Abstract:With the arrival of the big data era, more powerful computers and more mature big data platform tools for enterprises from the massive data mining data value has become possible, especially based on Hadoop Big Data Platform, which can even handle TB, PB level of data with cheap commercial hardware. In the initial construction process of Hadoop Big Data Platform, the first step often starts with the building function, ignoring the security control strategy. The Yahoo team proposed Kerberos-based authentication scheme in 2009, which led to the Hadoop Big Data Platform security control work in full swing. This article introduces the history of the Hadoop Big Data Platform. Then, it describes the traditional security issues existing in Hadoop Big Data Platform before 2009. Finally, it tries to present the security of the Hadoop ecosystem components in the industry and the security solution for each component. We hope to provide reference for the construction of Hadoop Big Data Platform security, so people can reasonably use advanced security control program to protect the enterprise's and user's privacy data.

Abstract:Moving object detection algorithms are widely used in video surveillance and other fields. But due to noise, illumination changes and other interference, traditional algorithms are often ineffective. To get a better performance, we transform the problem into a pixel-wise segmentation problem, and propose a novel algorithm based on deep encoder-decoder neural networks. We train an encoder-decoder network offline to learn the differences between the background and the video frame. We firstly use the Gaussian Mixture Model (GMM) to generate a background, and then feed video frames and the background into the encoder-decoder network to get detection results. This method utilizes the advantages of deep convolution network in anti-noise and feature learning, and performs well without complicated parameter tuning. We experiment on the CDnet2014 dataset, and results show that the algorithm we propose performs much better than the original GMM algorithm, and even outperforms some top algorithms in some scenes. Due to the simple network architecture, our algorithm is capable of almost real-time processing using a GPU, which shows its great practicality.

Abstract:In recent year, with the development of various network platforms, there are always a lot of similar local community structures between users in different networks. In consideration of some single-view community detection algorithms cannot find the multi-factor community structures, in this paper we present a Multi-view Local collaborative Selecting Clustering model (called co-MLSC). This model can solve many constraints problems (like nodes, clusters, and sufficient information) and over adjustment problems. Firstly, the model can build a choice regulate matrix that can train the common part of the node set, and converge its common structure. Then we also build a local optimization matrix that regards the node structure as a training set, and uses the KRR algorithm to complete the division of isolated nodes. Finally, we use the UCI and DBLP data sets to demonstrate the effectiveness and applicability of our algorithm.

Abstract:Micro-blog user behavior prediction aims to study user behavior habits. This paper mainly studies the factors that affect the behaviors of users of microblogging from three aspects: the user attribute, user interest, and user's emotion. We extract the characteristics of the user behaviors, training and forecasting the model. The experimental results show that the average accuracy of forwarding behavior can reach 82.56% in the prediction, the average prediction accuracy of behavior in the comments reaching 84.59%, the prediction average accuracy of likes behavior rate reaching 79.35%, which indicates the effectiveness of user interest and emotion characteristics in the promotion of microblogging user behavior prediction.

Abstract:This study researches on the visual detection and following of object people by indoor monocular robots，which includes scene change detection algorithm, visual object people detection algorithm, visual object tracking algorithm, and robot following, with focuses on scene change detection algorithm and visual object tracking algorithm. A high speed scene change detection algorithm judges whether the scene changes by constructing scene models. If the scene changes, the algorithm outputs the change region, which is used by the visual object detection algorithm. The experiment shows this algorithm speeds up the system and alleviates the latency of robots. The visual object tracking algorithm combines the appearance model and map information obtained in SLAM process. The map information can judge which part of object bounding box is actually the background, which can reduce the effect of occlusion and object scale change on appearance model. This algorithm improves visual object tracking performance in the experiments. This paper applies the latest deep neural networks to do visual object people detection. We train a small deep neural network with enhancement on indoor people, which achieves a good balance between running speed and detection performance. Based on the visual detection and visual tracking of target, we accomplish robot following. Since monocular robots can only get the bearing information of target, the goal of robot following is keeping the target in the horizontally middle point of image plane. The robot can successfully follow human even if the person is partially occluded.

