This PDF file contains the front matter associated with SPIE Proceedings Volume 12566, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

In Chinese sentiment analysis, sentiment words are just a drop in the ocean compared with the whole corpus. In order to solve the problem of insufficient emotion lexicon and prior knowledge, proposes a method to predict the emotion intensity of target words based on neural network model (Neural Network Emebdding Score, NNES). By training a small number of labeled samples, using clustering algorithm to find the seed words, calculate the similarity between the target words and the seed words, and using it as the input of neural network to predict the emotional intensity of the unlabeled words. Compared with the traditional machine learning regression models, it has smaller mean square error. Meanwhile, a BiGRU model based on attention mechanism and convolution is proposed by integrating the predicted emotion intensity with word vector (Neural Network Emebdding Score with CNN and Attention-BiGRU, NNESC-Att-BiGRU). To compare several popular models on product and hotel review data sets, and the proposed model has better classification effect on Chinese sentiment classification task.

Aiming at the problem of intelligent classification and recognition of radar jamming signals, the convolutional neural network structure is studied. By optimizing the basic network, the normalization layer and activation layer is added to the LENET-5 structure to improve the accuracy of recognition results. The linear frequency modulation signal and amplitude modulation interference, frequency modulation interference, comb spectrum interference, slice reconstruction interference, intermittent sampling and forwarding interference are analyzed. Six signal models are used to generate data sets, and intelligent methods are adopted to realize classification and recognition.

Parking lot selection is an important part of drivers' parking behavior. Building a good parking lot recommendation system can reduce the time of parking lot selection and effectively alleviate parking problems. Firstly, the process of parking lot selection and the principle of collaborative filtering are analyzed. Then, the data of drivers' preferences are collected, and the cosine similarity method is used to find similar drivers and parking lots. Finally, based on the collaborative filtering theory, using MATLAB to realize the construction of parking lot recommendation system. The results show that the recommendation system can better recommend parking lots for drivers and facilitate drivers to make decisions quickly.

Existing vehicle detection methods lack the fine vehicle detection algorithm. In order to improve the accuracy and applicability of anchor-based object detection models, a novel and practical vehicle Fine-grained identification network (EFDet-SPP) based on the EfficientDet is proposed. The improved network adds a Spatial Pyramid Pooling module (SPP) after the feature extraction network for concatenating features to enhance network learning capabilities, and multi-scale extraction of highly semantic features of images. Anchor-based predictions are converted to pixel-based predictions by combining FCOS's head network, eliminating the hyperparameters associated with anchor boxes. And with Mosaic, Copy-Paste data augmentation methods scale small object samples to achieve data sample balance. Experimental results show that the improved network has achieved 94.8% in the actual collected fine vehicle detection dataset, which is greatly improved compared with the EfficientDet network, and does not significantly increase the training parameters and calculation amount of the network.

A new network super-resolution algorithm is proposed in the light of multi-feature fusion to surmount: the problem of manual feature extraction as well as low information throughput in traditional super-resolution algorithms. The algorithm adopts a learning method of the indirect mapping from the hazing image to the clear super-resolution image, which calculates the similarity between the point and other points, normalize the similarity to obtain the weight between each point, and then multiply it by the feature map value of the corresponding point and use the similarity. In the algorithm, utilizing multi-scale feature fusion enables to rebuild details of the image. Different depths of field combine the contour information obtained by shallow convolution with the detail information obtained by deep convolution to enhance the overall super-resolution’s effect. The experimental results and related data show that the channel throughput has been significantly improved, and compared with other algorithms, it has better detail information and contrast, which provides a new idea for dehazing methods.

As the main way of underwater data transmission, acoustic communication is still limited by the low-level signal-to-noise ratio and channel capacity. The Orbital Angular Momentum (OAM) based wireless optical communication provides a new dimension to data transmission with an expanded channel capacity. We introduce the basic OAM multiplexing communication system, analyze the basic principle of its transmission process, and verify the feasibility of underwater vortex optical communication by simulation. On this basis, the influence of transmission distance and underwater turbulence on the communication system is studied. Our results show that the transmission performance of the system is better in the transmission range of 30 m, and the influence of turbulence on the system performance increases gradually with the increase of signal-to-noise ratio.

Cancer diagnosis, prognosis, and therapeutic response predictions are based on data from various modalities, such as histology slides and molecular profiles from genomic data. In cancer clinical treatment, the technology of intelligent diagnosis for cancer patients has become an essential research domain with the rapid growth of various pathological data. In this work, we propose a multimodal fusion method for cancer survival analysis based on Cross-Attention Transformer. Compared to similar bimodal work, our work greatly reduces the number of parameters in the feature fusion model (our fusion model has 7625 parameters), and achieves the State-of-the-Art effect (81.85%) in bimodal cancer survival analysis task with histology images and genomic features data of Glioma cancer from TCGA database. （Previous bimodal Sota work in this task is Kronecker Product which achieves 81.40% with 170130 parameters）In addition, our experiments show that Cross-Attention can not only increase the correlation between the two modalities but also offer a better bimodal feature representation for the final fusion.

With the rapid development of smart grid technology, more and more attention has been paid to the stability of the power grid. When the power grid fails, the dispatch center will receive various types of data, including the PMU data collected by the PMU (Power Management Unit) device used by the Wide Area Monitoring System (WAMS) and the data collected by the Supervisory Control System for Data Acquisition and Control (SCADA). Alarm information data, they describe the same grid fault from different dimensions. At present, the power grid fault diagnosis based on PMU data and alarm information texts has been fully studied, but the information source is single, and it is difficult to ensure the accuracy of the diagnosis results when complex faults and multiple faults occur in the power grid. This paper starts with the information source of power grid fault diagnosis and develops a power grid fault diagnosis method that integrates the dual data sources of PMU and alarm information: 1. effectively connect alarm message text and PMU based on fault events. 2. complete the fusion of PMU and alarm information to improve the scope and accuracy of power grid fault diagnosis. Finally, the simulation fault data is used to test. The experimental results show that the accuracy of fault diagnosis of dual data source fusion is higher than that of single data.

At present, generative adaptive networks (GAN), has become a popular module in deep learning. GAN is very effective in image generation and image style migration. At present, the research on the migration of image style is focused on images such as oil painting and landscape painting, and lacks the research on the conversion between object images. This paper extends the style transfer technology to object recognition, uses CycleGAN method to learn the mapping relationship between zebra and horse, and realizes the transformation between zebra and horse. Viewing the generation effect of different learning methods by changing the learning times and learning rate policy. This work realizes the conversion between zebra and horse, and shows the generated pictures under different training times and different learning situations. Under the same training times, the conversion effect from horse to zebra will be better. After a certain number of trainings, the training effect will gradually decline. The conversion effect of the same type will be improved with the increase of training times. Different learning rate policies will bring different generation effects.

In this study, we provide an approach named TRFit for unstructured 3D point cloud normal estimation. It handles noise and uneven densities point clouds well. Recently, learning-based normal estimation methods have significantly outperformed traditional methods on benchmark normal estimation datasets. In order to estimate normals, they frequently employed neural networks to learn point-wise weights for weighted least squares polynomial surfaces fitting. However, existing methods often ignore local geometric relationships, which will make the fitted surface significantly different from the real. To this end, we propose to use graph convolutional to learn local structural information. Meanwhile, we suggest the Geometric Relation Transformer (GRT), a transformer-based scale aggregation module, to fully utilize points from various neighborhood sizes. It can adaptively capture the relations between different regions. We achieve state-of-the-art results on the baseline normal estimation dataset, and experimental results show that TRFit obviously improves the accuracy of normal estimates, preserves their details. Moreover, it exhibits robustness to noise, density variations, and outliers. Besides, we demonstrate its application to surface reconstruction and denoising.

The security of government and enterprise data sharing is very important and critical. To increase the number of data sharing, create a more stable transmission and processing environment, and reduce network threats, this paper studies the security sharing method of government and enterprise data based on differential privacy protection. Firstly, the government and enterprise data encryption are described, and the Tendermint differential overlapping interactive sharing nodes are deployed in the preset area. Then, based on this, the interactive differential privacy protection data sharing model is designed, and the threat identification method is used to achieve data security sharing. The experimental results show that compared with the traditional proxy encryption data security sharing test group and the traditional CP-ABE data security sharing test group, the differential privacy protection sharing test group designed in this paper achieves relatively more times of one-way data security sharing, which indicates that the proposed method has a small error and fast speed in the actual data transmission process. The data processing in the region has less restrictive conditions and has practical application value.

Under the background of today's era, the development of power grid intelligence is an inevitable trend in the development of power companies. In this process, what the power industry needs to do is to do a good job in the prevention of smart grid information security risks, so as to avoid power companies due to excessive power grid information security risks. bring certain losses. The power marketing system is a vital part of the power system. Starting from the marketing system, combined with the current business situation, we will study the development trend and overall construction plan of the new generation of marketing business application systems, and further study the security architecture of the system. From the perspective of multi-faceted security management and control, to implement the relevant national and company-related personal information protection requirements, and to improve the information security of the new generation of marketing systems has become an urgent problem to be solved.

Based on Microsoft Kinect equipment, this paper adopts motion capture technology to collect human motion data in the actual training process of rhythmic gymnastics, designs and develops the functional module of motion data analysis combined with ASP.NET development environment, and integrates and packages other auxiliary functional modules to form an auxiliary training system of rhythmic gymnastics. Aiming at all kinds of problems in the teaching practice of rhythmic gymnastics in colleges and universities, the system will put forward a comprehensive application solution, which takes the physical movements, equipment movements and self-selected complete movements of rhythmic gymnastics students in daily training in colleges and universities as the research objects, takes the transformation and analysis of movement information data as the method, and takes the visual interaction at the network end as the display form. It greatly facilitates the practical application of students and teachers, effectively improves the training effect of students, achieves the purpose of auxiliary training, optimizes the teaching process of rhythmic gymnastics education, and further promotes the process of information teaching construction in colleges and universities.

In order to solve the problem of small-scale key point positioning accuracy in multiplayer pose estimation, using top-down method and advanced yolov4-tiny, this paper proposes an improved multi-person pose estimation method based on stacked hourglass deep learning network. The coordinate attention mechanism is introduced in the original hourglass network residual module to perform feature enhancement, which suppresses useless features and improves useful features, thus improving the recognition rate of small-scale joint points. From experiments, the index PCK@0.5 reaches 88.9% on MPII dataset, which verifies the effectiveness of our proposed method.

Smart industrial parks are an important part of smart cities and building an information platform using the Internet of Things and cloud computing has become the development direction of smart industrial parks. According to the current needs of related users in the smart industrial park, the architecture of the information system of the smart industrial park was proposed. The design scheme realized the functions of intelligent analysis, interconnection and optimal decision-making of the smart industrial park. And based on the Internet of Things, the application in the smart industrial park was introduced to provide guidance for the improvement of the intelligence level of the smart industrial park.

Security and verification applications are primarily developed in C or C++, so reverse engineering is prone to analyze and invade programs based on C or C++. As a result, protecting the source code of C or C++ as well as guaranteeing the safety of those security and verification applications are vital tasks in the network security area. This paper proposes a new framework enlightened by dynamic heterogeneous redundancy architecture, which originated from the mimic defense field and mixed with a state-of-the-art obfuscation method to not only protect the code property but also improve the security level of those applications. Three key characteristics of the framework are redundancy, heterogeneous and dynamic. This article firstly builds an LLVM optimizer to generate redundant binary code. Besides, we import some random perturbation in the optimizer to realize heterogeneity between generated redundant codes. Finally, exploiting these heterogeneous redundant codes constructs a dynamic architecture to detect and defend anomaly requests.

Taxi destination prediction can grasp the flow direction of the taxi, facilitate the taxi dispatches. There has always been a long-term dependency problem in taxi trajectory prediction. Although LSTM can solve the long-term dependency problem to a certain extent, it does not have a good ability to deal with the deep correlation between long trajectory sequences. To address the above problem, we propose a taxi destination prediction method based on LSTM with Tree Memory Module (TMM-LSTM). TMM-LSTM stores the state of the input trajectory through an external memory structure. It uses a tree structure to process more historical information and better deal with the long-term relationship between trajectory points. TMM-LSTM can better solve the long-term dependency problem in the taxi trajectory sequence. Experiments demonstrate that the average error distance is 6% lower than traditional LSTM model.

Human pose recognition is a current research hot spot, which aims to detect the information of human joints, orientations and scales from images or videos and predict the specific pose of the human body based on all the key point information. Benefiting from the rapid evolution of deep learning, target detection algorithms based on convolutional neural networks have achieved breakthroughs in both accuracy and efficiency. On the basis of detailed literature research and analysis, this paper provides a comprehensive evaluation of the research progress of human pose estimation. Specifically, we mainly introduce the classification of human action poses and the methods of human action pose recognition, including the design ideas, basic framework and advantages and disadvantages of representative recognition methods. Finally, we summarize the main challenges and give an outlook on the future research development of object detection.

