge detection and Hough circle detection methodologies was employed to designate the positions of the table tennis balls. Through analysis of circle centers and quantities, the specific locations of the table tennis balls were ascertained for the pick-up robot. This automated system, grounded in machine vision, holds the potential to significantly enhance the level of automation involved in the process of collecting table tennis balls.

With the development of urban construction, the power infrastructure continues to expand and improve. Power transmission lines, in particular, are numerous and widely distributed. The maintenance, inspection, and troubleshooting of these lines are essential for the stable operation of the power grid. However, the proliferation of birds has expanded their activity range, posing a significant threat to the power system. Therefore, this paper combines an image brightness enhancement algorithm with YOLOv5 to effectively detect nest images under low-light conditions. Experimental results have demonstrated the effectiveness of the proposed algorithm in this paper.

Aiming at the management problems in the construction plan of high-level higher vocational schools and majors with Chinese characteristics, we have designed and developed a monitoring platform for professional group development. The platform includes several main modules: the school big screen visual data analysis module, the professional big screen visual data analysis module, the data management module, and the account management module. By utilizing the API provided by the back-end server to acquire relevant data, and integrating the use of ECharts, Vue, Element UI, Axios, and other front-end libraries and plug-ins, we successfully developed each module in the monitoring platform for professional group development. Through the implementation and utilization of the project, the platform demonstrates characteristics of simplicity, universality, practicality, and effective operation. However, it is important to provide further explanation and background information, ensuring the readers' full understanding of the design and development of the monitoring platform. Additionally, clearer explanations of the functionalities and roles of each module in the platform can be provided to improve coherence and logic in the article.

This paper presents Railsight, a data visualization and analytics platform for the UK rail industry. The platform is divided into several components, with a back-end webserver facilitated by the Django framework, a flexible front-end web interface developed using React and Echarts (for data visualization), and a database powered on MySQL. Railsight users have the freedom to select from different data sources and visualize them according to their personal preferences, explore specific data slices, choose from different visualization modules, export the data and dynamically adjust the time-series, and to record and view the history of the tables. These features could facilitate a deep understanding of the modern rail system in the UK and in turn improve its operational efficiency, passenger satisfaction, and sustainability.

Deep learning has emerged as a pivotal tool in the domain of image dehazing. Deep learning models can learn the mapping between foggy and haze-free images from extensive datasets, thereby achieving the dehazing effect. However, in contemporary deep learning-based dehazing algorithms, the presence of low-quality images engenders challenges, giving rise to issues such as spatial distortions during the dehazing process. Consequently, deep learning-based image dehazing has evolved into a formidable challenge within the realm of computer vision. In order to address this issue, we propose a two-stage conceptual network termed JDN. In the first stage, we employ the image restoration algorithm BOPBL for defect rectification. Subsequently, in the second stage, we employ the non-paired dehazing algorithm RefineDnet. Through experimental validation, we ascertain a significant enhancement in dehazing efficacy. This enhancement can be attributed to the propensity of dehazing algorithms to yield erroneous mappings, there by culminating in distortions. When incorporating image restoration algorithms as a preprocessing step, the acquisition of mappings becomes more amenable, consequently leading to superior dehazing effects. The introduced JDN network is capable of end-to-end training. Moreover, we demonstrate the efficacy of this approach in both image dehazing and quality enhancement. Experimental results on the SOTS-Outdoor dataset reveal that, in comparison to paired algorithms such as FFANet, hardGAN, and PSD, as well as unpaired algorithms including DCP, CycleDehaze, YOLY, and RefineDNet, the JDN exhibits superior performance.

In the realm of software engineering, defect prediction stands as a vital and valuable research domain. With the emergence of powerful models like ChatGPT, the potential for enhancing software defect prediction becomes significant. Leveraging the GHPR defect prediction dataset, this study delves into the application of ChatGPT for source code-level defect prediction. Different prompt paradigms, including general-purpose prompts, prompts that emphasize minority class, prompts that include few-shot examples, prompts that include target explanations, and prompts using chain-ofthoughts (CoT), are investigated for their impact on predictive outcomes. Furthermore, a comparative analysis is conducted between specific defects predicted by ChatGPT and defects rectified by developers in the GHPR dataset. While direct application of ChatGPT in source code-level defect prediction shows results akin to random guessing, the study reveals that employing CoT prompts unveils a propensity for ChatGPT to identify a higher number of issues, proposing questions with heightened rationale. The findings emphasize the necessity to move beyond predictive metrics when utilizing ChatGPT for source code-based defect prediction, advocating for a more profound analytical approach.

