Abstract:Due to the rapid development of process mining technology, the variety of process mining algorithms has increased rapidly, and the introduction of existing algorithm research articles is no longer comprehensive. In view of this, we systematically analyze and summarize process mining algorithms so far. Firstly, we analyze the current situation of process mining algorithms in general and then classify them into two categories according to their characteristics: traditional process mining algorithms and process mining algorithms based on computational intelligence and machine learning technologies. Meanwhile, we briefly introduce the basic ideas and related steps of each subclass of representative algorithms and discuss the current advantages and disadvantages of the algorithms. Finally, suggestions regarding algorithm research and improvement in the next step are proposed. The classification and summary of algorithms can help beginners to sort out relevant algorithm knowledge in the field of process mining, and the analysis of the development status and algorithm comparison can guide researchers in areas that need to be broken through.

Abstract:In virtual reality, the accurate cognition of egocentric distance and the reasonable design of pointing cursors play crucial roles in the user’s interactive experience and task execution efficiency. Given the extended research of Fitts’ law in three-dimensional pointing tasks, the interactive experiment of pointing cursors in virtual space was carried out taking the types of pointing cursors, egocentric distance, target size, and azimuth angles as variables. First, six hypotheses were proposed. For example, the difference among pointing cursors, egocentric distance, target size, and azimuth angles could affect the cognitive accuracy of egocentric distance and target pointing speed. Second, the target pointing accuracy and target pointing time were obtained based on the experimental data, and the change law of the cognitive accuracy of egocentric distance and target pointing speed was discussed. Experimental results showed that in virtual reality, the pointing cursor and target size all had a significant effect on the cognitive accuracy of egocentric distance and the target pointing speed, while the target azimuth angles and egocentric distance could only significantly affect the target pointing speed. The pointing cursor and egocentric distance had a two-way interactive effect on the cognitive accuracy of egocentric distance and the target pointing speed. The target pointing time of the two kinds of pointing cursors was linearly and positively correlated with the difficulty coefficient, which was consistent with the extended model of Fitts’ law in three-dimensional pointing tasks. This conclusion reveals the influencing factors of egocentric distance cognition, which can effectively guide the design of interactive methods and three-dimensional pointing tasks in virtual reality.

Abstract:Semantic segmentation is a very challenging task because of the complexity of parsing the scene, the diversity of segmented objects, and the differences in spatial positions of objects. To tackle this dilemma, this paper proposes a novel architecture named double branch and multi-stage network (DBMSNet) based on dense dilated convolution. Firstly, four feature maps (De1, De2, De3, and De4) with different resolutions are extracted by the backbone network, and then the feature refinement maps of De1 and De3 are output through the feature refinement (FR) module. Secondly, the output branch is processed by the mixed dilation module (MDM) to extract rich spatial location features, while the De4 branch is processed by the pyramid pooling module (PPM) to extract multi-scale semantic information. Finally, the two branches are merged and the segmentation result is output. Comprehensive experiments are conducted on two public datasets of CelebAMask-HQ and Cityscapes, on which our model achieves mean intersection-over-union (mIoU) scores of 74.64% and 78.29%, respectively. The results show that the segmentation accuracy of this study is higher than that of the counterpart method, and this method has fewer parameters.

Abstract:Detecting P300 signals from electroencephalograms (EEGs) is the key to the realization of P300 brain-computer interface (BCI) systems. Because EEG signals vary greatly among different individuals, the existing P300 detection methods based on deep learning require plenty of EEG data to train the model, and there is still no satisfactory solution for learning from limited data of patients. In this study, we proposed an improved prototype network for P300 signal detection of samples with a small size, which extracts features with a convolutional neural network (CNN) and utilizes the cosine similarity of the measurement method to classify and recognize P300 signals. This method achieves a good recognition performance with an average character recognition rate of 95% on the data set II of the third BCI competition. Furthermore, we applied this method to diagnose the consciousness of a small number of patients with disorders of consciousness (DOC). Ten patients with DOC and five healthy subjects participated in a command-following experiment. All healthy subjects achieved significant accuracy (100%) and the results of consciousness diagnosis of the DOC patients were consistent with clinical evaluation. Our findings suggest that the model is of great significance to the improvement of P300 BCI systems for limited data.

Abstract:In data sharing scenarios where users access data resources across domains, their identity legitimacy and secure communication need to be ensured. To this end, this paper proposed a two-factor, i.e., biometrics and passwords, cross-domain authentication and key agreement scheme based on blockchain. Fuzzy extraction technology is used to extract the key and public information of users’ biometrics for authentication participation, avoiding biometric information leakage. The blockchain ledger is used to store users’ identity information including biometric keys and biometric public information, ensuring the consistency of users’ identity information without any tampering. In cross-domain authentication, the authentication server in the authentication domain does not need to communicate with the authentication server in the user registration domain. Instead, it is completed by directly querying the blockchain ledger to obtain users' identity information. Security and performance analysis show that the proposed scheme can provide stronger security with less computational overhead.