Cyber Mimic Defense (CMD) is a new generation of active defense technology after firewall, intrusion and other traditional defense technology. It aims to deal with uncertain threats in the network environment with visual uncertainty. This paper briefly introduces the architecture and principle of CMD and defines four main elements of the current simulation platform: system architecture, heterogeneous policy, scheduling policy and voting policy. Combined with examples, the four elements are respectively summarized. The system architecture is divided into C mode and D mode, and the heterogeneous strategy includes implementation mode, implementation method and synchronization mode. Scheduling policies are classified into offline policies and online policies. Voting policies include voting algorithms, voting levels, and delay control.

In the fire detection system, the evidence of multiple sensors often conflicts with each other. If the classical DS (Dempster-Shafer) evidence theory is used for information fusion, it will lead to a high rate of misjudgment. In this paper, an improved DS evidence theory algorithm based on standard deviation is proposed and applied to the fire detection system to judge the fire state. This paper first describes the working principle of the system in detail, and then uses the overall standard deviation of sensor evidence to measure each sensor, judge the conflicting evidence items, then redefine the weight of each sensor according to the degree of conflict, modify the probability distribution of the conflict evidence. Finally, the DS combination rule calculation is performed to corrected Basic Probability Assignment (BPA) to obtain the fusion result. The experimental results show that the method proposed in this paper can effectively improve the evidence problems in the classical evidence theory, improves the fire identification ability and the speed of fire prediction, and greatly enhances the reliability of the system.

Stock investors have been making accurate predictions about the stock market in search of maximum profits. However, the stock market has a high degree of uncertainty, which makes it difficult to predict the development trend of the stock market. Existing stock prediction models generally improve the accuracy by changing the network structure and lack in-depth research on abnormal stock data. To solve this problem, we propose a logit-based stock prediction network LogNet, which uses the correctly predicted logits to measure the reliability of stock data, then calculate the confidence interval of the stock data, and use the credible data to make stock predictions. In addition, the model uses the theory of the Extremely Randomized Trees (ExtraTrees) theory to select the historical price data features of stocks. Experimental results show that LogNet has state-of-the-art performance on Twitter data and historical price datasets.

With the rapid development of society and the continuous change of the times, my country's Internet technology has achieved great development space. Cloud computing, Internet of Things, etc. have formed the current era of big data. At present, there are a lot of network access points, and the networked devices among them will generate a lot of data, so there are higher requirements for the security of the cyberspace, and it is also a golden opportunity. Traditional cyberspace security research is relatively inefficient in data processing, machine learning has strong adaptability, and strong learning capabilities also ensure the security of cyberspace. Under such a social background, it is necessary to give full play to the role of machine learning. After applying machine learning, it can effectively solve the problems existing in cyberspace security research. Based on this, this paper discusses the application of machine learning in cyberspace security research.

Multi-hop question answering aims to predict answers to questions and generate supporting facts for answers by reasoning over the content of multiple documents. The recently proposed Semantic Role Labeling Graph Reasoning Network (SRLGRN) has achieved excellent performance on multi-hop QA tasks. However, SRLGRN is lacking in modelling the textual relationships, which are import cues for reasoning. To this end, this paper proposes an enhanced SRLGRN multi-hop question answering approach by modelling textual relationships at different granularity (document relationships and sentence relationships). By modelling document relationships, a novel document filter based on document relationship threshold is designed for SRLGRN to dynamically select documents relevant to the question from multiple documents. By modelling sentence relationships, a sentence relationship-aware answer type prediction module is added to SRLGRN, which models sentences in documents as sentence graphs and then uses graph convolution network to predict answer type. The obtained answer type further guide the answer reasoning module of SRLGRN to obtain question answer with supporting facts. The experimental results show that the proposed scheme outperforms SRLGRN in terms of answer prediction and supporting fact prediction, with a 2% improvement in answer F1 metrics and a 3.1% improvement in joint F1 performance.

Personalized recommendation systems have high speed requirements, and most of the previous methods rely on powerful hardware to meet the real-time requirements. We algorithmically propose a fast recommender system. Our proposed method abandons the previous global comparison method, but uses a local comparison method, which greatly reduces the computational complexity. The experimental results also show that the effect of the method is not decreased, but improved.

The use of blockchain technology in high-elastic power grids often requires the deployment of blockchain nodes in charging piles, substations, new energy vehicles and other equipment, rather than in traditional servers. Limited by equipment performance and deployment conditions, the technical implementation of blockchain also needs to be optimized accordingly. In this paper, a novel hash scheme without multiple iterations which is suitable for high-elastic power grid is proposed. The hash scheme is constructed using large scale bool functions, which meet the requirements of balance, nonlinearity and SAC. The characteristics of the hash algorithm are also suitable for hardware implementation, which can build hardware micro services in the deployment architecture of high-elastic power grids.

This paper proposes a real-time vehicle location method based on multi-source information fusion of vehicle wheel speed features and visual images for the problems of weak texture features, high similarity of regional features, low accuracy of feature recognition and detection, and jump of vehicle location results in ship scene. Firstly, in order to solve the problem of inertial navigation sensor failed and get accurate pose estimation of vehicle motion, the article builds a vehicle motion prediction model based on vehicle wheel speed features. Secondly, according to the significant geometric features of the actual scene, Hough transform, edge extraction, template matching and inverse projection transform algorithms are used to identify and detect visual features of the scene. Thirdly, an extended Kalman filter is used to fuse vehicle wheel speed pose estimation and visual features to identify and detect the results to obtain reliable and high-precision real-time location of the vehicle. Finally, experiments are employed to verify the real-time vehicle location method based on multi-source information fusion of vehicle wheel speed features and visual images. The experimental results show that this method can effectively achieve vehicle localization in weak texture, similar regional features, and non-stationary road scenes.

In the information times, with the rapid development of network technology, emerging technologies such as big data and artificial intelligence have gradually entered daily life, but various network security risks have also followed. In recent years, as a new security strategy, mimic defense technology has been a new force suddenly rises, providing a new idea and direction for protecting network security. In the actual application scenario, the resource consumption of the mimic defense service is large, and it is generally deployed and provided by the cloud service provider. Although the mimic defense can play a good role in defending against external attacks, the data transmission in the current system is based on plaintext, which cannot guarantee the privacy of user data. Homomorphic encryption technology, as an encryption supporting data ciphertext operation, can protect data privacy on the premise of completing data operation requirements. As an important field of homomorphic technology, homomorphic hash technology has been widely used in cloud storage scenarios. Aiming at the problem that the cloud service mimic defense system cannot protect data privacy, this paper introduces homomorphic encryption technology into the mimic defense system to protect the privacy of user data. The scheme proposed in this paper is based on the mimic defense architecture, which optimizes the adjudication process of the adjudication module through the homomorphic encryption technology, make full use of the characteristics of homomorphic operation, protects the privacy of user data, and improves the system security. Moreover, this paper builds a mimic routing platform for experimental analysis. The experimental results show that the malicious attacker can see the user’s private information under the traditional defense mechanism. However, after homomorphic encryption optimization, each user authentication is still less than 1 millisecond, and the user information data become invisible to malicious attackers.

Trajectory prediction is crucial for collision avoidance and path planning in the autonomous driving task stacks. Numerous models have been proposed to learn the complicated spatiotemporal and social interactions among traffic agents. However, we find that some models perform badly in some corner scenarios, and it is related to insufficient supervision using single mean squared error (MSE) loss. Therefore, we propose the orientational MSE (OMSE) loss for better trajectory prediction. We introduce the global proxy matching strategy in OMSE loss to enable model to optimize with supervision from both Euclidean distance and orientation information. We evaluate OMSE loss on two public trajectory datasets for pedestrians and vehicles. Both quantitative and qualitative results illustrate the improvement on forecasting performance on cases of interest. The proposed method works in a learnable manner and surpasses traditional post-processing methods with more flexibility.

The software development of wheeled crane load spectrum calculation will involve a series of database design problems, especially the characteristics of many data table types, and large number of fields. Based on the above situation, a method of batch and fast database creation based on high-level programming language is proposed. Firstly, the requirement analysis and detailed design of the wheel crane load spectrum database are carried out, then the different database creation methods are analyzed and compared, the specific database creation algorithm flow is given, and finally the database creation software is developed in C++ language and a running example is given. The proposed database design method is efficient and labor-saving and is conducive to the rapid construction of data tables for large number of fields in the crane load spectrum.

End-to-end task-oriented dialogue systems rely heavily on an understanding of the dialog history. This often faces the challenge of inferring which dialog history information is critical to generating responses. In this paper, we address this challenge by leveraging a dialog history manager component that dynamically focuses on dialog history memory. It performs multiple add and forget operations by fusing an enhanced entity representation of dialog history and Knowledge Base (KB) information as queries, remembering entities relevant to responses and filtering out unimportant information. Experimental results on an open task-oriented dialogue dataset show that our model outperforms the baseline system in terms of effectiveness and produces contextually consistent responses.

The long sequence time-sequence forecasting problem attracts a lot of organizations. Many prediction application scenes are about long sequence time-sequence forecasting problems. Under such circumstances, many researchers have tried to solve these problems by employing some models that have proved efficient in the Natural Language Processing field, like long short term memory networks and Transformers, etc. And there are a lot of improvements based on the primary recurrent neural network, and Transformer. Recently, a model called informer which is made for the LSTF was proposed. This model claimed that it improves prediction performance on the long sequence time-series forecasting problem. But in the later experiments, more and more researchers found that informers still cannot handle all the long sequence time-sequence forecasting problems. This paper is going to look at how datasets effect the performance of different models. The experiment is carried out on the Bitcoin dataset with four features and one output. The result shows that the Informer (transformer-like model) cannot always perform well so that sometimes choosing models with simple architecture may gain better results.

With the rapid development of intelligent driving and on-board intelligent applications, the computing power of on-board units is gradually inadequate. Intelligent networked vehicles offloading tasks to cloud servers through the Internet of Vehicles is considered to be a promising method. However, long distance deployment of cloud servers and the instability of return links also bring high time delay. While Mobile Edge Computing (MEC) effectively solves this problem by deploying computing resources to the network edge. Therefore, based on the idea of mobile edge computing, this paper first constructs the local edge collaborative computing model. By comprehensively considering the factors such as user psychology, vehicle speed, acceleration, location, communication ability and computing ability, the utility function of task vehicle and service vehicle is established. Then, according to the Stackelberg game strategy, the interaction behavior between task vehicle and service vehicle is modeled, the Stackelberg cyclic iterative task offloading algorithm in the Internet of Vehicles environment is proposed. It is proved that there is a Nash equilibrium point between service vehicle and task vehicle. Finally, the simulation results show that the algorithm has achieved a balance between task delay and expense, task vehicle utility and service vehicle utility, and has higher performance than other algorithms.

Fisheries IoT realizes real-time online detecting and precise modulation of aquaculture process, which greatly enhances the informationization, automation and intelligence level of farms, reduces production costs, promotes the scale of fishery industry, and improves the quality level and market competitiveness of agricultural products. However, the key to large-scale fishery farming lies in accurate water environment modulation, the IoT gateway system is the key to realize the ability of intelligent fishery to achieve accurate water environment modulation. Once the aquatic product networking is attacked by malicious IoT commands, it may cause rapid disintegration of the farming environment ecosystem at any time, resulting in huge economic losses. Due to the size and power consumption of fishery IoT gateways, traditional security protection means are often difficult to be applied in such scenarios. Therefore, this paper proposes an improved malicious command detection method for smart fisheries sensor gateways based on deep learning techniques, aiming to accurately and efficiently identify malicious control commands of smart fisheries sensor networks on IoT gateways. The main contributions of this paper are three: (1) Transforming various types of signals from gateways into feature texts, which are used as inputs for deep learning for command classification. (2) Based on the BiLSTM model, it is improved by adding a multi-headed self-attention mechanism to obtain a smaller size and higher accuracy malicious command detection model applicable to smart fisheries gateways (3) Experimental results show that compared with other models, highly confusing malicious commands elaborated for the command characteristics of short, small size and high change rate of smart fisheries sensor commands The improved BiLSTM model has a higher recognition rate.

Distribution network has complicated grid structure and various kinds of faults，which make the current measures hard to accurately locate fault zones in field operation. This paper proposes a differential evolutionary algorithm to locate the fault based on the transient wave record data from the distribution network, using the mechanism of cooperative coevolution with penalty factor to optimize the solution set and punishment factor. The simulation result shows that the proposed method has good convergence and fault-tolerant ability when single point or multi-points fault occurs in the distribution network. In addition， it can provide important technical support for the fault position in the distribution network based on the transient wave data.

Exploration and development in the far-reaching sea of the South China Sea faces many challenges, such as great difficulty in well control. Once an oil spill accident occurs, it is necessary to deal with the oil spill from submarine wells in time to avoid major environmental pollution. In order to study the spreading range of submarine blowout oil and its influencing factors, this paper uses the Fluent platform, combining with the standard 𝑘െ𝜀 turbulence model and the multiphase flow VOF model, to preliminarily establish a submarine blowout oil spill model. This paper also analyzes the influence of ocean current velocity, blowout velocity, oil spill density on underwater migration trajectory and oil spill diffusion range. Numerical simulation results show that under the conditions of higher ocean current velocity, lower blowout velocity and greater oil spill density, the oil has a long underwater migration time, and the position where the oil first floats to the sea surface is far away from the wellhead level and will cause a greater range of oil spill pollution. This research is of practical significance for the rescue and rescue of blowouts and the emergency prediction and disposal of oil spills on deep water offshore platforms.

