This PDF file contains the front matter associated with SPIE Proceedings Volume 12456, including the Title Page, Copyright information, Table of Contents, and Conference Committee Page.

With the continuous development of science and technology, robots and artificial intelligence have made great progress and progress. The scope of application in the production and life practice of today's society is also more and more extensive. How to perfectly combine artificial intelligence and intelligent robots to create more wealth and convenience for society and life is the focus of people's attention and research. The State Council's "New Generation Artificial Intelligence Development Plan" clearly puts forward the goal of reaching the world's leading level of artificial intelligence theory, technology and application by 2030. This paper firstly introduces the research background and significance of robotics and artificial intelligence. Then it analyzes the connotation, development status and main applications of robots and artificial intelligence. Finally, the development trend and development prospect of robots and artificial intelligence are prospected.

Passive positioning systems play an essential role in the field of electronic countermeasures. Many factors affect the efficiency of passive positioning systems, especially the deployment of its observatories station. In this paper, the genetic algorithm based on bi-level selection is proposed to solve the deployment optimization problem of the time difference of arrival for a passive positioning system in three-dimensional space. In this paper, three sets of comparison experiments are set up to determine the superiority of the proposed algorithm. The experimental results show that the improved genetic algorithm can be well applied to the deployment of three-dimensional time difference of arrival positioning, and the proposed bi-level selection operation is clarified which has a great effect on promoting the convergence of the algorithm.

In order to solve the problem of online inspection of distribution station operation and maintenance, a design method of automatic inspection system for smart distribution station based on cloud-edge collaboration is proposed. Firstly, the overall architecture of the smart distribution station automatic inspection system based on cloud-edge collaboration is designed. Then, the system software architecture is designed in detail through three major parts and 12 functions including cloud resource layer, system platform layer and application layer. Secondly, the hardware architecture of the smart station terminal is designed in detail through the construction of the secure access area and the configuration of the virtual server of the cloud master station. Thirdly, based on artificial intelligence speech recognition technology, the overall architecture of the voice inspection assistant is described. Finally, the software development and application show that there is important practical significance in improving the intelligent maintenance level of distribution network.

5G network layering technology is a new type of technology, which can ensure the service quality of the entire system. Currently, in 5G networks, the performance and security of slice layering technology has become a hot topic. To this end, based on SecPSO, this paper presents a new method for security separation of 5G slices. The SecPSO algorithm is improved and optimized through distributed edge computing 5G network slicing, selecting the fitness function, and listing the algorithm steps through the particle aggregation degree design. Through the improvement of the SecPSO algorithm, the security of the slice is guaranteed, and the algorithm of SecPSO is optimized, which shows the superiority and reliability of the method, and lays a foundation for the application of 5G network.

In large-scale social networks, rich-club phenomenon is prominent. The important nodes selected by the existing algorithms are often too clustered, and the overlapping influence between nodes leads to a poor effect of maximizing the comprehensive influence of the node group. To solve this problem, an important node group mining algorithm based on information entropy and overlap coefficient is designed inspired by iterative optimization ideas. The algorithm consists of two stages: importance initialization stage and importance adaptive iterative update stage. In the initialization stage, the information entropy is used to measure the amount of information of nodes, evaluate the local importance and global importance of nodes, and improve the accuracy of the algorithm. In the iterative update stage, the overlap coefficient is used to evaluate the influence of the currently selected most important node on its surrounding neighbors, and adaptively weaken its local importance. Through the simulation experiment of the SIR propagation model, on the four real network data, the average distance between the nodes selected by the algorithm is larger and the distribution is wider, which reduces the overlap of influence and obtains greater influence.

In a short time, how to use scientific and technological means to complete the allocation of vehicles in the implementation of tasks is of great significance to improve the decision-making ability of personnel equipped with vehicles. Based on KNN algorithm, python related tools are used to calculate, classify and visualize 25 test samples after processing. Finally, according to the Euclidean distance between the test sample and the nearest neighbor sample, it is concluded that the vehicle selection of sample 26 is ' selection ', which provides a more scientific auxiliary decision method for the rapid decision of vehicle selection and allocation, and has certain reference value.

The current text generation generally requires understanding the contextual derivation of corpus based on a large amount of data. In fact, it is very difficult because of the limited data. We develop a novel text generation algorithm based on few-shot learning data which uses a dual-channel decoder and can smoothly perform copying and generating for the predicted words. Experimental results show that this proposed algorithm’s efficiency.

To make full use of the pulmonary nodules features extracted by convolutional network and improve detection of pulmonary nodules, an improved Mask R-CNN pulmonary nodules lesion detection algorithm fused with attention mechanism was proposed. The improved multi-scale pyramid is used to enhance the contextual lung nodules information, so that network can obtain high-level semantic features without losing underlying texture information. In order to obtain richer pulmonary nodules lesion features, residual network fused with attention model is designed to extract nodules channel and spatial information. Experiments were carried out on the lung nodules analysis 16 (LUNA16) dataset. The improved algorithm achieved an average detection sensitivity of 96.8% and an average segmentation Dice coefficient of 96.2% for lung nodules lesion detection in the dataset. The results show that proposed algorithm can effectively detect and segment lung nodules lesions.

Landslide is a very serious geological disaster. When the heavy rain occurs in the area with loose soil layer, the mountain becomes loose after strong erosion of rain water. As soon as the threshold of adhesion force between mud and mountain base is broken, the powerful harmful force formed by landslides will bury houses, people and cars. Traditional detection methods, represented by InSAR or LIDAR, were capable of identifying landslides to a certain extent, but in-time and real-time performance were not good enough. In this paper, an intelligent identification method based on convolutional neural network(CNN) is proposed, which can in-time and real-time to identify landslides, thus greatly reducing the loss of life, health and property.

In view of the serious economic losses caused by the unreasonable investment decision-making scheme of the current medium and low-voltage distribution network project, an optimization method for the investment decision-making of the medium and low-voltage distribution network project based on genetic algorithm is proposed. Combined with the optimization of the multi energy topology of the distribution network based on Genetic algorithm, the investment income model for the multi energy storage, renewable energy power generation and line reconstruction of the distribution network is established. According to the model, the required investment is planned by stages, an investment evaluation system considering efficiency and benefit is constructed, the evaluation algorithm is optimized, and an investment decision-making model is constructed based on the calculation results. Finally, based on the actual operation data of a regional multi-energy power grid, a multi-energy investment decision-making simulation model of distribution network is established. The simulation results verify that the optimization method of investment decision-making of medium and low-voltage distribution network projects based on genetic algorithm can improve the efficiency and benefit of multienergy investment.

Blockchain is considered as a revolutionary technology that has recently fueled extensive attention across many industries. Except for Bitcoin, it has been applied into diverse areas ranging from finance, logistics to food traceability and medical care. It will not take long for blockchain to change the whole society. Although there have been a large number of papers concentrating on blockchain application in a variety of areas, there is lack of a comprehensive survey about blockchain from the perspective of technology and application. To this end, we present a comprehensive overview on the core technologies in blockchain, including cryptography, smart contract and consensus mechanism. In addition, this paper reviews blockchain application and discusses existing problems and bottlenecks in the development of blockchain technology.

As an important part of today's optoelectronic products, optical microstructure components are bound to restrict their development if they are still processed by traditional methods. Therefore, in order to promote the development of optical microstructure components, artificial intelligence technology is applied to make up for the insufficiency of manual processing methods and promote the development of optical microstructures in aerospace, information and other fields. Specifically, in the context of artificial intelligence, a single-point diamond is used to conduct machining experiments on micro-V grooves to explore the influence of spindle speed, depth of cut and feed speed on the machining accuracy of micro-V grooves, and to select appropriate cutting parameters. Finally, using these cutting parameters, the micro-V groove workpiece is processed, and it is found that the workpiece has the following characteristics: smooth surface, clear structure, clear array, etc. It has good consistency and can meet the requirements of optical microstructure machining accuracy. It can be seen that artificial intelligence can promote the development of optical microstructure processing and has a positive role in practice.

The Internet of Things technology has gradually become a new development field, and at the same time, it has shown great growth potential and has entered all fields of human life. This paper aims to explore IoT-based smart building design methods. We analyze domestic and foreign smart buildings and IoT technologies and propose smart buildings suitable for the IoT era. Introduce the most advanced Internet-based smart technology capabilities that fully integrate the two components of wireless sensor networks and traditional infrastructure. Secondly, use the wireless sensor network software and built-in installation program to provide access to the received data information, and develop services such as historical data center, real-time data update, historical database export, historical data analysis, etc., to provide data for the next stage of data visualization. The system is functionally tested, and the experimental results show that 23℃ can be used as the comfortable temperature for the intelligent building system to automatically set in summer. Smart buildings based on IoT technology further improve our quality of life.

In the field of artificial intelligence, the number of sports information sources on Internet platforms is increasing. The purpose of this paper is to study artificial intelligence technology and competitive sports. Firstly, the characteristics of the current sports event Internet data are studied and the corresponding collection framework and collection method are determined, so as to provide a basis for the selection of the recommendation model algorithm. Taking the problem-oriented marathon event in this paper as an example, it discusses the characteristics of its network information, and realizes the collection and analysis of its network data, This provides guidance and data support for building a suitable sports activity planning model. The experimental results show that the sports event recognition algorithm based on the LDA topic model is easy to achieve the best effect on a large amount of event text data, the system is clear, and the features are customized. The overall event reliability of the comprehensive security model based on LDA keyword model and Word2vec sequence is higher than 90%.

