Table of contents

The 2021 International Conference on Artificial Intelligence and Industrial Technology Applications (AIITA 2021) was planned to be held on June 18-20, 2021 in Nanchang, China. Due to the impact of COVID-19, many communities from all over the world were under strict health measures and strict travel restrictions, and participants of 2021 International Conference on Artificial Intelligence and Industrial Technology Applications (AIITA 2021) also meet with the difficulty of travel restrictions. For communities, if our conference was not held in a virtual form, it would add huge work on traffic system, Exit-Entry system, medical system, etc., which is quite a serious issue, to actively respond to the call of the government, to strengthen the protection work, to effectively reduce people gathering and prevent COVID-19, considering the situation that most of the authors would like to publish their articles and make academic communications as scheduled, AIITA 2021 was held online instead of postponing the conference. It was a challenge, not only to the organizers, but also to the participants who were delivering their speeches using videoconference tools, under different time zones.

AIITA 2021 was hosted by Jiangxi Province Robot Industry Alliance, organized by AEIC Academic Exchange Information Center. AIITA 2021 is to bring together innovative academics and industrial experts in the field of Artificial Intelligence and Industrial Technology Applications to a common forum. The primary goal of the conference is to promote research and developmental activities in Artificial Intelligence and Industrial Technology Applications and another goal is to promote scientific information interchange between researchers, developers, engineers, students, and practitioners working all around the world.

The conference brings together about 80 leading researchers, engineers and scientists in the domain of interest from different countries. The conference model was divided into two parts, including keynote presentations and online discussion. In the first part, each keynote speakers were allocated 30 minutes to present their talks via Zoom. After the keynote talks, all participants joined in a WeChat communication group to discuss more about the talks and presentations.

We were greatly honor to have invited two distinguished experts as our keynote speakers. Prof. Philippe Fournier-Viger, from Harbin Institute of Technology, China was the first one to perform a thought-provoking speech. His Research interests: Data Mining, Big Data, Artificial Intelligence, Pattern Mining, Itemset Mining, Graph Mining, Sequence Prediction. Another keynote speaker, Assoc. Prof. Md. Shohel Sayeed, from Multimedia University, Malaysia. He had outstanding research in Biometrics, Image and Signal Processing, Data Mining, Artificial Intelligence, Cloud Computing and soft computing. Their insightful speeches had triggered heated discussion in the second session of the conference. The WeChat discussion lasted for about 30 minutes. Every participant praised this conference for disseminating useful and insightful knowledge.

We were glad to share with you that we received lots of submissions from the conference and we selected a bunch of high-quality papers and compiled them into the proceedings after rigorously reviewed them. These papers feature following topics but are not limited to: Artificial intelligence, Intelligent manufacturing, Automation and other relevant directions. All the papers have been through rigorous review and process to meet the requirements of International publication standard.

We would like to thank the organization staff, the members of the program committees and reviewers. They have worked very hard in reviewing papers and making valuable suggestions for the authors to improve their work. We also would like to express our gratitude to the external reviewers, for providing extra help in the review process, and the authors for contributing their research result to the conference.

Lastly, we would like to warmly thank all the authors who, with their presentations and papers, generously contributed to the lively exchange of scientific information that is so vital to the endurance of scientific conferences of this kind.

Committee of AIITA 2021

List of Committee member are available in the pdf

All papers published in this volume of Journal of Physics: Conference Series have been peer reviewed through processes administered by the Editors. Reviews were conducted by expert referees to the professional and scientific standards expected of a proceedings journal published by IOP Publishing.

Type of peer review: Double-blind

• Conference submission management system: AI Scholar System

• Number of submissions received: 89

• Number of submissions sent for review: 85

• Number of submissions accepted: 54

• Acceptance Rate (Number of Submissions Accepted / Number of Submissions Received X 100): 60.7%

• Average number of reviews per paper: 2

• Total number of reviewers involved: 40

• Any additional info on review process:

• Step 1. Each of selected paper will be reviewed by two/three professional experts in the related subject area.

• Step 2. Review Reports received from the experts will be judged by one of the editors either Review Reports are logical or not?

• Step 3. If not logical, then editor can assign new reviewer or can also judge at his/her own.

• Step 4. If logical, then Review Reports will be sent to authors to modify the manuscript accordingly.

• Step 5. Authors will be required to revise their papers according to the points raised.

• Step 6. Revised version will then be evaluated by the editor for the incorporation of the points raised by the reviewers.

• Step 7. Then the editor will send the revised manuscript to the reviewers again for re-evaluation.

• Step 8. If the reviewers approve the revise version of the manuscript, then will be accepted for publication.

• Contact person for queries:

Xuexia Ye

publication@keoaeic.org

AEIC Academic Exchange Information Centre

In order to make the exoskeleton with better wearing effect, it is necessary to improve the exoskeleton motion intention recognition. In this paper, thin-film pressure sensors and inertial sensors are used to acquire human motion information on a wire-driven exoskeleton robot. The lower limb exoskeleton uses DTW (Dynamic Time Wraping) algorithm to recognize four kinds of human motion patterns by comparing the collected data with database features. The result shows that DTW algorithm can effectively recognize human motion patterns.

