Abstract:Traditional Brain-Computer Interface (BCI) systems have many shortcomings, such as BCI based on motor imaginary requires a large number of exercises; BCI based on P300 requires multiple repetitions of flicker; the number of control commands of BCI based on SSVEP is affected by stimulation frequency and other factors. To this end, researchers have proposed a hybrid Brian-Computer Interface (hBCI). This paper mainly discusses the research progress of hBCI, and reviews three common types of hBCI, such as hBCI of multiple brain models, hBCI of various sensory stimuli, and hBCI of various signals. By analyzing the general principles, stimulation paradigms, experimental results, advantages and applications of the latest hBCI system, we find that using hBCI technology can improve the classification accuracy of BCI and increase the number of control commands, which is obviously better than single mode BCI.

Abstract:Considering that decision maker may hesitate to give the assessed value in the scenario of time urgency and incomplete information, the emergency group decision making method of internet public opinion outbreak based on hesitant fuzzy set is proposed. Firstly, the weight determination model of each evaluation index is established by hesitant fuzzy information entropy and cross entropy. Secondly, the HFWA and score function are used to calculate the evaluation score of each evaluation index. Then, using the weight value and evaluation score of each index to calculate comprehensive harmfulness score of internet public opinion emergency to assist emergency departments to determine the disposal order. Finally, the effectiveness of the proposed method is proved by a case study.

Abstract:Modeling and reasoning about the multi-turn dialogue history is a main challenge for building an intelligent chatbot. Memory Networks with recurrent or gated architectures have been demonstrated promising for conversation modeling. However, it still suffers from two drawbacks, one is relatively low computational efficiency for its complex architectures, the other is costly strong supervision information or fixed priori knowledge, which hinders its extension and application to new domains. This paper proposes an end-to-end memory network with multi-head attention. Firstly, the model adopts a method using word embedding combined with position encoding to represent text input; Secondly, it uses multi-head attention to capture important information in different subspaces of conversational interactions. Finally, multi-layered attention is stacked via shortcut connections to achieve repeatedly reasoning over the modeling result. Experiments on the bAbI-dialog datasets show that the network can effectively model and reason for multi-turn dialogue and has a better time performance.

Abstract:Lattice Quantum Chromo Dynamics (LQCD) is a non-perturbative method for the study of low-energy strong interactions between quarks and gluons. The statistical and systematic uncertainties of the results from LQCD are in principle all under control and can be reduced steadily. Based on LQCD theory, larger volume of lattice grids can calculate physical processes in larger space. And one can divide the space more meticulously to obtain more accurate results. Therefore, large system LQCD calculation is of great significance to the study of QCD theory, but is demanding for higher program computing performance. In this work, the large-scale parallel analysis and performance optimization of LQCD configuration generating and glueball measurement program are studied. Based on the blocking and even-odd algorithms used in LQCD simulation, we design a parallel algorithm based on MPI and OpenMP, and design an optimized data communication module. Aiming at the bottleneck of configuration file output, the solution of configuration file parallel output is put forward. The simulation programs are tested and analyzed on an Intel KNL platform and the x86_64 queues of “Tianhe 2” supercomputer. The results verify the effectiveness of the corresponding optimization measures, and the efficiency of parallel simulation is also analyzed. The maximum size of the test is 1728 nodes (i.e. 41 472 CPU cores).

Abstract:In terms of the problem of Multi-Attribute Decision-Making (MADM) under the uncertain background, the interval grey number MADM method based on the prospect theory and Evidential Reasoning (ER) is put forward. Firstly, considering the property of interval grey number, an improved distance formula of interval grey number is developed. On this basis, taking the positive and negative ideal solutions as the reference point, the distance set between the attribute value and the positive and negative ideal solution is calculated. Secondly, the decision makers’ psychological risk factors are introduced into the interval grey number MADM to develop a prospect value function based on the improved distance formula. Thirdly, the alternatives are selected by ER and the comparison rule of interval numbers. Finally, an illustrative example shows that the proposed method has rationality and feasibility.