With the rapid development of sensors and integrated chips, traditional power grid inspection tasks are gradually replaced by drones and new sensors. However, the ability of robots or drones to collect and process information needs to be improved. The integration ability needs to be improved. This paper analyzes the shortcomings of the current UAV inspection robots, builds a wireless sensor network based on Lora, 5G modules and various sensor combinations, and completes the data collection work with the inspection robot at the same time. The information is sent back to the PC data center, and finally the purpose of drone inspection and assistance in inspection is achieved. After testing, the multi-sensor fusion UAV power line inspection can realize a full series of density curve distribution diagrams to meet the limited requirements of the line safety distance.

As an important part of breeding research, germplasm resources play a fundamental role in maize breeding. This study starts from the needs of corn breeding scientific research, according to modern breeding technology, using computer technology and systems engineering technology, to study the key technologies of corn breeding process informatization, and build a practical and reliable corn breeding intelligent auxiliary decision support system. The establishment of this system will help to mine massive germplasm resources information, shorten the breeding time of new varieties with high quality and high yield, promote the development of informatization and intelligence in the breeding process, improve the level of breeding technology, and improve the level of corn breeding in the future. and provide a reference basis for the utilization of excellent maize germplasm resources.

The position of artificial intelligence in human society has gained widespread attention, and the impact it has brought cannot be underestimated. The so-called deep learning is through the existing knowledge, experience or information to organize, and will be used in real life, it is the use of computer and artificial intelligence principles to achieve interactive learning process between the human brain and the machine and a new system model, now the depth of knowledge mining and acquisition has become the current trend of development. This paper will put forward relevant suggestions and give specific path analysis for future research in the field of artificial intelligence in China at the present stage for the deep learning mode of English in the digital background.

With the widespread promotion of intelligent manufacturing and the rise of cloud computing and artificial intelligence, big data analysis plays an increasingly important role in the production and operation of manufacturing enterprises. Mining and visual analysis of the massive data of manufacturing enterprises is conducive to discovering the hidden laws and causal relationships of the data, so as to extract useful information and help enterprises implement better decision-making. This paper expounds big data analysis from three aspects: data life cycle management, big data analysis types and key technologies involved, analyzes the application of big data analysis in the field of intelligent manufacturing, and proposes a big data analysis architecture for manufacturing enterprises. Finally, corresponding countermeasures and suggestions are put forward for the problems existing in big data analysis under the background of intelligent manufacturing.

We propose a deep learning model-based algorithm for the analysis of abnormal behaviour of violent crime video targets, for violent crime videos, to determine whether there are abnormalities in the targets. 2D-net, 3D-net and MIL are used to construct the model. A sparse sampling strategy is used to extract features and a continuous MIL is used to achieve the final classification. The model achieves satisfactory results on the UCF_Crimes dataset and performs well in comparison with similar algorithms in the field.

Traditional static defense methods based on threat detection, isolation and filtering are difficult to defend against increasingly complex and intelligent cloud network intrusions. New defense methods try to increase the uncertainty and complexity of the cloud environment. Aiming at the problems of increasing attack surface of web services and difficult security management and control, based on the mimetic web endogenous security system architecture, this paper presents the system physical frame and logical frame of Mimicloud, a web architecture of the DHR model based on the continuous time Markov chain Model and analyze the security mechanism, and finally conduct an evaluation test. This paper innovatively uses the DHR cloud endogenous security architecture to provide highly secure SaaS services, and then constructs a security mechanism with dynamic, random, diversity and other characteristics, which makes the attacker lose part of the latent resources and attack implementation conditions. It is difficult to maintain the continuous control and access to the successful attack. The mimic Web can also provide theoretical support for enhancing the endogenous security capabilities of the next-generation information cloud infrastructure and key technological breakthroughs of mimic security defense in the cloud environment.

The evaluation of teaching effect is composed of many important indicators, and learning adaptability is one of the more important indicators. Effectively analyzing the application of learning can more accurately evaluate the learning effect of students. Therefore, under the background of artificial intelligence, the analysis of college students' English learning adaptability should be strengthened to accurately understand the effect of college students’ English learning and lay a good foundation for optimizing college students’ English teaching activities. Based on this, this paper studies the English learning adaptability of college students based on artificial intelligence through experimental analysis, and provides support for further improving the results of learning adaptability.

With the intensification of the global energy crisis and the increasing shortage of fossil fuels, more and more countries have begun to strengthen the development and utilization of new energy. In recent years, with the help of policies and capital, the photovoltaic industry has achieved rapid development and occupies a leading position in new energy. The output power of a photovoltaic power station is affected by various factors such as solar radiation intensity, temperature, and installation method. The aging of photovoltaic modules, surface dust and component damage will also affect the power generation efficiency of photovoltaic power plants. The identification of abnormal data can not only help power station owners and operation and maintenance manufacturers to find potential equipment failures and other problems at the first time, but also can effectively avoid safety risks and economic losses. This paper proposes an abnormal data detection algorithm for small photovoltaic power plants based on machine learning. First, the operating data of photovoltaic power plants are normalized, and then the abnormal scores of the data samples are calculated by the iForest method, and then the classification center is calculated by the K-means method. This method can realize abnormal data detection in small photovoltaic power plants without irradiators, effectively avoid potential failures and risks of photovoltaic power plants, and ensure the operation safety of power plants and power grids to a certain extent.

With the increasing complexity of the power communication network structure, more and more business information is carried, and ensuring the safe and stable operation of the network has become a daily operation problem. Communication failures in power systems are often associated with power supplies, communication optical cables, etc. Power supply failures involve rectifier modules, batteries, and high-frequency switching power supplies. Optical cable problems are mainly due to partial or complete blocking of optical fiber lines. This paper mainly studies power communication failures. Common O&M and overhaul strategies.

Online educational resources are a new type of learning mode with information and knowledge as the core under modern Internet technology. It includes various elements such as computers and multimedia equipment, and uses modern information technology such as computers and multimedia equipment to collect and process information, and realize the simulation of various knowledge and behavioral skills needed in the learning process of the taught person. In this environment, people can access and interact with teaching resources remotely via the Internet. Due to some problems and shortcomings in the process of practical application, its development is limited to a certain extent. Therefore, this paper focuses on building a semantic model of online educational resources based on the discrete cosine transform, using ontological means, and using it to improve the quality of communication services of the whole learning community and provide users with better and more intelligent educational resources.

In order to further improve the efficiency and accuracy of translation, based on the existing computer-aided translation software modules, artificial intelligence technology is introduced to gradually design its architecture, functional modules and databases and other components, and build a test environment, and test the function of the translation system software by taking Chinese to English as an example. The test results show that the translation system software can complete the translation task, and all the indicators meet the standards, which proves that the computer-aided translation software based on artificial intelligence technology designed this time has achieved initial success.

At present, vehicle tracking has been realized in many fields. The applications of vehicle tracking include Advanced Driver Assisted System and live broadcast of automobile events. After reading the paper's Intelligent vehicle pedestrian tracking based on YOLOv3 and DASiamRPN, we learn that they propose a pedestrian tracking algorithm which combines YOLOv3 and DASiamRPN to realize the designated pedestrian tracking. Finally, the pedestrian tracking is successfully realized, which ensures the tracking accuracy. However, while exploring the guarantee of accuracy, there are few studies on how to achieve successful tracking at different tracking speeds. The most difficult part about vehicle tracking is fast vehicle tracking. High speed often makes the system lose tracking target. So, in this paper, we studied how to implement fast vehicle tracking. However, there are many algorithms for vehicle tracking, such as the Lucas-kanade algorithm, Faster RCNN algorithm, Yolo algorithm, SSD algorithm, and DeepSORT algorithm. More and more new algorithms have appeared in recent years. But through experiments, we found that there are some vehicles tracking algorithms such as the Lucas-kanade algorithm, which cannot achieve accurate vehicle tracking in the case of rapid vehicle movement. Therefore, we listed and studied a variety of algorithms for vehicle tracking. After querying the papers., comparing and analyzing, one of the appropriate algorithms was selected: DeepSORT and Yolo-V5 algorithm. By analyzing the principles and image processing flow of DeepSORT and Yolo-V5 algorithm, we know that this algorithm is updated and more reliable. After experiments, it successfully tracks fast moving vehicles in a given video. After statistics, the successful tracking time, testing time and accuracy of the video meet the successful tracking standard.

With the continuous innovation of information technology, the field of artificial intelligence has seen unprecedented development. At present, artificial intelligence is involved in all aspects of our life, and solving the problems in our life is essentially to optimize the problem structure. For artificial intelligence algorithm, the simplest search mode is to search the problem directly. The artificial intelligence search algorithm does not need the derivative information with the constraint function and the objective function, but only needs the function information. The algorithm is an effective way to optimize the solution of undifferentiable or relatively high cost of differentiable. In this paper, the optimization problem is described, and artificial intelligence search algorithm is studied, and the special use of artificial intelligence technology in the field of dynamic programming algorithm is analyzed. This paper studies the applicable fields and characteristics of breadth search algorithm and depth search algorithm, and finally tests the results of the algorithm research.