Abstract:At present, big data regarding education are increasingly growing. How to efficiently and accurately extract high-value knowledge from the massive data to meet the personalized education needs of learners or educators is a hot topic worthy of attention in smart education. As a visual analysis technology, knowledge graphs can effectively construct and mine knowledge and the interrelationship between knowledge, which has been successfully applied in many fields. The introduction of graph embedding technology is beneficial to significantly improve the processing efficiency of knowledge graphs in the context of big data. To meet the knowledge processing needs of personalized education, this paper first introduces the basic concepts of knowledge graph and graph embedding algorithms and then expounds the triple-based representation learning model from three aspects: vector translation, tensor-based factorization, and neural network-based representation learning. Then, from the perspective of seven application types, the practical application of knowledge graph and graph embedding in the field of personalized education is reviewed. Finally, the paper is summarized and the directions of future research are discussed.

Abstract:Remote sensing change detection aims to compare multi-temporal remote sensing images at the same location and identify significant as well as potential changes between them. Most of the related works focus on the chronological changes but perform poorly on anti-chronological detection. To avoid temporal effect, a common approach is to involve both chronological and anti-chronological data into datasets, but the model training time would be doubled simultaneously. Therefore, this paper proposes a 2-channel siamese network to ensure high accuracy as well as efficient training at the same time. Firstly, a symmetric model is constructed based on existing models to achieve fast training only using the original chronological datasets and to learn both chronological and anti-chronological features. Next, 2-channel siamese input model is designed to wrap the inputs for more robust feature extraction. Finally, attention mechanism is applied to further fuse and refine the extracted features. The proposed method is evaluated on the Onera Satellite Change Detection Sentinel-2 dataset. The Proposed model outperforms several existing models in terms of both accuracy and training validity. A further ablation study verifies the efficacy of proposed models.

Abstract:With the continuous development of the national pipeline network, the scales of system business and disaster data have increased significantly. A pipeline geological disaster monitoring and early warning system based on the microservice architecture is developed to prevent the occurrence of accidents that are vulnerable to geological disasters during long-distance pipeline transportation. The system adopts the mode of separated front-end and back-end development and has the characteristics of high concurrency, low coupling, high availability, and easy expansion. It integrates observation, reporting, research, risk assessment, and forecasting and early warning functions. At present, the system has been operating stably in national pipeline network branches for a long time, effectively settling disaster early warning, disaster handling, disaster information management, and other issues. The proposed system provides effective solutions for the long-distance transportation pipeline industry.

Abstract:Intelligent wheelchairs can effectively improve the mobility and self-care ability of people with mobility disabilities in daily life, but the current control schemes of intelligent wheelchairs often have problems such as difficult control actions and slow response speed. For the above situations, this paper put forward an equipment control scheme based on the coordinated control of electroencephalogram (EEG) and electromyogram (EMG) considering the existing bioelectricity mechanical interface technology and equipment control technology. This scheme allows users to use only four simple head actions, i.e. blink, right or left bite, and concentration, to move the equipment forward or backward, to stop or turn the equipment, and to control the speed. The experimental results show that the scheme can be effectively applied to the controlling of intelligent cars for daily movement and has a high control accuracy and a fast control response speed.

Abstract:Traditional big data trading markets focus on facilitating big data transactions, similar to ordinary commodity trading platforms. Their main functions lie in data catalog management and transaction management in the transaction process. This traditional mode has many drawbacks. Except for public data, considering data security, privacy, data abuse, or damage to own competitive advantages due to data sharing, data owners are still cautious about data exchange. At the same time, companies have limited understandings of data in many fields, which also restricts data exchange. The capacity output cannot be fully achieved by the data catalog alone. To effectively address the above drawbacks, this paper designs a big data trading market platform. The system uses Spring MVC to build the overall back-end architecture and integrates Redis, Mybits, and other technologies to provide data institutions with public Hive clusters for saving data sets. Moreover, it realizes the functions of data set purchase, sample data viewing, and user screening of data set content.

Abstract:In the learning of online course C Language Programming, the interaction between teachers and students is poor, and the teaching efficiency is low. It is difficult for students to solve the common code defects in programming themselves. To better help students solve the problems in learning and assist teachers to achieve the teaching purpose, this paper develops a practical system to assist students in programming. Firstly, the system classifies the common code defects, focusing on the analysis of syntactic, morphological and semantic defects that are not easy to detect by the compiler. Secondly, it builds an intelligent analyzer that integrates a variety of detection tools to store the set of knowledge rules and extend the abstract pattern of common code defects. Finally, it detects the codes and gives error reports and modification suggestions. This system is capable of assisting students in programming by cooperating with the student model. The experimental results show that the system can successfully detect common code defects and thereby assist students in programming practice.