At present, the mesoscale simulation of waste fiber recycled concrete is regarded as a two-phase medium composed of recycled concrete and waste fiber. It is only studied in two dimensions, the three-dimensional fine and microscopic model of waste fiber recycled concrete is not established. In this paper, mesoscale numerical simulations on the compressive strength of waste fiber recycled concrete are performed based on Python. The recycled concrete is considered as a multiphase body consisting of aggregates, old and new mortar, and interface. A random distribution model of recycled aggregates based on the Monte Carlo method is written in Python language. Thus, the mesoscale three-dimensional (3D) numerical model of waste fiber recycled concrete is established. By comparing the results of the compressive strength test with the simulation, it can be seen that the model based on Python can more accurately reflect the damage pattern and compressive strength of waste fiber recycled concrete under pressure.

The medical intervention after a disaster event is the focus of government and citizen's attention in the public health and clinical medicine industries. For breast cancer, which has a high incidence in Russia and Belarus, the author combines the IoT management system and big data analysis to build a model of the framework and a preventive medicine system for a framework analysis and outlook.

The impacts of the Xinhengsha Reclamation Project on saltwater intrusion and freshwater resources in the Yangtze Estuary is simulated by Ecom-si. It is found that the salinity in the North Branch decreases slightly, the salinity in the North Channel and South Channel decreases obviously after the implementation of the Xinhengsha Reclamation Project. In addition, the implementation of the Xinhengsha Reclamation Project has a significant impact on the water intaking of the three reservoirs (Dongfengxisha Reservoir, Chenhang Reservoir, Qingcaosha Reservoir). In a spring-neap tide cycle, after the implementation of the Xinhengsha Reclamation Project, Dongfengxisha Reservoir is shortened from a maximum of 7 days before the project to 6 days, Chenhang Reservoir is shortened from 3.5 days to 0 days, Qingcaosha Reservoir is shortened from 4 days to 2.5 days. The above research results show that the implementation of the Xinhengsha Reclamation Project is beneficial to the water intaking of the three reservoirs.

This paper designs a log analysis platform based on big data technology, which uses flume log collection in Hadoop, and combines with HDFS for distributed storage, uses spark streaming to process real-time data, and finally stores the results into MySQL database, and Java Web technology to provide query and visualization functions. In view of the huge amount of data, the diversity of data storage formats and collection methods of network security logs in the current era, the log audit system based on the background of big data provides an integrated solution from collection, analysis, and storage to presentation.

The long-term prediction of stock prices provides an important reference for quantitative investment decisions. Aiming at the problem of insufficient accuracy of long-term series prediction in existing stock forecasting models, this paper proposes a long-term stock price series forecasting method based on PSO-Informer. First, 43 kinds of technical indicator factors and K-line data were selected to construct the input data, and then the PSO-Informer model was used to predict the future 60 time points of the stock closing price. In the model training process, the particle swarm algorithm is used to optimize the parameters of the Informer network. Based on the five-minute K-line data of the SSE 50 stock index and the CSI 300 stock index, experimental research was conducted respectively. Taking the accuracy of the long-term stock price prediction overall trend as the evaluation index, and the prediction accuracy reaches 68.2% and 67.5% respectively. The comparison experiments with ARIMA, Prophet, PSO-LSTM and Informer prediction models show that the model has the best performance and is practical.

Finding contours of interest from medical images is an important task in the field of medical image analysis. The current deep learning-based image segmentation approaches have obtained promising results. However, most of these models do not take into account the anisotropy and asymmetric features which play an important role in describing the target contours. In order to address this issue, we propose new loss-function applied to the deep learning model with dense distance regression, which can benefit the edge-based features, thus able to improve the stability of the segmentation procedure and to reduce the probability of outliers in the segmentation results. The introduced loss function is embedded into the deep learning model, which can perform an end-to-end image segmentation procedure for medical images. Ablation experiments were done with other loss functions and three datasets were used to verify whether this loss function is effective. SOTA results were obtained for the proposed loss function in this paper compared to the recently designed method for reducing the boundary error.

Conditional Generative adversarial net (CGAN) can generate data according to the given constraints. Convolutional neural network (CNN) is a neural network that can be used to process images providing learnable parameters for efficient classification. However, there are few researches on whether it is more beneficial for CNN to classify from the images generated by GANs as the training set in the past studies. To solve the issue, this paper compares the training effect of CNN classification of Handwritten-number images using real images, images generated entirely by CGAN that are results through CGAN’s trained from real handwritten-number images, and images mixed with them as training data sets. The dataset used is a famous dataset called MNIST which consists of 10 categories. Based on it, the corresponding experiments are carried out and the final experimental result indicated that the images generated by CGAN can further improve the performance of the CNN.

With the popularization of electronic medical record, internet medical care, AI medical image and other applications, the wave of medical digitization is coming. The electronization of medical data has brought convenience to the sharing of medical data, but it also makes the protection of personal sensitive data in medical data face unprecedented challenges. How to effectively protect personal sensitive information while sharing medical data. In this paper, a hybrid encryption algorithm based on SM4 encryption algorithm and Ciphertext-Policy Attribute-Based Encryption (CP-ABE) with revocable attributes is proposed. The algorithm first uses the SM4 encryption algorithm to encrypt the medical data, and then the SM4 encryption key is used as the encrypted data for encryption by CP-ABE. At the same time, aiming at the requirement of CP-ABE algorithm to revoke attributes, this paper proposes to embed a revocation table in the ciphertext to realize the direct revocation and restore of specific attributes of specific users. Through theoretical analysis and experimental verification, the revocable attribute hybrid encryption algorithm proposed has great advantages for the security and efficiency of personal sensitive information in cloud medical data.

The talent structure has been flattened, from the pyramid talent structure to the olive talent structure. Jobs are more complicated, industries are upgraded from low-end to high-end, and the situation faced by jobs is more complex. Talent specification compounding, industry understanding, data thinking, programming ability, innovative thinking and comprehensive problem-solving ability. With the rapid development of science and technology, the demand for social talents is constantly changing, and the big data wave is coming,How to match the talent training of the school with the talent demand of the society. This paper uses the crawler technology to obtain data from the recruitment website, through data collection, cleaning, and data preprocessing, and finally carries out data analysis and visual display.

With the development of satellite remote sensing technology, the quality and quantity of remote sensing images are constantly improved. Remote sensing feature classification is also playing an increasingly important role in urban planning, resource exploration and other fields. In the early stage of remote sensing feature classification, machine learning algorithms such as SVM and K-means are mainly used. Nowadays, with the expansion of deep learning, various kinds of research in the computer vision field emerge in an endless manner. Remote sensing images are also mostly classified by different neural networks. According to the characteristics and advantages of U-NET, channel attention mechanism, ResNet, large convolution kernel and structural reparameterization, this paper proposes a network structure called RA-UNET. This paper uses the remote sensing ground object classification dataset LoveDA to conduct experiments. The results show that the network classification effect of this paper is better, with mIoU reaching 59.4% and mPA reaching 72.6%. And use the network in this paper and the four mainstream neural networks of FCN, SegNet, PSPNet and UNet to conduct comparative experiments. The comparative experimental results show that the classification effect of the network in this paper is better than the above four mainstream neural networks.

The rapid development of edge network devices has led to the explosive growth of their data, and the difficulty of dealing with heterogeneous data in edge devices has been further increased. To solve the problem of heterogeneous data fusion without interaction, this paper proposes a data heterogeneous model analysis based on federated learning. Preprocess the multi-source heterogeneous data to obtain the main features of the condensed data. Then, the multi-source heterogeneous data nodes are positioned to avoid multi-fusion results, and Spatio-temporal correlation degree of the multi-source heterogeneous data is calculated to improve the accuracy of fusion. Finally, a multi-source heterogeneous data fusion model is established based on federated learning to ensure the security of data fusion. Compared with the traditional model, the data fusion of the proposed model is more stable, and the error is smaller. The effectiveness of the proposed model is verified by the stability and accuracy of the fusion of the heterogeneous data. The multi-source heterogeneous data fusion model studied in this paper can improve the quality of Internet of Things data and promote the development of edge devices in China.

In recent years, deep learning has been increasingly used to analyze financial data. For deep learning to predict the buy, sell, and hold points of stocks are prone to over-fitting, unreasonable feature extraction, and other issues. This paper builds a CBAM-CNN model based on Convolutional Neural Network (CNN) and Convolutional Block Attention Module (CBAM) to predict the buy, sell and hold points. In order to verify the applicability and superiority of the proposed method, the shares of Dao 30 and SHH 50 from stock listing to August 11, 2021 are selected, and the accuracy of the deep learning algorithm is evaluated using confusion matrix, weighted F1 score, and Kappa coefficient. The analysis results show that this algorithm has a high classification prediction accuracy because it can identify most of the buy and sell instances and therefore has a better effect. In addition, compared with CNN that do not use the CBAM attention mechanism, classification performance is significantly improved. The results from this analysis can help investors determine their better investment strategies.

In order to make the endpoint temperature prediction hit rate and velocity meet the requirements, in this paper, the Extreme Learning Machine (ELM) is used to build a model which predicts the endpoint temperature of the AOD refining furnace, and the model is trained with the real production data to search the nonlinear variation law between the input quantity and the endpoint temperature. The Sparrow Search Algorithm (SSA) is used to adjust the parameters of model to improve the hit rate of ELM. By comparing with other prediction models, found that the model is better than the pre-optimized temperature model in prediction hit rate and better than Support Vector Regression (SVR) in prediction velocity, which is more suitable for guiding actual production.

By replicating the CNN part of Galnet network and inspired by Swin Transformer, the article constructs a Transformer-based model called SwinT Galnet by adding the Swin Transformer block into Galnet thus introduces the attention mechanism into that traditional model, trying to investigate the improvement of Transformer over CNN in a single Sérsic galaxy model data. This article compares Sérsic parameters of SwinT with those of CNN. The paper compares the parameters trained by Galnet and SwinT Galnet and shows that the Transformer has an improvement by calculating global attention in accuracy over the pure CNN-based network in galaxy parameters regression tasks, indicates that Transformer has a large potential in the field of galaxy morphology.

The DNase I Hypersensitive site (DHS) is the chromatin region that exhibits a hypersensitive response to cleavage by the DNase I enzyme. It is a universal marker for regulatory DNA and associated with genetic variation in a wide range of diseases and phenotypic traits. However, traditional experimental methods have limited the rapid detection of DHS as well as its development. Therefore, effective and accurate methods to explore potential DHSs need to be developed urgently. In this task, a deep learning approach called iDHS-DPPE to predict DHSs in different cell types and developmental stages of the mouse. iDHS-DPPE uses a dual-path parallel integrated neural network to identify DHSs accurately. First, the DNA sequence is segmented into 2-mers to extract information. Then, the DHSs accurately-attention model captures remote dependencies and the MSFRN model enables hierarchical information fusion. The dual models are trained separately to enhance the feature information. Finally, the ensemble decision of two models yields the prediction results, enabling the integration of information from multiple views. The average AUC across all datasets was 93.1% and 93.3% in the 5-fold cross-validation and independent testing experiments, respectively. The experimental results demonstrate that iDHS-DPPE outperforms the state-of-the-art method on all datasets, proving that iDHS-DPPE is effective and reliable for identifying DHSs.

The innovative integration of beach sports and the tourism industry, as a breakthrough in sports tourism, is the current development direction of the tourism industry. With the development of the economy and the improvement of living standards, artificial intelligence technology is gradually applied in various fields. Using the data accumulated by the tourism industry, the application of big data technology in the tourism industry can be realized. First, the beach sports and the tourism industry are integrated; second, the risk assessment model is used to determine the economic profit trend brought by beach sports in the development of tourism; then, the validity of the model is verified by experiments.

In today's era of big data, the logistics supply chain generates massive amounts of data at all stages, and the privacy issues of logistics data are increasingly prominent. In order to efficiently utilize the logistics data of each enterprise to meet the needs of the enterprise and achieve secure data sharing, a federated learning-based logistics data sharing scheme is proposed. Using federated learning to federate multiple sources of data for modelling, the reputation value of each enterprise is stored on the blockchain and the enterprises that provide high quality data sharing are rewarded. Finally, the effectiveness of the scheme and the impact of data quality and algorithm selection on model training are verified through simulation experiments.

With the continuous development of the information age, big data technology has emerged in the context of a large number of information-derived clusters, and the advantages of its application in image processing have become increasingly prominent. This paper takes medical image enhancement technology as the landing point, based on the current situation and development trend of big data technology in image processing applications, analyzes its application principles, compares the effects of different medical image enhancement algorithms, discusses the aspects of big data technology that can be improved such as image transformation, restoration, compression, etc., focuses on the application of medical image enhancement and proposes countermeasures for the application of big data technology in the future, so that researchers can fully exploit the value of big data, promote the application of big data technology especially for image processing, and promote new developments in image processing technology, especially for medical applications.