Abstract:To solve a problem with a comprehensive and integrated vision, we must analyze different aspects from the perspective of locality and consider it in a divide-and-conquer manner. The multi-viewpoint requirement engineering is getting more and more attention, but there is a lack of unity among viewpoints. This not only fails to capture the complete system requirements, but also causes difficulties in changing requirements. In order to solve these problems, this study first builds a multi-viewpoint based modeling process framework, for a reasonable modeling framework is conductive to obtain complete system requirements. Secondly, the tracking meta-model under the new multi-viewpoint modeling process framework is set up, aiming to illustrate the smooth transition of system requirements across multi-viewpoint meta-model. Finally, a tracking relationship among the tracking meta-model with the requirement tracking matrix method is established, thus relevant elements can be tracked and changed by calculating the change tracking matrix as the requirement changes. The above research can solve the problem of difficulty in requirements change.

Abstract:At present, with the development of Next-generation Sequencing Technology (NGS), new MSI detection methods and software tools are emerging around high-throughput sequencing data. However, existing methods have problems that they need paired normal tissue sequencing data for reference or a large number of other microsatellite stable (MSS) samples’ normal tissue sequencing data to construct baselines. And this will cause the inconvenience to useness. Regarding the issue above, a new MSI detection model based on the information entropy theory for tumor tissue sequencing data is proposed in this study. First, built on original detection software MSlsensor1.1, the MSI detection module based on only tumor tissue sequencing data is proposed and optimized. The augmented software can perform MSI detection for tumor-normal paired sequencing data and single tumor sequencing data. Second, benchmark performance is evaluated on the extended module. For this module, the software performance indicators were evaluated using exon sequencing data of samples. The results show that the performance of the detection module with information entropy theory for single tumor sequencinge data is better, which provides theoretical basis and technical support for the subsequent iteration of more complex mutation signal detection process.

Abstract:With the widespread use of deep learning and the popularity of smart mobile devices, it has become a new trend that migrates deep learning applications to mobile devices. This study designs a bird identification system based on Android platform and lightweight convolutional neural network. The system does not rely on any external computing and storage resources. This study also proposes three model stacking methods based on lightweight convolutional neural network as the basic model, which is weighted average, bilinear stacking, and multi-picture and single model stacking. In this paper, we introduce three stacking methods’ structure, advantages, and disadvantages in detail. And we also give some selection methods of hyperparameters through experiments. The experimental results show that the model stacking is much better than the single model, and the accuracy of the model has been significantly improved, which can be better applied to Android mobile devices.

Abstract:Breaking isolated information island, integrating heterogeneous data, gathering and sharing exchanges, conducting in-depth analysis and mining, and providing industry-side decision-making and situation analysis have far-reaching theoretical and applied value. Based on the actual demand of the situational awareness service of the Chinese Academy of Sciences, this study designs and implements a Hive-based Hadoop/Spark dual computing engine big data warehouse supporting OLAP analysis in multiple ways, and carries out an optimization design of usability, load balancing, and resource management, which provides platform support for the subsequent data aggregation and mining, knowledge map construction and discipline situation analysis. Experimental results show that the system is flexible, efficient, available, and scalable, the resource scheduling is scientific, and the load balancing effect is obvious.

Abstract:As an important network architecture for solving network security problems such as route hijacking, RPKI has two main components: data synchronization between the supply side and the relying party, and data transmission between the relying party and the demand side. At present, the research content worldwide mainly focuses on the data synchronization link between the supply side and the relying party. The data transmission between the relying party and the demand side is still in the initial state of exploration. In view of the shortcomings of the current RPKI theoretical architecture, we design and use Jsonized RPKI cache data to implement an HTTPS-based RPKI cache update architecture, which can meet the needs of actual deployment. Experimental results show that the distribution architecture is stable. Compared with the current RPKI theoretical architecture, it can adapt to the needs of multi-layer transmission and large-scale data transmission.

Abstract:Low delay live is a challenging technical difficulty in video live broadcast. In this study, delay optimization is carried out from three perspectives: network push stream, first frame delay, and video coding delay. The details are as follows. Firstly, according to the network condition, the code rate of live pushing stream is adjusted adaptively. Secondly, the key frame caching algorithm is implemented to reduce the opening time of the first frame to 2~4 seconds. Then, a fast selection algorithm is improved for intra-frame prediction mode, the coding complexity is reduced by cutting down the number of candidate modes and the encoding time can be reduced by up to about 22% under the premise that video quality is basically unchanged. Finally, based on the above three optimization strategies, the screen live system is designed, implemented, and applied to classroom teaching to meet the low delay scenes of classroom screen sharing.

