This PDF file contains the front matter associated with SPIE Proceedings Volume 12602, including the Title Page, Copyright information, Table of Contents, and Conference Committee lists.

Compared with traditional weapons, high-speed weapons have the advantages of high speed, high precision, and strong maneuverability. In future wars, high-speed weapons will occupy an important position. While developing offensive highspeed weapons, countries are also considering how to deal with the threat of high-speed weapons. Due to their excellent strike effect and penetration performance, high-speed weapons can effectively break through the existing defense systems of various countries. Military groups represented by the United States and Russia are accelerating the construction of highspeed weapons defense systems. This paper first describes the development status of high-speed weapons, and summarizes the situation and development trend of the defense system construction by various military powers. Then combined with the relevant characteristics of high-speed weapons, from the aspects of early warning and detection, command and control, and interception weapons. At the same time, this paper sorts out the key problems faced by the existing defense system. Finally, some enlightenments and thoughts are put forward on building a defense system under the threat of highspeed weapons.

As the power industry develops, the scale of physical assets of the power grid is constantly expanding, which makes management difficult. To ensure the stability and security of power supply, it is worthwhile to study how to improve efficiency and achieve precise management. At present, some power grid enterprises lack effective means to fully understand their own physical asset management level so that they cannot reasonably allocate enterprise resources to achieve ideal performance. This paper suggests a scientific and effective comprehensive evaluation method of physical asset management to help power grid enterprises solve the current problems. A comprehensive evaluation index system of asset management applicable to power grid enterprises is established, and a method to calculate the weights of evaluation indexes is proposed by applying game theory ideas. The effectiveness of this evaluation method is verified by the example of State Grid Jibei Electric Power Co., Ltd. The results indicate that it has a high practical significance and can provide a basis for power grid enterprises to make scientific decisions in asset management.

With the acceleration of information construction, the network structure and technology of information system are full of diversity. How to effectively resolve security risks and effectively deal with various unexpected security incidents has become a problem that cannot be ignored. Different manufacturers and departments have different means of managing the software and hardware of the equipment, and it is difficult for the monitoring master station to correctly interpret the various types of messages they generate. First, the paper introduces the concept of SOC, secondly, it proposes a new SOC model, and finally solves two difficulties in the process of parsing and normalization: data parsing with multi-level structure and integration of data with different formats. The new SOC model can normalize the message content in nonstandard format and effectively improve the master station's advanced analysis and application of system and equipment events.

Time management is fundamental to an enterprise and can affect its development. An efficient time management system can play an important role in supervising employees and improving the productivity of an enterprise. However, today, an enterprise's working hours management suffers from reliance on paper carriers, poor real-time, data leakage, inefficiency and reliance on local servers. As local servers have limited carrying capacity, adding local servers will disguise the financial pressure on SMEs, increase SMEs' labour costs and have poor energy efficiency, thus failing to meet the man-hour management requirements of modern SMEs and seriously slowing down their development. Therefore, this paper applies the theory and method of cloud environment data technology to design and study a working hour management information system of an enterprise based on cloud environment data technology, which can manage the whole cycle and time of employees, improve the efficiency of employees, effectively control the project planning progress, and greatly enhance the core competitiveness of the enterprise.

In order to solve the problems in the monitoring of algae concentration, such as the traditional fixed-point monitoring method covers a small area, the remote sensing monitoring method has a large error, the processing algorithm is complicated, in addition, the vertical distribution and change pattern of algae concentration cannot be obtained, the three-dimensional monitoring system of algae concentration based on the unmanned ship is designed. First, design a long-endurance unmanned ship with fuel-powered, and onshore terminal can control it to reach the designated measurement position; Second, the unmanned ship carries the submersible, and the shore side can control the ascent and descent of the submersible, as well as the acquisition and transmission of sensor data. Combined with satellite positioning data to display real-time the navigation video of the unmanned ship, various sensing data such as measured data of algae concentration, water depth pressure and so on, to complete the three-dimensional monitoring of algae concentration. Finally, tests were carried out in the sea near Tielu Port, Sanya, through the measurement of 144 locations at 9 stations, it was found that the vertical distribution of algae had phototaxis, showing that its concentration gradually decreases from the surface to the depths of the ocean.

An improved MUSIC method is extended to solve the parameters estimation of cyclostationary sources. The method constructs two cyclic moment matrices from the data received by the centro-symmetric uniform linear array, and uses the propagation operator to estimate the noise subspace, then estimates the parameters by the root music method. In contrast with the traditional ROOT-MUSIC, although the accuracy is slightly worse, it has low computing complexity. Furthermore, the method is applicable to the case of additive Gaussian noise environment, whether white noise or color noise.

It is a major research focus of battlefield situation visualization to identify targets quickly when the background has a similar color. A 3D heat map display technique based on volume rendering is presented in this paper. The method uses an improved kernel density function of a dual-kernel heat map and an ellipsoidal ray casting algorithm. In this paper, the method is verified by experiments, and its superiority is illustrated from many aspects. This method can not only quickly identify the target when the color of the target is similar to that of the background, but also distinguish the importance of the target. In addition, the ellipsoid ray casting algorithm can be used to perspective the potential in 3D space to a certain extent.

Natural language processing is a research direction in many fields such as linguistics, computer science, and data fusion of study. The representation of word vector is a method to map words into the real vector space, which is the core technology of many current natural language processing tasks. This paper summarizes and studies some typical expression methods of word vector as well as research the word vectors of linguistics and mathematical principle. We first elaborate the process of mapping the words to the vector, namely, encoding natural language information to word vector according to semantics. Secondly, we analyze several typical methods such as co-occurrence matrix, Word2Vec, GloVe, ELMo on information carrying capacity. Thirdly, on the basis of analyzing the principles of these methods, this paper also uses SVD decomposition, neural network, and other methods respectively to reproduce the specific process of generating word vectors. Finally, combining word similarity calculation and text sentiment classification task, we compare the performance of word vectors trained by various methods in different tasks. Experiments verify the conclusion that different word vector generation methods have different emphases in carrying linguistic ability and perform differently in different tasks.

Aiming at the difficulty of radar signal detection in low signal-to-noise ratio (SNR) condition with traditional methods, a stacked auto-encoder (SAE) and support vector data description (SVDD) based detection method is proposed. Firstly, the radar signal with noise is extracted by SAE to obtain the representative features. Secondly, the SVDD is trained with the extracted features to obtain a spherical discriminative boundary for classification offline. Finally, the trained SAE-SVDD used as the one-class classifier to detect the signal by minimizing both the reconstruction error and the hypersphere volume simultaneously in a real-time manner. Simulation results indicate that the proposed algorithm can extract and identify the radar signal under noise condition effectively with a good robustness. It has practical significance for improving the accuracy of radar signal detection under low SNR.

Cardiovascular disease (CVD) is a serious threat for human health, its death rate outclasses other diseases. Open heart surgery, which is one of the main treatments for cardiovascular disease, is a form of invasive surgical treatment. Therefore, effective detection and analysis of cardiomotility based on body surface physiological signal have the advantage of noninvasive and nondestructive for CVD patients. Meanwhile, electrocardiograph (ECG) signal is considered a form of nonlinear, nonstationary and faint physiological signal on body surface. Heart rate variability (HRV) can be extracted and analyzed from waveform of ECG signal. In this paper, an ECG signal analyzing method based on EEMD and Poincaré plot is proposed, combined with an improved adaptive noise filtering method. First, IMF components are generated from ECG signal using EEMD. Through IMFs selection based on improved adaptive denoising method, reconstructed ECG signal is acquired. Next, RR interval sequences are extracted from original and reconstructed ECG signal respectively. Poincaré plot analysis is used on RR interval sequences and feature parameters are extracted. Finally, statistical analysis has been executed. The original data in this research were collected under open heart surgery at Far Eastern Memorial Hospital in real time. Preoperative, intraoperative and postoperative data were included. The experimental results illustrate, all features extracted from postoperative reconstruction ECG signal prove high significant difference (p<0.01) comparing to preoperation at least. The results have been discussed with physicians, and already been used for clinical auxiliary analysis. The method put forward in this paper is demonstrated feasible.

To meet the application requirements of low cost and high reliability of passive indoor localization system, this paper proposes a device-free indoor localization method based on channel state information (CSI) spatial awareness (BADIL). Firstly, we obtain the CSI through the 5300 NIC, analyze the spatial relationship between CSI fingerprints at different locations, and combine the amplitude and phase information to establish a more stable CSI spatial awareness (CSA) model. Then we design the representation method used to construct the training and test sets of this model. Finally, we combine BP neural networks to implement a device-free localization method in different indoor environments. The results of the experiments have proved the effectiveness of the proposed scheme.

Compared with biological nanopores, solid-state nanopores（SSN） have advantages in chemical, mechanical, and thermal stability, pore size tunability and integration, and detect a broader range of targets. However, the current problems of low signal-to-noise ratio (SNR), inefficient preparation, and poor reproducibility of solid-state nanopores remain unsolved, limiting their application to replace biological nanopores as the main sensor devices. In this paper, we address the low SNR problems of solid-state nanopores, introduce the recent research done by researchers to improve the SNR of solidstate nanopores, make a breakdown into the existing resistive pulse detection field, and give an outlook on various sensing detection methods based on nanopores.

Semantic segmentation of urban-scale point clouds is widely used in aviation, unmanned aerial vehicles, and autonomous driving. However, owing to the many points in the urban-scale point cloud dataset and massive computation in the learning process, traditional networks often have poor segmentation performance and high costs. In this study, we adopted the Point Transformer network as the baseline and integrated random point sampling and attentive pooling into a new transitiondown block, embedded in the encoder structure of the baseline to improve the speed and accuracy of semantic segmentation. On the challenging SensatUrban dataset, the Point Transformer network and the proposed network obtained mIoU values of 71.1% and 76.8%, respectively. The results show that the proposed network effectively improves the shortcomings of the Point Transformer network and achieves better semantic segmentation performance of urban-scale point clouds.

Weather forecast is very important for people's production and life. In recent years, with the rapid development of deep learning, the powerful feature expression ability of deep neural networks makes the effect of prediction obviously improved. In this paper, the Transformer combined with the Seq2Seq model to extract the spatial and temporal characteristics of data, and multiple meteorological elements in the future can be predicted from the historical data of multiple meteorological elements. Through comparative experiments, the effectiveness of the Transformer combined with the Seq2Seq model in multi-factor short-term weather forecast has been verified, and the accuracy of meteorological factors prediction has been improved.

Low imbalanced and fully matched active baluns have been designed for 21GHz up-conversion Gilbert mixer’s ports. It consists of a common-gate FET and a common-source FET with input and output matching networks. The balun has been realized in 0.25μm GaAs pHEMT process. Simulation shows that the phase error of LO balun could be controlled within 2 degrees over the frequency of 3-6GHz, and the phase error of IF balun could be controlled within 3 degrees over the frequency of 16-18GHz.All ports are matched at the impendence of 50Ω for mixer use. The inhibition of the Gilbert mixer using proposed active baluns is more than 40dBc.

The application of non-contact radar to monitor the parameters of human vital has become a topic of concern to many researchers in recent years, but the vital signals collected in the monitoring process can always contain interferences. This paper proposes a method to deal with the separable interferences with the original signal in the time domain or frequency domain: after the contaminated segment is cut off, the contaminated segment is reconstructed by using the estimation results of the non-contaminated segment, and iterative processing is performed to improve the accuracy of estimation of human respiratory rate and heart rate. The results of numerical simulation and measured data processing show that the proposed algorithm can effectively improve the accuracy of human respiration rate and heart rate estimation compared with the direct clipping of contaminated segments.