In the era of big data, the network is always full of massive data. Cloud computing provides huge technical support for processing massive data. The cloud environment stores a large number of important information of individuals, enterprises and even countries. It has high commercial value, so it has become the primary target of many network attacks. Therefore, it is necessary to monitor the traffic in the cloud environment in real time and block abnormal traffic in time to ensure a safe and stable network environment for users. The existing intrusion detection systems can be divided into software systems and hardware systems, which are deployed in the backbone network in the network environment for real-time traffic detection. It is difficult to meet the traffic detection of multi branches and massive data in the cloud environment. This paper combines the deep neural network model with Hadoop framework, and proposes a distributed intrusion detection system model based on CNN-GRU. The deep neural network model is deployed in multiple nodes in the cloud environment, and the data is stored through HDFS and mapped and integrated by MapReduce method, so as to realize the intrusion detection of multi node parallel cloud environment. Finally, through the open source intrusion detection data set, the experimental results prove the effectiveness of the proposed method.

With the improvement of the public’s living standard, the public has higher requirements for the whole house customization system, and it is more inclined to the whole house customization system with a higher degree of intelligence. Therefore, this paper studies the whole house customization system based on artificial intelligence, combining the different needs of consumers, designers and systems. In order to further improve the technical content of the whole house customization system, the basic idea and process of designing the whole house customization system based on artificial intelligence are discussed in detail from the aspects of system architecture, system function and algorithm application.

With the development and progress of the times, China’s scientific and technological strength has gradually improved. The communication between robots and humans is getting closer and closer, and humans have endowed robots with emotions and actions through repeated research and experiments. The research of robots has also attracted the attention of many experts and scholars in the fields of psychology and computer science. These scholars and experts combine computer science with psychological science and cognitive science to try to create a human-computer interaction environment with emotional feedback. Efforts to achieve harmonious and natural human-computer interaction. The research content of this paper is guided by artificial psychology theory and affective computing, and the technology of constructing service robot interaction platform is studied in detail. The following issues related to service robots are analyzed and discussed: control architecture, emotion expression, emotion modeling, robotic application systems, etc. It also lays a solid theoretical foundation for the establishment of a harmonious man-machine.

The comment data of e-commerce platform is an important basis for consumers and marketers to make decisions. With the development of platform related technologies, the scale of image comment data shows an increasing trend with stronger information expression ability. At present, the research on the related technology of comment image and comment text is relatively mature, and how to effectively integrate the image part and text part is the focus of image-text comment data research. This paper proposes an e-commerce image-text comment mining model based on depth feature fusion. According to different stages, the model is mainly divided into four stages: image-text comment data acquisition, pre-processing, feature extraction and feature fusion. The e-commerce image-text comment data set is constructed through the data acquisition and pre-processing. In the feature extraction and fusion stage, the graphic information fusion is realized by designing the structure of deep neural network model. The fused features are applied to downstream mining tasks. Finally, the experiment on the data set shows that the classification accuracy of the model is higher than that of single text or single image, which shows that the model can effectively improve the mining effect of image-text comments.

Smart metering was initially used for electricity and natural gas. Smart water meters (SWMs) have attracted much attention in recent years. SWMs are installed in users' homes to collect users with water consumption information. However, SWMs real-time data collection can reveal users' private information. The paper proposes a secure smart water meter data aggregation scheme based on edge computing. The solution adopts Boneh-Goh-Nissim cryptosystem, BLS short signature and edge computing to protect users' privacy information. It can resist various attacks and protect the water consumption data of users from being leaked. The solution can resist various attacks while protecting the user's water consumption data information from being leaked. The performance evaluation shows that the communication cost of this scheme are lower than those of the existing schemes.

Aiming at the problems of low detection accuracy and large weight files in the traditional traffic sign recognition algorithm, it is not suitable for practical application. A traffic sign recognition method based on the improved YOLOv5 algorithm is proposed. First, improve the loss function of YOLOv5, use the DIOU loss function to optimize the training model, improve the accuracy of the algorithm, and achieve faster target recognition. Combined with the lightweight convolutional neural network MobileNetv2, the lightweight improvement of the YOLOv5 network is achieved.

Mathematical symbols and formulas play an important role in the process of social development. The development of today's computer technology also benefits from the development of mathematical symbols and formulas to a large extent. This article uses mathematical symbols and formulas to turn the concept of responsibility in human society into mathematical units and mathematical formulas. This research will turn the natural language of responsibility into the language of the digital symbolic language of responsibility (the mathematical language of responsibility), and use the principles of mathematics and physics to digitize the microstructure of the traditional concept of the natural language of responsibility, so as to realize the digital symbolic language system of responsibility, and let human behavior , law, ethics and responsibility have the characteristics of digital symbolization to meet the development of big data artificial intelligence of digital identity and digital behavior.

This paper conducts a study on the application of sensor technology in intelligent substations. The network topology of the sensor network, the deployment of sensor nodes, and the security mechanism of the sensor network are analyzed. The final application scheme is proposed following the analysis above. Different network topologies are adopted for wired and wireless sensor networks. At the same time, the improved self-organizing map (SOM) algorithm is used for node deployment. The advanced encryption standard (AES) encryption improvement algorithm is adopted for data encryption and decryption to attain timely and efficient multifaceted sensing of information in intelligent substations under the guarantee of data security.

A supplementary controlled system can strongly show how intelligent a substation is. Nowadays, supplementary control facilities which can achieve functions such as safety protection, fire monitoring, water supply and drainage, heating and ventilation, and video surveillance have been set in substations, but those systems lack overall planning. Based on the complex network theory, the objective function for network stability evaluation is constructed in this paper to get the optimal layout of a video surveillance system during the later process. Constrained by the anti-interference of the connection network of supplementary controlled facilities, the automatic optimization layout of the supplementary controlled system can be obtained by particle swarm optimization (PSO).

The chaotic systems and DNA operations are used to encrypt images jointly. The fractional Chen Chaotic System is used to generate key DNA images, and the logical map is almost applied to all encryption steps. The values of all the image pixels are first converted from decimal to quaternary, and then represented by DNA sequences. And the DNA computing methods are applied to the diffusion step between the key DNA and the DNA image converted from the plain image. Experiments can prove that the image encryption algorithm proposed by us can meet the requirements of image encryption, and can resist various attack methods, including noise attack, differential attack, etc.

The EHV power grid has a large span and close electrical connection. To carry out the operation and maintenance of power grid in a wide area requires the coordinated operation of multiple branches of operation and maintenance enterprises. In view of the fact that the relevant technical support system is independently constructed by each branch, there are problems such as low management efficiency and many information islands, the study proposes a centralized technical support system construction scheme based on virtualization technology, which can effectively solve the contradiction between unified system management and individualized needs of branches.

The prevalence of nasopharyngeal carcinoma is the first in otolaryngology head and neck malignancies. The location of the disease is hidden, and most patients are already in the advanced stage of cancer when they are found. In order to explore the construction of an index model of early dyslipidemia in patients with nasopharyngeal carcinoma, this study used the laboratory information management system to obtain a large amount of blood test data and used statistical software to conduct correlation analysis with healthy people. Through the analysis, a dyslipidemia index model belonging to the early screening of nasopharyngeal carcinoma can be gradually constructed. In this study, 55 patients with nasopharyngeal carcinoma who visited a doctor at the same time were selected as the experimental group, and 1244 people with physical examination were selected as the control group. The study found that the serum total cholesterol, triglyceride, and high-density lipoprotein of patients with nasopharyngeal carcinoma were significantly lower than those of healthy people. Serum total cholesterol, triglyceride, and high-density lipoprotein can be used as basic data for early dyslipidemia index model in patients with nasopharyngeal carcinoma. In this study, a screening model of early blood lipid metabolism indicators in nasopharyngeal carcinoma patients was constructed by means of the blood test data stored in the laboratory information management system, which provided a theoretical reference for the early screening of nasopharyngeal carcinoma patients.

Esophageal malignant tumor is one of the common digestive system malignant tumors. It is difficult to detect early, and there are great individual differences. In order to further explore the differences in blood indexes between patients with esophageal malignant tumors and healthy people, this study used the LIS information system to collect a large number of blood test data for statistical analysis. In this study, 346 patients with esophageal malignant tumors who visited a doctor at the same time were selected as the experimental group, and 906 people with physical examination were selected as the control group. The study found that serum albumin, fasting blood glucose, total cholesterol, triglyceride, high-density lipoprotein and other blood test indicators of patients with esophageal malignant tumor were significantly different from those of healthy people. The above blood test indicators can build an early screening model for esophageal malignant tumors. This study uses the medical health data stored in the LIS information system to verify the differences in blood indicators between patients with esophageal malignant tumors and healthy people, and establish a digital model for early screening of patients with esophageal cancer, which is conducive to rapid and accurate early screening of esophageal cancer. Provide reference for early detection and timely treatment of esophageal cancer.