Abstract:Convolutional Neural Network (CNN), which is one of the deep learning algorithms, has been applied in many fields. Because the scale and structure of the network model are complex and the model has large amount of data, it is necessary to reduce the requirements for computational resource. Generally, it needs to use data parallel strategy to partition and calculate tasks with large amount of data. However, just using data parallel strategy which does not combine with the characteristics of computing tasks, it would result in high volume data transmission. Because of that, it is essential to design a reasonable data partitioning strategy for reducing the amount of data transmission through the analysis of the network structure and the computing characteristics of CNN. Firstly, this paper introduces the optimization of computing tasks in deep learning accelerator. Then, it introduces the architecture of the deep learning accelerator based on many-core BWDSP and designs the strategy of computing partition. And it compares and analyzes the experimental results based on VGGNet-16. The experimental results show that the proposed optimization algorithm can significantly improve the performance of data transmission and reduce the amount of data transmission.

Abstract:The need for social activities is thriving while there does exist security issues during the communication between strangers. In order to address those problems, street view message system is designed to connect someone who "commonly feel" in the same scene, which differ from the previous way of socializing and is better for people to get into this scene. People can share their thoughts of scenic spot on the message board. Those thoughts can form into a barrage attraction. This system consists of three modules: deep learning module, database module and display module. Deep learning matches the images of scanned scenes with those of existing scenes in the database, and finally displays them in AR mode. This system can provide a different way of life and bring convenience to people.

Abstract:Foodborne diseases have a long history and cause huge social and economic losses every year. Artificial intelligence technology has brought new approaches to the detection and warning of foodborne disease events. Based on Internet big data, this study develops an intelligent detection and risk prediction platform for foodborne disease events. The platform is oriented to the data automatic acquisition, data analysis and visual display of foodborne disease events in the Internet, through D-M-V layered models and modules. The platform solves the problems of data acquisition, data fusion, event detection, risk prediction and visualization of foodborne disease events. The platform can automatically collect social media data, social economy data and other data from the Internet, make heterogeneous data efficient fusion according to the spatio-temporal coordinates of the data, detect foodborne disease events from social media data and infer their key information; use multi-source data to predict foodborne disease risks, and provide efficient visualization methods and interactive means. In this study, we use the 2018 Beijing foodborne disease data as an example to verify the platform function.

Abstract:The peak of clamping force data of Metro doors can reflect the degradation of the transmission system to a certain extent. Based on this, this study uses the developed data acquisition system to collect, store, display, and query the clamping force of the new on-line metro door in real time. ARIMA model and EMD-ARIMA model are used to predict the trend of mean and standard deviation of peak clamping force with cumulative running time, and the probability of early failure of door transmission system is determined based on the prediction results. The comparison of the two models shows that EMD-ARIMA model can predict the change trend of peak clamping force of metro doors, and the improved prediction method can provide a new idea for predicting the deterioration of metro doors in debugging period.

Abstract:In order to study the dynamic diffusion mechanism and accident risk of tritium, a virtual simulation system for tritium diffusion based on the 3D graphics engine is developed. The accident simulation object is determined by the preliminary hazard analysis method for tritium system. The particle system is applied to simulate the dynamic process of the tritium diffusion, and the human-computer interaction function is realized based on virtual roaming and simulation technology. Finally, take the Tritium extraction system in the International Thermal nuclear Experimental Reactor (ITER) as an example, the dynamic diffusion process and concentration change of tritium is simulated when the hydrogen isotope separation column and pipeline leak. The simulation results provide a basis for the development of safety protection measures when tritium leaks.

Abstract:WiFi Access Points (AP) are widely deployed in public places such as schools, shopping malls, and airports to meet the network access demands of users at any time from any location. However, managing a large-scale WiFi network is still a difficult problem to be solved, and the diversity of devices makes it difficult to provide centralized access management in WLAN. Meanwhile, due to the promotion of various new network services and applications, the problem of low utilization of network resources in traditional WLAN and unbalanced load of access points is becoming more serious. In this study, we adopt the idea of Software-Defined Network (SDN) and apply the Protocol Oblivious Forwarding (POF) technology in the WLAN network. We propose a deep programmable WLAN architecture to realize centralized access management. Moreover, considering the factors such as Received Signal Strength Indicator (RSSI) and real-time network traffic, we design a wireless AP access management mechanism based on MultiPath TCP (MPTCP). It can make full use of the bandwidth resources of each AP, avoid the continuous transmission of data on the congested path, and hence achieve the balance of work loads of substrate APs.