Due to the diversity of garbage types in our daily life, we will encounter many difficulties in the process of classification. In this regard, I combine hog features and boosting algorithm to develop a SVM classification method. Firstly, the input image is preprocessed to make the image more recognizable. Secondly, the hog algorithm is used to extract the features of the image. Finally, the classification device is trained, and the relevant information is sent to the image set. On this basis, the classification situation is detected. The final results show that the classification efficiency of the algorithm is as high as 95% or even more, which is about 10% higher than that of single SVM classification method. It can accurately classify garbage and has certain feasibility.

To solve the problem of inadequate expression of action behavior features, this paper proposes an action recognition method based on attention mechanism. Firstly, in the feature extraction part, a CSE module is designed to model action features spatio-temporally, and then this module is incorporated into the residual network to improve the feature extraction ability of the model; after that, the LSTM network is used to solve the problem of temporal association of features; finally, the actions are classified by Softmax. The experimental results show that the improved recognition rates of this method on UCF101, HMDB51 and Kinetics400 datasets are 96.23%, 92.03% and 75.65%, respectively.

The gait of Amur tiger was studied through video images, and a more accurate individual recognition system of Wild Amur tiger was given on the premise of supplementing pattern recognition technology. Through further research on the common gait of Amur tiger, the basic information database was established to realize the physiological state research of Amur tiger. The characteristic structure and movement standard model of Amur tiger were constructed to complete the simulation and reconstruction of Amur tiger's routine movement. Through the simulation corridor of tiger movement trajectory, the ecological protection area can be divided effectively.

Semantic Role Labeling (SRL) is a shallow semantic analysis in the field of NLP, and a relatively basic and important step. Traditionally, SRL has been performed based on the results of syntactic analysis and has problems such as over-reliance on feature engineering. With the development of deep learning, many neural network models for NLP have been proposed and SRL tasks can be performed well by neural networks. Among these, long short-term memory networks form a very good fit with the SRL task by virtue of their sequence-based features. In this paper, in order, we first analyze the SRL task based on grammatical analysis and neural networks, then the SRL task based on LSTM and its improved models, then the dataset and model evaluation of the SRL task, and finally conclude and look forward.

The tracking accuracy of traditional algorithm decreases when the target is in a complex environment or the target itself changes, and an improved correlation filtering twin network algorithm combining the attention mechanism is proposed. Fast wavelet is used for filtering and multi-scale feature signal extraction, and loss function is used to minimize the squared error between sample and label. Then correlation filtering algorithm is used to calculate the similarity of image between target template and candidate region. Attention mechanism is introduced to improve correlation filtering twin network structure, and the GloU's online learning is advanced as an aid to determine template updating. The experimental results show that the algorithm have certain improvement effect for target shade, quick movement and deformation problems, so performance and accuracy of target tracking, the algorithm has better competitiveness.

Deep learning related to computer vision, speech recognition, and language processing has been developing rapidly over recent years. The applications of these models, however, have underlying risks. Recent studies have shown that small perturbation from adversarial examples could result in false interpretation of the neural network examples and false judgment. Therefore, understanding adversarial example technologies is essential for promoting the safety and robustness of neural network models. This paper summarizes current adversarial example technologies in different applications, discusses the current prospects and challenges, and envision potential future developments in related fields.

Sparrow search algorithm has good global search performance, but there is still a probability of falling into local optimum. In order to optimize the algorithm, K-means is proposed to make the population evenly distributed and improve the efficiency at the beginning. Then, the sine-cosine search and adaptive local search strategies are introduced to reduce the probability of falling into the local optimum in the middle and late stages, so that the optimal solution can be obtained easily. Finally, the two strategies are discussed and applied to SVM parameter optimization. The UCI dataset classification results show that the algorithm with the two strategies is suitable for SVM parameter optimization, and the algorithm with sine-cosine search has better optimization ability and stability.

In view of the low convergence speed of the LMS algorithm and the high computational complexity of the NLMS algorithm in wireless communication system, a selective partial update NLMS algorithm is proposed. This algorithm only updates part of the filter coefficients instead of all filter coefficients during each adaptation. It effectively reduce the complexity of classic adaptive filtering algorithms. In this paper, a selective partial update of the NLMS algorithm is deduced, and the simulation verification is carried out to compare with the LMS algorithm and the NLMS algorithm, and the superiority of the algorithm is verified.

Aiming at the low accuracy of the traditional Mask R-CNN applied to the image segmentation of different cats, an improved Mask R-CNN recognition and segmentation algorithm was proposed. The third channel of the FPN feature extraction path is added to obtain more comprehensive feature information, improve the accuracy of the segmentation mask and reduce the training time. The experimental results show that the method achieves 87.54% segmentation accuracy on the Kaggle dog and cat classification detection dataset, which is 13.57% better than the accuracy of the traditional Mask R-CNN algorithm on the same dataset, and has better detection and segmentation performance, providing a new method for the study of instance segmentation.