In order to study the growth status of Haematococcus pluvialis in the cell proliferation stage, BG11 was selected as the culture medium, and the light intensity, pH and cell radius were taken as the characteristic variables of the model, and the growth status of algal cells was taken as the target variables of the model to establish a regression prediction model.Through the relevant data of the cell proliferation stage of Haematococcus pluvialis measured in the experiment, the regression tree prediction model is adopted, and the cart algorithm is used for data mining. The growth status of Haematococcus pluvialis is predicted according to the relevant indicators, and finally the growth prediction model of Haematococcus pluvialis is obtained. The use of this model can effectively predict the growth of Haematococcus pluvialis cells, provide a unique idea for the efficient culture of Haematococcus pluvialis, and have a certain guiding significance for the large-scale culture of Haematococcus pluvialis.

With the rapid development of computer graphics and image technology and 3D modeling, great progress has been made in the field of 3D digital relief. Hollowed out relief is a form of concave relief, which is obtained by carving out the background plane on the basis of concave relief. Digital relief is generated by computer-aided modeling, which overcomes the shortcomings of traditional manual relief, such as low production efficiency and difficult to modify products, and is easy to save. In this paper, the design and implementation of image relief based on computer 3D modeling, the basic modeling process of computer 3D modeling, plane relief and the generation method of fractal pattern concave relief based on distance transformation are discussed. A differential algorithm is proposed to realize the concave convex effect of image relief. Experiments are carried out on different models on laptops configured as main frequency and graphic display card to verify the rationality of the above algorithm. The time consumption of viewpoint selection of different models is tested, and the modeling efficiency of different input models is counted. The test results show that without program optimization, the viewpoint selection time of the model with the number of mesh patches of 15K ~ 50K ranges from 14 to 58s, among which the development of model visibility judgment and significance calculation is the largest, accounting for about 50% ~ 60% of the running time of the whole process. Therefore, the algorithm optimization of visibility judgment will help to improve the operation efficiency of the system and reduce the waiting time of users. Under the same input model, the computational efficiency of this algorithm is hundreds of times higher than that of the literature method. For the models in the table, due to the complex occlusion degree of patches, the modeling time of shallow relief is up to 3.778 seconds; For the mesh model with the number of patches below 35K, the efficiency of bas relief modeling is about, which basically meets the requirements of real-time.

With the rapid development of deep learning, the research focus of speech synthesis has gradually shifted to artificial neural network technology. The speech quality has been greatly improved and has been introduced into many application scenarios. However, the existing synthesis systems need to use rich and high-quality parallel data sets when training models, and the synthesized speech is also weak in personalized performance. This paper describes an improved Mel spectrogram acoustic feature sequence prediction model based on Tacotron2 and a StarGAN-VC model. The model uses the predicted Mel spectrogram as input to generate Mel spectrogram sequence of the specified speaker and synthesize speech. StarGAN-VC model can train the model in non-parallel Mini dataset, generate Mel spectrogram sequence of designated speaker in real time and synthesize speech, which can well solve the problem of lack of non-parallel dataset and enrich the speech content generated by StarGAN -VC model. The experimental results show that StarGAN-VC model can generate relatively smooth Mel spectrogram by using the Mel spectrogram sequence predicted by the improved model, and have stronger expressiveness in dealing with the details of Mel spectrogram, so as to synthesize smooth and high intelligible speech. The model uses the speech data of the designated speaker for about 27 minutes to train the model and synthesize personalized speech, which provides an effective reference for the synthesis of personalized speech.

Self-potential method is a common method of environmental geophysical exploration. Researchers often use finite element method (FEM) to carry out forward modeling. Forward modeling is an important part of research, but the FEM has some problems, such as large amount of calculation, long time and so on. In this regard, we propose a method based on physics-informed neural networks (PINNs) for forward modeling of a kind of self-potential called streaming potential. The particularity of streaming potential is that its formation involves both flow field and electric field. In order to verify the feasibility of the PINNs method, we set up a two-dimensional (2D) scene and the corresponding boundary conditions, and tried to calculate the hydraulic head distribution and the corresponding self-potential distribution in the saturated and unsaturated flow region under stable conditions by using neural network. We have adopted two different network structure designs. One is to use the same neural network to calculate the hydraulic head distribution and electrical potential distribution in the region at the same time. The other is to use two different neural networks to calculate the hydraulic head and electrical potential respectively. In both methods, we have added the corresponding partial differential equation (PDE) to constrain the network training process. Then, we compare their results with the traditional FEM. We find that such a neural network with physical constraints can effectively obtain the spatial distribution of the unknown quantity we need. Compared with the FEM, PINNs method can be used in the case of incomplete boundary conditions. In addition, compared with other deep learning (DL) methods, the results of PINNs method are more interpretable.

To improve the accuracy of object classification and recognition, one of the methods is to add recommendation boxes, recommendation boxes are generated by region proposal algorithms. Because of the inherent ratio, the size of the candidate box generated by Faster RCNN's RPN is usually large, which would easily cause a great number of overflows in sliding search. This is unfriendly for multi-objects detection. To improve the precision of multi-objects detection, we introduce a multiscale proposal box that could predict object bounds and object scores at each position. Then, in order to increase the positive sample range of the foreground, the weighted cross entropy classification function is used for binary classification in the RPN network. In the ROI network, the candidate frame reset algorithm is proposed to realize the position regression of the prediction box and multi-classification of objects, which further improves the accuracy of object detection. The experimental result achieves the mAP of 76.2% on the VOC 07 classic dataset, which is 2.7% higher than the Faster R-CNN. On the VOC 12 test, the mAP of 75.6% is improved by 2.5% compared with the Faster R-CNN.

There are many factors that influence global temperature changes, and some methods have been proposed to find their causes and make predictions of global temperature changes. Identifying the relationship between population and temperature change and predicting the future temperature is the top priority of this study. To achieve this stated goal, this paper proposed to construct a data structure using an artificial neural network model and normalization and show the linear relationship between population size and temperature change using Pearson correlation and normalization. To make the analysis more accurate, this paper adjusted the dataset by removing useless data and eliminating the negative effects of odd sample data. The temperature data range was restricted between -1 and 1 in the normalization step of the Artificial Neural Networks model (ANN) and used the data from 1961 to 2018 as training datasets to predict the temperature in 2019. The accuracy of the ANN model, in terms of data prediction, was verified by comparing the value with the actual temperature in 2019. The world average temperature changes from 1961 to 2019 were calculated in Excel and obtained a Pearson Correlation Coefficient of 0.92727517, which indicates a strong linear relationship between the two datasets of human population and temperature change.

Transmission line inspection in most parts of the country is mainly manual inspection. This kind of inspection method is not only inefficient, but also poses a threat to the safety of inspectors. With the development of UAV and artificial intelligence technology, the power industries have begun to patrol transmission lines on a large scale using UAV running object detection algorithms. However, the existing algorithms are complex, which are not suitable for deployment on embedded devices because of low detection speed and high power consumption. In this paper, we propose a lightweight model to detect cranes, tower cranes, construction machinery, smoke, flame, and foreign matters based on You Only Look Once (YOLO) v5s, which can realize real-time high-precision object detection. Experimental results show that the mean Average Precision (mAP) of our proposed model is 72.23%, and the detection speed is also very fast. It can be concluded that the model is very suitable to run on embedded devices to detect potential safety hazards of transmission lines.

As new rural roads' construction program are being completed in China, the number of traffic accidents that occur on rural roads has increased steadily in recent years. With certain dangerous road sections having a high incidence of accidents, which has become a point that cannot be ignored in the process of rural construction. This paper designs a set of STM32 microcontroller as the core, combined with a variety of sensors and road control facilities of the dangerous road warning system and practical application in rural areas with prominent traffic safety problems, the results show that the designed early warning system can effectively reduce the probability of traffic accidents on dangerous roads.

Objective: To analyze the salary status of printing-related job postings on the 51job recruitment website, and to analyze the main factors affecting the salary level. Method: The method takes the recruitment information obtained from 51job recruitment website as the original data set, selects the work city, working experience, company nature, educational background as the characteristics, and the average salary as the classification variable of the prediction, and constructs the prediction through the Random Forest (RF) classification algorithm. Model, perform one-hot encoding on the two features of ad(work city) and character(company nature) to get the size of the influencing factors of each city and each company nature, and use grid search and cross-validation methods to select the input parameters of the model, use accuracy, confusion matrix and classification report to verify model accuracy, use feature_importance in sklearn to get the importance of model input feature parameters. Conclusion: The constructed RF model has good accuracy, and the accuracy rate is 0.89; working experience; educational background; private and foreign capital in the nature of the company; Shanghai, Dongguan and Shenzhen in the work city are highly important.

A pedestrian detection method based on YoloV5 RGB and thermal image data fusion is proposed in this paper. This method uses two yolov5 branch structures to learn the features of RGB and thermal image data respectively, and finally uses multimodal features for fusion detection. It includes the following stages. Firstly, we use yolov5 network as a branch network to learn features from paired RGB and thermal image data. The two yoloV5-based backbone networks extract the features of the two modes for preliminary fusion, and then obtain the importance parameters of the two modes through feature compression and extraction. The effective information is enhanced, and redundant information is eliminated by multiplying the initial features. Finally, the fusion feature is used for target detection. Through this method, the detection effect is improved. We have done a lot of experiments on the public KAIST and VOT2019 pedestrian data set, and the experiments show that our method is better than the advanced method.