With rising pressure on energy and the environment, clean and renewable new energy sources have received a lot of attention in recent years. A hybrid energy storage system made of flywheel energy storage and battery energy storage is offered in this work as a technical way to increase the sustainable energy performance for low-carbon power grid development. In this paper, a simulation model of a hybrid flywheel-battery energy storage system based on autocoupled PID for a wind turbine-based power generation device is built in Matlab/Simulink, and the proposed control strategy of a triple full bridge topology based on autocoupled PID is simulated and verified.The results show that the hybrid flywheelbattery hybrid energy storage system can effectively reduce the amount of grid frequency variation and lower the wind turbine output power fluctuation; the autotransfer PID control strategy is used for optimization of the power distribution coefficient in the hybrid energy storage system, and it is discovered that the autotransfer PID control strategy can make the hybrid energy storage system maintain good power variation adaptation.

In traditional signal processing, signal sampling has to meet the Nyquist sampling theorem, so a extremely huge amount of data has to been processed during the measurement process of testing target’s radar cross section (RCS) in the microwave anechoic chamber. Based on compressed sensing theory, the RCS data of azimuth domain is sparsely sampled by the proposed method at a rate lower than Nyquist’s sampling and then the original data of full sampling amount is reconstructed by the reconstruction algorithm. When the sampling compression rate is more than 30%, the data reconstruction error from the standard target object doesn’t affect the evaluation of the target’s RCS performance. The proposed method of sparse sampling and reconstruction can significantly reduce the amount of measured data in the RCS test of target object, thereby effectively improving the test efficiency and shortening the test period.

Traditional interaction techniques such as mouse and keyboard or gestures for a touch screen are utilized for a 2D electronic sand table，which don’t have enough degrees of freedom (DoFs) for operations on a 3D graph. Gestures interaction method has received increasing attention for 3D electronic sand tables with solid terrain construction and virtual-real fusion display techniques. This paper researches an interactive feedback method with gestures based on contextual information, gestures controlling virtual targets and a gesture interaction modal are designed. Users can control virtual targets with intuitive and feasible interaction methods through the real calculation of a gesture recognition engine and a gesture understanding engine, and the displaying status of the virtual targets can be updated in real time.

Synthetic aperture radar (SAR) system and communication work in the same frequency band, SAR signal will be interfered by communication signal. In this paper, a method based on the combination of intermittent Interrupted SAR and communication is proposed to realize the simultaneous work of radar and communication without affecting each other, so as to realize the integration of radar communication. By analyzing the time domain characteristics and frequency domain characteristics of communication signals, the communication signal is a narrow-band signal in frequency domain, part of the bandwidth of the SAR working frequency band is given up for communication. At this time, the SAR signal is interrupted. The interrupted part of the SAR signal is recovered by linear prediction extrapolation to obtain a complete SAR signal, which solves the interference problem of the SAR system from communication and realizes integration. The effectiveness of this method is proved by SAR point target experiments.

Focusing on the differentiated and combat needs of maintenance support capability evaluation for an early warning detection equipment, the process and influencing factors of maintenance support for this type of equipment are analyzed, and the four level maintenance support capability evaluation index is constructed. Based on analytic hierarchy process (AHP) and fuzzy mathematics theory, the construction and solving process of multi-level fuzzy comprehensive evaluation model are studied. Then, the key technical problems such as different indexes normalization are discussed. Finally, the maintenance support capability evaluation system is researched and developed, which can provide decision support for equipment maintenance support improvement.

The real-time problem of signal processing of high frequency surface wave radar (HFSWR) has become a hot point. The key of digital signal processing is the method based on two-second fast Fourier transformation (FFT). In this paper，the application of GPU to accelerate the signal processing based on CUDA is introduced. The principle of signal processing and GPU parallelization is explained and elaborated, and the comparison of performance is showed in this paper.

At present, there is a lack of efficient numerical calculation methods for electromagnetic scattering from an infinite ground by a long duration source. Based on the two-dimensional frequency domain finite difference (FDFD) method, the surface impedance boundary conditions (SIBC) suitable for TM mode are derived, and an efficient FDFD-SIBC hybrid method is developed, which can be used to quickly calculate the ground electromagnetic scattering caused by long-term current source radiation. Firstly, the finite-size ground is intercepted and equivalent by using impedance boundary conditions to form a half-space calculation region. Then, according to the electric field component of a grid adjacent to the ground, combined with the ground reflection coefficient and the wave propagation phase delay factor, the FDFD formula of the electric field at the impedance boundary is derived. Finally, the FDFD method is used to construct the matrix equation between the electric field and the source of all grids in the half-space, and the complex conjugate gradient method is used to solve the equation to obtain the electric field distribution. The numerical simulation results of half-space ground electromagnetic scattering are in good agreement with the traditional FDTD method, which verifies the correctness of the FDFD-SIBC hybrid method.

The TCR high-voltage static dynamic reactive power compensation device is widely used in municipal distribution networks such as smart construction sites. The purpose of this article is to propose a smart construction site reactive power compensation device system, which can ensure high-precision reactive power compensation device state feedback control.

This study mainly uses virtual reality technology (Virtual Reality, VR) to create the basic theoretical knowledge and analysis of real-time data processing and GPU parallel computing in Spark+GPU heterogeneous environment. The quadratic programming algorithm and time sliding window algorithm are analyzed in detail from theory to practical application, and in the Spark+GPU cluster environment built in the virtual environment, the feasibility of the above algorithm is verified by real-time data processing, GPU scheduling and parallel computing. Through the numerical experiments of test functions and practical engineering applications, the GPU optimization after interval segmentation is more powerful than CPU, the algorithm is sensitive, and the global optimization of surface model provides a new solution.

Phase-sensitive optical time-domain reflectometry (Φ-OTDR) can monitor vibration signals in real time, but lack the ability to identify multiple events such as human activity and vehicle activity. With the help of machine learning methods, events can basically be identified. For the recognition of events such as human activities and vehicle activities, and to improve the classification accuracy, this paper adopts an event recognition method based on the combination of deep learning and traditional classifiers. The spatiotemporal data matrix from Φ-OTDR is directly sent to a convolutional neural network (CNN), and data features can be acquired automatically. The data features are sent to traditional classifiers for further classification. In this paper, support vector machine (SVM) is selected as the classifier to improve the classification accuracy. The experiment collected 4428 sets of data of four events. Experiments verifies the feasibility of multi-event recognition system of φ-OTDR based on CNN-SVM, and realized the recognition accuracy rate of four events was 68%- 81%.

The free space optical communication information transmission rate is high, which is attractive for high-speed communication scenarios. Compared with the near-Earth communication link, the satellite-to-ground laser communication link has the characteristics of long link distance, cloud layer and atmospheric turbulence. Aiming at the problem that the spread of star-to-ground laser communication pulses produces inter-code crosstalk and reduces the quality of the communication system, this paper analyzes the relationship between gauss optical pulse time domain widening and intercode crosstalk, and uses adaptive filters to balance the channels. The results show that the laser pulse broadening amount increases with the decrease of the input light pulse half wide, and the use of adaptive filter can effectively reduce the influence of code-to-code crosstalk caused by pulse broadening on the communication quality.

In 2021, the airline industry was affected by COVID-19, and many airlines suffered losses. The main reason for the loss were the decline in revenue and the surge in costs. Therefore, in terms of creating the competitive advantage of airlines, "price war" is no longer applicable, and improving service quality has become an effective means. Customer satisfaction is the most effective indicator to measure service quality. In this study, a satisfaction evaluation system is established based on structural equation model and customer satisfaction importance matrix. Then, a questionnaire is designed to analyze the influence of different factors on customer satisfaction. The research finds that brand image and perceived quality have a great impact on customer satisfaction. In addition, some suggestions for airlines to improve customer satisfaction are given.

In order to solve the problem that there are many multi-way valve test items and the test steps are cumbersome, a set of multi-way valve intelligent test system is designed in this paper.According to the test requirements, the corresponding hydraulic system and electronic control system are designed, and the measurement and control software with multiple functional modules is developed based on Labview.The software has a login module, a sensor parameter setting module, a sub-menu module, a data query module, and a main menu module.The test results show that the function of the test system is perfect, and the test results meet the user's requirements.

In view of tunnel engineering characteristics, such as "high ground stress, high gas concentration, many fault-fracture zones/unfavorable geological conditions/groundwater corrosiveness types and large buried depth," research on key tunnel digital construction technologies was performed to integrate engineering construction, geological investigation report and geographic information 3D models into 3D visual data information. In the research, large BIM models are lightened with the mass cutting method based on the combination of BIM and GIS of the Cesium engine. A heterogeneous model integration method is proposed to convert BIM into FBX format. Then FBX is converted into 3DTiles standard format suitable for Cesium engine, thereby combining BIM and GIS technologies. In the end, a B/S architecture-based case is demonstrated, which solves the tunnel safety warning problem by realizing a visual tunnel safety warning. Site data is loaded in the full element 3D model in a real-time manner with the combination of Cesium engine and tile & lightweight technologies to achieve back-end "data middle platform" integrated operation and 1:1 mapping of front-end "digital construction site" and "real construction site," which has reference value to the application of BIM + GIS technologies in the railway tunnel industry.

The popularization and development of the Internet has prompted people to enter the era of informatization and digitization. With the development of new media technology and the influx of massive amounts of information, the problem between human information needs and information has changed from insufficient information to "useful" information acquisition difficulties. As a form of new media, short video has been favored by many people because of its unique characteristics since its emergence, but if the short video platform wants to get healthy and sustainable development, it needs to meet the personalized information needs of users to the greatest extent. At this time, the acquisition and analysis of users' personalized needs has become one of the focuses of the platform's work. Taking the short video platform as an example, this paper uses the K-means clustering analysis algorithm in data mining and the Apriori association rule mining algorithm to analyze the user's personalized information needs, and then according to the users requirements analysis, a user personalized information recommendation system for short video platforms - "KA-R" system is proposed to meet the purpose of meeting the personalized information needs of users.

The LADI-FDTD (leapfrog alternative direction implicit finite difference time domain method), was widely used to solve the problem of electromagnetic calcula-tion on large size because it overcomes the stability condition of traditional finite difference time domain method, and thus improving the efficiency of calculation. There are a lot of problems are axisymmetric calculatio in the simulation especial-ly in cylindrical coordinates and spherical coordinates. So LADI-FDTD is applied to calculate axisymmetric problems, however it’s unconditionally stability is destroyed by iterative matrix. The instability of propagation located at iteration equation on axis line of symmetry by observing the electromagnetic wave spatial propagation. Based on this phenomenon, an improved stability method of ax-isymmetric LADI-FDTD is proposed, then numerical result proved that given method still has excellent stability under the condition of large time steps.

Heart sound is an important basis for analyzing heart health. The feature extraction and classification model of heart sound signal are optimized, and an auxiliary diagnosis algorithm for congenital heart disease based on multi-modality and dense residual network is proposed. This method does not need to segment heart sounds, but only needs to extract double features in the time domain, which simplifies the process of preprocessing and feature extraction. Double features are learned using a parallel double-branch convolutional recurrent neural network with dense residual connections. The proposed classification algorithm was trained, validated, and tested on a total of 4050 clinical congenital heart disease heart sound samples, and obtained a classification accuracy of 97.60%, which is expected to be used for clinical auxiliary diagnosis and screening of congenital heart disease.