Abstract:Solar-Induced chlorophyll Fluorescence (SIF) is a functional proxy of Gross Primary Production (GPP), and it is crucial to monitor the global or regional vegetation productivity and climate change. However, there is no available original dataset with global continuous coverage at high spatial resolution. Although there are some reconstructed datasets, but they are not specific enough to a region of interest. This disadvantage thus will limit their application for research related to SIF in such a region of interest. In order to explore the method to reconstruct SIF dataset in a region of interest, we built models on MODIS reflectance data and original OCO-2 SIF, combining machine learning, remote sensing technology and ecological principles. The reconstructed SIF dataset was built based on the spatial-temporal features with contiguous spatial coverage and higher spatial resolution. Based on the validation performance, this framework is capable for providing efficient and specific SIF data for a region of interest, and it can support the research related to SIF in this area.

Abstract:Aiming at how to dig out useful knowledge from the massive meteorological data and improve the accuracy of meteorological forecast, this paper proposed a weather forecast method based on the genetic neural network algorithm on Hadoop platform. The method combined genetic algorithm with neural network algorithm, which could avoid the problem of local optimization in traditional algorithm. Then, the genetic neural network forecasting model is established, and the daily data of the ground climate from 1951 to 2006 of 13 stations in Tianjin is used as experimental data. Finally, the experiment is performed taking the rainfall level as decision attribute, and the results show that the method proposed in this paper can get better prediction accuracy for all rainfall level than traditional neural network algorithm. It has the highest prediction precision for the rainfall level R0 and reaches 87%, which can not only effectively deal with mass meteorological data, but also has high prediction precision and good scalability, it proposes a new way of thinking and method for weather forecast.

Abstract:With the increase of investment in ecological protection, the application of infrared camera technology in natural reserves has developed rapidly. Species recognition, which is particularly important in how to fully mine photo information, is the premise of other work. In image recognition, with the outbreak of deep learning, the image recognition has been revolutionized. Convolutional neural network as the representative network structure almost completely overcomes the traditional method in accuracy. However, due to the huge impact of the network structure on the accuracy of the final image recognition, people often choose a network structure suitable for their own dataset from some classic network structures, such as VGG16, VGG19, ResNet50, and so on, in practical applications. Nevertheless, it may need to re-select network structure for different datasets. Therefore, in the species recognition of protected area, this study proposes an automatic construction network structure technology based on AutoML. The technology can automatically build appropriate network structures for different datasets of protected area to avoid manual selection of network structures. At the same time, the technology achieves an accuracy comparable to manual selection of network structures.

Abstract:In order to solve the problem that the low efficiency of traditional frequent patterns mining algorithm, an improved Apriori algorithm based on FP-tree is proposed. Firstly, the join preprocessing process is added to the join step of Apriori algorithm. Secondly, the CP-tree is extended to construct a new tree structure, ECP-tree. The new tree structure can construct a compact prefix tree with only one scan of the database, and support interactive mining and incremental mining. Then, the improved points are combined with the APFT algorithm for mining frequent patterns. Finally, experiments are performed using two datasets in the UCI database. The experimental results show that the improved algorithm has higher mining efficiency and the frequent pattern mining speed is significantly improved.

Abstract:Aiming at the data sparsity problem of collaborative filtering model, a variational autoencoder with clustering latent variable is proposed to process the implicit feedback data. The deep generative model can not only learn the feature distribution of latent variable, but also complete the clustering of features. The original data is reconstructed by multinomial likelihood, the parameters are estimated by Bayesian inference, and the regularization parameter is introduced into the model. By adjusting its size, it can avoid excessive regularization and make the model fit better. A nonlinear probability model has a better ability to model the prediction of missing scores. Experimental results on three data sets of MovieLens show that the proposed algorithm has better recommended performance than the other advanced baselines.

Abstract:The No-Fit Polygon (NFP) can be used for handling of stock packing and cutting problems with two-dimensional non-regular shape. Previously, NFP has not been widely applied because it is difficult to be implemented and is lack of generic approaches that can cope with all problem cases without specific case-by-case handling. This paper introduces an orbital method. The method can handle the typical degenerate cases, such as holes, interlocking concavities. And we make benchmark for ESICUP datasets which from the literature, proving that this approach can be efficient with almost every situation. It has certain reference significance for the research of the packing and cutting problem.