With the enhancement of people's investment awareness, more and more people invest in bitcoin and gold for profit. Therefore, accurate prediction of Bitcoin and gold price movements is very important for investors. In view of this, this paper innovatively proposes a combined model that uses Complementary Ensemble Empirical Mode Decomposition (CEEMD) to reduce noise and separate long-term trends; and uses a Particle Swarm Optimization combined with Long Short-Term Memory (PSO-LSTM) model to predict price changes. Compared with traditional models, PSO-LSTM has better robustness and generalization ability, and can better extract temporal features. In order to verify the validity of the model, this paper selects the bitcoin and gold price data from September 11, 2016, to September 10, 2021, uses the sliding window method to divide the data set, and finally calculates the MSE, RMSE, MAE, DTC of ARIMA, BP, SVM, LSTM and CEEMD-PSO-LSTM models. Eventually found that the CEEMD-PSO-LSTM had the best accuracy and stability.

Online popular restaurants are those that are widely concerned by the society and sought after by the public through we media platform or internet marketing. Online comment is the product of the information age. The daily life of Internet users is to exchange information, express views and communicate with others through major Internet platforms. The outbreak of COVID-19 in 2020 has hit the catering industry in China. According to the statistics of the existing literature, it is found that there are few studies on online popular restaurants, and the research methods are relatively simple and traditional. The research on online comments of online popular restaurants can explore the emotional tendency of consumers, find the problems existing in online popular restaurants and put forward corresponding development suggestions. This paper uses Python technology to obtain the comments of 30 popular restaurants in Dalian on the public comment website, and puts forward corresponding opinions and suggestions on the operation of online popular restaurants through data mining. It is concluded that consumers care about the following aspects in the consumption process: taste, service, decoration style, waiting time in line. In this regard, we put forward the following suggestions: improve the taste of food, constantly push through the old and bring forth the new, and the primary task for the sustainable development of the restaurant is to ensure the taste; Improve service quality and create a high-quality service culture; Create a unique decoration style and resolutely resist and crack down on piracy; Reduce waiting time or provide better service during waiting time.

Education-related search produces a large number of time series data. Effective data mining is of great significance to education decision. A new model was applicated in this paper to predict the parental education anxiety by comparing seven machine learning methods. To ensure the reliability and performance, data were collected from State Statistics Bureau. Results demonstrated that the PCA-LSTM model successfully captures the uncertainty of related policies and events that cause parent education anxiety and has better performance than other results. Our model provides early demand for dynamic real-time analysis and accurate prediction.

This paper presents a novel CNN model called comparison prediction network for apparent age estimation. The algorithm is structured by feature extraction and Face feature database. Compared with the existing methods, our algorithm can better deal with the problem caused by differences between apparent age and actual age, which improves the prediction precision of the model with the increasing credibility and robustness of the model prediction results, enhancing the generalization ability of the model. The algorithm has fewer parameters and is lighter than other methods, which is suitable for mobile deployment.

Printed circuit board (PCB) manufacturing is one of the most important parts of electronic production, where a small defect may cause the final product to fail. Therefore, the industry urgently needs a system to detect and locate all manufacturing defects. In this paper, we propose Generative Adversarial Networks (GANs) based learning defect system with an extremely low bit per pixel (BPP) for feature compression. The system includes an encoder, generator, and multi-scale discriminator for generative learned compression and a comparator to distinguish defected components from a compressed feature map generated by GAN. The model synthesizes images at extreme low bitrates where traditional methods such as JPEG show strong artifacts, resulting in a proportional reduction in the storage of feature comparison. Experimental results demonstrate the effectiveness and efficiency of our model with 97.8% mAP at 72FPS.

Nowadays, more and more users are playing an important role on the Internet. The more comments they post, the more information they contain and the more informative they are. In order to analyze the sentiment orientation of users' comments more accurately, the text Vit-BiGRU-Attention sentiment classification model is based on BiGRU and Attention mechanisms. First, the CBOW model is used to train the word vector. Second, BiGRU is combined with Viterbi algorithm to extract contextual features of the text by combining forward and backward hidden layers. Then, different weights are assigned to words by the attention mechanism to enhance the understanding of emotions and determine the polarity of emotions. Finally, the output is passed through a softmax classifier. The experimental results show that the accuracy of the model has been greatly improved.

Cigarette retailers usually rely on past sales data, intuition and experience to select products. Such selection strategy suffers from some problems, such as low efficiency, single selection index, serious homogenization, and focusing only on immediate interests. As a result, it is unable to respond to market changes sensitively, resulting in a loss of store profits. Under the background of big data, this paper mainly studies how to use machine learning algorithms to make intelligent and efficient store selection of cigarette commodities, by combining internal data of the enterprise and big data of external people, goods and stores. Six stores involving a total of 30 tobacco retailers were taken as the experimental objects, the location of which cover Kunshan, Zhangjiaxiang, Taicang, Changshu and Wujiang districts. The commodity sales in the second and the third quarters were compared. By using the selection system, the customer unit price of tobacco retailers increased by an average of 9% year-on-year, the cigarette profit increased by an average of 5% year-on-year, and the cigarette inventory turnover rate increased by 12% year-on-year. It shows that the model has satisfactory performance. Furthermore, the cigarette selection model can be dynamically updated and optimized. It provides the most suitable and real-time cigarette selection suggestion scheme for retailers and helps the retailers to improve cigarette sales.

The recognition effect of news text sequence data is strongly related to the importance of each word and the dependency relationship between them. Although the capsule network can learn the correlation information between news text as a whole and local, it lacks the attention to the key words in the text and ignores the distant dependencies in the text. To remedy the above shortcomings, this paper proposes a news text classification model which is based on multi-head attention and parallel capsule networks, using a multi-head attention layer for feature extraction and then a parallel capsule network module as the classification layer. The model can retrieve wealthier text details. Experimental results demonstrate that the proposed model of this paper works better than the mainstream capsule network based text classification models in both single-label and multi-label classification tasks of news texts.

In the process of short text classification, the problems of sparse features and fuzzy semantics are often encountered, and the word vectors trained by traditional methods cannot accurately express text features. To enhance the expression effect of text features, this paper introduces part-of-speech features to enhance the semantic expression of features. On this basis, an ACBiGRU classification model based on enhanced word vectors is proposed. The model takes the enhanced word vector as input, uses the CNN network to extract the features initially and highlight the local features of the text, a then uses the BiGRU network to learn and combined with the attention mechanism to highlight key information.

There is a high correlation between user behavior and user features in recommender systems. User review texts reflect user preferences and item feature information. However, the current research on CTR prediction models based on user behaviors fails to fully mine user features. As a result, the prediction accuracy of the model is not high. To solve this problem, we propose a click-through rate prediction model that fuses user comment text and behavior sequence. The model uses a text convolutional neural network to extract the features of user review text to obtain the feature vector of user comment text, and uses an attention mechanism to capture the user's interest points from the user's behavior sequence to obtain the user's interest feature vector. A multi-layer perceptron is then used to fuse the user's comment text feature vector, interest feature vector and item feature vector for click-through rate prediction. The experimental results show that the proposed model has better prediction performance than current click-through rate prediction models.

The semantic segmentation model needs to output a category for each pixel, which is an intensive prediction task, so it needs more time. Real-world scenarios, such as autopilot, require fast reasoning speeds. Previous work has been to speed up model reasoning in one way or another, so the speed increase is limited. In this paper, we combine multi-level methods to speed up reasoning in a network and obtain remarkable results.

News and publishing industry had encountered dramatically changes in the last two decades. The traditional companies in the industry are facing challenges in the so-called new media era from two aspects. On the one hand, technology companies consistently occupy their territories, on the other hand, the traditional publication lifecycle suffers a long supply chain management and can hardly cope with the fast-changing internet realities. In order to deal with these problems, this paper proposes an intelligent publishing model. The model integrates modern information technologies into the publishing workflow in which artificial intelligence plays an important role. By machine learning technologies, the vast amount of intelligent properties accumulated by publishing houses can be converted to machine understandable knowledge and further to enhance users' experiences. The proposed model refines the publishing process and helps the industry entities facing the process of incoming media convergence.

Cloud computing, as a current research hotspot in the field of information, is a new storage method combining distributed database technology and web server technology, which achieves specific goals mainly through the integration of a large number of resources in a virtual environment. In this paper, we explore the application of cloud computing to computer data processing and design a data processing system. User requests are first returned to the corresponding server via a network connection, and then algorithms are used to derive the relevant information, which is fed back to the next interactive interface for users to browse, query and compare. The system demonstrates that cloud computing technology can ensure effective data processing after initialisation and can improve the response time to user requests in some way.

To deal with the problem of low accuracy of sports performance prediction, and to obtain ideal sports performance prediction results, this paper proposes a sports performance prediction model based on the selection of influencing factors and support vector machine. In this model, particle swarm optimization is introduced to determine the most relevant influencing factors related to the change characteristics of sports performance, which reduces the number of input vectors of sports performance prediction model and speeds up the modelling speed of sports performance. Then, the support vector machine is used to learn the historical data of sports performance, which overcomes the defects of traditional models such as artificial neural network and improves the prediction accuracy of sports performance. Experimental results are provided to verify the advantage of the proposed algorithm with respect to the traditional methods.