Deep learning has achieved remarkable results in image recognition, semantic segmentation and other fields. However, the primary premise of training deep learning model is the support of a large number of data sets and labeled samples. Therefore, deep learning is still faced with difficulties in medical, military and other areas where high-quality samples are scarce. Based on the image classification task as the application background, this paper improves a measure learning algorithm based on covariance representation. In the original Few-shot Learning algorithm based on measure learning, Firstly, from the perspective of second-order statistics, the covariance matrix between the feature vectors of each sample is constructed to realize the class representation; Then, an attention adaptive module is introduced to adjust the feature vectors of the class representation and query samples to make them closer to the class representation of the corresponding class. Finally, the experiment is carried out on the public data set(Omnight) to verify the effectiveness of the model. The experimental results show that the classification accuracy of the model designed in this paper is 6% higher than that of the Few-shot learning algorithm based on metric learning.

Icing prediction and early warning depend on icing prediction model and weather element forecast. In order to improve the accuracy of weather forecast, we test and analyze the forecast improvement of the temperature, wind speed and precipitation using numerical model and data fusion and assimilation of satellite-ground data. Through comparative analysis, data fusion and assimilation can significantly improve the forecast level of temperature and precipitation. The accuracy of 72-hour forecast does not decrease significantly compared with that of 24-hour forecast. Therefore, data fusion and assimilation using satellite-ground data has a better guiding significance for the prediction and early warning of 24-72 hour ice cover.

Based on the pavement temperature data of Sutong Bridge, a prediction model based on Long-Short Term Memory neural network was proposed and optimized in this study, in an effort to improve the prediction and early warning reliability of bridge state. Next, sufficient time was guaranteed for the sake of fault processing by dividing the data model into three parts: observation window, warning window and prediction window. The experimental results show that in comparison with the traditional time series prediction model, the proposed prediction model based on LSTM network is more accurate in the peak prediction, and it can effectively reduce the occurrence of false early warning. Moreover, the comparison results of root mean square error manifest that the proposed model displays a better stability in the long-term prediction.

Combat simulation test is the main means of evaluating command information systems. In order to improve the intelligence and efficiency of simulation experiment, the method of applying intelligent speech interaction technology (ISIT) to the simulation experiment automation process is proposed. The application processes of speech instruction recognition, instruction element analysis, data instruction generation and speech synthesis feedback are designed based on human-machine ISIT. Meanwhile, taking the air combat command and guidance scene an example, the command and guidance voice instruction database is constructed, and the speech interaction simulation experiment is designed and verified. The results show that ISIT can effectively reduce the number of commanders and operators in air combat simulation test, and improve test efficiency.

The rise of a new generation of artificial intelligence technology, represented by deep learning, has promoted the vigorous development of natural language processing technology. As a typical application of natural language processing technology, human-machine intelligent dialogue system, coupled with its commercial value in the fields of voice assistants and chat robots, has become a hot topic in the current academic and industrial circles. In the past, the responses generated by intelligent dialogue robots is single and universal, and even the content is inappropriate. Therefore, this study proposes to use the knowledge graph as the background knowledge when the dialogue model generates the responses[1]. In order to meet the needs of users more closely, it also proposes to introduce the user information participating in the dialogue and the dialogue scene into the model. The model is trained and evaluated on the DuRecDial public dataset, and the optimized model is compared with the original model. The experimental results show that the model with these two modules has better effect than the original model, especially in the generation index, the F1, BLEU2 and DIST-2 indexes have been improved by 0.91%, 0.5% and 0.9% respectively.

In the process of music therapy, in order to solve the problems of short music segments with poor emotional adjustment effect and poor emotional perception of patients, this paper proposes an emotional music reconstruction method that combines Deep Belief Networks (DBN) and Gated Recurrent Unit (GRU). According to the emotional music clips fed back by EEG, the high-dimensional features of music were extracted from the DBN network, and the music features were input into the GRU to construct the emotional music reconstruction model. The accuracy of note prediction of reconstructed music is analyzed, and the results show that this method is better than the reconstruction method using GRU alone. An anxiety evaluation experiment of the subjects was designed, and the Self-rating Anxiety Scale (SAS) of the subjects before and after playing reconstructed music was analyzed, thus verifying that the music generated by the emotional music reconstruction method can effectively alleviate anxiety, which is an excellent tool in the field of music therapy.

3D face reconstruction based on 3D morphable models (3DMM) uses single or multiple 2D face RGB images to reconstruct the 3D information of the target and restore the spatial structure of the face, which has important research significance for face recognition, film industry, medical field and so on. This paper introduces the main technical methods and research status of 3D face reconstruction based on 3dmm in recent 20 years, summarizes the advantages and disadvantages and applicability of various researches on 3D face reconstruction using 3dmm, and analyzes the current research hotspot and future development trend of 3D face reconstruction.

In order to improve the throughput of sub-users of cognitive radio networks and reduce the interference to primary users, firstly, we consider the spectrum sensing time, decision threshold, and power allocation factors of joint sub-users, and construct the average network throughput and average interference as a multi-objective optimization problem, and secondly, propose a multi-objective optimization algorithm based on improved sticky bacterium algorithm (SMA) to model and solve this problem to obtain the best the sensing time, decision threshold and power. Finally, the proposed method is compared with the NSGA-II algorithm, and the simulation results show that the proposed method can effectively search for the optimal solutions of the spectrum sensing parameters to maximize the average network throughput and minimize the average interference.