Abstract:Good learning motivation plays an important role in defining learning objectives, straightening learning attitudes, and stimulating learning potentials. Existing strategies of strengthening the learning motivation of students are promoting peer interaction, introducing game-based teaching activities, or simply combining the before-mentioned two strategies, and they all have their limitations. In view of this, in this paper, a novel game based peer learning strategy is designed by comprehensively considering both the competition mode and the cooperation mode in peer interaction and employing a rich game mechanism. Then, a game based peer learning system is developed on the basis of our proposed strategy. The proposed system can create a game based online collaborative question-answering environment for students. In this environment, the system assigns students to different levels according to their historical question-answering records so that students can freely team up according to the hierarchical income rules and answer questions cooperatively. Finally, on the basis of the grade levels and question-answering results of the team members, the system calculates their question-answering incomes and updates their grade levels accordingly. Teaching practices show that the proposed game based peer learning system enhances the learning motivation of students.

Abstract:Given that the existing design of experiments methods are unable to perform the efficient design of experiments for complex systems, this paper proposes a design of experiments method based on the variational auto-encoder. First, experimental historical data are used to train the variational auto-encoder to encode the complex experimental sample space into a relatively simple latent variable space. Then, samples are obtained from the latent variable space. Finally, new experimental samples are generated by the decoder through restoration, and the design of experiments is achieved. The performance of the proposed method in fitting the hit model of the straight-running torpedo is compared with those of several benchmark design of experiments methods. It is shown that with the same number of samples, the proposed method can optimize the design of experiments and improve the efficiency of the experiments.

Abstract:In this study, the current situation and characteristics of the three-dimensional (3D) geographic information system (GIS) and the large-screen display system are analyzed. Then, with a general comprehensive display platform as an example and the digital-twin visualized display of Nanning City as the subject, a large-screen visualization strategy based on the 3D GIS is proposed from the perspectives of spatial information scene design, expression object selection, page layout, platform color design, symbol design, chart design, and dynamic effect design. The system architecture and key technical points of such platforms are expounded in view of this strategy. A platform of this type is developed to provide a reference for similar visualization cases.

Abstract:As an important network infrastructure for service, the domain name system (DNS) is a necessary link for terminals to access the Internet. In recent years, more and more attempts have been made to trick users into malicious servers through DNS, posing a huge threat to Internet security. It is of great practical significance for both operators and network regulators to prevent and resolve access to malicious domains or IPs, including phishing websites, spam, ransomware, and pornographic websites. Therefore, this paper describes the working principle of Response Policy Zones (RPZ), builds a DNS RPZ security protection system, and then configures the related core software. Then, experiments are conducted on the system to verify the protection effect against malicious domains and IPs.

Abstract:In view of the low positioning accuracy of a single sensor and the importance of the environment map in the mobile robot positioning system, this paper proposes a mobile robot localization method based on multi-sensor fusion. Firstly, in an unknown environment, this paper respectively uses a single odometer and fused odometer and inertial measurement unit (IMU) by extended Kalman filter (EKF) algorithm to estimate the position. Experiments show that they have cumulative errors. Then, in a known environment, adaptive Monte Carlo localization (AMCL) algorithm is used to integrate odometer, IMU and lidar for positioning. The experimental results show that the method can correct the cumulative errors. Compared with the single odometer positioning and fusion positioning based on the EKF algorithm in an unknown environment, the proposed method has the average positioning error reduced by 68% and 30% respectively, which proves the effectiveness of multi-sensor fusion positioning and the importance of environment maps.

Abstract:To make up the gap of short answer grading systems in multilingual teaching, this paper proposes an automatic short answer grading system based on siamese network and bidirectional encoder representations from transformers (BERT) model. First, the question and answer texts of short answers yield sentence vectors of texts with the natural language-preprocessed BERT model. The BERT model has been trained on a large-scale multilingual corpus, and the obtained text vectors contain rich contextual semantic information and can deal with multilingual information. Then, the sentence vectors of question and answer texts are subjected to the calculation of the semantic similarity in the siamese network of a deep network. Finally, a logistic regression classifier is employed to complete automatic short answer grading. The datasets used for automatic short answer grading tasks are provided by the Hewlett Foundation, and the quadratic weighted kappa coefficient is used as the evaluation index of the model. The experimental results show that the proposed method outperforms other baseline models for automatic short answer grading in each data subset.