Convolutional neural networks are more and more widely used in our lives. Computer vision tasks dominated by traditional algorithms are also trying to introduce deep learning networks. Image dehazing is a good field. At present, the most widely used image dehazing field is the unsupervised dehazing algorithm. By decomposing low-quality images, image dehazing (separation into clear image and haze map), it can be performed expansion operations. At present, due to the lack of paired high-quality images and hazy images in the field of image dehazing, this further intensifies the need for unsupervised dehazing algorithms. In order to further improve this problem, this article proposes an extractor DIPV1 based on a new U-Net-like structure, and based on this, combined with the biased fitting problem of the network, the existing Double-DIP structure is improved. we call it Better Double-DIP(BDouble-DIP). This network is an end-to-end network and is jointly trained by 2 DIP-V1.By stopping early, we can get better results. By applying the physical dehazing model, we can finally get a good dehazing result. We conduct experiments on the open source dataset O-Haze dataset, and compare BDoubleDIP with existing traditional algorithms: SIHR, EIDBC, supervised neural networks like DehazeNet: NTDF, NLD, URIE, and BOPBL, as well as unsupervised neural networks: DoubleDIP

Reservoir dams serve as critical infrastructure. However, during their prolonged operation, cracks tend to appear on surfaces, particularly on slopes. The development of cracks poses risks to the stability of reservoir dams. This paper employs computer vision and deep learning methods to analyze images of reservoir dams. The primary directional characteristics of dam cracks are identified by leveraging deep learning method, and the distinctive features of cracks are calculated by utilizing the OpenCV platform and filtering techniques. Additionally, the fundamental features of cracks are determined through analysis and calculations using crack images of a dam as an example.

Automatic Speech Recognition (ASR) is an important technology in modern society, since it acts as a great tool for humans to communicate with computers, visually impaired people, and deaf people. However, existing speech recognition methods are still facing many problems. Some of the methods require a large model and excessive parameters, others cannot achieve reliable accuracy. Therefore, our study utilizes an Efficient Transformer Model with Convolutional Network to conduct an ASR task. Our model significantly improves the accuracy of speech recognition while does not become a huge model with a large number of parameters.

In order to overcome the shortcomings of low efficiency and subjective factors in manually analyzing and evaluating weld defect images obtained by the ultrasonic TOFD method. In this study, combined with artificial intelligence image recognition technology, an automatic detection algorithm of ultrasonic TOFD scanning image weld defects based on Yolov6 of PyTorch platform was developed. Through continuous optimization of parameter configuration, a recognition model with good recognition effect is obtained. At this time, the recognition and classification of five common defects, namely stoma, slag inclusion, incomplete penetration, lack of fusion, and crackle, are correct on the test set. The correct recognition confidence can reach above 0.9, and the positioning is accurate.

Traditional positioning methods are difficult to meet the requirements of indoor positioning accuracy and power consumption, UWB technology is gradually popular because of its strong penetration and anti-interference ability in indoor environment. This paper describes a set of low-power indoor positioning system based on UWB, which mainly includes positioning tags, positioning stations, positioning engine and front-end monitor. It generally introduces the whole positioning process and the design of each module. In particular, it describes how to realize the control of power consumption of the tags, the setting of distance between the stations and the working mode of the positioning engine. Compared with the traditional positioning means, the design is more suitable for indoor environment. The engineering test proves that the design has good application effect, both in power consumption control and in the actual positioning accuracy.

Empirical studies in the field of machine learning based on Monte Carlo methods explore a powerful statistical technique aimed at solving the core problems of uncertainty modeling, model evaluation, and decision making; Monte Carlo methods provide new perspectives and tools for the application of machine learning models through stochastic sampling and simulation; however, Monte Carlo methods are also facing a number of challenges, such as computational efficiency, high-dimensional space problems, and sampling bias, etc.; in order to overcome these problems, this paper explores future research directions that may involve improved sampling methods, innovative applications that incorporate deep learning, etc. Therefore, the continuous innovation in this field will provide new opportunities and challenges for the development and application of machine learning models.

With the progress of military informatization construction, the combination of architectural design and knowledge graph has become a key method to address issues such as insufficient knowledge level, weak correlation, and low reuse rate in architectural design. A high-quality ontology model can improve the quality of the knowledge graph for architectural design, and promote the knowledge, visualization, and intelligent application of architectural design outcomes. Based on data sources such as DoDAF Meta-Model, MoDAF Meta-Model, and Inf-proA Meta-model, this paper studied the relationship between architectural viewpoint models and knowledge extraction, and proposed an ontology construction method for architectural design. By summarizing and generalizing the ontologies in the field of architectural design, a seven-step process for ontology construction in architectural design was proposed, and the consistency of the ontology was verified using Protégé software. The verification results showed that there were no conflicts among the concepts, relationships, attributes, and logical rules in the architectural ontology, providing strong support for architectural design work. Starting from current military requirements, this paper introduces ontology model construction into the field of architectural design, providing reference for the future construction and application of architectural knowledge graphs.