Aiming at the problem of landslide geological disaster prediction, based on the analysis of geological structure and historical disaster data of a county in southwest Shaanxi, this paper puts forward a landslide geological disaster prediction model based on AdaBoost. Firstly, the data of landslide geological disasters are analyzed, and the main factors of landslide geological disasters are determined by principal component analysis and Spearman grade correlation coefficient method. Then, the main factors are used as feature data to train the model, and grid search is used to optimize the super parameters in the model. Finally, experiments on real data sets show that the prediction accuracy of AdaBoost model is 1.5% higher than the best prediction results of decision tree, logistic regression, support vector machine and k-nearest neighbor method, and AUC is 0.24% higher. It further verifies the effectiveness and feasibility of the model, which can provide a scientific basis for the prediction of landslide geological disasters.

With the increasing level of intelligence of logistics enterprises in China, the mode of artificial intelligent logistics distribution has become a new strategy for the development of modern logistics industry. In the logistics industry of artificial intelligence, logistics robot has occupied the market and become the key link of logistics distribution. It is of great significance to study the distribution mode of artificial intelligence robot. With the development of our national economy and the improvement of people's living standard, the demand for logistics service quality and distribution efficiency is higher and higher. At present, most enterprises in our country abandon the traditional manpower distribution operation mode. This paper expounds the basic concepts of "artificial intelligence" and "intelligent robot" through the literature research method and the research method of combining theory with practical cases, and obtains the cognition of the current technology of intelligent logistics by collecting, processing and sorting the relevant documents. The emergence of artificial intelligence makes large amount of data fast and accurate, while artificial intelligence robot makes logistics management system, object transportation, distribution and order quantity processing intelligent, and produces the integration of production, the production chain is greatly reduced, the communication between logistics enterprises and intelligent robots is closer, the corresponding cost can be greatly reduced, and the enterprise can manage and operate effectively.

General machine learning requires a large number of training samples, meaning a lot of manpower and material resources. To solve this problem, active learning and its algorithms are often used. Active learning is a form of semi-supervised machine learning where the algorithm can choose which data it wants to learn from and then use the smallest and most effective labeled data set to make predictions or classifications. This article will show an example of using active learning to predict red wine quality. The predictive modeling approach the author chose was the K-Nearest Neighbor and the active learning algorithm was ranked batch-mode sampling. By observing the learning curve, the author found that generally the prediction accuracy of active learning would increase as the number of iterations increased. The author compared the experiment with another case using classic iris flower data set, and concluded that the prediction accuracy of active learning for different data sets depends on many factors, such as the correlation between the independent and dependant variables, the size of the data set and the number of iterations.

Cardiovascular disease (CVD) is a dangerous disease, which can be effectively prevented by detecting arrhythmia early. In order to detect arrhythmia accurately, more and more researches use artificial intelligence methods to realize the classification and detection of electrocardiogram (ECG) signals. However, most of these designs are unfriendly to the hardware design due to too many parameters which lead to large calculation power and data accessing power. In this paper, we present an efficient arrhythmia classifier based on the convolutional neural network with the incremental quantification. This more efficient design can classify ECG signals accurately with lower capacity of parameters. The simulation results show that the recognition rate of the network with the incremental quantification has reached 92.76% with 39.34KB memory footprint, which is beneficial to hardware design and has better accuracy than other advanced quantification methods.

With the popularization and use of three-dimensional data acquisition equipment, the acquisition of three-dimensional data is more convenient. Three-dimensional data contains rich shape and scale information. How to effectively and accurately classify, segment and identify three-dimensional data is a research hotspot in the field of computer vision. Aiming at the particularity of 3D point cloud data and the neglect of local correlation between points, this paper proposes a new end-to-end depth network framework, namely KE-PointVNet. KE-PointVNet can directly deal with point cloud and construct deep convolutional network. KE-PointVNet extracts local geometric features of each point based on the EdgeConv module. By using the KNN proximity algorithm to construct local neighborhood of the point cloud, it can avoid its non-local diffusion, and finally obtain the high-level semantics of the point cloud through the local aggregation vector VLAD layer. The experimental results show that this method has a higher classification accuracy than most of the existing point cloud classification methods.

In this study, a hyper-heuristic named Sub-domain Elimination Strategies based on Firefly Algorithm (SESFA) is proposed. First, a typical hyper-heuristic is usually using the high-level strategy selection or the combination of the low-level heuristics to obtain a new hyper-heuristic, each round of optimization process is carried out in the whole problem domain. However, SESFA evaluates the problem domain through the feedback information of the meta-heuristic at the lower level, eliminating the poor performance areas, and adjusting the underlying heuristic or adjusting the algorithm parameters to improve the overall optimization performance. Second, the problem domain segmentation function in SESFA can reduce the complexity of the objective function within a single sub-domain, which is conducive to improving the optimization efficiency of the underlying heuristic. Further, the problem domain segmentation function in SESFA also makes there is no direct correlation between different sub-domains, so different underlying heuristics can be adopted in different sub-domains, which is beneficial to the realization of parallel computing. Comparing SESFA with Firefly Algorithms with five standard test functions, the results show that SESFA has advantages in precision, stability and success rate.