Abstract:For the lack of processing power and backward technology, satellites cannot achieve real-time processing of multi-source data in orbit in the space-based system, so this paper presents a design of a space-based network universal service platform for micro satellites in the medium and low orbit. This design provides a new hardware architecture for the development of current satellite technology. At the same time, this heterogeneous architecture cannot only achieve the independent operation of the satellites, but it can also provide hardware acceleration for the data processing of the onboard computers. This paper mainly introduces the hardware design of space-based network universal service platform, including chip selection, external relations, and the required storage capacity. In order to realize the compatibility of the interface, the module of the MIL-STD-1553B bus is designed and implemented on the FPGA to send data from the bus controller to the remote terminal.

Abstract:With the popularity of Internet plus, both consumers and custom clothing enterprises have high demand for Internet online apparel customization. Interactive service is a new online service model, which provides a new concept of consumption. In view of the present situation of garment customization industry, we propose and study the design of interactive garment customization platform based on WeX5. Firstly, this paper introduces the key technologies used in the system, and points out the advantages of the application of mobile terminal mode customization. Then, it analyzes the requirements of customized apparel, and completes the design of mobile terminals.

Abstract:The quadrotorcraft attitude control is the core of the four rotorcraft control system. In this study, the attitude control system of four rotorcraft is designed by analyzing the flight principle and model of the four rotorcraft. In this system, the STM32 series processor is used as the main chip, MPU6050 triaxial acceleration set, the triaxial gyroscope inertia measurement unit, magnetometer, and other sensors to detect posture information. The system is based on the idea of modular design and the sensor uses a simple structure of the digital interface to exchange data. The closed-loop control of attitude angle is carried out by double closed-loop PID control algorithm. Finally, the experimental results show that the flight effect of the four rotorcraft is stable on the experimental platform, and the system meets the requirements of flight attitude control of four rotorcraft.

Abstract:In the design and development of high confidence software, the software fault tolerance is one of the techniques to improve the credibility of the system. Fault tolerance is the ability of software to guarantee the service when the fault occurs. And a processing method for fault tolerance is to rely on redundancy technology. Based on the analysis of the structural redundancy and its influence on the credibility of the system, this paper proposes a redundancy mechanism for the core components of the component-based trusted software. The redundancy structure includes dual redundant structure for single component and composite components, 2 out of 3 redundant structure and its extension. And the fault detection and the judgment method are given. At the same time, the reliability of the system is analyzed on the basis of various redundant structures.

Abstract:In the field of aerospace applications, a bootloader, aiming to satisfy the rapid development and iteration of the applications, is needed to reconstruct different applications and ensure its high reliability. Based on the SPARC architecture, this study designs and implements a booloader, which cannot only boot applications automatically according to the bootflag, but also can reconstruct or boot a special application under the control of the ground command. Also, the three redundant architecture, rebound wall, EDAC protection, and other reliability measures have been taken to ensure that the most software failure can be restored. Finally, we test the bootloader on the computer module, and the results have achieved the desired purpose.

Abstract:XML documents are semi-structured data, and XML frequent path mining can be divided into two steps: XML document serialization and sequence mining. The existing serialization method expresses the XML document as a set of Xpath paths with a plenty of node redundancy. Algorithms based on Apriori require multiple scanning of the database and can generate a large number of candidate sets. The PrefixSpan algorithm generates a large number of projection databases, occupying a lot of memory space. In view of the shortcomings of the existing algorithms used in XML frequent path mining, this paper proposes an efficient mining algorithm called Prefix-based XML Frequent Path Mining Algorithm (PXFP). The PXFP algorithm traverses the XML document tree in a breadth-first manner and represents each node as “node: parent node”, which reduces the node redundancy. The PXFP does not generate the projection database, but only gets the sub-node of the prefix, and then increases the length of the frequent pattern by the position information of the frequent sub-path, which reduces scanning the database. The experimental results show that the PXFP algorithm achieves higher time and space efficiency than the PrefixSpan algorithm.