Given the speedy development of Internet technology, the amount of information we get is increasing explosively, so the extraction and summary of information are extremely important. Text summarization technology, as one of the important research of natural language processing, can extract text expressing core ideas from massive data, to quickly obtain the required information. However, the traditional recurrent neural network has weak parallel computing ability, and there are cases of inconsistency with the original text content and fabrication of facts. To solve the above problems, an abstractive summary model combining a pre-trained language model and knowledge enhancement is proposed. The model incorporates the extracted factual knowledge into the summary generation process with the help of graph networks, to retain the original meaning to the greatest extent. Finally, the model is verified by standard text dataset CNN&Daily Mail and good experimental results were obtained.

Aiming at the multi traveling salesman problem(MTSP) of balancing workload, this paper proposes two local optimization operators for heuristic planning algorithm. According to the workload constraint, all the cities in the travel path are allocated to k sub tours, and a feasible solution is obtained. Then, the local optimal solution is found through the local optimization operator to optimize a single sub tour and the method of exchanging vertices between multiple sub tours. Genetic is used for global search, and there are two optimization operators in the local optimization process. Among them, eliminating the inclusion relationship operator is used to eliminate the inclusion relationship between sub tours, and the eliminating intersection relationship operator is used to avoid the intersection between different sub tours. These two operators tend to actively shorten the total length of travel, so the convergence performance is good. Data experiments show that the optimization operator in this paper has obvious optimization ability in solving similar problems of MTSP.

Due to the limited field of view of monitoring equipment and the high cost of radar equipment, the traditional method of image identification of ships is not effective. In recent years, with the rise of embedded devices, ship trajectory data has been widely collected, which provides the possibility of ship tracking method. If a simple convolutional neural network is used to extract the data information in the ship track, the timing information in the data cannot be obtained. Therefore, this paper proposes a ship path classification method based on deep learning(LSTM-ResNet model). This model combined with the feature of convolution network extract the path in the boat, LSTM model is utilized to extract implicit in the temporal feature vector data, the characteristics of the two models after the output through the model merge to form a new model, after the former sequence pretreatment process, the data is to heavy, cuts, after filling, ships track recognition model based on convolutional neural network, it has obvious effectiveness.

In order to explore the thermal instability of Silicon Carbide Based Vertical Double Diffused Metal-Oxide Field Effect Transistor（SiC VDMOS） at high temperature, it has been shown that the temperature coefficient of the device saturation current has a change from positive to negative, and it is speculated that this is related to the change of the threshold voltage. Based on this, this paper builds a high-temperature test platform for the electrical parameters of SiC VDMOS, and tests the output and transfer characteristics and threshold voltage of SiC VDMOS in the operating temperature range of 300K ~ 475K, which verifies the correctness of this view. It is preliminarily determined that the change of the threshold voltage is related to the interface state, A 1.2 kV SiC VDMOS high temperature simulation model based on Silvaco TCAD was built. By using high-K dielectrics with different dielectric constants including Al2O3, Si3N4, and HfO2 as gate materials, and comparing their respective zero temperature coefficient points, it is found that the thermal stability of the device is greatly improved after using high-K dielectrics. The correctness of the speculation is verified.

Liquid crystal is a tunable material, which has the characteristic of low loss in high frequency band, and its relative permittivity can be controlled by external voltage. This paper has proposed a 4×4 beam reconfigurable array antenna based on liquid crystal material in Ku-band. The array antenna is designed to be 40×40mm in size. Its scanning angle achieves ±30° at center frequency of 16GHz, while it reaches gain of 17.1dBi. The proposed antenna broadens the application of liquid crystal materials in microwave frequency band.

Although the detection accuracy of salient object detection (SOD) by deep learning has met most of the task requirements, how to learn the detailed information such as the edges of objects still has been challenging. Most of the existing SOD methods use interpolation operations in upsampling, which makes it difficult to capture the detailed information of salient objects. And this detailed information helps to improve the detection accuracy, so this paper proposes a high-frequency edge detail enhancement network of salient object detection to obtain the high-frequency detailed information of images and use this information to refine the result. To adaptively detect salient objects of different scales, the proposed method first uses a scale-varying detection module to spatially process the features of different sizes obtained by the backbone network. Then, the high-frequency residual block is used to obtain the high-frequency details lost due to interpolation operations. Finally, the high-frequency cascaded enhancement module utilizes the positional alignment mechanism to fuse high-frequency information into original salient objects features and obtains the final predicted results. Experimental results on five benchmark datasets demonstrate that the proposed method without any post-processing performs favorably outperform 12 state-of-the-art approaches.

With the development of globalization and information technology, space laser communication technology has gradually developed from point-to-point mode to point-to-multipoint mode of simultaneous laser communication. In order to solve the problem of high-speed data transmission in one-to-many laser communication relay network, laser with different polarization angles is affected by the atmosphere to different degrees, this paper built an all-optical wavelength conversion system under atmospheric turbulence environment based on OptiSystem simulation platform. The all-optical wavelength conversion technique of OOK modulated signal based on FWM parameter effect with double pump structure is studied. The results show that the all-optical wavelength conversion scheme with parallel dual-pump structure can realize the polarization insensitivity of the signal and reduce the influence of atmospheric turbulence on the communication system.

In view of the problems of ship target detection in remote sensing images, such as small ship targets, densely arranged ships, and arbitrary target directions, this paper proposes a target detection method based on improved YOLOv5. This method combines the improved DesnesNet and X-RFB modules to enhance the feature extraction of large-scale changing targets; it adds direction information on the basis of the traditional bounding box, and uses KFIOU instead of GIOU to perform curve optimization on the loss function; At the same time of quantification, high-speed and high-precision coexistence of ship target detection can be achieved. The experimental results show that: compared with the original YOLOv5 method, the map index of this method is improved from the original 91.0% to 96.5%, which is obviously better than the comparison method.

Few-shot learning(FSL) aims to generalize model to novel categoeries by few labelled samples, which is challenging for machine. Large-scaled pretrained models, especially vision transformers achieve excellent performances benefiting from numerous and diverse data. Researchers have exploited pretrained models in few-shot classification by simply updating the whole parameters and finetuning on few samples. In this paper, we explore two methods: vision prompt tuning and a reparameterization method called ‘scaling&&shift’ to leverage pretrained models in few-shot classification. Vision prompt tuning is for vision transformer only and we first evaluate the method in few-shot setting. ‘Scaling&&shift’ is originally applied in convolution neural networks(CNN). We extend it to vision transformer. The two methods are evaluated on standard benchmarks such as miniImageNet, CUB, CIFAR-FS, clipart and sketch. The results show that ‘scaling&&shift’ reaches the same level compared to updating the whole parameters. Vision prompt tuning is 0%~5% lower than updating the whole parameters over five datasets while it has quite smaller amount of parameters updated.

Today automobile industry, many studies on autonomous driving are being conducted, and the distribution of vehicles with autonomous driving functions is increasing. For the era of fully autonomous vehicles, connectivity of vehicles such as intelligent transportation systems, V2V, and V2X is required based on communication technologies such as LTE, 5G, and WiFi. However, this increase in the connectivity of automobiles has become an opportunity to raise the need for vehicle cyber security, and recently, ISO 21434 international standard has been established and announced in relation to cyber security. In this paper, we propose a threat analysis and risk assessment guide for cybersecurity so that we can derive cybersecurity requirements and develop a controller that can be safe against cybersecurity attacks. It also shows that the proposed method conforms to the requirements of the ISO 21434 standard. Through the result of the case study, cybersecurity requirements were derived to a level applicable to controller development, and it was confirmed whether the risk assessment requirements suggested in the ISO 21434 cybersecurity engineering standard were satisfied.

Named entity recognition of military equipment is an important task in the construction of knowledge graph in the military domain. It is a key technical means to improve the intelligence degree of military intelligence information retrieval, intelligence analysis, command and decision. There are many problems in the task of named entity recognition in the field of military equipment, such as fuzzy entity boundary, complex grammar structure and many professional words, which directly lead to the loss of accuracy in named entity recognition. Aiming at the above challenges, this paper proposes a BERT- BILSTM -CRF neural network model based on type labeling and part-of-speech labeling of entities in the field of military equipment. BERT's pre-trained language model fully considers the correlation between words when constructing word vectors, which is used to supplement the semantic relations embedded in words, and can solve the fuzzy problem in name entity recognition. BILSTM layer is used to carry out bidirectional semantic coding, which can solve the long-distance dependence problem. Finally, the output of BERT-BILSTM layer is decoded by CRF layer, and the optimal tag sequence is obtained. The experimental results show that compared with CRF model and BILSTM-CRF model, the F1 value of the proposed model is increased by 10%.Compared with the BILSTM-Attention-CRF model, the F value of this model increased by 10.48%, and the recall rate increased by 15.02%.Compared with the BERT-IDCNN-CRF model, the F value of this model is increased by 0.62%, and the recall rate is increased by 4.55%.

Current training systems for substations still lack the tracking and analysis of trainee’s personalized characteristics during studying and training. In this work, we propose a user model construction method for substation trainees based on behavior analysis. The user model establishes an electronic portrait for trainees from the two dimensions of performance and ability. Behavior data of trainees relevant to studying and training are recorded in real time, restored in uniform xAPI-based format, and further analyzed to quantify individual characteristics. A test of the user model is carried out using our substation simulation learning system. The results indicate that, the user model intuitively displays the method, process, attitude, skill and knowledge scores of trainees in one week. Moreover, individual characteristics like motivation, weakness, etc. can be reasonably speculated from the user model of each trainee, which act as important references for subsequent targeted teaching advice and personalized content recommendations.

Urban street lighting system plays an important role in urban traffic lighting and road safety. With it, the power consumption of urban street lights is increasing, and the cost is also increasing year by year. The purpose of this paper is to detect and classify targets through road monitoring radar, so as to achieve the purpose of real-time detection of pedestrians, vehicles and other targets, so as to control the brightness of street lights to save power consumption. This paper firstly establishes the radar echo model of various targets, deduces the model formula of radar echo, and analyzes the micro-Doppler characteristics produced by the micro-moving parts of the target. Secondly, Secondly, the ResNet residual neural network training method is used to analyze the target micro-Doppler information, extract the depth information, and realize the target detection and recognition task.

In recent years, radar high-resolution range profile target recognition(HRRP) based on deep learning has achieved good result in practical recognition tasks. However, due to the limitations of various conditions in the realistic environment, it is difficult to obtain enough samples, which greatly limits recognition performance. From the perspective of transfer learning, this paper proposes a target recognition method of few HRRPs based on distance-based distribution calibration and dynamic pseudo-label fine-tuning(FCPL). This method calibrates the distribution of the features of target domain samples according to the distance in the feature space. Moreover, it creates pseudo-labels with source data in the label space of the target domain to fine-tune the model. Experimental results show that the method this paper proposed achieves higher recognition accuracy than traditional transfer learning and meta-learning on four types of few-shot targets. In addition, the method can be built on most model structures or methods and achieve the smooth transition from few-shot target recognition task to many-shot task.