Abstract:The mayfly algorithm is a new type of swarm intelligence optimization algorithm inspired by mayfly flight and mating behavior. It has good optimization performance, but its efficiency is affected by failure mayflies when faced with high-dimensional and complex problems. In view of this, a migration evolutionary mayfly algorithm (MEMA) is proposed in this paper. First, the individual ability of the mayfly population is evaluated, and individuals with a long life-cycle but weaker evolutionary ability are eliminated from the population. At the same time, with those eliminated ones as strongholds, a global position shift is performed on the mayfly population to obtain new individuals. Then, directional dynamic evolution training is carried out on new individuals to improve the overall optimization ability of the population. Finally, in the Matlab environment, six benchmark test functions are randomly selected to design simulation experiments for the effectiveness verification of the MEMA algorithm. The experimental results show that compared with the other five comparison algorithms, the MEMA algorithm outperforms in both low-dimensional and high-dimensional function tests for the optimal solution search, and it has advantages in convergence accuracy, convergence speed, and robustness.

Abstract:Given the poor performance on the small target detection of clothing safety in video surveillance for oilfield operation, this paper proposes a standardized clothing detection method based on Cascade-YOLOv5 (C-YOLOv5), an improvement from YOLOv5. Firstly, a small target detection network cascading with YOLO-people and YOLO-dress is built to locate the pedestrian target. Then the pedestrian area is cut out and transformed in scale to detect the clothing safety of pedestrians. To fully integrate the shallow and deep feature information, this paper adopts four convolutional feature layers with different scales to predict the undetected targets. Finally, in the original image, different color frames are used to mark the types of pedestrians and their clothing parts, determining whether the pedestrians are dressed properly. Experimental results show that compared with the original YOLOv5 algorithm, the C-YOLOv5 method not only meets the real-time requirement but also improves the detection mAP by 2.3 percentage points. At the same time, the improved method of fusing deep and shallow information effectively enhances the representation ability of features and promotes the detection accuracy of small targets.

Abstract:The rapid development of 5G communication technology has led to a comprehensive enhancement of the industrial Internet of Things (IIoT). The scale of IIoT data will become larger, and the dimensionality of data will become higher. As a result, how to efficiently use stream clustering for IIoT data mining is an urgent problem. In this regard, this paper proposes an adaptive clustering method of an IIoT data stream. The algorithm exploits the high density between micro-clusters and calculates the local density peaks of each micro-cluster node to adaptively generate the number of macro-clusters. It employs a gravitational energy function to recursively update the micro-clusters on line and removes the computation between edge-intersecting micro-clusters to achieve a reduction in the computational effort required to maintain the macro-clusters. The theoretical analysis and experimental comparison show that the proposed method has higher-quality clustering results than the current mainstream stream clustering algorithms.

Abstract:To address the low security and identity authentication difficulties of education resource sharing, this paper proposed a cross-domain identity authentication scheme based on blockchain technology and certificateless signature. The user security, cross-domain authentication, traceability of malicious users, and non-tamperable registration information were realized during the identity authentication process by the sharing of high security and no need for key escrow in the certificateless signature technology with blockchain distributed networks. First, intra-domain blockchain and cross-domain blockchain were design to facilitate the cross-domain authentication by a blockchain-based identity authentication model. Then, user safety and traceability of malicious users were guaranteed by certificateless signature and trapdoor hash functions in the authentication phase. Through the analysis, this scheme meets the security attributes such as mutual authentication and user identity security. Compared with other schemes, it has advantages in computing overhead and communication overhead. The scheme can better meet the computing environment with limited computing power on the user side.

Abstract:The demand of intelligent vehicles for high-precision positioning is increasingly strong. In complex environments of urban buildings, overpasses, and so on, the number of visible GPS satellites decreases and the inertial measurement unit (IMU) in a fusion positioning system of the vehicle GPS andthe IMU produces a time accumulation error, leading to inaccurate positioning. This paper proposes a fusion positioning algorithm of an ultra wide band (UWB) and a GPS based on the unscented Kalman filter (UKF). The system architecture scheme is constructed. The data analysis algorithm for the UWB module is optimized, and the model of a nonlinear fusion positioning system of a UWB and a GPS is built. The complexity of the algorithm is analyzed, and the algorithm is written into the controller for real-time filtering. The noise error and variance of different algorithms are analyzed. The experiments show that the fusion positioning algorithm of a UWB and a GPS based on the unscented Kalman filter, with good real-time performance, high solution accuracy, and no filter divergence, can meet the needs of high-precision positioning of vehicles in complex urban environments.