Under the surveillance scenario, due to the influence of occlusion, deformation and illumination, the accuracy of person re-identification (ReID) task will be greatly affected. The extraction of effective pedestrian features has become the key of person ReID task. In view of this problem, this paper creatively uses frequency channel attention (FCA) network to carry out person ReID task. FCA network solves the problem of insufficient information representation caused by global average pooling (GAP) in traditional channel attention network. In addition, this paper conducts experiments on two datasets. The ablation experimental results show the effectiveness of FCA network in person ReID task, and the results of contrastive experiment show the superiority of FCA network in person ReID task.

At present, the target tracking algorithm for the video of the vehicle involved in the accident is not perfect, which causes the target to rotate and the target is blocked, and the tracking effect is not good. This paper uses a CamShift tracking algorithm that combines HLBP feature matching and unscented Kalman filtering. Firstly, the vehicle features are extracted through the algorithm to obtain more accurate features, thereby reducing the interference caused by the target rotation on the feature extraction. Secondly, the degree of occlusion of the target is judged by the Bhattacharyya distance, and finally the UKF algorithm is used to predict the target position. The vehicle efficiently solve the problem of poor tracking performance when the target is occluded. Experiments have proved that, in the actual application of tracking the accident vehicle, the algorithm can effectively reduce the influence of external factors on the tracking effect, and the tracking accuracy is greatly improved.

Facial expression can truly reflect people's inner activities, human emotions can be fully reflected through the expression, facial expression recognition in the field of artificial intelligence has important research significance, in daily life also has its application value. At present, facial expression recognition technology has become a very promising frontier technology, but also the current research focus in the field of computer vision.

In this paper, facial expression recognition based on convolution neural network is studied. The concrete work has the following several parts: For the part of face detection, this paper introduces the common methods of knowledge-based rules, feature-based, template-based matching, and statistical model-based, and V-J detector is used for face detection in this paper. In the part of expression recognition, we study the recognition of happiness, sadness, anger, depression, fear and surprise by convolution neural network. This paper uses keras to build a deep learning framework. The neural network consists of volume base layer, pool layer, activation layer and full connection layer. The classifier uses softmax. Using the image data in the standard database as input, after the processing of each layer in the neural network, and finally outputting the probability corresponding to six expressions through softmax, it is generally believed that the expression with the highest probability is the facial expression in the input image.

At present, VR technology has been widely used in R&D and manufacturing of product design projects because of its many advantages. How to further improve the application effect of VR technology in product design practice projects has become one of the research hotspots in the design field in recent years. Based on this, this paper explores the application of VR technology in product design. Firstly, it introduces the basic situation of VR technology. Then, taking the design of leisure and entertainment products in public facilities as an example, it discusses the basic application process of VR technology. Finally, it looks forward to the future development of VR technology based on the shortcomings of VR technology in product design.

Online search engines and other computer-aided tools are playing an increasingly important role in technical translation, especially in the translation of out-of-vocabulary new terms. Dictionaries, at times, prove to be of little help to term translation where out-of-vocabulary terms rely on polysemy for heuristic effects. This paper explores the role of online search engines in terminology translation by analyzing the effectiveness of four terminology translation extraction methods which utilize online search engines as search tools, thereby generating suggestions on how to use search result for reference. An experiment with some new terms was conducted to examine the effectiveness of each method. This study revealed that template-based term translation extraction and dictionary-based translation extraction have the highest accuracy rate of 86% and 89% respectively while semantic prediction search method is the most flexible method with the highest retrieval rate of 95% although its results are not always accurate. Therefore, a verification model was proposed, which utilizes quantitative-and-qualitative context analysis to verify the accuracy of translation results obtained from a search engine.

Aiming at the problem that the value of weighted matrix Q and R in linear quadratic optimal control (LQG) controller depends on the designer's experience, adaptive differential evolution (JADE) algorithm with optional external archive is introduced to optimize the control parameters of LQG, so as to improve the performance of LQG controller. The dynamics model of 1/4 vehicle active suspension was established as the object, and the LQG objective function was constructed. The simulation experiments were carried out on B-class and C-class roads with the help of Matlab/Simulink.The experimental results show that the RMS values of the vertical acceleration of the vehicle body, the dynamic travel of the suspension and the dynamic displacement of the tire of the active suspension based on JADE optimized LQG controller under different road surfaces are reduced to a large extent compared with that of the passive suspension, which improves the stability of the vehicle and the comfort of the ride.

The drawing cushion of press line in the press workshop of the whole car factory has been extensively upgraded from air cushion to hydraulic drawing cushion. The hydraulic drawing cushion is more accurate than the drawing force of the air cushion, however, the hydraulic drawing cushion is a complex of mechanical, hydraulic, Motion Control and detection technology, and coupled with the press, which makes the fault diagnosis of the hydraulic drawing cushion more difficult. This paper mainly introduces the problems of the coordination among the three systems of mechanical, hydraulic and electronic control of Rexroth Company's servo-hydraulic drawing cushion and its intelligent fault diagnosis method.