To improve the stability and maneuverability of the vehicle when steering, a joint deflection-slip control strategy is proposed for a 4WD electric wheel vehicle（EWV）model. A vehicle two-track 2-degree-of-freedom dynamics model is established, and a sliding-mode controller based on the isokinetic convergence law is designed with the vehicle's transverse moment as the control target, the transverse target moment is obtained by adjusting the left and right wheel hub motor drive and braking torque, and a combined Carsim/Simulink simulation model is built to complete the simulation verification. The results show that with the application of deflection slip steering, the vehicle steering radius and the mass lateral deflection angle and transverse sway angle generated during steering are reduced, which enhances the steering stability and maneuverability of the vehicle.

Aiming at the problems that current DDoS detection methods are difficult to capture long-distance time-series correlation between network traffic, low detection accuracy and insufficient ability to detect unknown types of DDoS attacks, a Transformer-based DDoS detection method XGB-Trans is proposed. First, the feature selection module based on the XGBoost is used to reduce the feature dimensionality of the data; then, the Transformer after the optimized architecture is used to extract the time-series related features between the low-dimensional data, and detect DDoS attacks accordingly. The experimental results on two network security datasets show that the detection accuracy and the generalization ability of the XGB-Trans are superior to other methods, and it has the ability to detect unknown DDoS attacks.

In recent years, with the rapid development of satellite technology, object detection technology in remote sensing images has become an important research direction in the field of image processing. In this paper, aiming at the problems of low resolution, slow detection speed and vulnerability to background interference in remote sensing satellite images of ground vehicles, a vehicle small object detection algorithm based on residual network is proposed. Based on the original SSD (Single Shot MultiBox Detector) algorithm, the backbone network adopts the residual network with stronger feature extraction ability and less parameters, which can effectively improve the detection accuracy of small targets; Meanwhile, the Dense Extra Feature Layers network is proposed to enhance Feature propagation. Experimental results show that the mAP value of the proposed method in vehicle remote sensing image detection reaches 0.573, and the detection speed is 30.1fps, which has a good improvement in the detection accuracy and speed.

In order to realize the effective dynamic obstacle avoidance of the robot when the indoor topographic map is unknown, this paper uses hardware such as lidar, panoramic camera, and software codes such as deep learning to build a mobile robot that meets dynamic obstacle avoidance. First, the robot uses lidar to obtain information about the surrounding environment of the robot, and uses the firefly algorithm to improve the traditional G-Mapping algorithm to draw a more accurate grid map to obtain the global optimal path. At the same time, the hyperbolic panoramic camera 360⁰is used to extract the information of the surrounding environment of the moving robot and transfer it to the PC, and then usetheYoloV5 deep learning framework combined with ShufflenetV2 to integrate the DWA algorithm to avoid obstacles in real time for the robot. Experiments show that the accuracy and speed of real-time obstacle avoidance of the robot are improved to a certain extent under this method.

Finite control set model predictive current control (FCS-MPCC) has the advantages of simplicity, fast response and excellent performance, but the motor is susceptible to parameter perturbation due to temperature and other factors. The parameter mismatch between the prediction model and the actual model will lead to the significant degradation of the performance of FCS-MPCC control method. In order to improve the robustness of FCS-MPCC to parameters, in this paper, a new model-free predictive current control based on extended sliding mode observer (MFPSC-ESMO) is proposed. According to the idea of model-free control, an improved current prediction controller is designed, which only uses the input and output of the system without considering specific parameters, and uses the extended sliding mode observer to observe the unknown disturbance in the hyperlocal model. The simulation results show that compared with the traditional FCS-MPCC, the proposed method has lower current ripple and torque ripple, and stronger robustness under parameter perturbation.

A new method for repairing point cloud model holes of aero-engine compressor blades is presented in this paper, which can be used to repair point cloud model holes of aero-engine compressor blades accurately. Firstly, by constructing the boundary diffusion mechanism, the generation judgment of hole repair points was defined, and the generation method of hole repair points based on boundary diffusion mechanism was designed, which unifies the diversified hole repair modes; Secondly, the implicit surface based on radial basis function was constructed, the position of repair points was corrected accurately, the repair area with uniform density was generated, and the hole repair was completed. Finally, it was verified by an example and compared with the central diffusion repair method. The results show that: By using the method in this paper, various holes in the point cloud model of aero-engine compressor blades, such as plane hole, curved face, island hole and cross-face, can be repaired evenly. It also avoids noise confusion caused by irregular hole boundary, and realizes better fitting between hole repair point and curved surface, thus providing accurate data support for aero-engine compressor blade damage detection.

With the rise of command city construction, the importance of pedestrian re-identification based on video surveillance has become increasingly prominent. The person re-identification task mainly studies recognizing images of the same person under different cameras. Due to the high cost of labeling data in supervised learning in common methods, and the lack of guidance for labeling data in unsupervised learning, the experimental results are unsatisfactory. In this paper, we propose a new weakly supervised person re-identification problem framework: a human-computer interaction-based sample active push and auxiliary labeling method is designed to quickly obtain a small number of identity-related labels with low labor cost; At the same time, the weak supervision information that is easily obtained in the real scene is introduced into the pseudo-label generation process to improve the clustering method, aiming to solve the problem of insufficient initial labeling data of images in the label estimation stage and weakly supervised labels are not well utilized; Finally, in view of the insufficient utilization of weakly supervised labels in the model training process, a weakly supervised loss function WSP loss is designed and added to the training to optimize the overall loss function. Compared with the current state-of-the-art unsupervised and weakly supervised person re-identification work, this paper shows better performance under the same conditions and even using less labeled information.

Aiming at the problem of untimely fault location in optical fiber maintenance and no early warning capability of optical fiber faults, a real-time optical fiber online detection device is designed and implemented. Based on the OTDR detection principle, ZYNQ is the main control chip, high-speed AD interface is integrated on the PL side of ZYNQ, and PetaLinux system is integrated on the PS side to realize real-time reception and processing of optical fiber signals. Real-time inspection of optical fibers. After a large number of tests and verifications, the equipment can detect the optical fiber status online in real time, report the fault information in a timely manner, and analyze the historical detection data to obtain the optical fiber attenuation trend and achieve fault early warning.

To date, the thin tungsten disulfide (WS₂) with its tunable band gap structure has renewed increasing attention in microelectronic design owing to its favorable electronic and optoelectronic properties. Chemical vapor deposition (CVD) as a bottom-up growth method of WS₂ is well known for its compatibility with the mature integrated circuit technology (CMOS), making it the most promising method to synthesize large-area and high-quality WS₂ crystals and uplift its integration into microelectronic applications-based field-effect transistor (FET). However, ultra-clean transfer of CVD-grown monolayer tungsten disulfide (WS₂) using polymer-assisted transfer with defect-free is still crucial for large-area WS₂devices' high performance. Additionally, the inhomogeneous distribution of the CVD-grown WS₂ drastically hampers the widespread usage of WS₂-based FET applications due to the alignment difficulty during the FET microfabrication. In this work, we present a new transfer insight of CVD grown WS₂ toward selective transfer and single chip multiple usages based on laser-cutting and conventional PMMA wet etching transfer. The transferred WS₂ surface structure and morphology were well characterized using optical microscopy, PL, Raman spectroscopy, and AFM, confirming the ultraclean and crackles of the transferred WS₂ crystals. While the WS₂ FET developed following this strategy led to high carrier mobility and on/off ratio of 71 cm²/Vs and 10⁵, respectively. The work described in this paper could broaden the CVD synthesis and transfer of WS₂ and their promising microelectronics application-based field-effect transistor.

The new Enhanced Plasma Detector (EPD) utilizes the plasma characteristics of dielectric barrier discharge and has a highly stable frequency control system composed of a focus stabilization electrode and an electron injection electrode. Compared with the existing plasma Technology, EPD detector has better signal-to-noise ratio and ionization efficiency, widely used in special gas, semiconductor, air separation, petrosal chemical industry and other industries. In this paper, through the application research on the detection technology of sulfur hexafluoride (SF₆) decomposition product EPD in GIS equipment, it is verified that EPD has the characteristics of low detection limit, good selectivity, wide linear range, and low operation and maintenance cost, which makes up for the existing detection. It is a new type of detector applied to power GIS equipment.

Minimally invasive surgery (MIS) has a wide range of applications in the medical field. Its emergence has brought many benefits to patients but has also brought great challenges to surgeons, requiring doctors with extensive surgical experience. The visual simultaneous localization and mapping (VSLAM) method can track the position of the endoscope during the operation and complete the 3D reconstruction of the scene, which can provide significant help to the surgeon through surgical navigation. Feature matching is an essential link in VSLAM, and the correct correspondence of feature points between two images determines the overall accuracy of VSLAM surgical navigation tasks. Aiming at the problems of nonrigid transformation and noise in medical image feature matching jobs, we propose a robust feature matching algorithm that can handle the effects of soft tissue deformation. This paper is validated using publicly available medical endoscopy datasets and compared with other state-of-the-art methods. The experimental results show that our method outperforms different approaches and has better performance in surgical scenarios.

A cooperative search path planning algorithm based on model predictive control (MPC) is proposed to solve the problem of multi-MAV cooperative search for dynamic target in unknown limited area. Firstly, the environment model, MAV state model, sensor model and search probability graph model are established. Secondly, based on the model, multi-target efficiency functions including target search profit and MAV flight cost are established. Considering that micro aerial vehicles (MAVs) detect each other independently, the integrated target probability fusion method is proposed to realize the information interaction between multiple aircraft and ensure the maximization of search benefits. Finally, the effectiveness and practicability of the algorithm are verified by simulation experiments.

In recent years, with the continuous input of high-power mining and transportation equipment and the continuous increase of mining depth, the power grid structure becomes more and more complex, the relay protector parameters are difficult to set, and power supply safety accidents such as switch misoperation and skip tripping are easy to occur. Through hardware design and software design, the power monitoring substation based on uCOS is researched and developed, and the fault monitoring and early warning system of coal mine power grid is constructed to achieve the purpose of intelligent identification and monitoring and early warning of common faults of coal mine power grid.

Because of the unclear boundaries and different shapes and sizes of breast masses, the accuracy of using traditional computer-aided diagnosis systems is low and it is difficult to meet the clinical requirements of physicians. In this paper, we propose a breast mass detection algorithm based on the combination of YOLOv5 and improved coordinate attention, to meet the clinical requirements of high accuracy and real-time. First, a novel backbone feature extraction network is constructed by combining the underlying backbone network and attention mechanism to fully learn useful features and suppress irrelevant features, thus enhancing the feature expression capability. Then a multi-path aggregation network is designed as the neck of feature fusion to fully fuse the feature information at different levels. Validation experiments are conducted on the DDSM breast mass dataset, and the results show that the network can accurately detect masses of different scales in different backgrounds with better real-time performance. Compared with the base YOLOv5, the network improves by 2.3% in accuracy.

Aiming at the problems of complex operation and low accuracy in psychological stress detection, we proposed a new method based on heart rate variability. We set the length gradient exploration for the length of the window and the window repeat part. Get the best window settings, and we extracted the time domain, frequency domain and nonlinear features of heart rate variability for model construction. We used the TOPT Automation Machine learning algorithm to obtain an optimal algorithm - extremely randomized tree algorithm, and the AUC on the test set was 0.934, which can be highly credible to detect psychological stress.