Abstract:With the wide use of Web application in recent years, its security has seriously affected the experience of relevant users. In particular, the SQL injection vulnerability attack is the most commonly used technique in attacking Web applications. In this study, a chopping-based method is proposed to detect the second-order SQL injection. Firstly, the static analysis based on chopping had been used to find the suspected first-order SQL injection paths. Secondly, the SQL statements in those suspected paths should be used to get the second-order operation pairs, which had been used to get the suspected second-order SQL injection paths. Finally, the attack vector is constructed to confirm the existence of the vulnerabilities which are hiding in Web applications. The experimental results show that the method we propose in this study can effectively detect the second-order SQL injection vulnerabilities thus prevent Web applications from SQL injection attack.

Abstract:Because of the sparse and dynamic crisscross characteristics, the short text makes the weight of traditional weighted method difficult to use effectively. This paper presents a new feature weight calculation algorithm based on part of speech. This algorithm is the quantum particle swarm optimization algorithm introduced into translation decision model which can calculate the probability of a feature with certain part of speech. Then it is tested by the text clustering algorithm. The test results show that the improved feature weight calculation algorithm on the clustering accuracy is better than TF-IDF and QPSO algorithm.

Abstract:To solve the SQL injection problem in the Web security, a new SQL injection filtering method named LFS (length-frequency-SQL syntax tree) is proposed in this study. The LFS includes two phases: the learning and the filtering phase. In the learning phase, the URL and the SQL statement mapping table are built based on the crawler and the database agent in a secure environment. In the filtering phase, the URL length, the access frequency, and the SQL syntax tree are detected to filter the user input to prevent SQL injection attacks. Simulation experiments and results analysis denote that the proposed LFS method can prevent SQL injection attacks more effectively than the traditional keyword filtering and regular expression filtering methods.

Abstract:When indoor robot adopts the odometer method for long distance navigation, the positioning accuracy decreases rapidly. In the past, the accuracy of the artificial landmark location scheme is low, and it is difficult to meet the real-time requirement of navigation. To deal with these problems, this paper designs the artificial landmark, which can be quickly and accurately identified, so as to correct the accumulative navigation errors by the robot odometer method, and the location information of the artificial landmark and the odometer method is effectively integrated with Calman Filter. The experimental results show that the recognition accuracy of digital artificial landmark is high, and recognition speed meets the real-time requirement of navigation. This method can effectively improve the robustness and accuracy of the mobile robot when it adopts the odometer method to navigate.

Abstract:Based on the analysis of the existing trajectory model, this paper proposes a trajectory similarity calculation model and a moving object trajectory prediction model based on moving object acceleration and trajectory deflection angle. The moving object trajectory prediction method is proposed by comprehensive calculation and prediction model. The method comprises the following steps. 1) The historical trajectory is clustered based on the similarity of the trajectory to form the training cluster, and the trajectory similarity of the trajectory data of the target moving object is calculated based on each training cluster to find the maximum similarity historical trajectory. 2) The trajectory prediction is made based on the historical trajectory and the prediction model of moving object acceleration and trajectory deflection angle. The experimental results show that the prediction accuracy of the method is more than 90% in 500 m range, and the prediction time is relatively short, which has high practical value.

Abstract:At present, in order to adapt to the coming of “big data and in-depth model” age, a feasible solution is put forward by using FPGA to realize hardware accelerator of various algorithms. In this study, by using Vivado HLS tools, a set of intelligent hardware acceleration architecture is designed based on the genetic algorithm, which can automatically generate TCL file by programming, and automatically call HLS tool to complete the simulation analysis and extract the data to analyze in the report. What's more, the case programs like FIR and DCT given by the Xilinx company are tested. A better solution is found in the experiments, and the efficiency is increased by magnitude compared with the manual methods. It has met the universality of the general algorithm in hardware acceleration.

Abstract:In order to solve the problems of reduced storage performance, reduced resource utilization, increased I/O response, which are caused by unbalanced load distribution of OpenStack, the weighted least-connection scheduling algorithm is improved. Researching load balancing scheduling algorithm for object storage, using CPU memory disk and I/O of the storage nodes information of resource utilization, considering the number of request connections, we calculate the storage node load capacity performance and weight. Load balancing determines task distribution direction according to the size of every weight. Experiments show that the improved load balancing scheduling algorithm can optimize the performance of storage, improve the data throughput storage performance and system stability.