Considering the problem that the imaging accuracy by traditional algorithms for two-phase flow is not ideal, this paper produces the electrical resistance tomography (ERT) algorithms based on compressed sensing (CS) theory, which focuses on three key issues of signal sparse representation, observation matrix optimization and signal reconstruction. Specifically, in order to improve the signal sparsity, Haar wavelets are adopted as the sparse basis, and the pixel index prior information is introduced. To make the sensitivity matrix in ERT better meet the CS theory requirement, SVD optimization is conducted. As for the imaging algorithm, the objective function based on the regularization items of L1 or L0 is selectively used according to the two-phase flow pattern. The experimental results indicate that compared with the commonly used image reconstruction algorithms such as Tikhonov algorithm, Landweber algorithm, and CG algorithm, the algorithm proposed in this paper has more comprehensive and balanced performance, and can be used as an efficient image reconstruction solution of ERT for two-phase flow.

Unit testing is a crucial quality assurance activity within the computer software development process. The efficacy of unit testing significantly impacts the final quality of software products. However, given the rapid pace of software development and release in today's industry, conventional unit testing is gradually proving inadequate in meeting requirements due to its limited effectiveness. In light of the existing circumstances and challenges associated with unit testing, this article introduces the evolving trajectory of unit testing while considering effectiveness. It delves into the attributes and benefits of Unit Test-Driven Development (UTDD) as a central focus for advancing the approach to unit testing. Additionally, it explores avenues for further enhancement in this direction.

In this paper, a high transmission metasurface for subwavelength focusing of terahertz waves was proposed. A full 2π phase coverage with high transmission at target frequency was designed by varying the lengths of the cross resonators. The high transmission characteristics of the resonators were analyzed and the performance of the focusing was also discussed. The results show that the maximum transmission of the resonators reaches 0.92 and the terahertz wave can be focused with at the focusing point of a full width at half-maximum of 143 μm, which agrees well with the full width at half-maximum of 139 μm obtained by Huygens' principle. This device with the characteristics of flexible, thin and easy-integration exhibits the potential applications in THz imaging and communications, and also can be extended in the design of other planar THz components easily.

Despite the great performance achieved by deep learning-based image recognition techniques in recent years, low-resolution image recognition is still challenging in terms of improving model performance and reducing cost consumption for practical applications. In this task, we propose a solution to achieve low-resolution image recognition by appropriately reducing the resolution of the images in the database so that there is a better match between them and the image to be recognized. In the task of face recognition, we use the Histogram of Oriented Gradient (HOG) and Support Vector Machine (SVM) to determine whether the image contains a face or not, followed by the KNN model for face information recognition. In the number plate recognition task, we use the Convolutional Neural Network (CNN) model to recognize number plates. The experiments were validated on two publicly available datasets. The experimental results show that it can perform better in accomplishing low-resolution image recognition tasks while saving computational resources and other costs.

In response to the issues of information loss in high-density regions and alteration of the original point cloud shape caused by voxel-based point cloud filtering, a voxel filtering method based on octree and point density is proposed. After voxelization using the octree, within each voxel, all points are arranged in ascending order based on their distances from the voxel center point. Starting from the nearest point to the voxel center point and at a preset interval, points are selected, and these selected points constitute the new point cloud, representing the filtered point cloud. Experimental results indicate that the proposed method not only better preserves the original point cloud's shape but also retains information about high-density areas. Compared to voxel filtering, the proposed method exhibits smaller point-to-point distance errors between the filtered point cloud and the original point cloud. This approach introduces a novel perspective for point cloud simplification.

In order to improve the performance of focused crawler, a topic crawling strategy based on Concept Context Graph(CCG) is proposed. User knowledge background is constructed from the initial topic text returned by zhishi.me Knowledge Graph, and it is transformed into a CCG to guide the crawling of focused crawler. During the crawling process, the crawler treats the hyperlinks it encounters as a Virtual Concept(VC), and then calculates the Sememe Similarity(SS) between the VC and the core concept in CCG based on HowNet. According to the relationship between the SS value and the given threshold, it is decided whether to keep the link or not, and the CCG is updated for the reserved VC. The corresponding hyperlinks of the preserved VC are then stored in the queue to be crawled according to the order of SS value from high to low. Through experiments, it can be seen that the topic crawling strategy can improve the performance of focused crawler.

Unsupervised learning of probabilistic models is a core problem in machine learning. Under the foundation that the model of easy data modeling is good, the NICE model is proposed. A deep learning framework for complex high-dimensional density modeling is presented in NICE. In this paper, CovnFlow is added to NICE to construct a new coupled flow model. Although the basic model of NICE has been changed, the calculation of the Jacobi matrix is not much more complicated, and the newly proposed model will allow for a fuller fusion of information. So using this model to generate pictures will be more efficient and the effect will be better.