Abstract:The development of traditional supply chain services faces many pain points, including forgery risk, enterprise information silo, failure to the cross-level transmission of core enterprise credit, and default risk. To tackle the problems of information silos of supply chain enterprises and unauthorized access to network resources by illegal users, this paper proposes an access control model based on smart contracts and enterprise credit to guarantee the data security and integrity of fruit quality traceability system. Combining role-based access control (RBAC) and attribute-based access control (ABAC), the model takes smart contracts as the underlying technology, enterprise credit values as reference attributes for cross-domain access, and subject attributes and credit values as decision bases to realize intra- and inter-domain access control. The experimental results show the effectiveness of the access control model proposed in this paper.

Abstract:Given that the traditional method of similarity weighted fusion in the process of multi-atlas based medical image segmentation does not consider the interference and redundancy of the atlas set, a method of brain magnetic resonance (MR) image segmentation based on a two-stage atlas selection strategy is proposed. In this method, a method based on minimum angle regression is used for rough atlas selection. Then, a method based on the Hausdorff distance is adopted for target-oriented precise atlas selection. The rough selection method can find the atlas similar to the target image on the whole, remove invalid variables, and reduce the interference and redundancy of the atlas set. The precise selection method pays more attention to the similarity calculation of the target tissue, and the similarity results are not affected by the size and location of the target tissue. Experimental results show that the proposed method is more robust and accurate than the traditional one-stage atlas selection method based on similarity calculation of the rectangular region.

Abstract:Machine reading comprehension and question answering has long been considered as one of the core problems of natural language understanding, which requires models to select the best answer from a given text and question. With the rise of pre-trained language models such as BERT, great breakthroughs have been made in natural language processing (NLP) tasks. However, there are still some shortcomings in complex reading comprehension tasks. To solve this problem, this paper proposes a machine reading comprehension model based on retrospective readers. The proposed model uses the pre-trained model RoBERTa to encode questions and articles and divides the reading comprehension section into two modules: an intensive reading module at the word level and a comprehensive reading module at the sentence level. These two modules capture the semantic information in articles and problems at two different granularity levels. Finally, the prediction results of the two modules are combined to produce the answer with the highest probability. The model accuracy is improved in the CAIL2020 dataset and the joint-F1 value of the model reaches 66.15%, which is 5.38% higher than that of the RoBERTa model. The effectiveness of this model is proved by ablation experiments.

Abstract:Image super-resolution reconstruction technology has always been a hot research direction in the field of computer vision. To improve the quality of reconstructed images, this paper proposes an upsampling technology based on content awareness for image reconstruction. The residual dense network is used as the backbone network, and the content awareness-based upsampling replaces the traditional sub-pixel convolution upsampling. In other words, in the stage of feature reconstruction, the convolution kernel will not share parameters in the entire feature map, but the neural network can generate a specific convolution kernel depending on the content of the feature map in each pixel. The algorithm reduces the number of parameters, thereby speeding up the network training speed. After multiple rounds of training and testing, the results show that the improved technology can yield a clearer reconstructed image and presents a great visual effect.

Abstract:The image principal contour contains important image features, and an accurate and effective extraction method of the image principal contour can not only reduce information redundancy but also reduce the time complexity of subsequent image analysis and processing. Depending on the information processing mechanism of visual neurons, this paper proposes an image principal contour extraction method based on spatio-temporal spike coding. First, the Gabor function is used to simulate the multi-scale and multi-directional information extraction from the image by the receptive field of ganglion cells. Second, the non-classical receptive field of the retina is simulated to construct an anisotropic suppression model to suppress the edge of the image background texture. Then, as for the visual images obtained by the receptive field of different scales, the small scale of the visual receptive field can help extract most of the texture information of the image, and the extraction under the large scale can make most of the texture of the image disappear, with only some characteristics of the principal contour retained for the adaptive adjustment of weights to perform spatio-temporal spike coding. Finally, the leaky integrate-and-fire neuron model is used to extract the principal contour of the image, and the principal contour of the final image is obtained by non-maximum suppression and hysteresis threshold binarization. From both subjective and objective aspects, the method proposed in this paper is simulated and verified on the RUG40 database, and its performance is compared with that of the existing mainstream image contour extraction methods. The experimental results show that the proposed method can reduce the redundant information of the principal contour of the image while effectively improving the accuracy of principal contour extraction.

Abstract:The existing parking lot classification methods are exposed to problems of low-level automation and high equipment and deployment costs, and the existing detection algorithms have low recall rates and poor detection accuracy. To solve these problems, this study proposes a vision-based parking space detection and classification algorithm to improve the utilization efficiency of parking lots. First, parking spaces are detected to help build a parking space table andincrementally expand the parking space classification model dataset. Then, the test dataset is used to train the support vector machine (SVM) model for parking space classification. Finally, real-time judgment of the parking space conditions is made one very parking space based on the surveillance video data. The experimental results show that under different lighting conditions, the recall rate of the line detection of parking spaces is above 94%, and the accuracy of the parking space classification model is above 95%. The algorithm boasts a high degree of automation, good accuracy, simple deployment, and high application value.