Abstract:As the most basic personal protective equipment, helmets are of great significance to the safety for workers. However, some workers lack safety awareness and often do not wear safety helmets. This study focuses on the detection of safety helmet in complex background. You Only Look Once (YOLO) is a state-of-the-art, real-time object detection algorithm, we propose to apply the YOLO detector for safety helmets detection, which achieves high accuracy. For the single-type detection problem without wearing a helmet, the classifier is modified and the output is modified to a tensor of 18 dimensions. We train YOLOv3 for safety helmets detection on the 2010 datasets based on the pre-training model in ImageNet. Then we optimize the model according to the loss function and IOU curve. The experimental results show that the safety helmet detector gets 98.7% accuracy and 35 fps on the 2000 detection test sets without GPU, which meets the real-time detection requirements. The effectiveness of the YOLOv3 safety helmet detection method is verified.

Abstract:In the era of big data, how to grasp customer needs through data analysis and increase the scientific nature of product optimization is of strategic importance to enterprises. This study applies online comment data to the assisted optimization of enterprise products, proposes techniques and methods for obtaining product optimization information, and realizes the acquisition of product optimization information. Firstly, we calculate the indicators such as customer attention and satisfaction in online reviews, and construct a weighting algorithm model for customer opinions. Next, the word pairs of product characteristics and customer opinions are extracted, and the weight of customer opinions is calculated according to the weight algorithm model. Then, the corresponding product optimization information is found through the correlation matrix. Finally, the feasibility of the method is verified by an example.

Abstract:For the problem of data consistency in the construction of distributed micro-services, this paper summarizes the principles of data consistency in the process of distributed computing, analyzes the importance of idempotent design for realizing micro-services, and proposes a method of using transactional message queue to solve the problem of data consistency in typical application scenarios of distributed micro-services, and gives RocketMQ implementation method and principle. Experiments show that transactional messages can better solve the problem of distributed data consistency. Finally, the advantages and disadvantages of the above methods are analyzed. The data consistency processing method proposed in this study can be used for reference in the construction of distributed micro-services.

Abstract:Copyright protection has always been one of the difficult issues in the field of book publishing. Digital watermarking has been widely used as an important method of copyright protection. According to the characteristics of the QR code and PCA algorithm, this study proposes a digital watermarking algorithm based on QR code book copyright protection. The algorithm firstly uses Principle Components Analysis (PCA) to décor relate the QR code image pixel to obtain the principle component which contains both high frequency and low frequency, then embeds the copyright information watermark after using Arnold scrambling to these principal component factor. The experimental results show that the new algorithm has strong robustness against geometric attacks, noise, image brightness and contrast increase or decrease attacks.

Abstract:Aiming at the blind separation problem of non-cooperative receiving PCMA signals, Improved Particle Filtering based on Genetic Algorithm (GA-IPF) is proposed. Based on the particle filter algorithm framework, the algorithm establishes multiple state distributions to approximate the true posterior probability density. At the same time, genetic algorithm is introduced instead of resampling to generate new particles, which increases particle diversity and avoids particle depletion during resampling. Simulation results show that when the carrier-to-noise ratio is greater than 9 dB, the separation accuracy is over 95%, compared with QRD-M Gibbs and other algorithms, the signal acquisition capability of the algorithm is improved by 4 dB, and the algorithm complexity is reduced by nearly 60%.

Abstract:This study proposes a semantic integrity analysis method based on recurrent neural network. By judging whether the sentence is semantically complete, the long text is divided into multiple semantic complete sentences. First, dividing the sentences into words, mapped to the corresponding word vector and labeled. Then the word vector and the annotation information are processed by the loop window and the undersampling method, and used as the input of the recurrent neural network. Finally we get the model by training. The result of experiment indicates that this method can achieve an accuracy of 91.61%. This method is the basis of automatic assessment of the subjective questions, and also helps the research of semantic analysis, question and answer system and machine translation.

Abstract:It is a difficult process for developers to use API and API sequences (APIs) correctly in software development. When developers are faced with unfamiliar function libraries or code repositories like Github that contains a large number of APIs, they need assistance of some recommendation tools or system. To the best of our knowledge, DeepApi can better understand the semantics of user’s query, but the RNN-based model has some problems: (1) it does not consider the weight of each word, (2) the input sequence is compressed into a fixed length vector, which loses much useful information, (3) long sentences lead to loss of key information. Therefore, this study uses a model based on attention mechanism to distinguish the importance of each word and solve the problem of long-distance dependence caused by long query input. We crawled 649 Java open source projects from Github and processed them to get a training set of 114 364 pairs of annotation-API sequences. The experimental results show that the proposed method can increase BLUE index by more than about 20% compared with DeepApi method on Top1, Top5, and Top10.