In recent years, with the large amount of data generated, data fusion and sharing have become more and more important. However, the expansion of the scope of data fusion and sharing has led to the increasing problem of data leakage risk. In this paper, we improve the FSC fusion algorithm to evaluate the sensitivity change during data fusion from three aspects: data set, data items and data values, so that the fused data sensitivity is more accurate, and later the fused data is anonymized for sharing. The algorithm reduces information loss and provides better privacy protection.

Through detailed formula derivation process, the 6-order modal vibration spectrum is obtained. The axial dynamic characteristics of the winding are analyzed. During the analysis, the "mass-spring-damping" dynamic analysis model of the winding is established. Combined with the modal analysis method, the dynamic displacement response of winding cake under different loads is calculated, and the frequency spectrum is analyzed. The influence of the preload on the dynamic force of the cake is obtained, and the appropriate axial preload is finally determined. This paper attempts to deduce the formula from the theoretical level to obtain the relevant physical concepts and process parameter visualization in the vibration process of high-voltage power equipment. The relationship between the displacement data of key components and time is obtained by physical model inversion. In this paper, at present, the sinusoidal excitation function is used to approximate the field complex excitation waveform, and the closed analytical solution has been expected to be obtained. For the field complex excitation waveform, numerical simulation results are obtained through numerical simulation such as boundary element method. The research content of this paper has good guiding significance for the deformation of power equipment under the action of vibration signals and operation and maintenance measures.

Under the background of high precision requirements of image search technology, new image search technologies emerge in an endless stream. Image search is one of the important research directions in deep learning. Among them, chromosome image search is a key step in medical imaging and modern genetic methods. In this paper, the principle of image search technology is analyzed, and a VGG16 neural network model using attention mechanism CBAM is presented to search for chromosome images. Finally, the optimization model is experimented and analyzed. The conclusion is that the addition of attention mechanism enhances the ability of the model to extract chromosome image features and efficiently completes the image search task. This method provides a new idea for the development of image search technology in the future.

As the oldest and most famous cryptocurrency, the price of Bitcoin has increased nearly 2 million times in the last decade. As a result, predicting the price of Bitcoin through machine learning has become a big hit in recent years. This paper analyzes the correlation between the price of Bitcoin and market or social factors that may affect the price of Bitcoin. Then the author uses these factors with higher correlation to predict the price of bitcoin by LSTM. The experiments show that the average absolute percentage error of the LSTM prediction of bitcoin price decreases from 11.52% to 10.16%, 9.79%, 9.73%, 9.59%, 8.82%, and 8.50%, respectively, after the introduction of external correlation factors.

In this paper, we use Uncertainty Quantification method to analyze the data from three types of success-failure type test sources: physical test, semi-physical test and computer simulation test. Bayes method is used to evaluate the data fusion of multiple source test, and the validity and applicability of the method are verified by experiments.

Today, driven by new technologies such as cloud computing, big data, the Internet of Things and artificial intelligence, the era of artificial intelligence has arrived. "Digital transformation" has become the only way for enterprises to seek breakthroughs and innovations. As the carrier of information, the applications of data are fundamental. This paper discusses the database construction, data mining, analysis methods and the application of optimizing the database. Each of these applications will be discussed in detail.

Combined with the actual needs of scoring in line dancing competition, an Android-based competition scoring system is designed. The system is mainly divided into Android terminal and PC terminal. The administrator can enter the main interface of the PC terminal by entering the correct username and password, and perform user management, parameter setting, data import, printing score table, and data export operations. On the Android side, the judges can perform user management, parameter setting, scoring and other operations after entering the correct username and password. The implementation of the system effectively improves the efficiency of scoring in the line dance competitions and ensures the correctness of the final score.

First released as open source in 2009 by the pseudonym Satoshi Nakamoto, Bitcoin is the longest running and best known cryptocurrency. Bitcoin's transaction history is characterised by openness and transparency, and Bitcoin has become an important part of financial transactions. Therefore, it is increasingly important to be able to make accurate predictions about the development of the Bitcoin market. In this study, we construct a prediction model for bitcoin trading strategies based on the LightGBM algorithm, and show that our model has an accuracy of over 95.1%. The results show that our model achieves an accuracy of over 95.1% and has a higher performance compared to popular machine learning models.

The power grid business has the characteristics of wide coverage, diverse terminal equipment types, large numbers, and rich business scenarios. The blockchain technology itself has the characteristics of non-tampering and traceability, which is highly compatible with the opening of the power grid business and can be powerful supporting the digital transformation strategy of the power grid company. In order to ensure the authenticity and reliability of electric power data, data security and reliable storage technology was studied based on blockchain technology. On the data source side of the power system, a set of oracles mechanism for the grid data on-chain process was designed to achieve data reliability and consistent transmission. In addition, to prevent data damage or malicious tampering after power grid data on the blockchain, a hybrid data storage scheme based on shadow chain was proposed to realize data safe storage and improve value-added services such as data storage and sharing.

This research is based on the attention mechanism English translation adaptive model. After analyzing the key factors that affect English language translation, the attention mechanism is used to extract the detailed features of such factors in each region to form a feature sample set, and the feature sample set is fused and normalized, so as to obtain a brand-new feature sample set. Input to build an English language translation model and output the translation results, According to the results, the overall translation effect of the model is predicted. The results show that the prediction model of this method has high prediction accuracy in training and testing.

The problem of small amount data and sample imbalance exists in the machine learning prediction of the molecular properties of anti breast cancer candidate drugs. Proposing a FL-Lightgbm prediction model based on WGAN-GP data enchance model in order to solve this problem. Firstly, WGAN-GP model is used for data enhancement to increase the sample size of the training data set. Considering the small difference between positive and negative samples, the enhanced data of positive and negative samples are generated respectively, and then combined them according to the original order to ensure that the generated data and the original data maintain the same distribution; Then the Focal Loss function is introduced into the Lightgbm model to increase learning ability for unbalanced samples, the model constructed is called FL-Lightgbm prediction model. After the training of the enhanced data set, the proposed model shows excellent prediction accuracy for 178 randomly selected validation samples in the experiment, and its highest accuracy, AUC and F1 values reach 0.882, 0.851 and 0.7272 respectively. In these three indexes, the proposed model has better prediction ability than the original Lightgbm model with over sampling algorithms such as BorderlineSMOTE and ADASYN.

Cointegration is an important topic for time series analysis, especially in finance pair trading and hedging area. Cointegration is a kind of structure in which a linear combination of two (or more) time series is stationary. Traditional way to identify cointegration is to use the OLS estimator, firstly run a regression and secondly run a unit root test on residuals. But such method is easy to lead to ambiguous and unstable result. Therefore, we developed a dimensionality reduction model based on automatically calculated common factors and adopted the Metric Learning method to find a method that can quickly reduce the dimensionality and test the cointegration relationship of stock pairs.

The target tracking and object tracking are defined in this paper and the difference between multi-target tracking and multi-object tracking is also be illustrated. The Bayes filter, Kalman filter, EKF, JPDA and Hungarian Algorithm are introduced with formulars and an example of moving camera to track the pedestrians used by Kalman filter are shown. In this example, the method which is based on Kalman filter that track pedestrians from a moving car which is installed with camera in the field of the multi-object tracking is analyzed with steps. The algorithm initializes boundary boxes to track the pedestrians and predict the pedestrians based on the previous position. Then, update the tracks and delete the useless tracks. The final step is creating the tracks. After displaying the result, the algorithm based on Kalman filter can successfully track the pedestrians with boundary boxes. However, when the camera is moving fast, some of the pedestrians cannot be recognized.

While extracting dynamic 3D human key-points data for human pose estimation using only one RGBD camera, some key-points may be covered by other body parts during users’ movement. In this case, some key-points’ depth data can’t successfully be fused with aligned RGB data and that will influence final estimation of movement in this period. In this article, we present a method to predict users’ movement when camera can’t read exact 3D information of human key- points. This method proposes a Kalman filter model with constraints for predicting movement while there is a short sensor failure. The experiments show that these constraints can enhance robustness of system and help the filter to have a smooth performance. During sensor failure, 60% location data is well predicted.

It is an important way to improve expressway monitoring management level to realize continuous real-time tracking and monitoring of abnormal driving vehicles such as overspeed, long time and low speed occupation of overtaking lane, continuous lane change, and dangerous goods transportation vehicles under different cameras. This article through studies the expressway ray regard convergence across monitoring area joint tracking technology, build the camera comprehensive control, camera control balance compensation and the monitoring area based on target motion continuity principle joint track model, such as through the accurate control of radar to detect vehicle more monitor cameras, solves the continuous tracking target vehicle monitoring technical problems. It has been successfully applied in Yanchong expressway, providing more convenient monitoring services for the expressway management department, and providing strong support for the decision-making of the expressway management department, which has high application value.

Cell migration assay is the most common research approach for cell migration. Quantitative research and analysis are carried out by measuring the migration of cells into the region that is artificially created among confluent monolayer cells. To improve the efficiency and accuracy of the analysis, the software/tools were developed to assist the image analysis process. However, these software and tools are still at the stage of measuring a single sample, which cannot satisfy the requirement of large sample size for cell migration assay device. In this paper, an image analysis tool based on Fiji is developed, which can segment multiple samples from a scanned image and then analyze a single sample in batch. In addition, the screening function should be added for the application scenario of large sample size. The samples can be filtered according to different conditions to improve the consistency of experimental conditions. The results show that the developed analysis tool ATCA has high accuracy in identifying cell-free zones, with a difference of 2.3% from the tool WHST and 2.9% from manual operation. The analysis efficiency of this tool is 15 times that of manual operation.

Sign language recognition can make the communication between deaf mutes and healthy people more convenient and fast. In recent years, with the continuous development of deep learning, the research method of deep learning has also been introduced into the field of sign language recognition. Compared with the recognition of isolated words, the recognition of continuous sign language is more time-dependent. The current research still has shortcomings in recognition accuracy. Therefore, we proposed a continuous sign language recognition method based on 3DCNN and BLSTM. Based on the spatial feature information extracted by 3DCNN and the short-term temporal relationship established, the global temporal modeling of the video information of continuous sign language is carried out by using the bidirectional semantic mining ability of BLSTM. The CTC loss function is constructed to solve the problem of time series label misalignment. At the same time, we add the calculation of auxiliary loss function and auxiliary classifier. Experiments show that the auxiliary loss function and classifier can effectively reduce the error rate of the network. The word error rate of the continuous sign language recognition algorithm proposed in this paper on the large continuous sign language dataset RWTH-PHONEIX-Weather 2014 is as low as 23.5%, which is lower than the previous algorithm.

Existing deep learning-based Super Resolution (SR) reconstruction algorithms achieve remarkable performance on images with known degradation. Most of the degradation models exists problems in self-adaptations when facing with the deviation of the degradation model of the image of the real scene, and the effect is not good. Therefore, this paper proposes a blind image super-resolution reconstruction algorithm based on dual regression, which aims to solve the problem of poor performance of super-resolution networks in real scenes. Firstly, the closed-loop network is used to constrain the mapping space, and the optimal reconstruction function is found to improve the network reconstruction performance. Secondly, the attention mechanism is adopted into the residual block of feature extraction to expand the receptive field of the feature map, improve the reuse of features, and strengthen the reconstruction of high-frequency information. Finally, the frequency-domain blur kernel map estimates the down sampling kernel and reconstructs the low-resolution image, adaptively extracts the feature expression, enhances the ability to restore texture details, and reconstructs the real-world image better.

For remote sensing image segmentation projects, it is usually necessary to segment the boundary of ground objects finely, but the fitting of the boundary is a difficult problem in deep learning image segmentation. At the same time, for remote sensing images taken from high altitude, usually many large-scale targets in natural scenes will become small target objects in remote sensing images. To solve this problem, in order to improve the recognition effect of remote sensing image scene targets with huge scale differences, this paper combines UNet and FPN. This paper uses the framework of pytorch1.7.1 to establish the UNet + FPN model. Through the experimental results and visual analysis of building data sets and road data sets, the model in this paper has obvious advantages.

Named entity recognition (NER) is an essential part of many natural language processing applications and serves as the basis of natural language processing. NER tasks aim to identify the entity boundaries and types of specific meanings in texts, which include person names, place names, organizations, and dates. Due to the complexity of the composition of agricultural entities and limited quantity of agricultural corpora, often with low quality, agricultural text named entity recognition (ANER) is a more difficult and challenging problem compared to general NER tasks. Existing models often have poor performance in ANER tasks. To solve the existing problems of ANER, we constructed a named entity dataset of agricultural context through manual annotation using collected massive agricultural texts. The dataset contains 8873 entities composed of six entity categories: crop type, disease type, pest type, pesticide type, fertilizer type, and crop part type. We propose an agricultural text named entity recognition model based on BiLSTM-CRF. Experimental results show that the F1 score of the model on the test set achieved 91.20%, and our model also achieved the highest precision of 93.23% and the highest recall of 89.39% compared to other existing models.

Semisupervised learning (SSL) algorithms have received extensive attention in medical images segmentation. However, most SSL methods ignore the significance of boundary regions, and they cannot fully extract the significant image features, resulting in unsatisfactory boundaries and nonsmooth objects. In this study, we propose an SSL network, named Multi-Task and Feature Attention network (MTFA-Net). Our network can produce segmentation map and signed distance map (SDM) to fully mine the geometric information of object boundaries. To extract much more meaningful features, we devise a feature attention module and embed it into V-Net to discover the significance of each neuron, strengthen the important features, and suppress secondary features. Our experiments demonstrate that the method can achieve more accurate segmentation results than current methods on 3D left atrial image database.