Abstract:With the rapid development of unmanned aerial vehicle (UAV) technology, UAVs are widely used in inspection tasks of various fields. In recent years, the scale and length of power networks have been growing rapidly, and UAVs have become the first choice for power inspection due to their unique performance and advantages. They can not only ensure safety, but also effectively improve inspection efficiency. Regarding inspection tasks, the path planning of UAVs is crucial in practical application. In this paper, a new hybrid meta-heuristic algorithm is proposed to solve the UAVs routing planning problem with multiple depots in power inspection. In the framework of adaptive large neighborhood search, the variable neighborhood descent strategy is added to enhance the neighborhood search ability and increase the possibility of finding a better solution. Experimental results show that the proposed algorithm can effectively solve the problem and has good stability and robustness. In addition, the proposed algorithm is compared with other meta-heuristic algorithms experimentally, and the comparison results verify that this algorithm can effectively reduce the number and time cost of UAVs used in inspection.

Abstract:For better use of high-quality graphite resources, this paper proposed a graphite classification and recognition algorithm based on transfer learning and focal loss convolutional neural network (CNN). The offline expansion and online enhancement of the self-built initial data set can effectively expand the data set and reduce the overfitting risk of deep CNN. With VGG16, ResNet34 and MobileNet V2 as basic models, a new output module is redesigned and loaded into the full connection layer, which improves the generalization ability and robustness of the model. Combined with the focal loss function, the hyperparameters of the model are modified and trained on the graphite data set. The simulation results show that the proposed method has the accuracy improved to above 95% with faster convergence and a more stable model, which proves the feasibility and effectiveness of the proposed algorithm.

Abstract:The application of artificial intelligent has been stimulating the productivity and technological revolution of industries. Traditional industries are facing small sample and imbalanced data problems due to the rarity nature of sample, cost and privacy issues. However, the sample generation results of existing methods are often limited to balancing generalization and validity. The purposed semantic meaning extraction of VAE’s latent variables based virtual sample generation method utilized the weights of encoder neural network as the measurement of dependency between input features and the latent variables. This method achieves flexible sample generation by controlling various dimensions of latent variables explicitly. The generated samples which satisfy the population distribution, are not necessarily included in the original samples. The results of sample expansion of civil buildings structural safety databases show that our method is capable of controllable generation of valid samples, and mitigating the problems of small sample and imbalanced data.

Abstract:In order to solve the dynamic and heterogeneous storage of distributed storage systems, this paper proposes a construction algorithm of heterogeneous fractional repetition codes based on node common edge (HFRC-NCE) is proposed in this paper. In particular, the data blocks encoded by MDS code are divided into cold and hot data blocks, which are copied and stored with different multiples in storage nodes. Moreover, combined with the characteristic of node common edge, the structure of the heterogeneous fractional repetition codes is more simple and intuitive, which can realize the precise non-coding repair of fault nodes. Compared with the fractional repetition codes constructed by complete graph and partial regular graph, theoretical analyses show that, although the storage overhead and bandwidth overhead of HFRC-NCE are a litter larger, its node repair options are larger and the node storage capacities are more diverse. Meanwhile, The reconstruction degree of HFRC-NCE is much smaller.

Abstract:When the traditional artificial potential field algorithm is used for path planning in an old warehouse, defects such as collision with obstacles far from the target, an unreachable target point, and local minimums, which originally appear infrequently, occur much more frequently. To improve the success rate of the artificial potential field algorithm in path finding in an old warehouse, this paper proposes an improved artificial potential field algorithm that corrects the above three defects and uses Matlab simulation to verify the effectiveness of the algorithm. In the improved artificial potential field algorithm, the problems of collision with obstacles far from the target and an unreachable target point are solved through the improvement of gravitation and repulsion. The local minimum problem is effectively solved by introducing temporary obstacles. In the experimental part, for different simulation environments, we use path length and program running time as evaluation indicators to compare the path planning effects of the traditional artificial potential field algorithm and the improved artificial potential field algorithm. Experimental results show that the improved algorithm always outperforms the traditional algorithm regardless of the presence or absence of defects in the environment.

Abstract:Fast And Efficient Subgroup Set Discovery (FSSD) is a subgroup discovery algorithm that aims to provide a diverse set of patterns in a short period of time. However, in order to reduce the running time, this algorithm selects a feature subset with a small number of domains. When the feature subset is irrelevant or weakly related to the target class, the quality of the pattern set decreases. To solve this problem, this paper proposes a FSSD algorithm based on ensemble feature selection. In the preprocessing stage, it uses ensemble feature selection based on ReliefF (Relief-F) and analysis of variance to obtain feature subset with diversity and strong correlation, and then uses FSSD algorithm to return high-quality pattern set. The experimental results on the UCI datasets and the National Health and Nutrition Examination Survey (NHANES) dataset show that the improved FSSD algorithm improves the quality of the pattern set, thereby summarizing more interesting knowledge. Furthermore, the feature validity and positive predictive value of the pattern set were further analyzed on the NHANES dataset.