Abstract:With the popularity of the Internet, microblogging as a representative of the social network has generated a lot of data. Exploring useful information from these data has become an important direction for today's research. According to the characteristics of microblogging text, this paper presents a method based on joint classifier to filter out noise microblogging, and then uses LDA model for subject discovery. The joint classifier model is composed of naive Bayesian, support vector machine and decision tree. The accuracy of the combined classifier is 87%, which can clearly show that this classification method is feasible and effective.

Abstract:The traditional adaptive genetic algorithm is slow in convergence and easy to fall into the local optimal solution. In order to resolve this problem, an adaptive genetic algorithm with density weighted is put forward in this study. Based on distribution density of population, this new algorithm can dynamically change the crossover probability and mutation probability of genetic algorithm, and combine with the best individual method. The results of the experiment show that the new algorithm can change the stability of local population, speed up the convergence, and improve its robustness and application.

Abstract:Genome-wide association studies (GWAS) refer to the method that uses correlation analysis to identify disease associated genes. Traditional research method did not consider the interaction between genes and had low accuracy and efficiency in the case of complex factors. Aimed at these aforementioned problems, this paper presents a key SNPs selecting algorithm based on mutual information. It constructs reversely the SNPs interaction network using simulation data based on the theory of mutual information and compares the difference of the statistics of SNPs interaction networks between case and control groups with the increase of the mutual information threshold. According to the selected threshold, we select the structural key SNPs. The results of experiments show that the method of parameter selection presented in this paper is useful to select the structural key SNPs.

Abstract:In the process of target tracking, the accuracy of tracker for single image source is not high enough and the target would be inclined to be lost when covered partially. A method of fusing the features of infrared image and visible light image is proposed in this study. First, the color information of the visible light image is extracted as a parameter in the target model, and the gray level information of the infrared image is taken as the other parameter. According to the two parameters, the target positions and its subgraphs can be acquired respectively. Then the corresponding Bhattacharyya coefficients are calculated by the anterior target subgraphs and the target models. The weights of the respective coefficients can be calculated on the basis of the weighting function. Finally, the target that is weighted with the Mean Shift algorithm could be tracked. This method makes full use of the advantages of infrared images and visible light images, improves the accuracy of tracker, and has solved the problem that the target is likely to be lost when partially covered.

Abstract:An automatic music classification method based on users' comments is presented in view of the few categories and limited search content for the existing music platforms. First of all, a linear CRF statistic segmentation model, n-gram word extraction and affinity analysis method are used to obtain a dictionary which can be adapted to music corpus segmentation. Secondly, we use linear CRF model to segment comments with dictionary above, and then we correct the segmentation result via split-merge testing. Thirdly, the optimized TFIDF keyword extraction model is applied to extract candidate tags, and we merge tags after that. Fourthly, candidate tags with fewer frequency are filtered from a global perspective. Finally, a probability classification network is established between the music and filtered tags to classify music. As the result shows, our music classification method achieves high accuracy. Furthermore, it can ensure the personality of music retrieval for generating music tags automatically in multiple dimensions according to users' comments.

Abstract:Scheduling is an important issue in the field of portfolio optimization, and batch scheduling is more complicated since it takes into consideration of workpieces sizes and machine capacity. In this study, to solve the batch scheduling problems with non-identical sizes, we propose a hybrid algorithm based on ant colony algorithm and fish swarm algorithm. By introducing the fish algorithm's swarm degree into the ant colony algorithm, the hybrid algorithm does not only avoid the premature, but also accelerates the convergence speed of the algorithm. In respect of load rate and utilization, the optimization algorithm has higher efficiency and achieves better results, because it can reduce searching time in finding optimal solution.