In order to improve the accuracy and reliability of abnormal heart rate detection, a method of abnormal heart rate detection based on double slope QRS beat location is proposed. Firstly, according to the singularity of QRS wave, the slope difference between the left and right sides is used for localization, and double thresholds are set to track the real-time change of the signal to ensure the robustness of localization. Secondly, wavelet coefficients are extracted as beat frequency features by wavelet transform. Finally, XGBoost algorithm is used for classification. In the case of the same data set, compared with the existing arrhythmia detection algorithm, the performance is further improved. The sensitivity of QRS detection is 99.65%, the positive prediction rate is 99.41%, and the accuracy of heart rate anomaly classification is 99.12%. This method can locate the beat more effectively, which provides a new method for the research of abnormal heart rate detection.

With the shortage of transmission channel resources and the increase of human activities, the inspection of transmission channel environment changes is of great significance to the protection of line breakage, the management of transmission channel resources and the detection of hidden dangers in advance. In order to explore a fast and intelligent extraction method of environmental change information in the transmission channel, this paper takes a transmission line section in Tongzhou District of Beijing as the experimental area, takes two high-resolution satellite images of Beijing No.2 as the data source, and conducts principal component analysis and vegetation index processing analysis on the two satellite image data. An intelligent patrol method for transmission channel environmental change based on multi-source feature multi-scale segmentation technology and object-oriented classification algorithm is proposed. The results show that the satellite remote sensing environmental change intelligent patrol method proposed in this paper is sensitive and effective to the human activity change patrol of the transmission channel, and can be extended to the large-scale and normal intelligent patrol operation of the transmission channel against external breach.

Based on the research of the existing intelligent scooter, this paper puts forward a more perfect scheme for the skateboard with STM32 as the core from the aspects of appearance, internal performance and whether the skateboard is safe in the process of running. Firstly, the performance and serial port of STM32 are analyzed, and then the corresponding gravity sensor is selected according to the module and connected to the host chip. Then the simulation debugging of gravity sensing was carried out. After normal operation, the assembly of the whole skateboard was started and the Bluetooth module and LED module were installed. At the same time, a detachable handle is also designed in appearance to ensure driving safety to the greatest extent. Finally, the wireless communication control system is used to achieve humanized design, and the intelligent direction and speed are adjusted through the gravity induction module.

The existing safety detection methods are not sufficient to achieve real-time accurate detection, and cannot perform real-time safety assessment. Taking the construction site as an example, this paper proposes a safety equipment wear detection method based on the joint constraint of linear distance and angle, and proposes the idea of establishing a knowledge database of safety protection equipment wear behavior, and proposes a safety factor evaluation model. In this paper, convolutional neural networks are used to identify the construction personnel and safety equipment in two basic poses, and the linear distance and deflection angle of the regional center point coordinates are used to jointly determine the wearing status of the safety equipment. The generated result matches the safety information knowledge base to obtain the final test results. We obtained the real-time safety factor of the construction personnel by correlating the detection results with the actual scene. Finally, the method used in the article is summarized and prospected.

The integration of Internet tech and mobile communication tech has changed the form of traditional info and communication tech, which can better serve the real needs of people for real-time communication and high-speed sharing and interconnection of data, info and resources. Based on this, this paper first analyses the connotation and characteristics of mobile communication and computer Internet, then studies the integration mode of mobile communication tech and computer Internet tech, and finally gives the trend of the combination of mobile communication tech and computer Internet tech.

To solve the problem of insufficient segmentation accuracy caused by the failure of UNET model to make full use of the weight relationship and semantic information between coding layers in lung nodule image segmentation, a lung nodule segmentation method based on BBClstm-Unet is proposed. In this method, parallel attention module is used to complete the information coding of a larger range of pulmonary nodules image and reset the weight of sampling correlation channel information. Bconvlstm structure is used to combine feature mapping, and forward and backward output sampling are fused to obtain the final segmentation result. At the same time, hybrid loss function is used to alleviate the class imbalance problem. The experimental results show that the average dice value of BBClstm-Unet network on luna16 data set reaches 90.83%, which is better than the original UNET network.

Road traffic safety has always been the focus of social concern, and the actual road traffic scenes, where only a single sensor is applied cannot cope with the interference brought by complex external factors, which makes vehicle detection extremely challenging. This paper focuses on a vehicle detection algorithm for the fusion of millimeter-wave radar sensor and monocular camera sensor, including the calibration of millimeter-wave radar and camera, the establishment of a temporal fusion model of the two sensors. Finally, the target information obtained from the two sensors is fused with the data using the adaptive Kalman filter fusion algorithm, which can reduce data ambiguity and increase the reliability and validity of the data. Experiments show that the method can overcome the shortcomings of single sensor in target detection, and the obtained target information is more comprehensive compared with the monocular camera detection results.