With the rapid development of distribution lines, which is also accompanied by many problems, the most common is the problem of the ground fault. When a ground fault occurs, it will greatly affect the normal operation of the distribution line. At present, the existing grounding fault location methods for distribution lines have a large amount of calculation, which leads to a large relative error of the location results in practical applications and cannot meet the actual needs. Based on this, this paper proposes a method of grounding fault location for distribution lines based on transient signals. The Kaernbauer transform matrix is used to decouple the three-phase lines of distribution lines, and the real wavelet transform is used to extract the amplitude information of transient signals. The fault location is estimated by using the extracted amplitude information. The experiment proves that the relative error of the design method for grounding fault location is less than that of the traditional method, which provides strong theoretical support for grounding fault location of distribution lines.

Face attribute editing, one of the important research directions in face image synthesis and processing techniques, aims to photorealistic editing single or multiple attributes of face images on demand using editing and generation models. Most existing methods are based on generative adversarial networks, using target attribute vectors to control the editing region or Gaussian noise as conditional input to capture texture details. However, these cannot better control the consistency of attributes in irrelevant regions, while the generation of fidelity is also limited. In this paper, we propose a method that uses an optimized latent space to fuse the attribute feature maps into the latent space. At the same time, make full use of the conditional information for additional constraints. Then, in the image generation phase, we use a progressive architecture for controlled editing of face attributes at different granularities. At last, we also conducted an ablation study on the selected training scheme further to demonstrate the stability and accuracy of our chosen method. The experiments show that our proposed approach, using an end-to-end progressive image translation network architecture, obtained qualitative (FID) as well as quantitative (LPIPS) face image editing results.

Without any processing, the synthetic image visually unrealistic due to the established differences in the appearance of the foreground and background. In view of this situation, the task of image harmonization arises at the historic moment, and its purpose is to adjust the foreground appearance of a synthesized image to be closer to the background, thereby eliminating local visual differences. However, due to the limitation of the spatial feature interaction range in the feature extraction process, the global appearance transfer effect is not good. Therefore, to solve this problem, we propose an enhanced spatial feature interaction module. Meanwhile, we propose a back-projection up sampling module, which refines the reconstruction error during the reconstruction up sampling process and better restores the details of the reconstruction foreground. Our experiments on a public dataset, iHarmony4, show that the method effectively generates synthetic images with consistent overall appearance and enhanced detail.

At present, gesture has become an important channel of human-computer interaction, and gesture recognition has been widely used in various fields. In this paper, the dynamic gesture recognition technology is studied from algorithm and system implementation for portable devices which require high real-time performance. The algorithm mainly uses the region of interest extraction based on face recognition, skin color detection based on HCrCg color space and gesture motion track marking based on scanline seed filling algorithm. The system is implemented by Xilinx ZYNQ, and a SOPC system architecture based on ARM Cortex-A9 hard core and ARM Cortex-M3 soft core and FPGA is proposed. The scanline seed filling algorithm with long running time is designed as a hardware accelerator to improve the running speed. Through the test of the prototype, the recognition accuracy can reach 95.75% in a simple background and 90.83% in a complex background. The average running time of the system is only 0.68 seconds, which is more than 30% faster than using pure software method. The system has good performance in recognition accuracy and running speed.

To make computers understand the molecules, the first and important thing is to represent molecules in a proper way, which will affect the efficiency of chemistry tasks like property prediction and molecular design. In this work, we introduce a molecular representation for noncrystalline small molecules based on the theory of quantum physics. This representation captures the microscopic spatial structure of the molecule, which ensures it reflects more visual perception information about the molecule. We use Drug3DNet as our baseline and test the efficiency of our representation. By comparing with several other representations, we prove that our representation performs better on most of the properties.

There are many subjective influencing factors, poor recognition effect and low efficiency in manual evaluation of pipeline weld defects. An intelligent identification method of pipeline weld defects based on improved DenseNet network is proposed. This method firstly uses the form of multi-channel convolution of different scales to improve the DenseNet network, thereby improving the generalization ability of the network. Then, the feature extraction ability of the network is improved by stacking two convolutions of the same scale. Finally, an attention mechanism module is introduced into the dense connection block of the network to achieve the effect of improving beneficial features and suppressing useless features. The experimental results show that the method can achieve 92% accuracy in the identification of pipeline weld defects, which is about 13% higher than the original method, and has high efficiency, which can fully achieve the purpose of industrial application.

Recently, video surveillance system network security incidents have occurred frequently, all communication levels of video surveillance systems face severe information security risks and have many vulnerabilities. So, a vulnerability scanning system of video surveillance device is proposed. This paper presents the architecture of vulnerability scanning system, describes the process of detection model, research on the verification of vulnerability, and finally carries out an experiment of the vulnerability scanning system.

Entity extraction is an information extraction technique that aims to locate and classify named entities (e.g., organizations, locations, persons...), which is a very important and fundamental problem in natural language processing. On the research of entity extraction, numerous models ignore the learning of grammatical structure. Considering the shortcomings of previous models, this paper first proposes the PALC (POStag-Attention-LSTM-CRF) model, which adds POS (part of speech) features to entity extraction. Specially, PALC fuses POS features with other features through a multi-layer bidirectional LSTM network and attention mechanism to improve the effect of entity extraction. The experimental results show that the accuracy of the PALC model in this paper on the CONLL03 dataset can be 90.65%, on the CONLL03 dataset can be 84.86%, and on OntoNote 5.0 English dataset can be 86.99%.

With the rapid development of high-precision sensors such as LiDAR and Kinect, point cloud has become the main data format to represent the three-dimensional world. And point cloud registration is an important step in the processing of these data. However, there are problems that wrong corresponding points, low registration accuracy and slow speed in the process of point cloud registration. We propose a fast point cloud registration method based on ISS-Shot feature descriptors, SVD algorithm, and an improved ICP algorithm. The proposed method firstly uses a voxel grid filter for down-sampling. Next, the feature points are extracted by the ISS algorithm and described by the SHOT algorithm. Afterwards, the ISS-Shot features are used for rough registration with SVD algorithm. Finally, the ICP algorithm with normal vector constraints and KD tree acceleration is used for accurate registration. The experimental results show that the proposed algorithm has higher registration accuracy and faster registration speed than the compared algorithms.

In this paper, an multi-channel acquisition device suitable for domestic Intelligent Programmable Logic Controller (IPLC) is developed. The GD32F407RE of GigaDevice is used as MCU control system and the AD chip SGM5208 of SGMICRO is used. The upper computer based on Loongson CPU realizes data interaction with the acquisition device through CAN communication. In the experiment, the precision current and voltage source is used to verify and test the developed multi-channel acquisition device. The device can realize the acquisition and transmission of DC analog 0-5V / 0-20mA , achieving a measurement accuracy of 0.1%. The developed multi-channel acquisition device achieves reliable operation in engineering applications. The influencing factors of measurement accuracy and real-time are analyzed, and the software is optimized. The device is designed based on domestic electronic components, which ensures the safety of the chip supply chain.

The security of web components is increasingly concerned by the industry, and identifying web components is of great significance to both network attack and defense. Aiming at this problem, we propose a novel identification method of web components based on machine learning. We transform the Web component identification problem into a classification problem. On this basis, we further construct the framework for realizing this method, and implement the framework. The experimental results show that the application of this method can have a good recognition effect on Web components.

In the field of item recommendation, traditional collaborative filtering-based approaches face the problems of data sparsity and cold start. As a heterogeneous network with multiple entities and relationships, the knowledge graph has rich semantic association information to help explore reasonable associations between users-items and items-items and assist recommendation systems in alleviating the problems of sparsity and cold start. Graph neural network-based approaches can explicitly encode key collaboration signals for user-item interactions and have been proven to be effective recommendation methods. However, previous work has not explored how the entity superordinate words and synonyms of the knowledge graph affect the metrics. In this paper, we discuss and experiment from two perspectives of knowledge graph construction and graph neural network embedding to explore how to improve the recommendation metrics comprehensively. In this paper, we construct commodity knowledge graphs and propose the model called TransD-KGAT, which improves three indexes compared with the KGAT model in terms of precision, recall, and ndcg by 2.64%, 4.96%, and 3.9%. It is also concluded that adding knowledge graphs of semantically similar and superordinate words improves the recommendation metrics.

The domain of computer vision's most popular study area has always been face recognition which aims to identify different face images and predict the corresponding identity information through feature analysis and modeling. In practical applications, a number of variables, including lighting, posture, and clarity, have an impact on facial recognition accuracy. Among them, the most challenging scene is occlusion face recognition, which will cause feature loss, local coherence and alignment errors to greatly inhibit the accuracy and generalization ability of the face model. Based on detailed literature research and analysis, this paper provides a comprehensive evaluation of the research progress of occluded face recognition. Specifically, based on the introduction of classical face recognition technology, we further discuss the design ideas, basic framework, advantages and disadvantages of representative occlusion face recognition methods from two aspects: robust feature extraction and robust classifier. Finally, we summarize the main challenges and give an outlook on the future research development of object detection.

Gender recognition of face images is an important research field in image classification. Convolutional Neural Network (CNN) has a good performance in this field. To obtain successful results, large and high-quality data sets are essential. However, data scarcity has always been a common challenge in deep learning. The Deep Convolutional Generative Adversarial Network (DCGAN) has a powerful application in data enhancement. The face image generated by DCGAN can be used to enhance face data. In this study, the fake face images generated by DCGAN and the real face images of CelebA datasets are used to train CNN classifier for gender classification of human faces. In the experiment, the training set and test set both are divided into three types, including real face images, fake face images, and real-fake mixed face images. The type of test set is the same as that of the training set, and the number of face images in both the training set and test set is small, including 1000 face images. By comparing the training accuracy and test accuracy of datasets under various conditions in CNN classifiers, it can judge whether the fake face images generated by DCGAN have a good data enhancement effect. The experimental results show that the data enhancement ability of DCGAN is also well applied to CNN classifier, and the face images generated by DCGAN can effectively improve the performance of the CNN classifier to a certain extent.

Wild animal recognition is important for wild animal protection. Because the number of different wild animals is different in the wild. The wild animal image dataset collected in field by using camera trap is a typical long tail dataset. This paper proposes an Effective-Class-Balanced Softmax Loss (ECBSL) to solve the long tail problem of self-built wild animal dataset. Firstly, a new cross entropy loss function is obtained by using pointwise mutual information instead of conditional probability for modeling. Then the improved effective number of samples calculation method is used to approximately calculate the prior probability distribution of different animal species. Finally, the effectiveness of ECBSL is proved by experiments. Experiments on the self-built wild animal dataset show that the proposed method improves the recognition accuracy of the tail classes and the whole dataset. The comparison experiments with other methods show that the proposed method is superior to other methods.

Person Re-Identification (re-ID) is a task that involves matching individuals captured by various cameras and in various poses. And because of the uncertainty of input image resolution in practice, traditional Person Re-Identification tends to have poor performance. Cross-Resolution Person re-ID is a hot branch of Person re-ID that aims to address the image with a different resolution. Besides, Person re-ID has been applied in lots of fields because of its practicality, including public security systems. Therefore, this poses a challenge to the security of the re-ID model. This paper discusses the results of attacks on PS-HRNet, VGG16, and ResNet networks using the Fast Gradient Sign Method. The experiment result of this paper proves the effectiveness of the Fast Gradient Sign Method on Cross-Resolution Person Re-Identification. It tests the transferable of the adversarial samples generated by the Fast Gradient Sign Method.

Intelligent anti-jam communication is a new generation of anti-interference technology combined with artificial intelligence, and the identification of interference signals is the basis of the technology. It is required to achieve better identification results with lower computational complexity in engineering applications. However, previous research has shown that they cannot balance these two sides. Here, we report an interference signal identification algorithm based on Extreme Learning Machine (ELM). Five typical oppressive interference signals were recognized based on ELM which is based on feature extraction. The overall correct identification rate is more than 96% under the condition of 40 neurons in a single hidden layer, and it has certain generalization ability. This study objectively promotes the engineering application of this technology.

With the development of Industrial Internet of Things, the types and functions of components are increasing, the application environment is becoming more and more complex. Also, the quality management of components is becoming more and more important. In order to understand the knowledge related to component quality management more conveniently and build an intelligent system for component quality management, this paper proposes a method to construct component quality management knowledge graph based on BERT word embedding model and entity relationship joint extraction method based on annotation strategy. Combining entity extraction and relationship extraction parts into one not only reduces the consumption of computing resources, but also reduces the propagation of wrong entities. In this paper, the sequence to sequence model of Bert-BilSTm-CRF is adopted. Through the BERT word embedding layer, the context information can be better utilized and the accuracy of extraction can be improved. Experimental results show that compared with other classical deep learning term extraction models, this model has a significant improvement in accuracy, recall rate and F1 value.