A certain ultra supercritical unit has a rated load of 670mw and has been undertaking industrial steam supply for a long time, with a maximum steam supply of 500t / h. During the load change process of the unit, the main steam pressure and temperature fluctuate greatly, and the load regulation does not meet the requirements, so the unit cannot operate safely and stably. It is optimized from ratio of water quantity to coal quantity(WFR) strategy. The logic related to the correction of the intermediate point temperature on the fuel quantity is deleted. The position of the intermediate point is changed from the inlet of the primary superheater to the outlet of the steam water separator. The adjustment of the WFR through the intermediate point temperature is changed to the adjustment method based on the calculation of enthalpy value. The heating load is converted into electrical load and added to the boiler main control command. After optimization, the speed and accuracy of load regulation are greatly improved, and the main steam pressure and temperature fluctuate within a reasonable range. This optimization strategy can be used for reference for the same type of units.

Foreign objects such as bird’s nests, honeycombs, kites, and plastics threaten the safety of electric transmission lines. Training a foreign object detection model with high generalization and accuracy needs to collect numerous data, which is usually unavailable because of data privacy and security. Federated learning is a promising solution but the total calculation amount is huge, and large number of gradient data will be transmitted in federated learning model updating. And because the data amount, training efficiency, and model accuracy are different at edge clients, it is not appropriate to aggregate clients’ edge models equally. To address the above mentioned problems, a prompt fine-tuning based efficient federated foreign object detection method of electric transmission line, which includes prompt federated fine-tuning mechanism and self-adaptive dynamic weighting aggregation strategy, is proposed in this paper. Prompt federated finetuning freezes the model backbone in edge client model training, and only the unfrozen parts of the model are transmitted and updated in the model aggregation phase. Self-adaptive dynamic weighting aggregation mechanism takes local data amount, edge model’s accuracy, and aggregation rounds into account when aggregating local models. Comprehensive experimental results demonstrate that the proposed method has not only better model accuracy but also faster training speed compared with other federated learning baselines on transmission line foreign object detection.

Digital twin is a key technology for realizing the mapping of digital model from physical system to information space. This paper studies the application of digital twin in component quality management, and proposes a digital twin system dedicated to component quality management. Model the appearance of components, state information management, fault prediction and detection, and space-time synchronization detection. Among them, the general system model library stores the appearance model, mechanism model and mathematical model of the components, and uses the relational database to access the data. By determining the nominal area and inputting the benchmark clustering, on this basis, the Fisher's quasi-measurement is used to calculate the abnormal probability, and the abnormal probability is used to realize the prediction of the failure of the components and equipment.

The time-domain characteristic of an acoustic signal produced by underwater discharges is clarified under the conditions of no reverberation and reverberation and the time-frequency analysis is conducted through HHT. Furthermore, the application of the plasma acoustic source to the underwater interference is discussed. Results show that a single signal generated from plasma acoustic sources has the feature of high amplitude, short pulse width, and broad frequency band. Furthermore, the amplitude and the pulse duration of the acoustic signal could be enhanced in the case of reverberation, which is beneficial to acoustic interference. In application, enough pulse width is easily obtained by the acoustic pulse trains generated from continuous underwater discharges. Therefore, the plasma acoustic source could be used as an effective underwater interference source.

Hepatic tumor thermal ablation surgery is currently one of the most effective methods for the clinical treatment of liver cancer. There are many studies on auxiliary systems for thermal ablation of liver tumors, but few studies on the design of constraints for thermal ablation path planning for clinical needs. Therefore, this study has conducted detailed research and improvements on the key technology of precise liver tumor thermal ablation path planning for clinical needs, and experimentally verified related algorithms based on the investigation of various of domestic and foreign literature.
In this paper, the OBBTree algorithm is applied to the liver tumor ablation path planning for clinical needs, which is often used in collision detection research, and three weak constraints for liver tumor ablation path optimization based on clinical needs are proposed, two implementation schemes for obtaining the start and end points of the puncture path, a multi-constraint optimization mathematical model for thermal ablation of liver tumors. At last, the evaluation score is calculated to obtain the optimal ablation planning path that meets the clinical needs.
After evaluation, this method is considered by the professional physician of hepatobiliary and pancreatic surgery (from the First Hospital of Jilin University) to be more in line with clinical needs. The operating efficiency of the algorithm exceeds the efficiency of manual planning by clinicians and its accuracy is no less than that planned by clinicians. It can minimize the penetrating damage to the body caused by the ablation process to the patient so that the patient has a better prognosis after surgery.

This paper focuses on the application of electrical machines in public transportation and investigates the potential benefits of auxiliary slot design on this kind of high-speed low torque machine. A preliminary surrogate-based optimization is conducted to the shape of auxiliary slots on the stator for minimizing its cogging torque ripple while maintaining its Total Harmonic Distortion (THD) at an acceptable level. The results show the optimization achieves its best performance while the edges of the auxiliary slot are parallel to the edges of the stator teeth, which could reflect a deeper resonance among the geometries of the stator magnetic field.

In this paper, a novel recognition method based on random matrix is proposed for different turbulence intensity. To reflect the degree of atmospheric turbulence, the continuous product form of phase screens is taken into consideration. After calculated the statistical distribution using random matrix theory, the fitting effect of statistical distribution determines the differences of turbulence. Also, based on the data-driven idea, the eigenvalue and singular values of phase screens are described as a whole. According to the adaptability of large dimensional random matrices, the Ring Law and M-P Law breaks through the assumptive restriction of infinite sample, thus building a significant model for potential changes of weak turbulence. The simulation experiments validate that the big data technology is effective attempt for atmospheric turbulence recognition.

In the segmentation of skin lesions, there are several difficult phenomena, such as blurred edges, hair occlusion, circular field of vision and diagnostic markers. In response to the above problems, we propose the Gated Axial Transformer with Comprehensive Attention (CA-GAT) to segment skin lesions. First, the U-Net encoder-decoder structure is used as the main framework, and the encoder primarily consists of axial transformer layers. The axial attention mechanism is able to efficiently grasp long-distance dependence while greatly reducing the computational complexity. The gating mechanism allows the model to learn accurate positional encodings without pre-training on large-scale datasets. Second, the triple attention mechanism is introduced into the decoder, thus enabling the model to better differentiate the lesion boundary. Finally, the Local-Global training strategy (LoGo) enables the model to better exclude external interference based on contextual information while improving model performance. We conducted experiments on ISIC2018 dataset. Compared with U-Net, CA-Net, MedT and CA-GAT without LoGo, the Dice coefficient of our model increases by 6.3%, 0.46%, 1.2% and 1.49% respectively, and other indicators are also improved. As indicated by the experiment, the model CA-GAT exhibits favorable segmentation performance.

This paper proposes a synthesis method for nonuniform linear antenna arrays that combines Gaussian process regression (GPR) and genetic algorithm (GA).In this mthod, the GPR model can be used to calculate the array radiation pattern in presence of mutual coupling effects, and then the GA is used to optimize the exciations and locations of the elements, so as to generate the desired radiation pattern. In this paper, taking a 9-element nonuniform linear array as an example and the desired radiation pattern corresponding to a Chebyshev distribution as the optimization objective, Optimize the exciations and locations of the elements. Finally, the optimization results are verified by electromagnetic simulation software CST, which shows that the method is effective.

In the process of straw utilization, there are some unreasonable factors and remote sensing image can help to solve the problem of straw storage site selection. A model mainly considering the location and yield of crops, together with relevant assumptions and constraints, is established. As an efficient tool to solve complex non-linear problems, the improved Immune Algorithm is proposed and utilized. Taking a remote sensing image in Xian County in China as an example, one, two and three straw storage site(s) are located respectively using the proposed method, which is proven to be effective.

This paper proposes a closed-loop fiber optic gyroscope control method based on the Sigma-delta modulation technique to improve the fiber optic gyroscope proportional-integral control method. Firstly, the sampling control system of digital closed-loop fiber optic gyro is established. Secondly, the fiber optic gyro digital closed-loop Sigma-delta modulator is designed and combined with the original system after the integrator. Simulation analysis is performed for both closed-loop control methods. Simulation results show that the maximum output noise amplitude of the original system is 2.8 deg/s and the maximum noise amplitude of the improved system is 0.8 deg/s under the input of a ramp signal with 100dB white noise. The noise amplitude is reduced by one order of magnitude, and the noise density is also significantly reduced. The proposed closed-loop control method of fiber optic gyro based on Sigma-delta modulation can effectively suppress the high-frequency noise of the weak signal of the gyro output, which has a promising application for the dynamic performance and noise suppression of fiber optic gyro.

In recent years, the rapid development of information technology has continuously brought risks and challenges to network security. How to effectively detect and classify malicious codes and avoid serious damage to information systems has become a current research focus. However, the existing malicious code analysis methods have insufficient semantic information and poor feature readability, making it difficult to discover the potential relationship between malicious codes. In order to solve these problems, this paper applies knowledge graph technology to the field of malicious code analysis, constructs malicious code behavior knowledge graph based on dynamic analysis design, and uses the representation learning model TransH to conduct knowledge graph embedding representation research, so as to use KNN algorithm to classify malicious sample families. The experimental results show that the method proposed in this paper has more advantages in the classification of malicious code families than the comparison model.

With the COVID-19 outbreak in 2019, the world is facing a major crisis and people's health is at serious risk. Accurate segmentation of lesions in CT images can help doctors understand disease infections, prescribe the right medicine and control patients' conditions. Fast and accurate diagnosis not only can make the limited medical resources get reasonable allocation, but also can control the spread of disease, and computer-aided diagnosis can achieve this purpose, so this paper proposes a deep learning segmentation network LLDSNet based on a small amount of data, which is divided into two modules: contextual feature-aware module (CFAM) and shape edge detection module (SEDM). Due to the different morphology of lesions in different CT, lesions with dispersion, small lesion area and background area imbalance, lesion area and normal area boundary blurred, etc. The problem of lesion segmentation in COVID-19 poses a major challenge. The CFAM can effectively extract the overall and local features, and the SEDM can accurately find the edges of the lesion area to segment the lesions in this area. The hybrid loss function is used to avoid the class imbalance problem and improve the overall network performance. It is demonstrated that LLDSNet dice achieves 0.696 for a small number of data sets, and the best performance compared to five currently popular segmentation networks.

Integrity verification can effectively protect cloud data. Currently, users entrust third-party auditors to verify the integrity of their outsourced data. Unfortunately, multiple auditors participate in the integrity verification, which increases the risk of data privacy disclosure. Therefore, more users expect a specific verifier to check the files in cloud storage. We proposed a scheme with a designated verifier, which can effectively protect data privacy when auditing. In the process oey generate tags. In the evidence generation, bilinear results of data blocks and tags are directly calculated to realize privacyf tags generation, we utilize the designated auditor public k protection. Based on the computational Diffie– Hellman assumption in the Bilinear Maps, we prove the security for our scheme.