Abstract:The realization of model data exchange in power system is of great practical significance. In order to solve this problem and improve the interoperability of the system, a solution of analyzing CIM/E file using Java language is proposed, which is based on the study of the Common Information Model (CIM) in the IEC 61970 series standard developed by the International Electrotechnical Commission (IEC), the analysis method of XML file and the description of related equipment class attribute of grid physical model. The solution means to analyze the CIM/E grid model data into the structured data in memory, then store the data into database to prevent analyzing it again the next time to improve the reusability. This database we use is MySQL, and finally the exported grid model CIM/E file validates this solution. The analytical scheme has much significance of guidance and reference to the analysis of the model data in the actual power system.

Abstract:The buddy system implemented in the Linux kernel is studied, and the process of buddy algorithm is demonstrated with examples. There are three levels of data structure dedicated to manage page frames in physical address space. Firstly, these data structures and their relationships are presented. Then, the buddy algorithm used to allocate and deallocate page frames is described in detail. As far as a memory block to be deallocated is concerned, how to calculate the index of its buddy and coalesced memory block is the key point of deallocating operation. Several conclusions pertaining to this calculation are dealt with and proved.

Abstract:Point cloud splicing has a wide application in the three-dimensional object reconstruction. The scanning equipment may be limited by light, occlusion or object size, so that the scanning equipment cannot obtain all point cloud information of the object from the same angle. The accuracy of traditional ICP is influenced by the initial pose of the cloud with poor robustness. Aiming at this problem, this paper proposes a point cloud stitching algorithm with multi-view cloud data. When the feature points are selected, the coordinate axes are combined with the thresholds to set the search range of a threshold constraint candidate point, and the nearest point set of Euclidean distance is obtained. The point cloud stitching is carried out by ICP algorithm. The experimental results show that the algorithm is superior to the traditional ICP in time consuming, and the splicing accuracy has obvious advantages.

Abstract:Aiming at the problem of large scale terrain rendering in 3D visualization system of transmission line, a virtual terrain rendering algorithm based on static LOD is proposed. Firstly, the model of terrain elevation data in transmission line corridor is established. The viewing angle is moved into the evaluation factor at the detail level. And the evaluation function is improved by combining the view distance and terrain complexity. On this basis, the variance of the elevation values of the data points in the grid is determined by the threshold value. Different linear interpolation functions are used to simulate the terrain surface. The experimental results show that the proposed algorithm can effectively reduce the number of terrain triangles in motion. The virtual scene frame is more smooth and has a good effect of terrain simulation.

Abstract:Path planning is one of the focuses of soccer robot research. The previous path planning algorithms ignore the influence of the players' movement on their surrounding areas, leading to the insecurity of the actual optimal safe path. In order to solve this problem, an A^* algorithm for dynamic obstacle avoidance is proposed. The method divides the football pitch according to the influence of the dribbler and the opposing defensive players, and sets the risk value for each area, and then uses the improved A^* algorithm to plan the path. The experimental results show that this method can effectively reduce the possibility for the player with the ball to be surrounded by the other defensive players, and the length and the security of the optimal security path are considered synthetically. The planned path's performance is proved to be better.

Abstract:Many convenient wearable devices are being used for medical purposes, like measuring the heart rate (HR), blood pressure. With the sleep quality monitor problem, the key point is how to discriminate the sleeping state from waking one out of these signals. This paper proposes a Bayesian approach based on dynamic time warping (DTW) method for sleeping and waking classification. It uses HR and surplus pulse O₂ (SpO₂) signals to analyze the sleeping states and the occurrence of some sleep-related problems. The DTW is used to extract features from the original HR and SpO₂ signals. Then a Bayesian classification method is introduced for the discrimination of sleeping and waking states. Finally, a case study from a real-world applications, collected from the website of the Sleep Heart Health Study, is presented to show the feasibility and advantages of the DTW-based Bayesian approach.

Abstract:At present, the BWDSP104X compiler deals with the conditional branching in the program by using the traditional predicate optimization method, and each instruction is associated with a predicate. Only when the predicate is true, can the instruction be executed. But its existence is limited when the conditional predicates do not eliminate jump branches, and there may be control dependencies between multiple conditional predicates, which is detrimental to instruction parallelism and instruction flow. Therefore, in the framework of the existing compiler, this paper proposes a multi-condition predicate compiler optimization method based on BWDSP104X architecture in view of the shortcomings of traditional predicate optimization method. The experimental results show that the optimization algorithm can achieve an average speed of 5.62 on the BWDSP104X compiler compared with the traditional predicate optimization method.