Abstract:In recent decades, people’s living standards have improved significantly, but health awareness is still weak. Poor living habits and eating habits have led to a sharp increase in the number of people with diabetes. The complications caused by diabetes are a serious threat to people’s health. Because awareness rate of diabetes is low, many patients with diabetes fail to detect the disease in time, leading to complications. In this study, by analyzing the characteristics of diabetes, according to the characteristics of small sample size and easy to be missing, the IV value analysis is used for feature selection, and CatBoost, a new type of Boosting algorithm, is used to predict diabetes patients and achieves significant predictive effects.

Abstract:Crowdsourced live video streaming, which attracts vast number of users by its rich viewer-broadcaster interaction mechanism, has flourished and expanded over the past few years. The analysis of live video streaming platform has become a research hotspot in the field of streaming media services. Automatic extraction of highlights in live video streaming is crucial for tag generation, video classification and content recommendation. However, the existing highlight detection analysis mostly focuses on audio or video data itself, such as video semantic analysis, audio emotional perception, etc., lacking the rational use of user interaction attributes. In this study, we take Douyu live video streaming platform as a case study. Through analysis of the viewer’s danmu posting and virtual gift donating behavior, we propose an automatic content highlight detection method based on the time series of danmu quantity and virtual gift value in the broadcasting. Firstly, we use z-score method to detect the sequence highlights, then we conduct highlight sample labeling and feature constructing. Finally, Random Forest is used to classify sequence highlights and identify the content highlights. The results show that the model we proposed can accomplish the task of automatic content highlight detection with high accuracy.

Abstract:With the increasing popularity of blockchain in recent years, smart contract has gradually attracted people’s attention. Smart contract is a piece of code that can automatically run, allowing developers to develop personalized according to specific protocol, business or logic, while the decentralized and distributed features of blockchain provide a sound platform for the application of smart contract. Firstly, the architecture and development of blockchain and smart contract are elaborated. Aiming at the military aviation missile business registration scenario, the super ledger is used as the development platform to propose the smart contract design scheme based on the blockchain, and the smart contract development, deployment and commissioning are carried out for the airtightness inspection of an aviation missile. Finally, the security analysis on the effectiveness, security and privacy protection means of the smart contract is carried out to provide reference for the application of blockchain technology and smart contract in the army.

Abstract:While installation of the slit lamp drum, there are some problems such as uncertainty in the evaluation of the clarity of the drum image by the human eye. In order to solve these problems, a digital correction method based on the definition evaluation algorithm is proposed to improve the drum quality. By studying the current mainstream image sharpness evaluation algorithm, several algorithms that are in good agreement with the subjective evaluation are selected, and some appropriate preprocessing method are used to process the image. Finally, the program is written to test the drum image. Three sets of experiments are designed to simulate the focal length change of the camera, the change of drum magnification, and the influence of illumination variation on the algorithm on the production line. The calculation results are compared and analyzed from four aspects: unimodality, unbiasedness, sensitivity, and real-time. The results show that the selected algorithm can accurately evaluate the drum image under different magnifications and different illuminations, and can meet the real-time requirements of the production line.

Abstract:Due to continuous pooling and down sampling, the resolution of the final feature hotspot map is seriously lost in the traditional full-convolution neural network, which leads to the loss of detail characterization ability of segmentation results. In order to make up for this defect, the feature map of middle layer is often fused by jumping connection to restore spatial information. Due to the failure to make full use of the low-level feature information of the network, the feature fusion stage of the traditional full-convolution network has some defects. This study makes an in-depth analysis of this phenomenon. The feature information enhancement method based on feature pyramid is adopted before the upper sampling path to overcome the deficiency of semantic information of shallow feature graph, so that the entire network can make full use of the feature graph generated by forward calculation and improve the segmentation result. The algorithm proposed in this study achieves 75.8% average pixel accuracy and 83.9% weight frequency crossover ratio on the Pascal VOC data set, effectively improved the classification accuracy.