Rapid detection and identification of landslide areas are very important for disaster prevention and mitigation. Aiming at the problems of time-consuming and labor-intensive traditional landslide information extraction methods and low recognition efficiency, a remote sensing landslide recognition method based on LinkNet, and convolution attention module was proposed. The model adopts the coding-decoding structure to improve the operation efficiency. The Convolutional Block Attention Module (CBAM) is applied to optimize the weight allocation from both channel and spatial dimensions to highlight the landslide feature information. And compared with the traditional U-Net and LinkNet models. The results show that the CBAM-LinkNet model has excellent performance in remote sensing landslide identification, which provides the possibility for rapid and accurate landslide identification.

Aiming at the problems that the method based on U-shaped network for medical image segmentation cannot capture the long-range dependencies and could lose some detail information, a multi-scale context-aware segmentation network for medical images is proposed. The model extracts the last three layer features of the encoder, and then introduces a global circular convolution transformer module to solve the problem of long-range dependencies capturing by modeling the global context information. Then, an attention guidance module is introduced to fuse features of different scales, so as to solve the problem of losing details while reducing the introduction of noise information in the low level features. The experimental performance on Synapse multi-organ segmentation datasets indicates that the model produces more precise segmentation results.

Autonomous Underwater Vehicles (AUVs) are important equipment for ocean development and exploration. To ensure the task implementation of AUV, faults shall be detected in time. We propose a fault detection method based on Multiscale Spatiotemporal Feature fusion (MSF) for the time-varying characteristics and multiple correlation characteristics of AUV monitoring data. First, we apply a variety of sampling and data processing methods to generate monitoring windows with different scales along the time axis. Then, a composite feature extraction method is proposed to obtain temporal and spatial features simultaneously, and a feature pyramid of temporal and spatial information is formed. We use Bidirectional Long Short-Term Memory (BiLSTM) to obtain the time-series characteristics of a single monitoring variable, and Convolutional Neural Networks (CNN) to obtain the implicit spatial relationship characteristics among multiple monitoring variables. Next, we use an adaptive feature fusion method to solve the inconsistency in different feature scales, which can adaptively suppress the possible conflict information of different scale features. Finally, we use a fully connected network to detect the fault of the fused features. The fault detection experiment of Haizhe AUV shows the effectiveness and superiority of the proposed method.

With the continuous development of research in the field of machine learning, especially the progress in deep learning and the continuous improvement of arithmetic power such as image processors, the recognition technology using biometric big data has gained wide attention and has been well applied in many fields such as human-witness matching, intelligent monitoring and epidemic prevention and control. The development trend of big data biometric identification technology is analyzed, the types of biometric features and the development and application of big data-driven biometric identification technology are summarized, and the future development trend of big data biometric identification technology is discussed.

High-standard and high-quality products are not only the foundation of the enterprise, but also the core competitiveness of the enterprise, which is a key factor for the enterprise to achieve sustainable development in the context of the new era. In recent years, as the production scale of Enshi Cigarette Factory has been expanding, the traditional detection and evaluation mechanism of cigarette quality stability, to a certain extent, can no longer meet the production environment well. Therefore, this paper introduced machine learning technology into the study, selected four cigarette types as modeling objects:HUANGHELOU (81235), HUANGHELOU (81247), HUANGHELOU (81183) and HUANGHELOU(01103), and took factors affecting cigarette quality as modeling objectives, so as to find the key factors affecting cigarette quality.

The use of LIDAR at sea is not only influenced by the light absorption properties of the water surface. The returned point cloud data is sparse, and the resolution is constantly changing. Nevertheless, it also produces invalid point cloud clutter due to the dynamic waves and bubbles on the water surface. For this reason, we propose a method for LIDAR obstacle detection after removing water surface clutter. The preprocessed point cloud data are grouped and sorted, and the DBSCAN algorithm is used to cluster the point clouds within different groups. The surface waves and bubbles are also removed according to the intensity variation values of each clustered point cloud, and then the obtained point clouds are mapped to fit the obstacle point cloud clusters to present a bit-pose closed frame. Simulation experiments and actual marine experimental tests prove that our method can effectively remove surface clutter waves and improve the positive detection rate by 86.36% compared with the conventional DBSCAN algorithm while reducing the false detection rate and the missed detection rate. The rate is improved by 12.91% and 11.71%, respectively, and can detect various water surface obstacles around unmanned vessels stably and accurately in the real environment.

Agricultural pest identification occupies a key position in agricultural economy and development. The accurate identification of pests is the premise of agricultural pest control. In recent years, image processing technology and deep learning technology have rapidly developed. Some researches have been applied to the field of insect recognition. Thus, some insect recognition deep learning models with good recognition accuracy and speed have been established. However, there is still much room for improvement when they were applied to the insect monitoring system deployed in the field. Considering the target recognition accuracy and speed, this paper selects the target detection algorithm YOLOv4-Tiny as the base model for insect recognition. The major advances are the attention mechanism and Spatial Pyramidal Pooling (SPP) structure as shown in: applying Convolutional Block Attention Module (CBAM) reduce computation and number of parameters; adopting SPP structure multi-scale pooling of input feature layers which increases the perceptual field and improves the robustness of the model. The experimental results show that the improved YOLOv4-Tiny model can significantly enhance the insect recognition accuracy.

Cracks in the rails will lead to great safety hazards in railway transportation. Aiming at the problems of low detection accuracy and inconspicuous part of cracks in crack detection, an improved model based on YOLOX-Nano is proposed. The SA-NET lightweight combined attention mechanism is added to the model to generate a feature map with channel attention and spatial attention, which strengthens the model's attention to target features and location information. Secondly, use Alpha-CIoU Loss to replace IoU Loss to increase the accuracy of the model's prediction box. The comparison experiment was carried out on the self-built data set, and the mAP of the improved YOLOX-Nano model reached 77.58%, the detection speed reached 42.2FPS, and the calculation amount and parameter amount of the model were only 0.508G and 3.5MB respectively, and the overall performance was better than other models.

Powered exoskeletons are a kind of wearable robotics system attached outside the limbs, providing additional force for the limbs, which plays an important role in limb rehabilitation and patient assistance. As the patient gradually recovers, the patient’s gait will change over time. Therefore, the author hopes to get real-time gait data from patients in order to provide medical guidance and help the patients recover better. Inspired by the real-time monitoring of industrial robots, the author puts forward a method of medical monitoring using digital twin onto powered exoskeletons. For cost reasons, the author uses commercially available sensors to build this system. Further, the author fabricates a demonstration system for the exoskeleton to achieve this goal together with collecting and analyzing the data.

In this paper, we consider the problem of scheduling an automatic guided vehicle (AGV) in an automated storage and retrieval system (AS/RS). In this system, the AGV acts as a transfer station and performs the task of depositing and taking out goods. It is assumed that the number and location of shelves are known, as well as the number and type of goods on each shelf. The AGV performs transport tasks with decisions that include finding the optimal transport sequence between different shelves and the optimal sequence of goods within each shelf. Therefore, these decisions jointly minimize the total transport distance. Using mixed-integer linear programming (MILP), the AGV can not only complete the task with the shortest working route but also return to the sorting station with the shortest path after completing the task. Using MATLAB software to simulate the two cases, the results of obtaining the optimal solution verified the correctness of the algorithm. The results show that the algorithm has high performance for optimizing AS/RSs.

The synthesis of human movement is used in a wide range of applications, such as in military, gaming, sports, medical, robotics and many other fields. The current common approach is based on the acquisition of human motion data by motion capture devices. However, employing this method to gather information on human movements is expensive, time-consuming, and subject to space constraints. In order to avoid these problems, our goal is to create a system that can generate a variety of naturalistic movements quickly and inexpensively. We assume that the creation of human motion is a complicated, non-linear process that is amenable to modeling with deep neural networks. First, using optical motion capture equipment, we collect a range of human motion data, which we then pre-process and annotate. After that-we combine Transformer with Conditional GAN (Cgan) to train this human motion generation model with the collected data. Finally, we evaluate this model by qualitative and qualitative means, which can generate multiple human motions from a high-dimensional potential space based on specified labels.

Wi-Fi is a popular wireless local area network technology, which has the characteristics of convenient networking and easy expansion. The existing data remote monitoring system mainly uses ZigBee technology to transmit monitoring data, and the response of the monitoring system takes a long time. Therefore, this paper proposes a remote monitoring system based on Wi-Fi technology. Firstly, a framework including intelligent perception layer, data communication layer and data integration layer is designed to realize the real-time data acquisition of the Internet of Things. Then, a data communication mechanism with high transmission rate is established by the Wi-Fi technology to realize the wireless transmission of monitoring data. Finally, the abnormal data judgment module is designed by using BP neural network to further analyze the real-time data of the Internet of Things. The abnormal monitoring results of the real-time data of the Internet of Things are obtained, and the monitoring results are presented through a visual interface. The system test results show that the total response time of the proposed system is 7440ms, which is reduced by 37. 2% and 42. 89% compared with the CAN-based and PLC-based systems. At the same time, the system realizes the intelligent analysis and efficient monitoring of Internet of Things data and promotes the development of data remote monitoring technology.

Collaborative knowledge innovation activities are conducive to the rapid development of knowledge economy. However, because the collaborative knowledge innovation is usually hidden in the complex network information transmission process, the efficiency and quality of knowledge innovation behaviors may be greatly affected. We take collaborative innovation participants, collaborative innovation teams and collaborative innovation achievements as the constituent elements and construct the research framework of collaborative knowledge innovation mode mining. Then, we use Apriori algorithm to mine the collaborative knowledge innovation mode and obtain the transformation mode. The results show that when the main contributors to a project are high active users, the project has a greater probability of showing a trend of high innovation activity and is more likely to become a high-quality project. The findings will help to improve the collaborative knowledge innovation ability of online community platforms and the efficiency of knowledge diffusion.

This paper designs an intelligent blind people guide cane based on the analysis of basic functions of conventional guide canes. The hardware components of the cane design includes environmental monitoring modules, acoustic distance measurement module, vibration alert module and GPS positioning module etc. The system can provide real-time alerting of obstacles ahead and route selection through analyzing road conditions by means of GPS positioning module. The cane’s hardware system is designed with monitoring and analysis software to enhance the user’s using experience, while ensuring the functions of user security.

The increasing structure of neural networks makes it difficult to deploy on edge devices with limited computing resources. Network pruning has become one of the most successful model compression methods in recent years. Existing works usually compress models by removing unimportant filters based on the importance. However, the importance-based algorithms tend to ignore the parameters that extract edge features with small criterion values. And recent studies have shown that the existing criteria rely on norm and lead to similar model compression structures. Aiming at the problems of ignoring edge features and manually specifying the pruning rate in current importance-based model pruning algorithms, this paper proposes an automatic recognition framework for neural network structure redundancy based on reinforcement learning. First, we perform cluster analysis on the filters of each layer, and map the filters into a multi-dimensional space to generate similar sets with different functions. We then propose a criterion for identifying redundant filters within similar sets. Finally, we use reinforcement learning to automatically optimize the cluster dimension, and then determine the optimal pruning rate for each layer to reduce the performance loss caused by pruning. Extensive experiments on various benchmark network architectures and datasets demonstrate the effectiveness of our proposed framework.

Identifying influential nodes is one of the crucial issues for controlling the network propagation process and exploring network properties in complex networks. Nevertheless, the accuracy of existing methods is still a challenge. In this paper, we rank influential nodes by considering tow aspects. On one hand, a normalized degree centrality is proposed to measure the local influence of each node. On the other hand, an improved fine-grained K-Shell decomposition is defined to measure the spreading ability of neighbors of a node. Further, a novel ranking measure is proposed by combining the normalized degree centrality and fine-grained K-Shell (NDF-FKS). The Susceptible-Infected-Recovery (SIR) model is used to simulate the network propagation process. Experiments with the model are performed on eight synthetic networks and four real networks. The NDF-FKS compared with six measures for accuracy and resolution. The results show that the accuracy of NDF-FKS outperforms existing six measures and has a competitive performance on distinguishing influential nodes.

MRI and CT are both important medical imaging modalities, but MRI and CT imaging are done in different ways, each with its own advantages and disadvantages. Obtaining both images at the same time can help physicians make better decisions about treatment options. However, due to various limitations, some patients can only obtain one type of image. Therefore, it is necessary to find a well-performing GAN to transform MRI and CT images. In this paper, the effect of Cycle-GAN with different activation functions is compared, such as LeakyRELU, and different number of layers in MRI-CT conversion. Also, this article compares the effects of Cycle-GAN and UNet-GAN. The results indicate that the Cycle-GAN model using LeakyRELU as the activation function is better than the Cycle-GAN model using RELU as the activation function. Second, the effect of deepening the layers of the GAN model is worse than that of the base model. And the effect of UNet-GAN is similar to that of Cycle-GAN. This is not quite as expected, because Cycle-GAN has one more discriminator than UNet-GAN, and the effect should be better. But the experimental results do not confirm this conclusion.

It is a challenge to detecting insulators quickly and correctly from complex and variable images of power line. An insulator detection method based on DN-DETR model is proposed in this paper. DN-DETR is an end-to-end object detection model based on the transformer model. Self-attention of transformer can obtain the global information of image quickly and correctly. Experiments show that our model converges quickly, and precision achieves 89.6% at 50 epochs.