Aiming at the decision problem of electric vehicle charging station site selection, six principles of site selection andfivekey factors affecting site selection are proposed, and suitable evaluation indexes are selected from various aspects suchas natural environment, economic environment, power grid conditions, and a comprehensive evaluation model isestablished based on fuzzy comprehensive evaluation method and single factor rotation method. By establishingtheevaluation system of electric vehicle charging station site selection, the key factors of charging station site selectioncanbe understood. To solve the bottleneck problem that restricts the industrial development of electric vehicles, make anin- depth analysis and scientific evaluation of the site layout of existing and planned to charge stations. It also providestechnical guidance and reference for the site selection of charging stations under the large-scale industrial development of electric vehicles.

SSD is a single-stage target detection model that uses different scales feature maps to make predictions. However, due to the insufficient nonlinearity of shallow feature maps and the lack of semantic information, the detection effect of small targets is much lower than that of medium and large targets. To address this issue, this paper proposes a strategy of deep supervision and feature fusion method based on the idea of a feature pyramid network on the basis of the SSD. Through a deep supervision strategy, the same supervisory signal is applied to multiple scales of features before fusion, narrowing the semantic gap between different scale features and improving the representation ability of multi-scale features for better fusion. The top-down fusion of features extracted from the SSD backbone network is used to improve the problem of lack of semantic information in shallow feature maps. The experimental results on the NWPUVHR-10 remote sensing dataset show that the average detection accuracy for small target detection is 88.3%, higher than the average detection accuracy of DSSD (78.7%), FSSD (86.7%), and SSD (84.2%).

In the cigarette production process, detecting defects in the outer packaging is a crucial step. To address the issue of high false recognition rates in factory environments, we propose an improved YOLOv7-tiny target detection algorithm. Our algorithm is based on the YOLOv7-tiny network structure and enhances feature extraction and location by adding the SA module at the end of the Backbone. Additionally, we integrate the GSConv module into the network Neck part to improve information flow and feature fusion ability. Our experimental results show that adding the SA module and the GSConv module to the network increases average accuracy by 2.2% and 3.5%, respectively. After fusing the two, the average detection accuracy is 94.6%. Compared to the original YOLOv7-tiny, our improvements result in a 5.4%increase in accuracy.

In the past 10 years, many models of target detection based on convolution networks have been proposed, which are mainly divided into single-stage target detection models and two-stage target detection models. The main feature of single-stage is fast detection speed, and the main feature of two-stage is high detection accuracy. The purpose of this experiment is to select the most suitable target detection model for Qinglitchi by comparing the performance of the single-stage target detection model and the two-stage target detection model. After many and sufficient model training, the experimental results show that the YOLOX-S model has the fastest detection speed, and FPS is 37.965; The average accuracy of detection is the highest, with an average accuracy of 89.98%. Therefore, the YOLOX-S model is selected as the target detection model of green litchi.

Convolutional Neural Network (CNN) is the most successful mathematical model of an artificial neural network used in the fields of computer vision, image recognition, and classification. High-performance neural network structures are large in size, and a large number of multiplicative-additive calculations are required for a complete inference process. Training out a high-performance convolutional neural network model requires tens of times more computation than the inference process. The current theoretical and technical level cannot achieve a model that can be universally applied to all domains, and different application scenarios require designing specific neural network structures and collecting specific data sets. Large computing power requirements and high-quality data acquisition are two key components for training to obtain high-performance convolutional neural networks.

Although deep learning has developed rapidly in recent years, the existence of adversarial examples casts a shadow over its future. Deep neural networks (DNNs) can be unreliable when facing these carefully crafted adversarial examples, limiting their real-world applications. While adversarial examples usually lack robustness and tend to lose effect when the viewing angle is changed. Moreover, it is impossible to control the angle at which adversarial examples are seen by the airborne detector. Therefore, we propose a multi-view robust adversarial attack by introducing perspective transformation and feature similarity loss to mitigate the negative effect that viewing angle changes bring to the features. We evaluated our adversarial patch on our multi-view test dataset and the results show that our proposed method reduced the mAP by 73.4%. Furthermore, compared to other methods, our approach performs best in the data with low depression angles, demonstrating its excellent multi-view robustness.

For Simultaneous Localization and Mapping (SLAM) systems in static environments, the algorithms that have matured so far have obtained more accurate trajectories. Still, the optimization results of the back end in dynamic environments could be better due to the influence of various dynamic targets. Existing algorithms could perform better in dynamic environments with high illumination variation, low texture, and many dynamic targets. This paper uses a YOLOv5-based target detection module to screen potential dynamic targets in the front end and verify and reject dynamic feature points by subsequent polar line constraints. Experimental results show that such an approach can improve the system’s stability. The experimental results also show that the results are improved compared to ORB-SLAM2 for both indoor and outdoor dynamic environment datasets.