Privacy protection in shared data stream publishing witness its thriving in recent years. Different privacy preserving methods commonly provide different protection effects. Unified measurement of protection intensity is the basis of privacy evaluation. Existing privacy measurement methods depend too much on background knowledge, and the measurement effect is strongly related to privacy information. They in common fall short in privacy measurement in privacy-preserving stream histogram publishing. Concerning these issues, a Bayesian-theorem based privacy measurement model is proposed for privacy-preserving stream histogram publishing. By analyzing the correlation degree between background knowledge and sliding window, the correlation between background knowledge and publishing results is established. The concept of correlation histogram is introduced to analyze the correlation between histograms containing the same user's state information, and a sliding window-oriented privacy leakage measurement mechanism of correlation histograms is proposed to measure the degree of privacy leakage between correlative histograms. Further, corresponding weights are set for both error probability of attacks and privacy leakage extent, to realize privacy measurement of privacy-preserving histogram publishing algorithm. Theoretical analysis and experimental results show that our solution can effectively measure the protection strength of streaming histogram privacy protection method.

To control the spread of the virus, mask detection is crucial in public areas, especially after the outbreak of Covid-19 pneumonia. This paper aims to improve the accuracy and precision of mask detection. This study improves mask-wearing detection by adding data augmentation, using the smooth label to replace the one-hot vector, and customizing the network connection of the YOLOv3 network. Through these targeted improvements, the average precision of face with mask detection has been increased by 0.9%, and the average precision of face without mask detection has been increased by 2.9%, which implies that it is a better strategy to do mask detection based on YOLOv3. By inputting photographs, the network can check, with high accuracy, whether the pedestrians in the picture wear masks or not, which will be a good supplementary to epidemic prevention and control.

This paper, originally published on 30 November 2022, was retracted from the SPIE Digital Library on 11 January 2023 by the publisher and in agreement with the authors, upon verification that a substantial portion of the paper was copied from the following work without attribution or permission: “Design of new energy vehicle operation monitoring system based on convolutional neural network” SiZhou Du1, YuBo Wang2, 1. National Technical University, Kharkiv Polytechnic Institute (Ukraine) 2. Beijing Jiaotong University (China), published on 10 November 2022.

CNN (convolutional neural network) is a classical research method of Deep Learning. It can obtain the characteristics of a picture through the information transmission between convolution layers and pooling layers, and generate some output (such as image classification) after final processing. Since the end of the 20th century, scholars have proposed various convolutional neural networks, which have their own characteristics. In this paper, we choose LeNet, AlexNet, VGG, ResNet and GoogLeNet to complete the cat and dog recognition task on kaggle, so as to identify and explore the performance of different networks in different situations. The results show that these classical algorithms have generally become more and more advanced over time, but this cognition is not completely correct. For example, when the number of samples is limited, the more advanced ResNet did not perform as well as the relatively primitive VGG network. Such characteristics help us choose the right algorithm in different situations and guide us refine the algorithm in the future.

Rectal cancer is the most common malignant tumor of the digestive tract, and its incidence ranks third among all types of tumors. To investigate the differences in lipid metabolism between patients with rectal malignancies and healthy individuals, this study used the LIS information system to obtain a large amount of blood test data and used statistical software to perform correlation analysis. The data of the experimental group in this study came from the detection information of 502 patients with rectal cancer extracted by LIS. The data of the control group were obtained from the detection information of 1251 physical examiners. Human lipid metabolism is mainly expressed by indicators such as high-density lipoprotein, triglyceride, total cholesterol, low-density lipoprotein, apolipoprotein AI and apolipoprotein B. This study found that the blood lipid index of healthy people was significantly higher than that of patients with rectal cancer. This study uses the medical information data extracted by LIS as the experimental object. Through comparative analysis, when the human body develops rectal cancer lesions, the lipid metabolism level of the body will decrease significantly. The evaluation of blood lipid metabolism can provide a theoretical reference for improving the quality of life and prognosis of patients with rectal cancer.

In this paper, the most popular micro channel small program is combined, based on "off-peak parking", using Hyperledger Fabric, Spring Boot and other advanced technologies to solve the problem of "difficult parking", so as to achieve a new shared parking mode. Based on the current private cars and private car occupancy rate data in China, this paper conducted a survey and questionnaire survey on "private car parking tendency of citizens" in Caidian District of Wuhan city, sorted out and analyzed the survey data, analyzed the demand, and designed and implemented a 5G shared parking system. The main function modules of this system include user login, user registration, parking information release, order submission, navigation management, payment order and so on. Through the testing of wechat application software for shared parking, it is verified that this application software can well meet the needs of users for parking space sharing.

Under the situation of deepening the reform of the electric power system and promoting the reform of transmission and distribution prices, the addition of microgrids and multiple loads has influenced the operational and economic benefits of the distribution network. How to evaluate the benefits of a distribution network is of great significance for guiding the subsequent planning and operation of the distribution network. The key to constructing an index system is the selection of an evaluation index and comprehensive evaluation method. This paper studies the evaluation index and complete evaluation method of distribution network benefit and uses the fuzzy analysis method to make entropy change weight for evaluation index weight based on scene data. Through example analysis, this method can effectively avoid the situation where individual indexes of the distribution network are low but comprehensive scores are high. It requires all indexes of the distribution network to be balanced, and there are no poor indexes, which can make the evaluation results closer to the actual situation. This further shows that this evaluation system can effectively evaluate the comprehensive benefits of a distribution network considering microgrid and multiple loads and guide the subsequent economic operation and planning of the distribution network.

While deep neural network technology brings high recognition accuracy to the field of synthetic aperture radar automatic target recognition, it also produces the problem of catastrophic forgetting. Currently, how to extract features for distinguishing new and old classes has become the main bottleneck for incremental learning performance improvement. In this paper, we propose a new incremental learning method to better distinguish between new and old classes. We use the trained neural network to extract the features of the old samples and utilize the k -means to select representative old samples in the feature space, and then train the new model with distillation loss. Through the experiments on the MSTAR dataset, our method has better incremental learning performance on SAR images under the same training time.

This article uses CiteSpace 5.8 R3 analysis tool, taking 749 Chinese teaching and learning papers retrieved from Web of science collection database in recent 5years (2018-2022) as the data source, to systematically analyze the following seven aspects: chronological distribution statistics, visual analysis of authors, top ten countries and regions, research institutions, fund support statistics, keywords analysis, and research hotspot analysis. The research results are helpful to grasp the development and research hotspot of Chinese teaching and learning.

This paper deals with the problem of recognizing associated words in compound sentences. As the conventional methods rely on manually extracted utterance features, in this paper, an associated words recognition method based on neural networks is proposed. The proposed method fuses the features extracted from the compound sentence corpus into word vectors, which are fed into the constructed deep neural network model for training. To analyze the compound sentences in the modern Chinese compound sentence corpus, we extract four common utterance features to establish an utterance feature base. Then, the extract features from the utterance feature base are combined into a new feature, which possesses high discriminative ability. Based on the utterance feature base, a training set and test set is constructed for test the performance of the proposed method. Experimental results show that the proposed method both improves the efficiency of recognition and achieves a high correct rate with respect to the traditional method.

User portraits were originally used in the field of e-commerce, which refers to the use of big data to analyze the information and behavioral characteristics of existing users. They are to explore user preferences and needs through user information to achieve precise marketing. Power-user portrait is an important measure for a power supply service system to realize "data-driven operation". This paper starts with user characteristics and constructs an improved word vector model for electronic users to describe the behavior characteristics of electronic user groups to bring about correct marketing methods and services. The experimental results show that the improved word vector model can better meet the aggregation needs of data collection such as electricity cost, electricity cost and power loss of users.

Because the characteristic selection and promotion performance of the traditional transmission network is not strong, the access accuracy of the network is very low. Therefore, a research on the network segment optimization scheme based on 5G access transmission bearer network is proposed. According to the index normalization, index weight and importance determination, it'll help construct and make an evaluation of the importance of the network segment of the transmission bearer network, find the link risk and node risk of the bearer network segment, and optimize the functions of multiple objectives. This paper further expands the analysis to the scenario where the 5G network is connected to the transmission bearer network, and describes the performance of the network. The analysis results in this paper are of great significance for the 5G network design that needs to optimize the transmission bearer network segment.

In order to overcome the influence of subjective factors such as rater effect and improve the speed and accuracy of largescale composition scoring, the research on automated essay scoring has increasingly become a hot topic in language assessment. At the same time, the characteristics of oversea Chinese test make it different from Hanyu Shuiping Kaoshi (HSK). With the help of the word list and vocabulary for oversea Chinese, we trained and test four models. By comprehensively evaluating the performance of each grade, we can find that convolutional neural network (CNN) has the best prediction effect on compositions, and can be used as an auxiliary AES to provide reference value for oversea Chinese teaching.

As the friendliness of human-computer interaction has become a key ambition of handheld terminal application design, by adding an appropriate amount of cache,a fast display method of Chinese vector characters utilizing GPU's hard acceleration capability is presented for the demand of Chinese display on handheld terminal devices.The comprehensive test demonstrates that the display speed of application interface and the page refresh rate can be sped up by 30-45% and 40-60% respectively.