Abstract:At present, traditional deep learning-based relation extraction methods are unable to extract relations in complex contexts and fail to consider the impact of non-target relations in a context on relation extraction. In response, this paper proposes a control input long short-term memory (CI-LSTM) network that adds an input control unit composed of an attention mechanism and a control gate valve unit to the traditional LSTM network. The control gate valve unit can perform focused learning on key positions according to the control vector, and the attention mechanism calculates the different features of the inputs of a single LSTM network. After experiments, this paper finally chooses to use syntactic dependency to generate control vectors and build a relation extraction model. An experiment is then conducted on the SemEval-2010 Task8 relation data set and the samples in the data set with complex contexts. The results show that compared with the traditional relation extraction method, the CI-LSTM network proposed in this paper achieves further improvement in accuracy and better performance in complex contexts.

Abstract:With the development of Internet technology and the outbreak of COVID-19 in 2020, more and more students have chosen online education. However, due to the large number of online courses, students are often unable to find suitable courses in time. A personalized intelligent recommendation system is an effective solution to this problem. Considering the obvious sequential characteristics of users for online learning, an online course recommendation model based on enhanced auto-encoders is proposed. First, the auto-encoder is enhanced with the long short-term memory network, so the model can extract the sequential characteristics of data. Then, the Softmax function is used to recommend online courses. Experimental results show that the proposed method has higher recommendation accuracy than the collaborative filtering algorithm and the recommendation model based on traditional auto-encoders.

Abstract:With the rapid development of video streaming services, scenarios in which large-scale users share bandwidth links increase unceasingly. The existing adaptive bitrate (ABR) algorithm used in dynamic adaptive streaming over HTTP (DASH) video streaming is mostly used to improve the quality of experience (QoE) of single-client customers, while some other algorithms are only for multi-client situations. This paper proposes a bandwidth scheduling algorithm for large-scale client situations. A clustering algorithm is adopted to reduce the scheduling scale. Then, bandwidth allocation is combined with the ABR algorithm to make bitrate decisions for clustering clients and thereby to improve bandwidth utilization and ensure a maximum overall QoE. Our experimental results show that compared with the bandwidth-sharing scheduling method, the method of scheduling the clustering client bandwidth and applying it to all clients achieves a 99.4% increase in overall user QoE. The overall QoE increase is 10.7% on that of the best state-of-the-art scheme Minerva.

Abstract:To address the problem that most existing algorithms and models for calculating user influence in networks only consider topology and greedy algorithms and rarely take into account the importance of trust degree on node influence, this paper proposed a global trust model (GTM) for evaluating node influence. The trust relationships of a node with its neighbor nodes were calculated as the local trust degrees. Then, the Beta reputation model was used to obtain the global trust degree through the local trust degrees of the node. Finally, the node influence was evaluated according to the global trust degree of the node. Experiments were conducted on real network datasets to compare this model with classical influence algorithms. The experimental results show that the proposed method not only has lower time complexity but also demonstrates a favorable influence propagation ability in addition to ensuring node trustworthiness and accuracy.

Abstract:Premature birth is the primary cause of neonatal death and disability, which can affect the long-term health of newborns. However, the accurate prediction of premature birth is a difficult problem in the medical field. The early screening of premature birth in medicine is mostly based on special examinations, but it is difficult to be applied to large-scale clinical practice due to cost accounting and other problems. The popularization of electronic medical records and the development of artificial intelligence technology provide support for early risk assessment of obstetric diseases. This paper uses the diagnosis and treatment information of obstetric electronic medical records and proposes a hybrid model of GRU and GBDT to predict the risk of premature birth. The hybrid model uses GRU to explore the probability of premature birth in multiple antenatal examination information of pregnant women and integrates the results into the pregnancy data and the last antenatal data before 28 weeks. Finally, GBDT is used to predict the risk of premature birth for higher accuracy. The experimental results show that evaluation indexes such as AUC and ROC of the prediction method based on GRU and GBDT are better than those of other single machine learning models. The proposed method can provide a reference for the obstetric medical staff to judge the risk of premature birth in the early and middle stages of pregnancy.