With the improvement of domestic economic level and the development of high-tech technology, many computer technologies, image processing technologies, digital media technologies, etc. are constantly developing and becoming more and more popular. Among them, digital media technology has spread to all fields of society, and at the same time, the influence brought by digital media technology is constantly increasing and expanding. With the development of digital media technology, many works of art and works expressed in art form have gradually become diverse, adding many new tricks to the art world and enriching the expression and presentation types of works of art to a great extent.

Nowadays, the detection, identification and classification of targets behind buildings in the process of wall penetration sensing is one of the main solution directions for detection activities. In this paper, a random forest-based human pose detection system for through-wall radar is proposed, aiming at optimizing the traditional through-wall radar target detection and identification by machine learning methods. The actual data acquisition is performed by UWB-MIMO through-wall radar system to construct multidimensional data and identify the pose. The experimental results show that the random forest method has high recognition performance by identifying multiple poses and has a pose resolution that traditional target recognition does not have.

In recent years, generative adversarial networks have performed well in the field of dialogue generation to improve the information diversity of dialogue responses. Often overlooked, however, is that the query and response are not relevant on the topic. In order to improve the topic relevance of chat conversation, the paper proposed a topic-relevance adversarial response generation model, TR-ARG, which is composed of generator G, discriminator D and topic classifier T. The experiment was evaluated on OpenSubtitles, an open dialog dataset, and compared with the current baseline models SEQ2SEQ and GAN-AEL. The results show that our model can effectively improve the topic relevance of generated responses.

When the Internet of things system is applied, it is necessary to build a data information platform. The perception layer devices, intelligent terminals and network terminal devices are connected to form the system platform, and the corresponding software is installed on the software management information platform to manage the data of the Internet of things system. We set up a data and information processing software platform to achieve the tasks and objectives of the Internet of things system. The data collected by the perception layer is calculated, stored, classified, analyzed, mined and understood, and the data of the perception layer is analyzed. So as to achieve the purpose of real-time control, accurate calculation, accurate management and scientific decision-making of the Internet of things system. Through the intelligent man-machine interface, the Internet of things smart grid, environmental monitoring, warehousing and logistics, industrial control, smart supermarket, smart medical, smart home, smart agriculture and extended function access can be realized.

Aiming at the problems of dimensional disasters and low classification accuracy caused by too many features extracted in the emotion recognition process, an EEG emotion recognition method with optimized feature selection is proposed. The individual rhythmic signals of the EEG are obtained by wavelet packet decomposition and the sample entropy, energy and power spectral density are extracted as EEG features. A discrete binarization of the feature matrix using the Beetle Antennae Search (BAS) algorithm, while introducing a subset of features into the objective function and searching for the optimal subset of features. Finally, the SVM classifier is used for classification. The experimental results show that it achieves 89.72% accuracy on the DEAP dataset and significantly reduces the original feature dimension compared with the traditional feature selection method, which has good application prospect.

5G network has great application potential in the field of industrial Internet. However, the current lack of development of 5G network in the field of security technology restricts its application on the Industrial Internet. In this article, the author combines the industry's needs for Industrial Internet security, and summarizes the research status of 5G network security technology, gives the development direction of three 5G network security technologies with development potential. This paper found that due to the three end-to-end characteristics of the Industrial Internet: physical security, information security, and system self-healing, there are security requirements; the other three 5G network security technologies, namely lightweight encryption mechanism, physical layer security technology and network slicing security technology will help 5G network be more widely used in the Industrial Internet field.

Aiming at the problem that the existing fatigue driving detection methods cannot make full use of eye features and semantic information, a driver eye fatigue detection method based on parallel neural network is proposed. This method detects human faces through a multi-task cascaded convolutional network, determines the driver's eye area according to the face ratio relationship, uses the parallel structure of the convolutional neural network and the residual network to recognize the eye state opening and closing, and according to PERCLOS the criterion is to judge the fatigue state. The experimental results show that the method has a high accuracy rate and can effectively detect the fatigue driving state.

As an important tool of intelligent transportation and intelligent travel, automobile is becoming more and more electronic and software-oriented. The intelligence and connection of automobile have brought a new and huge economic growth point to the automobile industry. But a number of new software and network technologies introduced have also posed a threat to automotive security. Fortunately, these technologies have also given birth to and promoted a new business of the automobile industry, namely the security business of intelligent connected vehicles (ICV). At the same time, under the huge volume base of the automobile industry, the development of the security market in the automobile industry will have an impact on the overall market scale of network security. Investigation, analysis and research were undertaken in this paper to demonstrate the relevance among the development of intelligent network vehicles, the automotive security, and the network security market.

Encode-decoder structure is used in deep learning for real-time dense segmentation task. On account of the limitation of calculation burden on mobile devices, we present a light-weight asymmetric encoder-decoder network in this paper, namely LAENet, which quickly and efficiently accomplish the task of real-time semantic segmentation. We employ an asymmetric convolution and group convolution structure combined with dilated convolution and dense connectivity to reduce computation cost and model size, which can guarantee adequate receptive field and enhance the model learning ability in encoder. On the other hand, feature pyramid networks (FPN) structure combine attention mechanism and ECRE block are utilized in the decoder to strike a balance between the network complexity and segmentation performance. Our approach achieves only have 0.84M parameters, and is able to reach 66 FPS in a single GTX 1080Ti GPU. Experiments on Cityscapes datasets demonstrate that superior performance of LAENet is better than the existing segmentation network, in terms of speed and accuracy trade-off without any post-processing.