Named entity recognition (NER) is one of the fundamental technologies in natural language processing. NER task extracts the entities such as location, person and organization from unstructured texts, which plays an important role in machine translation and other tasks. However, lacking of effective solutions of nested named entities to fuzzy boundary makes Chinese named entity recognition much harder than other languages. In order to solve the challenges brought by Chinese nested named entities, a segment labeling method for Chinese named nested entity recognition is proposed in this paper. Firstly, every entity is treated as several non-separable units to improve the labeling method. Secondly, each unit will be tagged differently according to its grammar and function in a particular sentence. Finally, every main entity in the recognition sequence is considered as a starting point, and merges each sub-entity into a complete nested entity after a candidate test. To verify the effectiveness of the method, a Chinese military corpus is collected to test the performance of this proposed method. The accuracy value, recall value and F1 value of this method are higher than those of the ordinary methods, which shows that this method can effectively improve the performance of Chinese named entity recognition.

At present, people pay less attention to the diversity of the generated results about the generation of handwritten Chinese characters. The structure of Chinese characters is complex and the process of handwriting has strong freedom. So the image of handwritten Chinese characters has a certain diversity. In this paper, a handwritten Chinese character generation adversarial network is proposed. By adding the standard font information and feature vector into the network, the generation results of the network can not only achieve certain accuracy but also produce diversity. Improved loss function makes the network more inclined to generate a diversity of results. The method using the connected domain segmentation is used to better ensure the accuracy of the generated image. By training on the handwritten Chinese character dataset, it is verified that the network shows good diversity when the accuracy is similar to that of other methods.

This paper aims to construct a set of evaluation index system of in-hospital preparation of Chinese medicine in medical institutions in Jilin Province based on the hierarchical analysis method, and to provide reference for relevant departments to formulate policies related to in-hospital preparation of Chinese medicine. The indexes of the evaluation index system of in-hospital preparation of Chinese medicine in medical institutions in Jilin Province were determined by literature research method and expert consultation method, and the index weights were calculated by using hierarchical analysis method. The results showed that this paper constructed the evaluation index system of in-hospital preparation of Chinese medicine in medical institutions in Jilin Province and determined the weights of each index, including 6 primary and 25 secondary indexes. 6 primary indexes were hospital staffing, hospital management, review policy, scientific research capability, preparation room management and clinical demand. This study constructed a set of scientific, applicable and easy-to-operate evaluation index system of in-hospital preparation of Chinese medicine in medical institutions in Jilin Province, and reasonable allocation of hospital staff and change of review policy concept are feasible options to promote the development of in-hospital preparation of Chinese medicine in medical institutions in Jilin Province.

To understand the temporal and spatial variation of extreme temperature in mainland China and predict the high temperature change in the future, this thesis analyzes the temporal variation characteristics of high temperature in China by using the daily maximum temperature observation data of China from 1961 to 2019 and the daily maximum temperature data of GFDL-ESM4 model from 2016 to 2074 in CMIP6 scenario comparison plan. It is shown in the results that: 1. From 1961 to 2020, the maximum high temperature intensity and high temperature days were in 1967 (839.53 ℃, 23 days), the highest average annual high temperature intensity was 925.58 ℃, 343 ℃ more than normal, and the maximum average annual high temperature days were 26 days, 10 days more than normal. 2. The high temperature intensity is the highest in July, the high temperature intensity is higher from May to September, and the high temperature intensity is lower from October to April of the following year. 3. The increase rate of high temperature intensity is the largest under the ssp585 emission scenario during 2015-2074, while the increase rate of the ssp119 emission scenario is the smallest, and the highest temperature is mostly located in 2072.

In the contemporary society, forest tourism is one of the most essential types of tourism, which is in line with modern people's pursuit of green health and a fresh natural lifestyle. Forest tourism can bring great economic benefits. With the aim to promote the sustainable development of forest tourism in Henan Province, this paper uses factor analysis method to classify the selected 9 indicators into 3 common factors and evaluates the competitive advantages and disadvantages of Henan Province through the common factor score and comprehensive score, based on the actual situation of forest tourism resources in 31 provinces. The study found that Henan Province currently ranks 18th in the tourism economic competitiveness of forest parks; ranks 14th in tourism management; ranked 31st in the competitiveness of investment in forest parks; ranks 19th in terms of comprehensive competition.

Named entity recognition (NER) on electronic medical records (EMRs) is a prerequisite task of medical information extraction. However, due to the high data sensitivity and the labeling difficulty of EMRs, there are few available data resources, making it difficult for NER to reach a practical level on EMRs. To alleviate the problem of low-resource, we propose a label sharing-based cross-corpus NER (LSCC-NER) model, which consists of the shared and private encoders divided by the cross-corpus label similarity. We design a category-wise multi-head self-attention unit for each encoder and introduce the entity category prediction task (ECP) to realize the division. Besides, considering the nested entities and data noise in EMRs, we utilize span-based decoding methods and adversarial training to further improve the robustness of our model. The experiments on two evaluation datasets of EMRs show that our proposed LSCC-NER model can achieve higher recognition performance compared with common transfer learning methods.

In the context of the combination of the Internet of Things and the smart grid, how to build a cost-effective ubiquitous Internet of Things needs to solve the shortcomings of cloud computing and big data information interaction in complex network topology, and excessive reliance on manual data annotation. Therefore, it is necessary to incorporate the complex network topology into the ubiquitous Internet of Things construction process. In this paper, through the analysis of the power Internet of Things system structure design, the real-time interaction of data is completed through the access of gateway devices, and a ubiquitous Internet of Things based on Modules/TCP is proposed. The DCS gateway and the security-hardened DSC gateway information transmission mechanism, design a method for real-time interaction between the perception layer and the network layer of a given power network, and provide certain ideas for the construction of the ubiquitous power Internet of Things.

With the continuous improvement of manufacturing industry requirements for product quality and efficiency, given the separation of part design and development and manufacturing, this paper explores the manufacturability analysis system based on process knowledge, and classifies process knowledge into standardized knowledge, structural process knowledge and manufacturing constraint knowledge, and adopts object-oriented process knowledge representation. The implementation method and process of process review of parts by the manufacturable analysis system are described. The two manufacturability evaluation strategies based on rules/planning were studied and analyzed, the process review of the 3D model was carried out experimentally, and the system's feasibility was verified. It improves parts' design quality and efficiency and is of great significance to the design and development of new products in the future manufacturing industry.

In recent years, deep learning has developed very fast. Since the introduction of deep learning, good research results have been achieved in the field of computer vision. In the era of artificial intelligence, computer vision technology has been widely used in fields such as unmanned driving and security monitoring. The birth of these technologies is based on image segmentation. Image segmentation algorithms based on deep learning are constantly being proposed, which have made a qualitative leap in performance and effect compared with traditional image segmentation algorithms, but there is still a lot of room for improvement. This paper improves on the U-net, a variant of the classic fully convolutional neural network. Combining U-net with atrous convolution, a new network model for image segmentation is proposed. In order to verify the actual segmentation effect of the new network model, this paper is tested on the public image segmentation data set, and compared with other classic image segmentation algorithms. The experimental results prove that the network model has a good segmentation effect.

The hybrid domain attention model was introduced into deep learning network and has good performance in computer vision recognition and classification tasks for considering both spatial and channel domain information, which leads to higher network complicity, more parameters and much training time. In this paper, an Efficient Hybrid Attention(EHA) model was proposed. The spatial and channel domain information was extracted and fused into the EHA-block, that can be embedded in deep network Resnet50 and lightweight mobile network MobileNetV2 to improve their capabilities with low parameter- increment. The experimental results on CIFAR10&100 and miniImageNet show that the number of parameters in EHA model decreases by 15.4% compared with CA hybrid model, and only increases by 0.49% compared with ECA channel model. Moreover, the accuracy of EHA classification is also improved.

Because the vibration signals of cable reflect the dynamic characteristics of the cable-stayed cable, it is vital to identify and analyze the vibration signals of cable in bridge health monitoring. Currently, there is little research on the vibration signals of stay cables using pattern recognition algorithms. In this work, we proposed a pattern recognition algorithm combining empirical mode decomposition (EMD) with binary image and modified LeNet-5 to analyze cable vibration signals. Experimental results show that this method has good anti-noise performance and is effective for vehicle load identification.

Detecting image splicing has become essential to fight against malicious forgery. To solve the problems of low accuracy of authenticity classification and low detection efficiency in some image splicing detection methods proposed in recent years, we propose an image splicing forgery detection method based on modified SSD network (ISD-SSD). In this method, the residual network model ResNet-50 is used to replace the feature extraction backbone network VGG-16 in SSD, which solves the problem of degradation caused by the deepening of the network and enhances the feature extraction ability of the model. In the multi-scale detection part, a multi-scale feature fusion module is introduced based on the feature pyramid idea, which organically combines the low-level visual features and high-level semantic features in the network structure. Finally, Focal loss is selected as the loss function in the loss calculation part to solve the problem of positive and negative samples and the imbalance of difficult and easy samples. Experiments on standard image manipulation datasets demonstrate that our ISD-SSD algorithm is superior to the existing image tamper detection algorithms (such as MFCN, Faster R-CNN algorithm, etc.), the evaluation metrics AP and F1 reach 77.86% and 75.18% respectively. In addition, it shows robustness in terms of detection speed.