Typical challenges that aircraft overhaul enterprises faced are the lack of timely, accurate scheduling of operational and inter-operational tasks in shop-floor level during the whole overhaul process, resulting in long overhaul cycle, rising cost, and labor. Therefore, it is extremely important to establish rapid-response troubleshooting methods considering the complexity of overhaul process uncertainty and disturbances to improve shop-floor performance. To address these challenges, a novel agent-based Aircraft Overhaul Adaptive Scheduling Framework (AOASF) is proposed to realize the rapid assignment of aircraft overhaul tasks. Within this framework, a scheduling method using Q-learning is designed, which aims to rapid-responding the disturbances in time and achieve adaptive scheduling. A simulation system is implemented to verify the effectiveness and superiority of the AOASF and the scheduling method. The simulation results show that the method is superior to conventional scheduling methods in terms of make span and Percent Tardy (PT), especially when tasks arrive frequently.

In the treatment of substations, it is very crucial to make a reasonable arrangement of route used for the maintenance of each substation. Moreover, given the urgency degree of different substations, the priority of each substation should be carefully considered for a good arrangement of route used for the maintenance. In this paper, considering the complexity of the routing arrangement, Genetic Algorithm (GA) and Ant Colony Optimization (ACO) were adopted with the designed priority coding methods and priority constraints for a more reasonable arrangement of route. Moreover, with the analysis of the performances of GA and ACO on the priority-based routing arrangement, a fused method was designed to obtain a good routing arrangement in an efficient manner. The experimental results show that, with the designed priority coding method and the priority constraints, a more reason result can be obtained by the fusion-based method.

The unattended operation of substations has become the development trend of today's power system. The traditional substation management system has outstanding problems such as complex management process and inability to realize remote unmanned control. In order to solve the above problems, electronic visa technology is proposed to realize remote unmanned permit, to realize real-time detection of remote specific targets using detection algorithms, and to track and record specific targets and behaviors using target-tracking algorithms. This system has been promoted and used in various substations in Yuxi City, Yunnan Province, solving the problem of remote safety control in the process of power production.

Deep learning, the critical part of machine learning, has become influential in different fields, including natural language processing, computational biology, and computer vision. In the last decade, there has been a massive surge in computer vision research since its related application is so promising. Many have proposed various methods to fulfill the automation of driving based on deep learning, but, up until now, there is still a gap between the virtual and reality. This paper focuses on its application in autonomous driving. A new framework is proposed to fill that gap using a deep neural network. Specifically, instead of using the raw images captured by cameras to make decisions, semantic segmentation is applied first to get intermediate products that can better connect virtual and reality. Considering the road landscape needs to be mainly treated, the pre-trained model PSPNet is used to process the original image data. Then this data is provided as input to a deep CNN model for feature extraction and prediction. Compared to the traditional method, a semantic segmentation process is added to help extract useful information within an image and is expected to bring some positive effects.

With the widespread use of sensing elements in commercial equipment, action recognition technology is required to be more practical in people's life, especially the stable and accurate recognition. Among them, using sliding window for motion perception is an effective recognition method. However, most of the current recognition models are designed for a single action, which not only has poor recognition stability, but also cannot effectively recognize the action. This paper presents a method of action recognition based on adaptive window and broad learning, and designs an action recognition system EVM, the system effectively preprocesses the action data and realizes the accurate recognition of actions. Firstly, EVM smooth the source action data. Then, this paper proposes an extreme value filtering method to avoid the interference of peak/valley extreme points and ensures the effectiveness of action division through the adaptive window. Finally, a recognition model based on broad learning is used to classify action behaviors. According to the comparison and verification of a large number of experiments, the EVM system has a recognition accuracy as high as 97.91%, which is much better and faster than the CNN model.

In this paper, monthly frequency multi-factor data on valuation, momentum, turnover rate and technology of A-share listed companies in Shanghai and Shenzhen markets from January 2012 to July 2022 are selected and input to LSTM and LSTM model with fused attention mechanism respectively for training after data pre-processing. The sector-neutral layered-portfolios and the sector-neutral stock selection portfolios were constructed based on the model output, respectively. In the model evaluation section, it is confirmed that the Attention-LSTM model outperforms the LSTM model in predicting stock ups and downs. The single-factor layered back test under monthly position adjustment and stock selection strategy back test confirmed that the Attention-LSTM model significantly outperformed the LSTM model in terms of annualized return, sharpe ratio, and maximum retracement, and also significantly outperformed the CSI 300 and CSI 500.

Patients with early-stage skin cancer have a 99% survival rate, but once it spreads, the chances of survival are poor. Deep learning has greatly improved the quality of computer-aided diagnosis of skin diseases. However, the majority of existing methods do not take into account the influence of the extracted features on the classification results. In this paper, we proposed a deep neural network with Multichannel Attention amplification Fusion (MAF) module for skin diseases classification. First, the images go through our specially designed pre-processing process which retains more basic image information. The features extracted by the deep neural network are fed into MAF which is designed to amplify the role of key feature points in the classification to improve the effectiveness of classification. Our model achieved an accuracy of 88.13% and 82.83% on the ISIC2016 dataset and ISIC2017 dataset, respectively. We also performed ablation studies which showed that our proposed pre-processing method and the MAF module effectively improve the expressiveness of the model.

In the area of artificial intelligence, automatic driving is a significant study area. Self-driving cars are vehicles that can sense and respond to surrounding environments to guarantee a safe and machine-controlled drive. However, the cost of a driverless car is considerably higher than a typical car, preventing the further promotion of such technology. The situation could be improved if there is a way to reduce the cost of sensors. This paper uses Machine Learning Agents to create a self-driving agent. The agent is trained under the Proximal Policy Optimization policy. In addition, a randomly generated map is constructed to improve the robustness of the agent. After training, the agent can drive the car without hitting walls and obstacles with six sensors. As tracks and blocks are randomly created, the driver agent applies to real-life situations and could be used as simulations for real self-driving cars.

With the encouragement of national policies and the improvement of the public’s awareness of environmental protection, the number of electric vehicles in our country has increased significantly. It is necessary to reasonably add new charging stations. This paper uses machine learning methods to remove private charging pile data from the original vehicle driving data and combine the characteristics of the charging points to obtain the location of the charging demand point, and then use hierarchical clustering algorithm to select the location of the new charging station and use the machine learning model to provide a reference for the capacity selection of each new charging station. Using the data of 1,000 vehicles operating in Zhengzhou as a case analysis, the results show that the charging demand satisfaction rate has increased from 62.27% to 80.8%, and the average distance between the new charging stations and the poi is as low as 143m. The results show that the method proposed in this paper can meet the needs of as many electric vehicle owners as possible while having better charging convenience.

Automatic and accurate arrival time pickup of microseismic first-arrival waves is an important prerequisite for high precision microseismic source location. Aiming at the low efficiency of the traditional manual pickup method and the low accuracy of the long, short window energy ratio (STA/LTA) method commonly used in automatic pickup for low signal-to-noise ratio signals, an automatic picking method of microseismic first arrival based on support vector machine based on particle swarm optimization is proposed. Firstly, according to the amplitude and energy of microseismic signal and the energy ratio of adjacent time, the signals are marked with different categories. Then the parameters are optimized by particle swarm optimization algorithm to construct the support vector machine model of microseismic first-arrival. Finally, the data is substituted to extract the microseismic first-arrival. The experiment is carried out with the microseismic monitoring data of underground roadway in a gold mine. The experimental results show that, under the condition of low SIGNal-to-noise ratio, the picking accuracy of the proposed method is 96.4%, the average pickup error is 3.9ms, and the picking accuracy and accuracy are better than STA/LTA method.

MATLAB is a collection of computing algorithms. It has more than 600 mathematical operation functions used in projects, and it is also widely used in climate analysis. This paper is based on the daily and monthly data of temperature and precipitation from 1980 to 2021, through MATLAB software, supplemented by sliding average, Trend analysis, pole symmetric mode decomposition (ESMD) and standardized precipitation evapotranspiration index (SPI) methods were used to analyze the temporal and spatial characteristics of climate and multi-scale drought in Yinchuan, Ningxia. The temperature in Yinchuan presents a staged rise as a whole. (2) ESMD is used to decompose the temperature and precipitation change signals in Ningxia, and multiple fluctuation cycles are obtained. The temperature change is dominated by the potential term, and the precipitation change is dominated by the interannual fluctuation. The warming stagnation is mainly characterized by chronological changes, with a period of about 8.2a, and the precipitation is mainly characterized by interannual changes, with a period of about 3.4a. (3) In the past 41 years, the SPI-12 index of Yinchuan in Ningxia has shown an upward trend, namely It shows a trend of developing from drought to wetness.

Sequential recommendation systems attempt to use the sequence of user interactions to predict users' next behavior according to their recent based on their most recent actions. Markov chains, recurrent neural networks and self-attention have been widely developed because of their traditional ability to capture the dynamics of sequential patterns. However, most of these models make a simplified assumption that they treat the interaction history as an ordered sequence and ignore the time interval between each interaction. In other words, their model is a time series without considering the actual time interval. In this paper, we try to explicitly model the interactive timestamp in a sequence model framework and explore the impact of different time intervals on the prediction of the next project. We propose the model Ti-SSE which combines TiSASRec and SSE-PT model. Thus, we avoid the shortcomings of the traditional sequential recommendation models that ignore the absolute time intervals while the Stochastic Shared Embeddings (SSE) regularization makes our model far from overfitting due to excessive use of self-attention blocks. The results on several datasets show that our model Ti-SSE has advantages over the previous models.

Human activities are closely related to geographic location. It is proposed to combine spatial location and mobile terminal pose sensor data with meeting the characteristics of real-time accuracy and flexibility in outdoor mobile augmented reality and to realize the virtual-real superposition through the transformation relationship between 3D model coordinate system, world coordinate system, camera coordinate system, image coordinate system, and pixel coordinate system. For the limitations of the vision-based registration method in outdoor scenes, this paper derives the transformation from spatial location data to screen coordinates in detail. It gives the solution and optimization of the transformation matrix and parameters.

People used to invest in the stock and fund markets in the past and now have paid more attention to cryptocurrencies’ market, in which Bitcoin is the most famous and classic underlying asset. In this research, in order to improve the effects of stock prediction, difference of close price and moving average lines will be used as the labels of ensemble learning models. With the predicting results of each label, formulas are used to derive the close price in the next day. The last results show that models have less errors as the moving average prices are used as labels. Based on the models built in this research, people manage to predict the prices’ trends of Bitcoin more accurately and make investment decisions that will yield additional returns.

Since the National Development and Reform Commission (NDRC) Notice on the Pilot Electricity Transmission and Distribution Tariff Reform initiated the new electricity pricing mechanism in China in 2014, the tariff reform has continued to deepen and gradually formed the idea of electricity transmission and distribution pricing based on effective assets and by the principle of "permitted costs plus reasonable profit". Against this background, this paper reviews the status of other scholars' research on electricity transmission and distribution tariff reform, analyzes the factors influencing grid investment capacity and establishes an investment capacity measurement model considering transmission and distribution cost supervision. Then, this paper conducts a case analysis with the example of State Grid Corporation and composes ideas for the optimization of investment strategies from the perspective of investment structure and investment timing.

The accurate control of the project timetable is beneficial in ensuring the power grid's high-quality development. Substation engineering schedule control is currently based primarily on handwritten reports, which do not represent actual field progress. Recognition techniques based on YOLOv3 and ResNet50 are proposed to improve schedule management. The intelligent identification of substation project progress was performed by identifying seven common elements on the project site. The algorithm's accuracy and recall are both above 80%.

With the frequent occurrence of global security problems, violent crowd behavior endangers public security seriously. Meanwhile, intelligent surveillance video technology can be applied for violent crowd behavior detection as more and more surveillance cameras are installed in public and sensitive areas. In this paper, we propose a novel mean kinetic violent flow (MKViF) algorithm for violent crowd behavior detection by extracting the kinetic energy feature of video flow. Specifically, A is firstly calculating the mean kinetic energy by streak flow of each corner in each frame. Then, we obtain a binary indicator of kinetic energy change by calculating the amplitude change between sequence frames. Finally, the MKViF vector for a sequence of frames is obtained by averaging these binary indicators of each pixel in all frames. Experimental results show that the proposed MKViF algorithm behaves better in classification performance and real-time processing performance (45 frames per second) than the existing algorithms.

To address the problems of poor attention to the richness of dialogue content and emotional responses in human-computer interaction, it is valuable to study how to make the machine responses diverse and emotional. A Multi-Style Emotional Dialogue Generation (MEDG) model is proposed, which uses Sequence to Sequence (Seq2Seq) as the text generation model. Firstly, the implicit function in the Seq2Seq encoder is mapped using sentiment sentence vector mapping and introducing an attention mechanism, which increases the sentiment constraint of the model in the generation process. Secondly, penalty factors and output vectors are introduced in the decoder to make the decoded results have diversity. Finally, experimental tests were conducted on an open network dataset, and the results showed that the MEDG model had better scores.