Abstract:The identification of blend feature plays a very important role in the feature recognition and reconstruction of design features. In this paper, a method of blend feature recognition based on design intent reasoning is proposed. Firstly, the geometrical shape of the blend surface is identified, and then the design intention of the general blend feature is captured. The blend features are identified and their order is deduced. A directed acyclic graph is used to describe the relationship among blend features, and the sequence of blend features is obtained by topology sorting of the directed acyclic graph. Finally, an example is given to demonstrate the effectiveness of the proposed method.

Abstract:The mobile ad hoc network is a multi-hop wireless network, which relies on the cooperation and retransmission between nodes to enhance the performance of networks. If the content is deployed in the appropriate location in the network, it will reduce the nodes' cost to get the data, and improve the performance of the network greatly. Since nodes are rational, they do not want to work with others on content deployment if there is no profit. In our work, an incentive method based on virtual credit is used in content deployment to encourage the cooperation among nodes on content deployment, then the performance of networks can be improved and the cost can be reduced. The experimental results show that the incentive method can reduce the content deployment cost effectively.

Abstract:Diabetes is becoming a more and more serious health challenge worldwide with the yearly rising prevalence, especially in developing countries, where the vast majority of diabetes are type 2 diabetes. Scientific research has proved that about 80% of type 2 diabetes complications can be prevented or delayed by timely detection. In this study, we propose an ensemble model to precisely diagnose the diabetes in a large-scale and imbalance dataset. The dataset used in our work covers millions of people from one province in China ranging from 2009 to 2015, which is highly skew. Results on the real-world dataset prove that our method is promising for diabetes diagnosis with a high sensitivity, F₃ and G-mean, i.e., 91.00%, 58.24%, 86.69%, respectively.

Abstract:Aiming at the problem that training data is insufficient due to little training data in speaker recognition system, this paper adopts GMM-UBM as the background model which can identify the characteristics of the target speaker. And SVM is introduced to solve the problem of poor robustness of the system caused by GMM-UBM. It has much influence on SVM identification performance with different kernel functions. Aiming at the Characteristics of Polynomial kernel with good generalization ability and poor earning ability and Gaussian kernel with good earning ability and poor generalization ability, it structures a new combination kernel function which combines the advantages of each single kernel function by linear weighted method. The experimental results show that the recognition rate and Equal Error Rate of the combination kernel is more ideal than other kernel functions. And it achieves satisfactory recognition rate and robustness in the situations of different signal-to-noise ratio.

Abstract:Based on the experimental platform whose main controller is Samsung S3C2410 chip, this article designs and implements a more efficient baud rate adaptive CAN driver. In this paper, the CAN drive structure is introduced and the principle of baud rate adaptive CAN driver is explained in detail. This scheme is combined with the common polling method and sampling method. At the same time, a new scheme is proposed to improve the efficiency of the drive by adding the user's input to the baud rate adaptive process. Finally, this paper carries out the data communication test and the performance analysis for this driver.

Abstract:The routing protocol is one of the important members of the TCP/IP protocol family, which contains the internal and external network protocol. The common network protocol in college is static routing and OSPF routing. This paper analyzes and summarizes the related theories of OSPF protocol and static routing, and initiatively closes down the network services by introducing the origin of the static black hole road according to the demand of campus network security management. Based on this, the paper introduces the black hole routing network architecture into the security management of campus network, and gives the network architecture of the core and the specific configuration process. With the aid of black hole, the routing can quickly close down the accident point to prevent more hazards, and reduces the negative impact on the entire campus network. Eventually, through the network the validity of the black hole routing configuration is tested.

Abstract:The development of computer science makes military weapons and equipment update fast. In the highly-informed society, the intelligent information processing technology in military, is badly needed. This paper proposes an identification method with the model of deep neural network based on the character of word vector and state. It's for weaponry in electronic text, such as aircraft, tank vehicle, artillery missile and missile weapon. The experiment shows the value of F-1 which equals 0.9102 on the test corpus.