The frequency diversity array (FDA) is capable of producing a range-angle dependent "S" beampattern, unlike conventional phased arrays that only provide directional beams in the angular domain, but it can only scan in a fixed direction, while frequency diversity arc array (FDAA) has a beam scan-ning capability within a 360 degrees range, which can realize all-round monitoring of the target position and provide a more flexible method for radar communication. In this letter, a dot-shaped beampattern optimization method for FDAA is proposed, which uses amplitude weighting and symmetric non-uniform frequency offset to achieve flexible beam scanning

In recent years, object tracking depends upon deep networks or complex architectures made great progress in accuracy, but less attention has been paid to tracking speed. On edge platforms with limited resources are used on UAVs, trackers are difficult to meet the requirements of real-time processing. To address the above problems, we proposed UAV object tracking based on vision transformer. Firstly, a lightweight neural network ResNet18 is used for feature extraction. Secondly, a transformer encoder is used to fully enhance fused features of the template features and the search features to improve feature expression, then, the enhanced features output from encoder are used for target prediction. Finally, further weighted fusion at decision level is used to obtain final tracking result, and a multi-supervised strategy is used. The proposed method achieves better results on three challenging large-scale tracking datasets, UAV123, LaSOT, and TrackingNet, reaching the success score(AUC) of 66.82, 64.26, and 81.2, respectively. In addition, our model tested speed on UAV123's test dataset, running at 116.2fps on NVIDIA 1660TI and 20.2fps on NVIDIA Jetson Xavier NX, which basically meets UAV's demand for real-time performance.

Digital delay filter can realize signal delay of different receiving channelswhich is often used in the field of communication and many scenarios where delay affects signal analysis. At present, the existing delay algorithm is difficult to be directly applied to the scene because it can only delay integer sampling points, or because the algorithm is complex and the convergence speed is slow. For this, this paper improves the existing adaptive algorithm on the basis of Farrow structure and adaptive algorithm. The improved algorithm reduces the calculation amount of each iteration through the approximation method. The change of step length in the iteration process is analyzed mathematically, and the step length change function is constructed. The experimental results show that the improved algorithm reduces the number of iterative convergence by at least 4 times compared with the original algorithm, and has excellent performance in time delay estimation of unknown signals.

A method of building hull model with OpenGL and C++ is proposed, which can better realize 3D modeling. Compared with the traditional 2D method, 3D design can express products more intuitively. Compared with some major commercial modeling software on the market, although commercial software can establish a complete surface space, the entire system modeling cannot be separated from the operation of commercial software, and software application cannot be promoted without independent intellectual property rights. The model constructed by this method analyzes the complex curves and surfaces contained in the shape of the hull according to the characteristics of the hull, and proposes a geometric modeling algorithm based on key points, which uses quadrilateral patches to splicing key points. Implement model meshing. Finally, a parametric model is formed by OpenGL rendering. The hull parametric model system developed based on OpenGL has good stability and strong compatibility, and can better assist in the evaluation of ship damage effects and target vulnerability, which are two extremely important tasks.

With the rapid development of the Internet, data analysis and other information fields, the amount of meteorological data is increasing exponentially every year, and the types of meteorological data are increasing. The storage of meteorological data should not only reach the response speed of millisecond level, but also provide data products and services in a short time. By analyzing the characteristics of meteorological data structure, this paper puts forward different storage methods for different data types. According to the data requirements of the meteorological business operations, this paper proposes the use of hierarchical storage, to store frequently used or soon-to-be-used data stored in high-performance memory. At the same time, a record table based on file access characteristics is established to record data attributes, and quickly locate the data that needs to be migrated. By simulating the data response efficiency of querying historical similar weather, and comparing the performance of the original single storage mode with the optimized storage mode, it is verified that hierarchical storage can improve the demand for data response of meteorological services.

Similarity calculation of short text is an important part of text classification. In view of the problem that statistical learning methods can only extract shallow features of short texts, but cannot express their deep semantics through the extracted features, a new method based on two-way mirror hole connection is proposed. A short text semantic similarity calculation model combining long short-term memory network (BiP-LSTM) and convolutional neural network (CNN). The word vector is pre-trained with the word vector attention mechanism, and the distance factor is introduced to solve the influence of the distance between words on the word sense association. When combining sentence-level features, an attention mechanism is added to make it contain more representative features. The experimental results show that, based on the Rotten_imdb and Quora Question Pairs datasets, the proposed short text similarity calculation method has an accuracy of 84.37% and an F1 value of 83.67%, which is better than other benchmark methods.

The paper shows how to convert the pattern DB into the q-gram index structure with the 'next byte range' and how to filter the data item (e.g. packet) using this structure. We indexed the patterns of the given DB as 6byte-part to raise the speed of matching and the utilization of memory. In this paper, we aimed to largely decrease the full-pattern matching system's load(the number of the input data item) and improve the performance of pattern matching by using 'SigArrg', 'next byte range' and 'Bloom Filter'. We made an algorithm so that you can apply it easily to the fields connected with pattern matching, and the experiment result shows that our proposed method speeds up pattern matching. This paper has several scalabilities.

It is a technical difficulty to eliminate the influence of severe visual occlusion in the 3D visualization of urban buildings on the digital earth. Aiming at increasingly complex virtual scenes, we proposed a graph visualization algorithm named Urban Building Dis-occlusion (UBD). The UBD algorithm consists of three parts: 1. Inspired by the skyline, constructing urban building feature points through bounding box collision detection to draw building lines with bidirectional sampling; 2. Multi-type control point layout to ensure smooth building lines; 3. Visual pyramid lens (VPL）elimination neighboring building occlusion enhances visual attention. Compared with the traditional skyline, the experimental results show that UBD can effectively reduce the visual interference of buildings to the building line, and has good timeliness, which fully meets our needs in practical projects. The algorithm provides a novel building-line style for graph visualization, and a new VPL interaction for 3D obstacle avoidance using viewpoint.

In this paper, deep learning method is applied to study the estimation of human joint points in complex multi-person videos. Although there are many excellent research results on human pose estimation from single frame images, but as a combination of multiple frames of continuous images, video contains more complex temporal information. So human pose estimation for videos is more challenging. In order to make use of temporal and spatial continuity in video sequence, temporal attention and spatial attention are used in this paper. To better consider the amount of information in each frame and aggregate information across multiple features. Temporal attention fusion is adopted by calculating the element correlation between the features of the current frame and adjacent frames; in order to take advantage of the spatial relationship within the frame, spatial attention is used to achieve efficient fusion. Through experimental evaluation, the method in this paper can recover the accurate human pose skeleton with high accuracy.

In cloud-edge collaborative working mode, application services are affected by complex components, configuration and deployment conditions and other multi-dimensional factors. Therefore, it’s important to use simulation tools to conduct experiments reasonably and effectively. EdgeCloudSim is a simulation test platform widely used in edge computing. However, EdgeCloudSim's built-in task scheduling algorithm FF(First Fit) has problem of queuing and blocking, which makes some tasks fail to obtain computing resources in time. Besides, the processing time and load are not ideal. Another algorithm LL(Least load) optimizes above problems, but VM list needs to be traversed every time while scheduling. So the decision-making efficiency is low. Furthermore, EdgeCloudSim ignores migration cost of tasks. This paper improves the task scheduling framework of EdgeCloudSim for above problems and proposes a two-stage cloud-edge task scheduling strategy based on MH-LL(Max Heap-Least Load) algorithm and PP-LL(Position Probability-Least Load) algorithm. When resources of scheduled edge cloud are insufficient, task will be migrated to other edge clouds according to the migration cost calculated by distance factor α and distance coefficient d. Comparing with the original platform by experiments, the improved platform has an average optimization of 1.38%, 2.978s, 0.129s, and 19.26% in task failure rate, service time, delay, and load. It can improve work efficiency and system performance.

In order to improve the running time of image stitching and remove the phenomenon of ghosting, this paper proposes an image stitching method based on the improved optimal seam line.First, the two input images are preprocessed using a median filter algorithm.Next, image registration is performed using the ORB algorithm. Finally, the best seam method based on the Canny operator is used.The experimental results show that when stitching images, whether for large scale or small scale images, the algorithm in this paper has faster running speed and better stitching effect.

In order to prevent the grid microservice architecture from illegal intrusion, it is necessary to verify the API gateway authority in the architecture. However, the current API gateway authority verification method has the problem of long execution time. In order to solve this problem, a new authorization verification method of API gateway in grid microservice architecture is designed. Take the framework management of microservices as the smallest independent unit of gateway verification, extract the API gateway business requirements, set the authentication enhancement module in combination with the grid microservice architecture, adjust the operation mode of public services such as extraction authentication, optimize the identity authorization and authority verification mode by using encryption algorithms, and realize the API gateway authority verification of the grid microservice architecture. The test results show that the average execution time of the power grid micro service architecture API gateway authority verification method in this paper is 2.843s and 3.632s less than that of the other two methods respectively, which indicates that the power grid micro service architecture API gateway authority verification method designed this time has more advantages in execution efficiency.

Multipath matching pursuit (MMP) algorithm is a generalization of the orthogonal matching pursuit (OMP) in the sense that multiple child paths are generated for every candidate per iteration, which increases the chance of selecting the true support and leads to high possibility of exact reconstruction. In this paper, we show that with the proposed constraints on the restricted isometry constant (RIC) of a measurement matrix and the magnitude of the minimum nonzero element of the sparse signal, MMP with given stopping rules could find the true support after the iteration stops. Our bounds on the RIC improve the existing results.

According to World Health Organization, heart disease has remained the leading cause of death for the last twenty years.Patients with heart disease will have palpitations, chest pain, dyspnea and other characteristics. How to diagnose heart disease based on these characteristics is an important research topic for doctors and scientists. Classifying heart disease with machine learning is known as a reliable and accurate heart disease detection technique. In this study, we aim to find a good model for predicting heart disease. To find the best model, we compare three GBDT(Gradient Boosting Decision Tree) models i.e., XGBoost, LightGBM and CatBoost to identifywhether a person has heart disease or not. We compare the three models tuning the hyperparameters to improve heart disease predicting. To compare results, we create three base models with default hyperparameters. The findings demonstrate that the CatBoost with hyperparamer tuning model as the the most appropriate heart disease prediction model with accuarcy of 90%. Results indicated that tuning the hyperparameter improved modeling results.

Nowadays, with the outbreak of COVID-19, the prevention and treatment of COVID-19 has gradually become the focus of social disease prevention, and most patients are also more concerned about the symptoms. COVID-19 has symptoms similar to the common cold, and it cannot be diagnosed based on the symptoms shown by the patient, so it is necessary to observe medical images of the lungs to finally determine whether they are COVID-19 positive. As the number of patients with symptoms similar to pneumonia increases, more and more medical images of the lungs need to be generated. At the same time, the number of physicians at this stage is far from meeting the needs of patients, resulting in patients unable to detect and understand their own conditions in time. In this regard, we have performed image augmentation, data cleaning, and designed a deep learning classification network based on the data set of COVID-19 lung medical images. accurate classification judgment. The network can achieve 95.76% classification accuracy for this task through a new fine-tuning method and hyperparameter tuning we designed, which has higher accuracy and less training time than the classic convolutional neural network model.

In general, modern construction projects are characterized by large construction scale, high construction difficulty, and complex project management and data information analysis. Hence, various technologies should be fused and integrated to comprehensively and effectively manage the project construction. In this paper, a multi-source heterogeneous data fusion technology, which combined building information modelling (BIM), geographic information system (GIS), and Internet of Things (IoT) (BIM + GIS + IoT Multi-Source Heterogenous Data Fusion), was systematically explored to achieve a total factor, data sensing, and heterogeneous fusion concept. Then, the BIM + GIS scene-based 3D modelling technology was discussed from three aspects: total factor information 3D modelling, model tiled loading, and ultra-large model lightweight processing. With a tunnel construction project combined, a tunnels Internet-based construction management platform was developed on the basis of BIM + GIS + IoT multi-source heterogeneous production data fusion, thereby realizing the organic integration of virtual model information and total factor production information. Furthermore, the platform provided the capability of monitoring core indexes such as gas, geology, personnel, and progress involved in the tunnel construction, which facilitated the dynamic visual management of the construction process and enhanced management efficiency.