With the large-scale application of cloud storage systems, lots of attribute-based access control (ABAC) schemes have been introduced to protect data and user security in this insecure environment. ABAC can make data owners control their own data and protect data security and privacy. However, there are two serious privacy leakage problems, namely user attribute privacy and policy privacy, should be solved in the process of constructing ABAC schemes. In this paper, an ABAC scheme supporting privacy protection is constructed, which can solve privacy leakage problems of user attribute privacy and policy privacy in large universe attribute system. Furthermore, a framework is proposed. The framework combines our ABAC scheme, the Ethereum blockchain and blockchain-based storage system. In this framework, the security characteristics of blockchain technology are used to realize decentralization, tamper-resistant and avoiding single point of failure. Besides, the problems of attribute revocation and policy updating are solved by smart contract on the Ethereum blockchain. Finally, we established an initial implementation on Linux and Rinkeby test network, and the experimental results show that our scheme is feasible.

Aiming at the problem of poor real-time performance of cyclostationary feature detection, a fast cyclostationary feature detection algorithm based on the Unified Computing Device Architecture (CUDA) is proposed. This method moves the data processing part to the GPU, and uses the advantages of a large number of GPU stream processors and powerful multi-threaded concurrent execution functions to accelerate the FAM algorithm in parallel, thereby realizing fast signal processing. Experimental results show that compared with traditional serial algorithms, this method reduces the time overhead of the algorithm and improves the real-time performance of the algorithm.

This paper presents a novel fixed-time adaptive tracking control scheme for the hypersonic flight vehicles (HFVs) subject to asymmetric time-varying constraints, uncertain dynamics and unknown external disturbances. By incorporating the back-stepping technique and radial basis function neural networks (RBFNNs), the uncertain dynamics of HFVs are estimated. Note that most existing results only achieve practical fixed-time stability but not fixed-time stability, or require specific knowledge of all the dynamics of HFVs. To remove such restrictions, a fixed-time controller is newly constructed by means of a tuning functions and a projection operator-based adaptation mechanism. In consequence, the tracking errors can asymptotically converge to the preassigned compact set within fixed-time and the asymmetric time-varying constraints of HFVs never are violated. Finally, the effectiveness and superiority of the proposed control strategy is demonstrated by numerical simulations.

With the rapid development of the network, network transmission encryption technologies such as SSL and SSH have emerged. Network traffic has grown exponentially, and transmission encryption has become an important means of protecting data security and privacy. However, encrypted data also brings hidden dangers that are not easily detectable to network security. Identifying the encrypted network traffic can effectively solve this problem. However, the current recognition probability is not high enough and the time delay caused by the recognition together makes it impossible to accurately detect and warn the network traffic. An encrypted network traffic recognition method based on deep learning is proposed. Experimental verification shows that the method is applied in the network. The accuracy of encrypted network traffic identification is 97.02%, which can meet actual needs.

In this paper, a practical method and process for evaluating the importance of the links in combat system oriented to soft killing is proposed, which fully considers the current mission requirements faced by combat commanders, and integrates the theory of operation loop. Decision makers can easily use this assessment method to prioritize communications link upgrades for combat systems. Firstly, the network model of the combat system is constructed by analyzing the structure of the combat system and the correlation between the components. Secondly, the subtask composition and relevant information of the mission needed to be executed by the current combat system are analyzed, and the evaluation index of mission participation on the link is introduced. Then, the combat capability evaluation method of combat system based on operation loop is used for evaluating the influence degree of combat network link on the survivability of combat system. Finally, the importance degree of link is analyzed and evaluated by combining these two indexes, and the key communication links between entities within the combat system are found by using these two indexes as evaluation indexes.

In order to cope with the impact of intelligent building-type virtual power plant on system frequency, this paper proposes a primary frequency control strategy. Firstly, a typical demand-side resource model of intelligent building-type virtual power plant is established, which considers the travel uncertainty of electric vehicles and resident's comfort preference. Secondly, the control parameters of electric vehicles and air-conditioning loads are designed respectively, and a control strategy based on variable coefficient droop control is proposed. Finally, a simulation example verifies the effectiveness of the primary frequency regulation strategy, which can effectively reduce system frequency fluctuations while ensuring user demand.

Traditional steam turbine maintenance work has many shortcomings, such as poor effectiveness, high cost, and long cycle. In order to solve this problem and improve the professional skills and maintenance efficiency of steam turbine maintenance personnel, we first collect steam turbine equipment drawings and 3D dimension data of parts, and use 3D modeling software to establish augmented reality scene model. Then combined with the virtual reality technology and image fusion technology to realize the maintenance procedure training and maintenance auxiliary guidance of the steam turbine unit. The practical application results show that this technology can realize the virtual and real operation of the model by users in augmented reality scenes, bring interactive and immersive experience to operators, and improve the work efficiency of maintenance personnel.