Abstract:Divide and conquer algorithm is widely used for tridiagonal matrix eigenproblems while computing efficiency and storage limitation are always bottlenecks for large scale problems. In this study, the proposed eigenproblem algorithm based on hybrid parallel paradigm with MPI/Cilk optimizes the divide and conquer algorithm both at data and task levels. The introduced task-based parallelization mechanism inside computing nodes solves the problem in data dependence and thread starvation by directed acyclic graph model. By coarse-grained partition of tasks the overhead of data communication among MPI nodes is also optimized, which helps to improve load balance. The numerical test is carried out and the result is compared with the pure MPI and MPI/openMP parallel algorithm, which shows the performance and efficiency of the algorithm.

Abstract:As the core component of the information system, the database stores a large amount of important data information and is vulnerable to the most harmful SQL injection attacks. Traditional database defense methods require prior knowledge such as the characteristics of attack behavior to implement effective defense, and have the defects of static, transparent, and lack of diversity. In this context, based on the dynamic heterogeneous redundancy principle of mimicry defense, the reserved word mimicry module, fingerprint filtering module and mimetic middleware module are used to realize fingerprinting, de-fingerprinting and similarity judgment of SQL injection instructions. A mimetic database model with endogenous security is proposed, and the model is tested using the SQL injection module in the penetration test rehearsal system DVWA to verify the availability and security.

Abstract:Traditional sorting operation can not be adjusted with the change of working environment. In view of this shortcoming, a sorting robot is researched based on machine vision. By introducing image processing and feature engineering technology into the visual module, the sorting system can be adjusted in time. Unlike these methods, this research is based on the industrial sorting system in the laboratory and applies the deep learning method to it. By introducing faster RCNN detection algorithm into visual module and improving of Region Proposal Network (RPN), the detection process of faster RCNN model is accelerated, so that the system meets the real-time requirements of industry. faster RCNN, as an end-to-end method, can automatically generate more expressive features for input images and extract corresponding features for corresponding targets. This avoids the manual design features. Its automatic feature generation ability makes it suitable for various scenarios, which improves the environmental adaptability of industrial sorting robots.

Abstract:To solve the systematic error estimation problem for a two-coordinate radar, the least squares radar systematic error estimation method by a joint ADS-B (Automatic Dependent Surveillance-Broadcast) device is proposed. In the proposed method, the ADS-B measurements are first transformed from geographic coordinate system to radar local rectangular coordinate system for building a unified observation space. Then, the radar track and ADS-B track are applied to calculate the corresponding straight-line equation by the line fitting algorithm, and these two equations are further utilized to calculate the angly between the two straight-lines for compenlating the azimuth component of the radar track. Finally, the least squares algorithm is applied to estimate the radar systematic error. The real-data experimental results of real data illustrate that the proposed method can effectively estimate the radar systematic error compared with the traditional straight-line algorithm and the least squares algorithm. After registering the radar track by the proposed method, the average range error and the azimuth error can be reduced by 71.7% and 52.7%, respectively.

Abstract:In order to solve the problem that traditional target tracking cannot be accurately tracked after occlusion, a target tracking algorithm combining YOLO and Camshift algorithm is proposed. Building a model of target detection using YOLO network structure, before the model is constructed, the image frame is preprocessed by image enhancement method, while maintaining sufficient image information in the video frame, improving the image quality and reducing the time complexity of the YOLO algorithm. The target is determined by the YOLO algorithm, and the initialization of the target tracking is completed. According to the position information of the target, the Camshift algorithm is used to process the subsequent video frames, and the target of each frame is updated, so that the position of the search window can be continuously adjusted to adapt to the movement of the target. The experimental results show that the proposed method can effectively overcome the problem of tracking loss after the target is occluded, and has good robustness.

Abstract:Genetic algorithm is often applied to optimization problems in industrial production, but facing non-linear, multi-extreme, and multi-variable problems, it is easy to fall into the local optimal range in the early optimization process. By adding catastrophe operation to the traditional genetic algorithm, this study reduces the common “premature” phenomenon of genetic algorithm, and sets the adaptive change of genetic operation with the change of iteration times of catastrophe operation, which enhances the optimization ability of the later period of the algorithm. The algorithm is tested with the actual production data of a steel enterprise in Hebei Province. The experimental results show that the algorithm can effectively reduce the cost of raw materials, reduce the occurrence of premature phenomena, improve the overall search ability of the algorithm, and can play a significant role in the optimization of industrial production under the premise of guaranteeing the performance and quality of sinter.