This article introduces the construction of an auxiliary translation system for Chinese and Kazakh bilingual terminology in the field of information technology and the design of each functional module. And the overall frame structure of its design, the working principle of each functional module, the collection, preprocessing, classification and translation process of bilingual term corpus are explained.

Multisequences over finite fields play an important role in applications that related to parallelization, such as word-based stream ciphers and pseudorandom vector generators. It is interesting to study complexity measures for multisequences. In this paper, we propose three constructions of multisequences over finite fields from rational function fields and Hermitian function fields. We also analyze the joint nonlinear complexity of these multisequences. Moreover, the length and dimension of these multisequences are flexible.

According to the shortcomings of the weighted mean of vectors, such as low convergence accuracy and easy to fall into local optimization at the end of iteration, a hybrid weighted mean of vectors with multiply strategies is proposed. Firstly, an adaptive probability search is proposed to solve the problem that the development and exploration of two state transitions are not flexible enough; Then, the soft besiege strategy of Harris hawks optimization is introduced to enhance the development ability of the algorithm; Finally, Cauchy mutation is used to disturb the population. The results of material balance engineering application show that the improved algorithm has higher convergence accuracy.

Fire early warning is one of the technical methods of ecological security, and computer technology is widely used in fire simulation.Fire simulation software has been more and more widely used in buildings.This article uses PyroSim software to simulate fire simulation of Bai folk houses. Firstly, a model of Bai folk house was constructed by on-the-spot investigation and used to simulate the different stages of a fire. In the model, it sets up grid slices and fire source points to simulate the fire, which is divided into three stages: before, during and after the fire, and the data of each simulation stage is collected. Secondly, it deeply analyzes the data characteristics of temperature, humidity, smoke concentration and so on which are obtained the corresponding changes. Finally, it determines the locations of temperature, humidity, smoke perception and other sensors are analyzed and determined to be 2-3 meters in Bai folk houses. This paper provides a reference for the improvement of the Fire Warning System and it is beneficial to the intelligent management of Bai folk houses.

In recent years, demand for collaborative model training has increased significantly. The key to realize decentralized model training without data disclosure is Federated Learning(FL). However, current research has not yet implemented a concrete method taking into account both efficiency and fault tolerance. And the incentive has not been fully considered. In this paper, the article proposes a novel voting-based blockchain-enabled federated learning platform named as VB-FLChain. The platform can be made incentive-aware by reward distribution in blockchain. With FL combined voting methods, the platform is more secure besides the high efficiency. Finally, the performance of VB-FLChain is measured by applying the UNSW-NB15 dataset.

This paper gave a plan for an error correction coding technology scheme based on adaptive data transmission for a series of problems, which included a large reduction in transmission characteristics caused by network delay triggering the network delay retransmission system, in research on the transmission control protocol (TCP) in wireless networks. The scheme reduced the loss of data segments through the error correction code of adaptive network transmission, thus preventing the triggering of TCP network packet loss retransmission system, further enhancing the level of error correction coding technology of TCP for adaptive network transmission, and providing a certain reference for research on TCP adaptive network transmission.

Because of the discontinuity at the node of broadband multisection directional coupler, its performance is poor at high frequency. Based on this structure, the gradient line coupler is developed. In order to avoid discontinuity and achieve the minimum design error, this paper uses optimization algorithm to calculate and optimize the structure of the tapered line coupler. A 6-18GHz gradient line coupler is designed by mathematical formula and genetic algorithm, and uses the three layers bias stripline as the transmission line. After optimization and simulation, the results show that the return loss is above 16dB, the coupling is 3.09 ± 0.25dB, and the phase difference is 91 ± 2°. This design method has better results than other methods.

By applying the dynamic Bayesian network theory, a dynamic Bayesian network evaluation model is established according to the structure, damage level and functional characteristics of military airport buildings. Based on the construction of the airport model, a Bayesian network diagram for evaluating the damage effect of military airport buildings is established by using the powerful inference function of dynamic Bayesian network. On the basis of determining the damage evaluation network nodes of military airport, the network structure and network parameters are determined. The simulation experiment verifies the correctness and effectiveness of the calculation. This model can not only evaluate the damage of the whole airport building, but also evaluate the damage effect of the airport building structure and other facilities. It provides scientific assessment for the decision makers' flexible strike methods and strike tasks.

Aiming at the problem of slow running speed of pedestrian single-target tracking task using SiamBAN algorithm under limited hardware conditions, this paper proposes a lightweight pedestrian single-target tracking network GhostSiamBAN. Specifically, we use GhostModule to build a new lightweight backbone network to replace the original Resnet50. The evaluation on OTB2015 and VOT2018 shows that our method can well balance the tracking accuracy and speed. Compared with SiamBAN, the test results of pedestrian single target tracking sequence on Nvidia GTX 1050 show that the average running speed of our method is about 26fps with an accuracy loss of about 5.46%, and the speed is increased by nearly 271.4%

Based on the basic characteristics of walking gait of bionic robots, a general gait algorithm is proposed through mathematical derivation, which can be applied to the practice and simulation of bionic robots with similar mechanical structures.

In modern electronic warfare, in order to correctly identify the intercepted enemy signals, it is first necessary to identify the modulation system used by the enemy signals and correctly demodulate them in order to obtain the information transmitted by the enemy and seize the opportunity in informationized warfare. Aiming at the low recognition rate of digital signal modulation recognition under low signal-to-noise ratio, the recognition model suitable for signal constellation diagram is improved on the basis of the classical convolutional neural network GoogLeNet network, and a recognition model suitable for signal constellation diagram is constructed. Experimental results show that when the signal-to-noise ratio SNR=5dB, the average classification accuracy of the 12 hybrid modulation types can reach 95.28%, and the classification accuracy of the improved GoogLeNet is increased by 1.1%.

The Industrial Internet of Things with positioning service has been applied in many fields, such as: intelligent Internet of Things warehousing system, which realizes the identification, positioning, statistics, and management of cargo information, as well as on-site personnel safety monitoring. This kind of network usually needs to complete the requirements of positioning, information transmission, and emergency message processing, and the priority of these different information transmissions is different. How to optimize resource allocation has become one of the urgent problems to be solved in related fields. This paper realizes the resource allocation of a specific M/M/1/N model by controlling the M/M/1 service resource models with different priorities, so that the relationship between the allocation of different priority service resources and the buffer size can be easily and clearly obtained. The research model provides a theoretical basis for the resource design of the MAC layer or the transport layer in the relevant network.

A microlens fluorescence imaging system and a panoramic image reconstruction algorithm for biological cell fluorescence detection are presented in this paper. Biological cell samples are labeled with fluorescent dyes and made into sections in a short time. The imaging system in this paper can detect the fluorescence of the samples. By modeling the optical path of the microlens imaging system and the two-dimensional mobile scanning imaging mode, the better fluorescence detection effect is obtained with the panoramic image reconstruction algorithm. The experimental results show that the microlens fluorescence imaging system can successfully collect the fluorescence information, and can reach the resolution of 2.68μm in the image sensor frame area. At the same time, the field of view area of the imaging system can reach at least 9 times or more of the sensor frame area after the reconstruction algorithm, which realizes the rapid and convenient cell fluorescence detection with

As the core of artificial intelligence, image recognition technology has been widely used in various fields and has brought great convenience to modern life. This paper expounds the image recognition technology and its principle, analyzes the process of image recognition based on artificial intelligence algorithm, and studies the application path of image recognition technology based on artificial intelligence algorithm. Finally, the advantages of artificial intelligence image recognition and its future development are pointed out.

Rational drug use prediction, which is used to estimate whether the drug use is reasonable in clinical medical treatment, is of great significance in combating excessive medical treatment. The feature interaction model based on electronic medical records is widely used in the field of medical behavior prediction. However, the features recorded in electronic medical records are very complex, and medical behavior has an obvious long-tail effect. The problem of feature sparseness makes traditional prediction models based on feature interaction impossible. It works well, we propose a novel business domain-based second-order relational graph embedding neural network model (SORGE-NN), which can distinguish the scope of features and mine the implicit propagation relationship, through the residual-multi-head attention layer , and perform high-order weighted combination of features, which effectively alleviates the challenges brought by feature sparse and complex features. We conduct experiments with real datasets, and the experimental results show that our proposed SORGE-NN achieves better results than current state-of-the-art prediction models.

In the context of the increasing volume of privacy data, the requirements for storage security are also increasing. Some privacy data cloud storage security-aware methods suffer from excessive storage overhead, which limits the subsequent utilization of space resources. To this end, a deep neural network-based security-aware method for privacy data cloud storage is designed. Identify proof of attribution for cloud storage services, complete read and write data operations on the blockchain with the help of smart contracts, set up privacy data encryption, decryption and authentication mechanisms, hide cryptographic statistics, build a data chunking encoding model based on deep neural networks, combine storage addresses provided by cloud servers, and optimise security-aware models using attack prediction algorithms. Experimental results: The average value of storage overhead of this designed privacy data cloud storage security-aware method is: 6050 KB, which is 460 KB and 487 KB less than the other two respectively, indicating that this designed privacy data cloud storage security-aware method is better when combined with deep neural network.

Facial expression recognition has always been a focus of study in emotion computing. With the aim of solving the problem that previous facial expression recognition studies are susceptible to individual differences and background noise, as well as the loss of facial expression features in deep networks of deep convolutional networks, we propose a facial expression method based on convolutional difference contextual feature fusion. The convolutional difference module can carry out difference operation between emotional image and neutral image in the input image, eliminate individual difference and background noise in expression features and improve the accuracy and generalization of the network. Finally, convolutional difference and contextual features are combined to improve the accuracy of the network further. Experiment results of 4 public facial expression datasets illustrate that compared with the existing facial expression recognition methods, this method has higher accuracy, can extract facial expression features more effectively, and has a better generalization ability.

Vehicle to grid (V2G) network plays an important role, which can realize the electric energy transaction between electric vehicle (EV) and smart grid (SG). However, sensitive information on the identity of EVs and the transaction bills may be leaked. In this paper, a privacy-preserving energy trading scheme based on blockchain is proposed, which can protect private information such as transaction amount and identities of EVs. Specifically, elliptic curve encryption and digital signature are used to realize the confidentiality and integrity of transaction information, respectively. At the same time, we also use blockchain and smart contract technology to ensure the immutability of transaction amount between EVs and SG. Finally, we prove the security of the proposed scheme and evaluate its efficiency.

Over the past few years, many universities are starting to set up the college Japanese course for undergraduate students. But many new situations and problems arise. So far, there has been relatively little research on these issues, not to mention research using an empirical approach. To address these issues and gain practical insights, this paper first collects the learning dataset of all first-year students taking a college Japanese course at a university. The program has a stable learning population, and a good research foundation that lasts for one year. Second, a clustering algorithm called KMeans was used to evaluate the student learning analysis. Specifically, we use the elbow technique to identify the Kvalues that work best. The empirical analysis showed that clusters performed best when the number of clusters was 2 which means students should be divided into two groups：the students evaluated as Good (49.43%), and the students evaluated as Fair (50.57%). In addition, we analyzed the learning performance of both groups and made practical suggestions for teaching. The findings of our study can provide a practical reference for higher education institutions and teaching staff.