As the main load-bearing component in building structure, the damage of Reinforced Concrete (RC) column determines whether the building structure collapses. In this paper, the influence of column structure and material parameters on the damage of RC columns is analyzed by ANSYS, and the damage degree of RC columns under different equivalent TNT is calculated and analyzed. According to the numerical results, the damage degree laws of RC columns with different parameters are summarized, the consumption process of explosion energy in RC columns is summarized, and the methods to improve the explosion resistance of RC columns are given.

In this paper, the balance problem of pulsating assembly line under mixed production mode is analyzed, which is decomposed into the balance problem within and between stations. On the basis of the balance between stations, the balance problem within stations is studied. Considering the versatility of personnel and the different characteristics of personnel's mastery of skills, the mathematical model is built with the goal of minimizing the total idle time of the assembly line. In view of the constructed mathematical model, an improved genetic algorithm based on two-bit coding is proposed to solve the problem. Finally, an example is given to verify the effectiveness of the algorithm.

Hidden digital watermark is one of the important means of anti-counterfeiting to maintain copyright. By extracting the hidden digital watermark from images, audio and video through specific codes and algorithms, it can provide very strong and powerful evidence to prove one's copyright ownership. The idea is that the image is regarded as a two-dimensional matrix, and then according to the two-dimensional matrix to expand the corresponding operation, at the same time select the appropriate point, and according to the appropriate point in the digital watermark image, digital watermark embedded operation. In this paper, DCT is used to insert digital watermark data, and DCT reverse transformation method is used to realize the extraction function of digital watermark.

Face recognition and tracking is a new topic of image analysis and computer vision systems study. We propose a face tracking method based on feature point tracking in this research. Face detection and face tracking are the two aspects of the proposed face tracking system. Before utilizing the KLT technique to track faces, use the Viola-Jones face detection algorithm to detect all faces in an image or video series. After detecting face objects in consecutive frames, the KLT algorithm tracks them and maintains long-term monitoring as faces come in and out. According to the experiments, the Viola-Jones algorithm is effective at detecting frontal faces. The KLT algorithm was put to the test by tracking human faces in a variety of films shot by a laptop camera. The KLT application is successful, and even if the identified face turns left or right, the face may be tracked.

With the quiet change of people's consumption habits, the used car market has been unprecedentedly prosperous and developed. However, the current second-hand car market does not have a unified price evaluation standard, which leads to frequent occurrences of wanton price listings and excessively low or high transaction prices, which create great obstacles to the transaction process and seriously affect the health and order of the second-hand car market development. This paper selects 13 index variables that affect the price of used cars and analyzes the correlation between each index and the current price of used cars. Through the random forest prediction model, GBDT prediction model and SVM prediction model proposed in this paper, the used car prices are predicted and compared. It is hoped that the research in this paper will have important significance for promoting the standardization of car pricing in the used car market.

The study of Artificial Intelligence (AI) has developed in a rapid speed, which offers technological support for intelligent translation and converts traditional translation practice. This paper, with a qualitive method, analyzes the possibility and necessity of translation application research. It presents a specific introduction to “Xun-Translation” application and discusses the application of artificial intelligence in translation application. The study finds out that with AI installed in translation application, research on translation application is possible and necessary, and that “Xun-Translation” application could be a sample for the following related research to refer to. Lastly, the study illustrates that artificial intelligence could be widely applied in researching translation applications so as to perfect the functions of the applications.

This paper proposes a song accompaniment generation method that combines audio analysis and symbolic music generation so that human music theory can be used to build a reinforcement learning model, training an agent to create music. The key to this algorithm is to extract music theory concepts from audio and a reward model that works well in reinforcement learning. However, some music theory rules are complex and challenging to describe. It is difficult to achieve competitive results only by hardcoding the reward. Therefore, to build an effective reward model, a neural network is used to evaluate the perceptual part of composition quality, and program discrimination is used to model easy-to-describe music theory, and the two work together. Experiments show that the proposed algorithm can generate accompaniment arrangements close to human composers, is compatible with various musical styles, and outperforms the baseline algorithm in multiple evaluation metrics.

Mobile edge computing (MEC) is one of the key technologies of 5G. Due to the limited computing resources of MEC server, it’s a quite hard problem for task offloading decision and resource allocation when multiple users share the resources in MEC. To solve this problem, an improved genetic algorithm is proposed, and a multi-segment chromosome structure and a mixed fitness scaling method are designed with the target of maximizing overall system benefit in terms of time delay utility. Simulation experiments show that this algorithm has good convergence and high computational efficiency.

In response to the problems of complex implementation, low accuracy, poor applicability and high latency of some contemporary human fall detection algorithms that do not achieve real-time results, this paper proposes a fall detection based on Tiny-YOLO target detection algorithm, Kalman target trajectory tracking, Alphapose human pose recognition and spatio-temporal convolutional network, a combination of multiple algorithms. It solves the problem that human target features are difficult to extract and subsequent tracking is lost. And the training speed is accelerated to achieve fast detection of human falls through video and improve the accuracy. The algorithm is tested on the publicly available datasets UR Fall Detection Dataset and Lei2 fall Detection Dataset, and the experimental results prove that the algorithm in this paper has high detection accuracy and effectively reduces the false detection rate.

Text classification is an important research area in the field of natural language processing. In view of the low efficiency of the traditional CNN and RNN algorithms for multi-label classification of long text due to timing and spatial displacement problems, this paper proposes a long text classification model with improved Transformer attention mechanism, and combines LDA topic classification algorithm to achieve multi-label classification of document length text. Firstly, the paper introduces the industry's solutions to the problem of multi-classification of long texts. Compared with traditional algorithms such as truncation method and pooling method, the PAPER proposes the LDA topic classification model combined with the improved Transformer-XL text classification model to extract text features with fine granularity, so as to classify texts with higher accuracy. Finally, comparative experiments show that the proposed solution has a significant improvement in P value, R value and F1 value compared with the traditional classification method in the field of long text multi-label classification.

In view of the low block propagation efficiency in the blockchain network, which leads to the low throughput and high acknowledgement delay of the blockchain system, a dynamic node neighbor selection strategy is designed. According to the node elimination rate, neighbor nodes are divided into reserved nodes and replacement nodes. First, all possible combinations of neighbor node replacement are found by using the combination rule. Then, when determining the strategy of each round, the neighbor node with good transmission timeliness is selected as the reserved node to improve the block transmission efficiency. The selection process of reserved nodes is regarded as a multi arm bandit problem, and a mathematical model is established by using Boltzmann selection strategy. The parameters of the strategy, such as node elimination rate and average propagation delay, are analyzed and verified by simulation experiments. The experimental results show that this strategy can optimize the topology of the blockchain network and effectively reduce the average propagation delay of the blockchain network.

Computer-aided diagnosis of COVID-19 from lung medical images has received increasing attention in previous clinical practice and research. However, developing such automatic model is usually challenging due to the requirement of a large amount of data and sufficient computer power. With only 317 training images, this paper presents a Classic Augmentation based Classifier Generative Adversarial Network (CACGAN) for data synthetising. In order to take into account, the feature extraction ability and lightness of the model for lung CT images, the CACGAN network is mainly constructed by convolution blocks. During the training process, each iteration will update the discriminator's network parameters twice and the generator's network parameters once. For the evaluation of CACGAN, this paper organized multiple comparison between each pair from CACGAN synthetic data, classic augmented data, and original data. In this paper, seven classifiers are built, ranging from simple to complex, and are trained for the three sets of data respectively. To control the variable, the three sets of data use the exact same classifier structure and the exact same validation dataset. The result shows the CACGAN successfully learned how to synthesize new lung CT images with specific labels.

The sensor for the SF6 characteristic decomposition gas based on the non dispersive infrared technology(NDIR) is designed. The system adopts the single light source and the dual wavelength optical structure, and further develops the optical sensor detection system for SF6 decomposition products. The system adopts single light source and dual wavelength structure to eliminate the unstable factors of the measurement system and has the high measurement accuracy. On the basis of effectively obtaining the time series of SF6 decomposition product volume fraction, the SLFRWNN(1, N) system model is applied to fit and predict it, and the high-precision fitting of gas volume fraction and prediction and analysis of gas concentration in wide area time domain are realized. The combination of the NDIR technology and system SLFRWNN(1, N) model provides novel method for the rapid measurement and prediction of SF6 decomposition gas in gas insulated switch-gear.

To resolve the issue of limited data during the deep learning model training, various data augmentation approaches that increase the amount of data based on existing data are proposed. In this paper, the possibility of using fake face images generated by Progressive Growing GAN (ProGAN) was examined, a type of generative adversarial network, based on the existing limited data set to improve the final accuracy of Convolutional Neural Network (CNN) for image classification. This paper used ProGAN to generate 100 fake faces and added two CNNs to construct two classifiers that determine the gender of the face images. The first gender classifier only used 100 real faces, namely 50 male and 50 female faces, that were randomly selected from CelebFaces Attributes Dataset (CelebA) for training. And for the second gender classifier, this study added 100 fake faces, namely 50 male and 50 female faces, based on the original 100 real faces. It can be observed that the accuracy and fitting degree of the two CNN classifiers through 50 epochs to see the influence of adding fake images. The experimental results showed that adding 100 generated fake images to the original real face images improved the final accuracy of the classifier and reduced the problem of overfitting. Thus, it can be concluded that using images generated by ProGAN is a valid method for improving the performance of the CNN classifier when data is limited.

The emergence of cryptocurrencies has promoted the development of blockchain technology. However, due to the low performance and poor scalability of the blockchain, it is difficult to apply the blockchain technology to production. Analysis of its essential reason is mainly caused by the distributed consensus protocol. Distributed consensus protocols provide data transparency, integrity, and immutability in a decentralized and untrusted environment, but good security greatly sacrifices scalability. In order to improve the performance and scalability of the system. This paper first improves the Byzantine consensus protocol and improves the throughput of a single shard; on this basis, an efficient shard formation protocol is designed, which can safely assign nodes to shards. This paper relies on trusted hardware (SGX) to achieve consensus and sharding protocol performance improvements. Second, we design a transaction protocol that ensures transaction security and flexibility even when the transaction coordinator is malicious; finally, our research is extensively evaluated on local clusters and on Google Cloud Platform. The results show that the consensus and shard formation protocol in this paper outperforms other advanced solutions in scale and can well scale the blockchain system through sharding and consensus formation protocol. More importantly, the scalable blockchain system based on the sharding strategy proposed in this paper achieves high throughput and can handle Visa-level workloads.

With the rapid development of Service-Oriented Computing (SOC), Web services have become the preferred technology for realizing Service-oriented computing problems and other related goals. How to find a cheap and high-quality Web service composition from a large number of Web services that provide the same function, that is, the research based on Quality of Service (QoS) is the most important problem in the Web service composition optimization model, which is also very important to improve the efficiency of the service composition. In this work, we use the intelligence optimization algorithms to search for the best combination of web services to achieve the functionality of the workflow’s tasks. And we propose a novel approach, called A Hybrid Optimized Multi-Population Flower Pollination Algorithm (AHOMFPA) to solve this problem. Empirical comparisons demonstrate AHOMFPA has advantages over other existing algorithms in efficiency and feasibility.

With the progress of the times, the concept of the quantum computer has been put forward. Encryption algorithms developed based on the discrete logarithm problem are also facing more and more challenges, However, the Elliptic Curve Digital Algorithm (ECDSA) used by bitcoin is an encryption algorithm developed based on the discrete logarithm problem. Therefore, the author wants to design a method that can prevent the password from being quickly cracked by the quantum computer and can also be compatible with the original encryption algorithm system to reduce the maintenance cost of updating the encryption system. For the two points of appeal, the author thought of improving the original encryption algorithm ECDSA. In this work, secondary random encryption on cQ through the randomness of signature is performed to improve ECDSA. The improved ECDSA encryption algorithm is not an asymmetric algorithm. This is a combination of two encryption algorithms. The combination of asymmetric and symmetric algorithms makes it difficult for quantum computers to crack. If the attacker does not know which algorithm to use, he may be forced to conduct violent attacks, which greatly increases the difficulty of cracking the original algorithm. At present, the author has realized the improvement of ECDSA, and can re-encrypt the original asymmetric ECDSA encryption algorithm so that it does not belong to the asymmetric algorithm. In this way, the security of the huge cryptocurrency market can be further guaranteed, and the algorithm can be greatly compatible with the cryptocurrency system using ECDSA as the encryption algorithm.

In harsh environments such as rain, snow, fog, and haze, the target perception capability of the optoelectronic system is seriously reduced. At the same time, the embedded image processing hardware platform completes high-definition video image preprocessing, target detection and other image processing tasks with slow response and time-consuming. The key video frame decoding proposed in this paper reduces the requirement of the tracker on the computing power of the system, and the image enhancement reduces the influence of the environment on the tracking effect. At the same time, the target tracking problem is converted into a "detection-prediction-tracking" problem. The detection model obtains the target position of the current video frame in real time, and the prediction model introduces the historical motion information of the target to predict the current position of the target. The tracker determines the target position of the detection model and the prediction model. Confidence tracking results are obtained after scoring. The experimental results show that the method can solve the influence of target deformation and occlusion and harsh environment on the tracking results to a certain extent, reduce the loss rate of tracking targets, and improve the accuracy and stability of tracking.