To address the problems of overfitting judgment results and the impact of the parameter on the accuracy of intelligent algorithms in fault diagnosis of the dredger pump, based on the analysis of common faults of centrifugal pumps, in this paper, a fault diagnosis method combining Support Vector Machine (SVM) and Bacterial Foraging Algorithm(BFA) for dredger pumps was proposed. By using the optimization ability of BFA, the optimal SVM penalty factor and kernel parameters are found, which improves the fault diagnosis ability. The superiority of the proposed method is verified by comparative analysis through simulation and actual examples. The results show that BFA has better optimization ability than traditional SVM.

In response to the identification problem of debris flow gullies in mountainous and canyon areas, a technologically advanced solution, namely the Two-Channel Residual Network (TCRNet) was proposed in this paper. This network uses DEM data and remote sensing data as network inputs and adopts a novel dual-channel architecture to extract spatial and spectral features respectively. To further optimize network performance, the network embeds the ECA mechanism to emphasize image feature information and adds a residual structure improved by global average pooling to output more information within the receptive field. The prediction results are evaluated for accuracy using the confusion matrix, and evaluation metrics such as precision and recall are calculated for model evaluation. Experimental results show that using deep convolutional neural networks to train DEM and remote sensing data of gullies in Nujiang Prefecture can achieve an 80% recognition rate, 0.79 recall rate, and 0.83 precision rate for debris flow gullies, indicating good model performance. In this study, the saved optimal parameters of the model were also used to evaluate the danger level of 672 gullies in Nujiang Prefecture and obtained four danger level zones: high, medium, low, and extremely low, which were visualized using ArcGIS software. These experimental results demonstrate that it is feasible to use deep convolutional neural networks to extract gully image features for rapid identification of debris flow gullies, providing important references.

The leakage of source code will lead to a serious information security crisis. It is of great significance for electric power mobile micro-applications to detect key codes and then protect them. Aiming at the poor representation ability of JavaScript code, we suggested a key code detection model based on a semantic convolutional memory fusion network. The fusion network was constructed by combining TextCNN and Attention-based BiLSTM to extract code features. The lexical unit sequence that is abstracted from the abstract syntax tree will be input to the fusion network for identification and classification. According to the experimental results, the proposed model shows improvement in the representation ability and overall performance in key code detection.

As the primary industry that can change the way of socio-economic development mode, the network information security of new energy power stations must be ensured. This article first designs a horizontal boundary exposed by pass monitoring scheme for the horizontal cross-district access method of new energy power stations. The K-nearest neighbor classification algorithm is used to match, identify, and classify the key factors of network flows with and without protection. A dedicated network remote monitoring method for automatic discovery process based on network topology information structure is designed for cross-district access process caused by dedicated network remote operation and maintenance. The dedicated network is located by the change of network topology. Based on the monitoring schemes of the above two scenarios, a dual-network card violation external monitoring system based on an improved agent is deployed in the network to detect high-concealment illegal cross-domain access processes in real-time in new energy power stations. The system verification finally proves the effectiveness of the proposed method.

With the rapid development of computer networks, a vast amount of information is transmitted and stored within them. However, each node within these networks could potentially be a source of information leakage. In order to protect data security, network security scanning is a necessary practice. This paper presents a Python scheduler for optimizing the performance of scanning multiple target IP addresses using the open-source network scanning software, nmap. The paper also examines the efficiency of multi-threaded calls to nmap by experimentally testing different numbers of target IPs and threads. By adjusting the number of threads appropriately, the scanning efficiency of multiple target IP addresses by nmap can be improved.

In opportunistic networks, the randomness of node mobility can significantly reduce the success rate of data delivery, while the buffer congestion of forwarding node can also cause network performance degradation. Considering above two issues, we design and implement a novel routing algorithm based on probability prediction (NRPP) for mobile opportunistic networks. NRPP is an improvement on the basis of ONRWF (our previous work), which can decide next hop routing by the improved variance probabilistic formula. Meanwhile, the proposed multicast RTS-CTS handshaking probing approach and efficient scheduling based on data priority which can significantly improve the success rate of data delivery. Finally, a test-bed is set up to validate the results in real scenarios. Experiment results demonstrate excellent performance of message delivery for opportunistic networks.

In recent years, the development of underwater and sub-ice wireless sensor networks has been promoted by advances in wireless sensor network technology. A key issue to improve the efficiency and performance of underwater robot networks is adaptive trajectory planning. This paper studies adaptive trajectory planning of multiple AUVs in the underwater environment for estimating the water parametric field of interest, using a reinforcement learning system where a fixed access point (AP) on the ice serves as a communication gateway. The water parameter field is modeled as a Gaussian process (GP) with unknown hyperparameters, and a field hyperparameter estimation method based on the maximum likelihood method is employed for more accurate estimation. Compared to the traditional Monte Carlo(MC) algorithm, the proposed algorithm demonstrates superior performance in terms of resource consumption and exploration rate.