As the current mainstream social networking method, online social networking brings convenience to people, but also generates negative problems such as language violence. Therefore, offensive text style transfer research has become an arduous task. Research has shown that the semantic loss is caused by the discreteness and timing of the text data, resulting in a low degree of content retention after style transferring, and the lack of relevant parallel corpus data and specific style tag keywords, therefore, the accuracy of text style transfer is limited. Aiming at the existing problems, this paper proposes an Offensive Text Style Transfer based on the Unsupervised Learning (OTST-UL) model. First, the input text is encoded through the bidirectional encoding attention mechanism to retain the core content of the text. Then the generatordiscriminator is applied for adversarial training, and a reconstruction loss algorithm is constructed to ensure the accuracy of the offensive language style transformation and the integrity of the text content. Experimental results show that the OTST-UL model outperforms existing text style transfer models on offensive language datasets.

Domain texts usually have significant domain features and long text lengths. Text classification models suitable for general fields cannot well meet the task of text classification in specific domains. Therefore, this paper proposes a domain text classification method based on BERT. First, segment the long domain text to obtain the sequence combination, and input it into the BERT pre-training model to obtain the word vector. Then the vector is compressed and encoded to obtain the pooled sequence feature vector. Finally, it is sequentially input to the Encoder layer for domain feature extraction, and the text is divided into various categories in the output layer. The experimental results show that the proposed BERT_VCA model has an average improvement of 1.12% in F1 value compared with the BERT_BASE model in the domain text classification task.

Compared with traditional methods, the quality of reconstructed images has been greatly improved by super-resolution image reconstruction methods based on deep residual networks. However, in order to make full use of shallow image features and solve the problem of instability during network training. This paper proposes a method for image super-resolution reconstruction that can better reconstruct texture details based on generative adversarial network (GAN). Dense blocks containing residual scaling (RRDBs) is used to construct the generative network to extract more image features. WGAN is introduced to construct the discriminative network to solve the problem of unstable training of generative adversarial network. The experimental results show that the proposed model in this paper has better visual effects and improved SSIM and PSNR values compared with other models.

Short news text classification plays an import role in natural language processing as the popularity of mobile phones. In this paper we propose a Chinese short news text classification method based on BERT and sparse autoencoder, regarding the overfitting caused by pretrained BERT. We use the BERT for text representation, the output vectors of BERT are dimension reduced through the sparse autoencoder, and then the Softmax classifier takes the reduced vectors as input to get the prediction of the input text. Experimental results show that our method mitigate the unbalance of the performance of different categories, raises the overall classification performance by six percentage, effectively alleviates the overfitting of text representation of BERT, and achieve a better Chinese short text classification performance than using naïve autoencoder and without autoencoder.

The image features are directly input into the decoding part of the model, which leads to the insufficient use of feature information and makes it difficult for the model to better express the image information. We introduce a Modular Co- Attention Transformer Layer (M-CATL) to efficiently model high-order intra-feature and inter-feature interactions for single and multiple input features to mine the details of image features. And construct a Deep Modular Co-Attention Transformer Block (DM-CATB) according to M-CATL and integrated into the encoder part of the model. Furthermore, we present a Deep Modular Co-Attention Transformer Network (DM-CATN) to fully model the spatial information and position information of image features and improve the ability of features characterization, in order to provide richer image information for decoding part. Experimental results demonstrate that DM-CATN significantly outperforms the previous state-of-the-art. Our best single model delivers 133.2% in CIDEr.

Aiming at the reasonable planning and effective utilization of massive multi-source heterogeneous quality data resources generated in the process of small-batch manufacturing, this paper proposes a quality data resource management platform with agility and dynamic integration. With the goal of business generalization and reuse, the typical quality business resources are summarized to form each data business service module of the platform. We propose a whole-process management scheme for general document management, data cleaning, and data analysis in quality data resource management. And based on the idea of componentization, separation and condensation are carried out. Relying on the enterprise's platform application practice, this paper verifies the feasibility and practicability of the proposed quality data resource management platform design through application examples.

With the wide application of knowledge graph such as recommendation system and text analysis, it is particularly important to create high-quality knowledge graph, it requires precise knowledge graph fusion. As a key part of knowledge graph fusion, entity alignment can provide more prior knowledge for knowledge graph and improve its usability. In order to obtain a more global graph structure feature, this paper designs a subgraph matching based method for entity alignment, named SMGNN. It based on the two features of map structure information and local relation semantics, and captures relationships between entities through GNN capture. Firstly, the entity is encoded by the subgraph information of the target node through the GNN based on two entity-aligned knowledge graphs. Secondly, the subgraph is the graph composed of the target node and all neighboring nodes connected to the node. Then, the alignment between the two graphs is regarded as a mapping on the hyperplane, and TransH model is used for alignment. Finally, we do experiments on DBP15K, a crosslanguage entity alignment dataset, the results show that SMGNN can effectively improve the alignment accuracy of knowledge graphs.

Patients with the loss of limb motor function often suffer from neurological damage of brain in the presence of normal limb neural pathways, resulting in reduced or lost limb movement. By observing the mirror image of the healthy limb movement, the relevant motor perception areas of the patient's brain can be stimulated and the motor and brain functions of the affected limb could also be remodeled. Based on the above theoretical support, this study designed a visual feedback upper limb rehabilitation system based on an attentional neural network, which analyzed and modelled the electromyography (EMG) of the upper limb through a new attentional neural network named MCAT-net. The proposed MCAT-net could determine the movement intention of the healthy upper limb with an accuracy about 91.04%. Further, based on MCAT-net, a visual feedback upper limb rehabilitation system is designed, which could receive and generate the augmented reality models of the affected upper limb for the patients who need the purpose of visual rehabilitating stimulation.

Thanks to the rapid development of Generative Adversarial Networks (GAN) in recent years, great progress has been made in image translation using GAN. Image-to-image transformation is one of the important applications in the field of computer vision, and its scope includes image inpainting, image colorization, super-resolution and image style transfer. In recent years, there are many classic research on GAN-based image transformation, such as CycleGAN, UNIT, AGGAN, etc. This paper focuses on researching and testing the U-GAT-IT unsupervised image style transfer method. The authors introduced a new attention module and a new learnable normalization function (AdaLIN), which enables flexible control of the amount of change in shape and texture during image conversion. This paper uses a new dataset for testing and verification, trying to achieve bidirectional conversion between real face photos and sketches, and qualitatively and quantitatively analyzes the method by calculating PSNR and SSIM.

This paper studies the Kazakh language in Kazakhstan (hereinafter referred to as the Kazakh language). This paper takes 14 Kazakh students as experimental objects and uses Mini-speech Lab as a research tool to investigate the first-level vowel pattern and positioning characteristics of Kazakh language. The study draws a map of the Kazakh first-level vowel pattern, which can enrich the theoretical achievements of Kazakh phonetics research at home and abroad, and provide high guiding value for Kazakh phonetics teaching.

As a challenging problem in the field of image analysis and computer vision, face recognition has garnered a lot of attention in recent years due to its many applications across a wide range of areas. In recent years, face recognition technology has been focused on humans, but considering the expansion and improvements of the cat industry, face recognition of cats should also receive some attention. This paper proposes a face recognition system for cats using Siamese and VGG16. We will send two images of cat faces to the Siamese network, which will be converted to a vector by mapping their features to the space and then get their probability of similarity by calculating their losses. Our training model on 13,106 cat images from a dataset of 509 different cats shows that our method can recognize cats’ faces with an accuracy of 72.91% on a test dataset with 1702 image pairs containing 851 pairs labelled true, and 851 pairs labelled false. Experiments have demonstrated that this method is convenient and contactless and has a high recognition rate.

Human pose estimation is a technique for estimating human pose from input information obtained from a sensor. It has been widely learned in the field of computer vision. And the human body information provided by human pose estimation has been widely used in many applications and it helps a lot with artificial intelligence. With the development of deep learning, human pose estimation has made significant progress and achieved significant performance using deep learning techniques. But there still are some problems such as occlusion, insufficient training data, etc. are still challenges that need to be handeled. Because of the rapid development of human pose estimation and the gradually emerging problems, this article tries to refine three types of problems in human pose estimation and summarize some teams’ latest methods to deal with these problems.

Wildlife recognition is the task of matching and outputting similar probabilities given two images, with twin networks being a popular framework in the field. In this work, we use the Siamese architecture as the main framework. As for the backbone, we compare two popular backbone networks, including VGGNet and ResNet, respectively. We compare those two networks with respect to the convergence speed, robustness and maximum accuracy to exploit the effectiveness of our method. We test those networks on AFHQ dataset. Our method with ResNet achieves 98.86% accuracy. Our experiments have confirmed that, to a certain extent, the larger the number of layers in the network, the higher the stability and accuracy of model training.

According to the specific situation of the designated area, the appropriate search route is formulated to complete the comprehensive search of the predetermined area in the shortest time. In this model, the search problem is transformed into an approximate Hamiltonian problem, that is, each grid center is regarded as a point in the graph, and a search route with the shortest total distance and as balanced as possible for everyone is designed. The optimization principle is used to solve the ground search problem intuitively, and the optimal search time is obtained.