For the detection needs of small target size and obscured targets in the pedestrian detection human task in complex scenes and the shortcomings of existing deep learning models, this paper proposes an improved pedestrian detection algorithm based on YOLOv3. First, the three-scale detection layer is increased to a four-scale detection layer to improve the accuracy of small target recognition through the retention of high-resolution feature map position information; second, add Spatial Pyramid Pooling layer to improve the generalization ability of the model; finally, the Convolutional Block Attention Module attention mechanism module is added to reinforce the important feature information and ignore irrelevant information.

Countersink quality detection using point cloud is an effective way to detect defect of countersink with high precision of point cloud. This paper proposes an algorithm for segmenting surface of countersink using point cloud. For features of countersink and requirement for Countersink quality detection, this algorithm based on normal vector estimation and smoothing operator. Experimental results demonstrate that the proposed algorithm can be considered as an accuracy, precision, and real-time approach for point cloud segmentation with inevitable disturbances in terms of the segmentation quality and CPU time.

The density-based spatial clustering of applications with noise (DBSCAN) method is exceptionally sensitive to the selection of parameters, making it difficult to obtain more accurate clustering results. To solve the above problem, we propose an adaptive TSO-DBSCAN algorithm using the excellent global search performance of the Tuna Swarm Optimization (TSO) method. We take the clustering radius parameter as the search objective of the TSO method. Then, the obtained evaluation metrics are used as the fitness function of the TSO method to adaptively search for the best parameter value and the optimal clustering result. The experimental results show that the proposed method has better clustering accuracy and effect on the processing of image segmentation.

Since the commercial use of 5G, the largest 5G network is built in China. By the July 2022, 870,000 5G base stations had been jointly built and shared, accounting for 46.9% of the total number of 5G base stations. Co-construction and sharing network not only achieves rapid coverage of the network, but also bring a great improvement in user experience. In the case of Co-construction and sharing network, whether there are differences in service performance and network coverage need in-depth analysis. This paper is to introduce the basic theory of 5G Co-construction and sharing network, service perception evaluation and analysis method. Based on the evaluation of the short video testing, this paper analyzes the reasons of the user perception differences in the 5G Co-construction and sharing network scenarios from the aspects of wireless coverage of 5G network, QoS scheduling of the core network and the attribution of the application server.

By applying the fish swarm algorithm to the generation of the multicast tree under the dynamic satellite network, a method for generating the multicast tree with good adaptability and inheritance was found, and the generation efficiency of the multicast tree was improved.

In order to solve the identity authentication and security communication problems of people, vehicles, roads and clouds in C-V2X communication, it is necessary to carry out the research on security authentication technology of C-V2X shortrange communication in intelligent connected vehicles. Based on the V2X scenario of Hongqi brand automobile, this study introduces the C-V2X short-range communication security authentication technical scheme, analyzes the requirements of the V2X PKI authentication system, combines the domestic commercial cryptography algorithm, adopts, and realizes the flexible addition or reduction of each module through hierarchical design and component-based development technology, which can quickly adapt to the different needs of users for network communication security, and provides comprehensive security services such as identity authentication, access control, and data integrity protection for network applications.

The International Classification of Diseases (ICD) is a set of classification codes for medical records. Automated ICD coding, which assigns the International Classification of Disease codes to patient visits, has attracted much research attention since it can save time and labor for billing. In this work, we proposed an automatic ICD coding method for clinical records based on a co-occurrence disease perception network. It uses a multi-scale residual network to capture text patterns of different lengths of clinical text and Personalized PageRank to capture co-occurrence between codes to enhance the ability of automatic coding (MSRPPnet). As shown in the experimental results on the real medical dataset MIMIC-III, the P@N and Micro-F1 of this method are 72.2% and 55.1%, respectively, which significantly improves the prediction performances.

With the development of optical communication, optical fiber sensing technology has come into people's view. Among them, the phase sensitive optical time-domain reflectometer φ-OTDR because of its high detection accuracy, good dynamic performance and has been widely studied. In this paper, a new signal processing algorithm is proposed to reduce the computational resources needed for the disturbance signal processing, so as to improve the detection speed of the system.

To accelerate the speed of finding new materials and to predict the attributes of materials, Machine Learning has emerged as a very useful tool in recent times. Predicting the band gap of a new material plays a significant role in evaluating the unknown material. To predict the band gap, several regression methods of supervised learning in machine learning were applied, such as linear regression, k-nearest neighbors-regression, random forest regression, support vector machine and decision tree regression. Besides, comparative results illustrated that the predictions of the random forest and the decision tree methods were very good. This enables the possibility to understand thousands of unknown band gaps of non-metals, especially in case of semiconductors.

Tunable diode laser absorption spectroscopy (TDLAS) technology is one of the spectral detection technologies, and its advantages include high selectivity, high sensitivity, and fast response speed. The detection accuracy of the TDLAS gas detection system is closely related to the calibration curve. When calibrating, the least-squares algorithm is often used for fitting. However, the fitting algorithm is inaccurate in the inversion of the coefficient matrix, which causes the system to detect excessive gas concentration errors. Therefore, an improved LS algorithm is proposed to solve the above problems. The experiment is verified by detecting the concentration of carbon dioxide gas. The central wavelength of the gas is selected at 1573.2 nm, and the range to be measured is 10%~30%. The error sum of the improved fitting algorithm is 3.1955×10^-5, the mean square error is 5.3259×10^-6, the standard deviation is 2.3×10^-3, and the correlation coefficient is 0.9998. In the whole concentration range, the absolute value of the relative error of the gas concentration obtained by its inversion is less than 3.0479%. The reliability of the improved fitting algorithm is verified, the fitting effect is improved, and the accuracy of TDLAS gas detection is improved.

With the rapid development of deep learning, many areas have achieved impressive progresses. However, deep learning often suffers from complex computation paradigms and huge memory occupancies, lightweight networks attract gradually increasing attention recently. Various lightweight network models have emerged in recent years, and are also widely used in small mobile devices. Then, there is still a lack of expressive power for lightweight networks compared to deep networks. This study leverages multiple statistics as representative of a feature map and proposes a Multi-Style Recalibration Module of a channel attention mechanism to improve the accuracy of lightweight network model recognition without increasing their storage space. This mechanism is embedded on MobileFaceNet as a study case. We test the new model on LFW dataset for demonstration and find that its occupancy only increases 0.2 MB, while the accuracy is increased by 0.18%.

The World Health Organization pointed out that diabetes is a non-communicable disease that needs attention at all times. From 2006 to 2016, the number of deaths from diabetes in the world increased by 31.1%. Diabetes is slightly less harmful to the human body than cancer, which ordinary people are afraid of. It is a disease with the largest number of known complications. Once complications occur, it is difficult for the patient's body to recover with drug treatment, and it has a high direct death rate.To apply random forest algorithm and Xgboost algorithm to study the risk factors of diabetes and build a high-quality risk prediction model. The data comes from the Pima Indians diabetes dataset, which includes a total of eight features, such as the number of pregnancies, blood pressure, plasma glucose concentration, body mass index, serum insulin concentration, etc. The response variable is whether the patient has diabetes. Random forest algorithm and Xgboost algorithm were applied to compare the prediction effects of the two models, and the importance scores of features were given at the same time. Results: In the importance score of the random forest model, the scores of plasma glucose concentration, body mass index, age, and diabetes spectrum function were particularly high, and the prediction accuracy of the random forest model was higher than that of the Xgboost model. Conclusion: Several strongly correlated features obtained by random forest algorithm can be used for a screening of high-risk groups of diabetes and the guidance of risk intervention measures, so as to find potential diabetic patients in the population, and achieve early intervention and early treatment. Reduce the waste of medical resources.

Maize is one of the important foods and feed crops with high agricultural value in the agricultural fields of northern China. Accurate detection and treatment of nightshade spodoptera frugiperda in agricultural fields is essential to increase agricultural production. Compared with the traditional manual detection methods, the convolutional neural network-based spodoptera frugiperda detection algorithm has the better recognition performance. However, the method of using convolutional neural networks for the recognition of spodoptera frugiperda still suffers from low accuracy and slow detection speed. To improve the detection accuracy and speed of the spodoptera frugiperda, the algorithm modified the information fusion based on the YOLOv3 model, and used quadruple upsampling fusion instead of the original double upsampling fusion to remove the redundant information brought by repeated fusion. In the experiment, the dataset consists of 4000 training images and 200 test images for training and testing. The experimental data show that the improved YOLOv3 algorithm improves the accuracy by 11.42% and the detection time by 0.0023s compared with that before the improvement.

Graph wavelet neural network, as a popular method of graph neural networks, has achieved good result in representation learning. However, graph wavelet neural network faces the phenomenon of over-smoothing, which means that repeated propagation of multi-layer network makes node feature representation indistinguishable. The current methods to alleviate over-smoothing phenomenon of graph wavelet neural network have the problem of add more model parameters, which leads to running time become higher and training process more difficult. In order to solve the above problem, this paper proposes a new method called Markov graph wavelet neural network. On the one hand, Markov graph wavelet neural network decouples feature transformation and feature propagation. First, multi-layer perceptron is used for feature transformation, which reduces the number of model training parameters and reduces running time. On the other hand, Markov diffusion kernel is applied to k-order feature propagation, effectively obtains node local information and node global information. Besides, the influence of initial feature is considered to update target node representation, which effectively alleviates over-smoothing phenomenon. Finally, the experimental result of Markov graph wavelet neural network model on Cora, CiteSeer, PubMed exceeds the representative baseline algorithms, which fully demonstrates the superiority of Markov graph wavelet neural network.

With the development of intelligent connected vehicles, especially L3 level automatic driving technology, it is very important to identify and avoid the impact of malicious jamming signals on the communication of intelligent connected vehicles. In this paper, we classify and identify the five kinds of common intelligent connected vehicle communication jamming signals: single-tone jamming signal, multi-tone jamming signal, linear swept-frequency jamming signal, partial band jamming signal and noise frequency modulation jamming signal. Firstly, six kinds of features of communication jamming signals are extracted: frequency domain moment kurtosis coefficient, frequency domain moment skewness coefficient, single frequency encircled energy, average spectral flatness coefficient, frequency domain parameter and time domain moment kurtosis coefficient, and then classified and identified based on convolution neural network (CNN). The simulation results show that when the jamming-to-noise ratio (JNR) is higher than -7dB, the recognition rate is over 92%, which indicates that CNN has good classification and recognition performance for communication jamming signals.

With the continuous development of intelligent connected vehicles, the information security risks are also increasing. In addition to the information security protection of vehicle external communication, the information security protection of vehicle internal network is also getting more and more attention. Based on the domestic commercial cipher algorithm, this paper formulates an overall information security protection strategy suitable for the scene inside the vehicle from the two aspects of vehicle external access security protection and vehicle communication network security protection.

Aiming at the many parameters and high computational complexity of video-based deep learning fall detection models, we propose a lightweight fall detection algorithm for 3D residual networks. In this approach, we design a low-rank depth-separable convolution structure. When performing deep convolution, the 3-dimensional parameter matrix is decomposed into 1-dimensional and 2-dimensional parameter matrices to reduce the model parameters and thus improve the performance. Meanwhile, the dataset is built by referring to the format of the URFall dataset and capturing videos of human falling and non-falling states from multiple angles using RGB cameras. The experimental results show that the lightweight 3D residual network can achieve 98.23% accuracy in distinguishing falls from non-falls, and the sensitivity and specificity are kept at a high and stable level.