Ensuring the safety and excavation efficiency of pumped storage power stations (PSPs) requires a focus on stability in the construction and design of underground cavern groups, along with the implementation of intelligent safety analysis and warning systems. This paper primarily focuses on the application of this system in Tiantai PSPs and highlights stability issues that may arise from uncertainties like joint fractures or poor rock conditions resulting in untimely support during construction. To accurately measure rock mass mechanics parameters, the paper recommends utilizing various mechanical responses of surrounding rocks and employing inverse analysis techniques for material parameter inversion. Additionally, the paper examines the deformation and stability effects caused by different factors such as lagged support and geological conditions through a case study that demonstrate support reinforcement practices at the powerhouse end. The paper also discusses suggested measures for dealing with these issues. Ultimately, this approach helps ensure both the safety and efficiency of the construction and operation of underground caverns while minimizing potential environmental and community impact.

Wordle is a popular word puzzle game offered by The New York Times, with two different modes: regular and difficult, which will affect the player to varying extent. Given that a report in New York Times shows that the number of reported results is changing every day, a time series forecasting model based on the ARIMA method was built on the data to account for the changes and create a forecast interval for the number of reported results on 1 March 2023. The Pearson coefficient was calculated to determine whether any change in the attributes of the language would have an effect on the percentage of the game scores gained in the difficult mode. For a given future solution word at a future date, a BP neural network-based model was developed to predict the relevant percentage for a future date. Based on entropy method and K-means algorithms to classify solution words according to difficulty, the attributes of a given word associated with each classification are identified. Experimental results show that the model is able to achieve good performance on most datasets.

The application of Internet provides a large amount of data basis for network vulnerability testing, but there are problemsof poor data quality and low accuracy. In order to fix this problem, this paper designs a network vulnerabilitytest method based on improved genetic algorithm. By improving and optimizing the encoding mode and fitness function, thetest data mutation library is integrated when genetic operation is carried out on the variable data segment of protocol message, which improves the efficiency of vulnerability mining. With the help of global search ability andcycliciteration of genetic algorithm, the test data with high code coverage is finally generated. Experimental results showthat this method is easier to trigger program anomalies than random search method, and improves the computational efficiency of genetic algorithm.

Analyze the encryption and decryption principle of KeeLoq and the shortcomings in security, propose an improvement scheme for KeeLoq algorithm to further improve its security; and conduct experimental verification by a BCM controller of a car, the experimental method is to combine the 027 service in the UDS protocol in KeeLoq algorithm to generate key authentication between the key side and the car side, the protocol generated temporary The key is passed with the factory key and serial number to derive a new password, and finally the key generation algorithm and the communication parties obtain the improved KeeLoq key in the learning process. The performance comparison with the original KeeLoq and triple KeeLoq algorithms concludes that while improving the security of the original algorithm, it reduces the complexity of the algorithm and increases the computation rate compared to triple KeeLoq. It is more suitable for PEPS system.

This article takes the recent notice from Guangdong Province on the "2022 Provincial Vocational College Satisfaction Survey" as the starting point and proposes that statistical analysis based on fuzzy mathematics theory and methods should be adopted to be more practical. After briefly introducing the importance of fuzzy mathematical methods and their applications, and aiming at the difficulties of multifactor fuzzy pattern recognition, we first start with the analysis of the fuzzy characteristics of the identified object, determine the mathematical model that conforms to the actual problem, adopt effective membership functions and reasonable model parameters, and then obtain the value of the comprehensive membership of each factor through the matrix operation of the membership of the relevant data in the fuzzy set, Finally, a comprehensive evaluation result is obtained.

With the deepening of Digital transformation in various industries, portal websites have become a medium for various industries to display their own image, and also a key platform for interaction and communication between industries and users. Some industry portal websites have complex website structures and unreasonable page link designs, resulting in insufficient user interest in browsing websites and difficulty in quickly obtaining the required information. This article is based on the Apriori algorithm, by mining and analyzing the user access logs of a certain industry portal website, grasping the browsing patterns of users accessing the industry portal website, analyzing the potential correlation relationships presented on the website's pages, providing a basis for website structure adjustment and page link optimization, and providing information reference for subsequent portal website reconstruction and optimization.

Network intrusion detection is a research hotspot of network security technology. To solve this problem, a network intrusion detection algorithm based on GrC-CVM is proposed. Firstly, through the idea of hierarchical granulation of Granular computing (GrC), the attribute features of the sample are reduced, the minimum feature set is found, and redundant attributes are removed. Then, using the advantages of Core Vector Machine (CVM) in processing big data, the intrusion detection model is constructed by using the reduced feature set. Finally, the model is tested by using test set samples and compared with other relevant algorithms. The experimental results show that the algorithm can greatly improve the detection efficiency while ensuring the detection accuracy, and is superior to other algorithm models, and can be used as a network intrusion detection tool.