Abstract:To tackle the problem that the existing video on demand (VOD) technology cannot be directly used in the intelligent mine management and control platform, this paper designs a VOD technology based on the adaptive bitrate streaming media transport protocol of HTTP and FFmpeg open-source library. The technology includes client module, web request processing module and multimedia processing module. In this technology, the client module sends the video request to the web request processing module through the set video source information. The web request processing module uses the video source information in the request to call the multimedia processing module to obtain the multimedia streaming file and returns the request address of the streaming file to the client. After receiving the streaming media address, the client requests and plays the video file according to the streaming media address. By introducing hls.js open-source library, the technology can achieve VOD in any client using HTML5 technology, realize the embedded video information of cross-terminal browser, multi-business module and plug-in removal of the intelligent mine management and control platform, and meet the needs of fusion display and collaborative analysis of video information and various business information of coal mine safety production.

Abstract:Telephone fraud is increasingly rampant, affecting the safety of people’s lives and property seriously. How to effectively prevent telephone fraud has become a great concern of society. This paper proposes a fraud phone recognition method based on the Attention-BiLSTM model. This method uses phone text as a data set and adopts a bi-directional long short-term memory (BiLSTM) model to extract the long-distance characteristics of a sentence. Furthermore, the attention mechanism is utilized to enhance the meaning feature weight of the words related to the fraud parts in phone text. The feature vector representation of phone text at the sentence level is achieved and inputted to the Softmax layer for classification prediction. The experimental results show that the BiLSTM fraud phone classification model based on the attention mechanism has the accuracy increased by 2.15% and 0.6% respectively compared with baseline models, possessing more excellent prediction performance.

Abstract:Cloud storage auditing for shared data refers to the integrity auditing of cloud data shared by a group of users. Since users may join or leave user groups for various reasons, it usually supports user revocation. In most existing cloud storage auditing schemes for shared data, the computation cost of user revocation is linearly correlated to the number of file blocks to be uploaded by the user group, which results in high computation and communication costs. How to reduce the computation and communication overhead caused by user revocation has become a key issue for realizing shared cloud storage auditing. Therefore, this paper proposes an efficient and revocable cloud storage auditing scheme for shared data, which uses the elliptic curve technology to achieve unpaired authentication and the Chinese remainder theorem to attain efficient user revocation. On the basis of ensuring safe user revocation, this scheme greatly reduces communication and computation costs. In addition, it uses identity-based cryptography technology to solve the complex certificate management problem of traditional public key cryptography. The safety analysis and experimental results show that the proposed scheme is both feasible and efficient.

Abstract:An optimizing method of operating routes based on the smallest spraying unit is proposed for assisting spraying operations in agricultural plant protection areas, with plant protection unmanned helicopters taken as the study object. The pros and cons of various full-coverage operation methods are analyzed in terms of the selection of return points, the number of swerves and the operating distance, which helps select the optimal operating approach of unmanned helicopters. The smallest spraying unit during plant protection operation is determined according to such parameters as the volumes of unmanned helicopter’s power, fuel and pesticide. The rectangular farmland is subjected to segmentation optimization in light of the rectangular cutting idea for the global planning of the plant protection operation area. The simulation results indicate that this optimizing method can effectively improve the spraying efficiency of unmanned helicopters, reduce the economic loss caused by useless flights and optimize energy consumption and time, which provides a reference solution for future high-standard farmland construction.

Abstract:In the process of building a personal credit risk evaluation model, feature engineering largely determines the performance of the evaluator. Traditional feature selection methods cannot fully consider the impact of high-dimensional indicators on the evaluation results, and most studies artificially determines the size of the feature set in the process of building the model, leading to high randomness and low credibility. Therefore, a random forest model (IV-XGBoostRF) based on traditional risk control indicators to optimize XGBoost is proposed. The traditional risk control indicators IV and XGBoost are combined to screen the original feature set to build a relatively complete credit evaluation model. The results of comparison experiments show that the accuracy of the improved random forest model is increased by 0.90%, and other evaluation indicators are better than the traditional credit evaluation model, which proves the feasibility of the feature selection method and has certain application value.

Abstract:With the continuous growth of business volume and functional requirements, applications of major business systems are gradually upgrading from the Spring Boot architecture to the SpringCloud micro-service architecture. Due to the significant version change, internal testing and external joint testing need to be conducted sufficiently before the applications officially go online. Under the constraints of limited machine resources in the DMZ domain and minimizing the exposure to the public network of the existing joint environment, this paper proposes a dual-architecture application paralleling and flow switching scheme based on the Nginx-F5, which provides the test system with both the external joint testing functions of Boot and Cloud architecture applications. The scheme splits the requests of external merchants according to attributes such as merchant code, business type, or province code and a certain percentage and forwards them to the micro-service application system. In this way, the application in the micro-service version can be fully tested, which provides a reference for system expansion and construction that needs large version parallel testing.