The problem of traffic congestion is a significant challenge faced by smart city development. The causes of traffic congestion are complex and diverse, involving both routine factors and random factors such as unexpected events. These factors present challenges for accurate prediction of urban traffic flow. In this paper, we propose a global factor-aware spatiotemporal neural network model called GFA-STNet which addresses the cyclic, spatiotemporal, and random characteristics of urban traffic flow. We utilize deep learning to capture the spatiotemporal correlations of urban flow, employ residual networks to capture the temporal features of nearby time, periodic time, and trend time in the variation of traffic flow over time. Graph convolutional networks are used to extract the adjacency relationships between regions and combine them with external factors to extract the global spatiotemporal relationships of traffic flow, which are used to predict the traffic flow of each region. Experiments are conducted on real-world datasets, and the results show that the proposed model improves the accuracy of predictions, compared to classical traffic prediction models.

In response to the difficulty of constructing scoliosis assessment indices to locate anatomical landmarks and the low degree of automation, this paper proposes a back asymmetry index extraction algorithm based on ICP alignment. The method firstly uses an automated scoliosis screening system to obtain equally spaced cross-sectional contour line point cloud data of the human back, secondly extracts the region of interest ROI by image processing techniques, then performs ICP alignment of the source point cloud and the mirror point cloud generated using the sagittal plane, and finally calculates the asymmetry index of the source and matching mirror point clouds according to the nearest distance principle. The experimental results show that the correlation between the asymmetry index extracted by this algorithm and the cobb angle is strong (r=0.879, p<0.01), the diagnostic value of the asymmetry index for the Cobb angle is high (AUC=0.833). the method does not require localization of anatomical landmarks, can be automated, and is suitable for the assessment of adolescent idiopathic scoliosis.

With the rapid advancement of artificial intelligence, facial recognition technology has permeated various domains, revolutionizing our daily lives. However, this convenience comes hand in hand with security concerns. The emergence of facial spoofing attacks has raised serious issues concerning information security, financial integrity, and personal safety. To mitigate the risks associated with facial spoofing attacks, this research paper proposes innovative approaches and strategies to tackle this problem. To solve the problem of face spoofing attacks, the following work is proposed in this paper. Firstly, this work presents a novel framework named self-adaptive feature enhancement for FAS. This framework used RGB, depth, and reflection channels together by a feature extractor module. What’s more, this paper proposed a Cross-regional Feature Fusion (CFF) network, which added self-attention from vision transformer to improve the classification efficiency. Finally, the effectiveness of the proposed approaches and strategies is demonstrated through experimental results. The conducted experiments on several public datasets validate the success and performance of this research work.

As an important middleware of distributed computing, message queue enables us to address issues like decoupling applications, asynchronous communication, fluctuating data traffic, and others. Kafka is a popular open-source message queue for collecting, processing, storing, and analyzing messages at scale, which is famous for its excellent performance, low latency, fault tolerance, and high throughput. In some scenarios like electronic trading, reliable consumption is the utmost concern. However, the native consumer client provided by Kafka cannot guarantee that messages are consumed successfully at least once. This paper proposes a two-phase submission algorithm using dynamic programming which can improve reliability of message consuming. This algorithm utilizes a two-phase submission to ensure that every message is consumed successfully at least once, and uses dynamic programming to improve submission efficiency. We conduct experiments to demonstrate that this algorithm is more efficient in contrast with the original submission strategy. In experiments, this algorithm must make sure that every message is consumed successfully at least once while various consuming failure rates are mocked.

In recent years, with the development of hyperspectral sensor technology, how to accurately capture different classes of ground objects from hyperspectral images in a large area is an urgent problem. This paper proposes a piecewise continuous learning method based on a spatial-spectral relational network, a circular classification process, including image cutting, image classification, and result map mosaic. This method can directly fine-tune each sub-region without modifying the network structure and can infer the whole region's figure class information. Taking Houston dataset as the experimental dataset, this paper verifies the effectiveness of the proposed classification process. Further, it reveals the application value of the proposed method in large-area hyperspectral image classification.

Physics-based cloth simulation is a classic topic in computer graphics, which aims to simulate realistic cloth effects in virtual environments. In recent years, the rapid development of virtual reality has put forward higher requirements for physics-based cloth simulation, and this technology has played an increasingly important role. This paper first introduces the fundamental knowledge of cloth simulation, then summarizes the existing mainstream methods in this field. We give a classification of these works and provide an overview of each one of them. We have also conducted a large number of experiments to compare the performance of several typical algorithms. Towards the end of the article, we identify some existing limitations in this field and provide targeted suggestions for future development.

With the rapid development of smart grid, the number of smart substations is increasing, and the establishment of a unified security isolation container is beneficial to the realization of unified information modeling and data management of the whole station, so the application research and development based on security isolation container is inevitable. The invention relates to an intelligent alarm APP system based on a safety isolation container established on the basis of a unified platform, which acquires alarm information data from the platform through an interactive information interface between the platform and the APP, forms a knowledge base based on fault logic description language modeling, and finally calls the interactive information interface between the platform and the APP to transmit inference information to the platform after fault reasoning and diagnosis. The practice shows that the intelligent alarm APP based on the security isolation container can perform independent start-stop operation, independent process white list, network white list and other security settings, and can intelligently process the fault information in real time and reliably, and support the integrated operation control and independent safe and stable operation of the smart grid.

To improve the performance of single-channel speech separation, this paper proposes a time-domain single-channel speech model based on the attention mechanism, utilizing the SANet network and a density-based spherical fuzzy clustering algorithm. The model incorporates the attention mechanism into the TCN embedding network to enhance the focus on important features and effectively capture the long-term dependencies in speech feature data. During the testing phase, the density-based spherical fuzzy clustering algorithm is employed to compute the positions of attractors, thereby improving estimation accuracy and reducing the impact of outliers and noise on cluster centers. Experimental results demonstrate that the proposed model achieves good performance in speech separation tasks.

Regarding the association between maternal mental and physical health and infant behavior, this paper proposes a method that integrates Grey Wolf Optimization (GWO) with Extreme Learning Machine (ELM) for the correlation and classification of infant behavioral features and maternal health indicators. Firstly, a dataset of maternal mental and physical health and infant behavioral features is established. Using the Random Forest method, physical and psychological indicators of the mother and their eight primary influencing factors are extracted from the dataset. Subsequently, the classification accuracy of infant behavior is used as the fitness function for the Grey Wolf algorithm. The GWO algorithm optimizes the kernel and regularization parameters of the ELM, which are then fed into the ELM to train the four key feature parameters of infant behavior. Finally, it is applied to the classification task of infant behavior. The results indicate that the classification method combining GWO and ELM has an accuracy of 86% and a precision of 91%. This outperforms traditional classification models such as XGBoost and GBDT in detection and classification. It also surpasses the performance of ELM optimized by traditional optimization algorithms like SSA and PSO, validating the feasibility and effectiveness of the proposed method.

This paper studies the prediction and performance comparison based on BP, PSO-BP and SA-BP models. Taking real estate as an example, because of the complex and nonlinear relationship between its influencing factors, the accuracy of traditional prediction methods is low. Therefore, this paper proposes an innovative prediction model based on SA-BP neural network, and verifies its effectiveness through 110 sets of real estate transaction data in Shenyang from 2021 to 2022. The experimental results show that the prediction accuracy of PSO-BP and SA-BP models is better than that of BP model, and SA-BP model performs best. This study fills the research gap in related fields and provides valuable reference for market decision-making such as real estate.

In order to reduce carbon emissions in the construction industry, it is crucial to study the factors that affect carbon emissions in this sector. Based on statistical yearbooks and the characteristics of carbon emissions in the construction industry, this study establishes a measurement method for national carbon emissions in the construction industry. The Logarithmic Mean Divisia Index (LMDI) model is utilized to calculate direct and indirect carbon emissions in the construction industry from 2010 to 2020. The study also explores and organizes the driving factors of carbon emissions in the construction industry. Based on the empirical research using LMDI decomposition, it is found that the total electricity consumption and carbon emission coefficient have a promoting effect on carbon emissions in the national construction industry. The effects of per unit area energy consumption and per unit electricity consumption on carbon emissions over a 10-year period are more promoting than inhibiting. Based on the LMDI decomposition results, the concept of "measuring carbon with electricity" is proposed to explore the relationship between annual electricity consumption and annual carbon emissions in the construction industry. This approach allows for the estimation of annual carbon emissions by knowing the annual electricity consumption in the construction industry. Three prediction models are compared to evaluate their effectiveness, and it is found that the Gaussian regression fitting model produces better results. According to the prediction results, it is projected that national carbon emissions in the construction industry will continue to rise over the next five years. Therefore, it is crucial to have sound national planning for the construction industry and to enhance awareness of low-carbon and environmental protection measures in order to achieve the goal of "dual carbon" reduction.

Facial Expression Recognition (FER) is a crucial research field, holding significant importance in understanding human emotions, communication, and social behavior. Although traditional FER methods perform well in laboratory settings, they often exhibit significantly reduced accuracy in the wild due to issues such as environmental noise, lighting variations, and individual visual differences, along with class-to-class and within-class variability in facial expressions. In response to these challenges, this paper proposes a method for FER in the wild based on Convolutional Neural Networks (CNNs) and Graph Convolutional Networks (GCNs). The method first extracts facial expression features using CNN, fully leverages the relationship between facial expression samples to model them, and then uses GCN to learn the generated graph structure for classification purposes. The experimental outcomes demonstrate that our approach outperforms traditional methods in the task of FER.