Aiming at problems regarding the recognition of motion states by existing smartphones, such as poor real-time performance, less movement category and complex algorithm, this paper proposes a method of using smartphone sensors to recognize six kinds of real time human movement states. Firstly, daily human movement data is acquired through smartphone acceleration sensors and gravitational acceleration sensors, and original data is handled with correction, smoothing, segmentation and direction-independent processing. Secondly, the footsteps identification algorithm is used to calculate peaks and troughs of footsteps from which the time-domain feature vectors are extracted. Finally, the movement states are classified according to feature vectors, and the Hierarchical Support Machines (H-SVMs) is used to recognize daily movement states. Experimental results show this method can effectively reduce the computational load of smartphones and improve real-time performance and accuracy of movement states recognition. This method is suitable for other similar behavior recognitions.

Submitted on January 21, 2018; Revised on February 15, 2018; Accepted on March 21, 2018
References: 26

The semi-supervised learning method has been widely used in the field of pattern recognition. Semi-supervised Extreme Learning Machine (SELM) is a typical semi-supervised learning algorithm. The graph construction result of the sample data has a tremendous impact on the SELM algorithm. In traditional graph composition methods such as Laplace graph, LLE graph and K neighboring graph, neighborhood parameters are specified by humans. If there are noises or uneven distribution in the data, the results are not very good. This paper proposes a SELM algorithm based on L1-Graph, which features no specifying parameters, is robust against noise, has a sparse solution and so on. The experiment confirms the effectiveness of the method.

Submitted on December 27, 2017; Revised on January 28, 2018; Accepted on February 24, 2018
References: 9

The vehicle routing problem (VRP) has become one of the focus issues in operations research and management sciences over the past two decades. One of its principal branches is the dynamic vehicle routing problem (DVRP), which can receive new order requests during the service process and make a timely response, unlike static vehicle routing problems (SVRP) where all information is known before the optimization starts. In this paper, we solve DVRP while using an enhanced genetic algorithm (GA) that tries to increase both diversity and global searching ability. The maximum saving method and the nearest neighbor method are adopted in the crossover operation to improve the path selection. Considering the near distance priority service principle (NDPSP) in the actual operation, a new assessment scheme with penalty factors is applied to our individual assessment. In addition, a paired-t test as a non-parametric statistical analysis is implemented to demonstrate the efficiency of the enhanced genetic optimization algorithm, based on a publicly available VRP benchmark, which includes 21 data sets. Analysis results show that our approach outperformed the published approached based on optimizing results.

Submitted on January 8, 2018; Revised on February 13, 2018; Accepted on March 22, 2018
References: 26

Convolutional neural networks (CNNs) have recently achieved remarkable success with superior performances in computer vision applications. In most CNN-based image retrieval methods, deep CNNs features are verified as discriminative descriptors for effective image representation. This paper exploits the best practice for CNNs application to national costume image retrieval. Several important aspects that affect the discriminative ability of deep CNNs features are investigated thoroughly, including layers selection, aggregation and weighting methods. Firstly, an effective weighting method for sum-pooling features aggregation is given, which is more suitable for national costume image than some typical aggregation methods such as SPoC and SCDA. Secondly, in view of the complementary strengths, compact multi-layer CNN features combined with low dimensions are proposed and proven to be effective for national costume expression. Finally, a re-ranking strategy of diffusion process is applied to further enhance the performance for national costume images retrieval. The experimental results show that the proposed method outperforms the existing methods remarkably, which will provide some new research ideas and technical references for researchers in the field of national costume image retrieval.

Submitted on December 17, 2017; Revised on January 29, 2018; Accepted on March 8, 2018
References: 33

Regarding SVM multi-classification problem, optimizing the parameters of SVM has become the key problem to improve the performance of the SVM multi-classification algorithm. In order to solve this problem, multi-chromosome genetic algorithm is proposed in this paper and used to optimize these parameters. In the SVM multi-classification decision tree, the algorithm constructs a chromosome for SVM parameter of each node and improves the corresponding rules of crossover and mutation in the genetic algorithm. The improved genetic algorithm optimizes the parameters of SVM in all nodes in the SVM multi-classification decision tree. The experimental results show that the SVM multi-classification decision tree algorithm using the multi-chromosome genetic algorithm has higher classification quality, compared with the traditional multi-SVM multi-classification algorithm.

Submitted on January 3, 2018; Revised on February 14, 2018; Accepted on March 25, 2018
References: 18

As more and more students fail in course studies, higher education is now facing challenges regarding increasingly lower course completion rates as well as overall graduation rates. However, failures in course studies is a comprehensive result of various factors, which is characterized by uncertainty. To deal with this issue, fuzzy sets theory and fuzzy logic are advantageous compared with traditional methods. In this study, based on dynamic student study process data, a fuzzy synthetic evaluation method for dynamic early warning student study failure risk is provided. For each student, three specific early warning factors: 1) student course participation, 2) assignment earned points, and 3) student attendance record, are selected as risk indicators, and the overall risk level is determined by a fuzzy synthetic evaluation approach, which can dynamically give the situation of risk as the evaluation time point changes. Finally, our obtained results show that the employed method is good for identifying at-risk students and exploring the risk reasons by showing the degrees of each early warning factors to the overall risk level. It is of significance for educators to timely apply corresponding strategic pedagogical interventions to help at-risk students avoid academic failure.

Submitted on February 1, 2018; Revised on February 26, 2018; Accepted on March 28, 2018
References: 34

Trie is fundamental to many applications, such as natural language processing, string similarity search and join, but it suffers from a high space overhead. Double array (DA) provides a new way to reduce the space overhead but suffers from low construction efficiency. There is an urgent demand to promote the construction efficiency of DA while maintaining a low memory overhead. To address this problem, we reveal that the collisions generated during DA construction process mainly contribute to the low construction efficiency. Based on this analysis, a partition double array (PDA) is proposed in this paper. PDA can reduce the number of collisions as well as the cost of handling collisions in DA, so higher construction efficiency is guaranteed. Experiments on real dataset indicates that PDAs have a construction efficiency 15x higher than DA. We also obtain a bonus 2.7x higher retrieval efficiency compared with DA.

Submitted on January 2, 2018; Revised on February 13, 2018; Accepted on March 26, 2018
References: 17

Recently, the keyword query in relational databases has received widespread attention. The traditional methods typically traverse the entire database for the final results. With the database, structure becomes more complex and its size increases quickly; the efficiency of above methods cannot be ensured. To solve this issue, we propose a hypergraph-based keyword query method. First, the concept of hypergraph is formally defined to model the relational database. Second, the strategy of multi-granular index construction is presented to prune the irrelevant supernodes. Then, a filtering-validating query method is put forward based on the above index. Finally, experiments are taken on the dataset DBLP to verify the validity of the proposed method.

Submitted on December 21, 2017; Revised on January 30, 2018; Accepted on March 3, 2018
References: 20

Authentication technology in the cloud environment cannot completely prevent the destruction of malicious users to cloud resources. The analysis and evaluation of the user behavior has become the key to effectively improve the security of cloud. This paper improves a multi-parameter behavior trust model (MTEM), which based on Beth model and Josang model. The MTEM model introduces a number of parameters involved in the transaction process of users and cloud service providers. Furthermore, we calculate the behavior weight value by using AHP, through fuzzy logic analysis of user behavior. According to the principle of maximum membership degree, we verified the feasibility and validity of the model by simulation. The simulation results show that the MTEM model can improve the detection accuracy of user's malicious behavior and the effectiveness of the users.

Submitted on January 3, 2018; Revised on January 31, 2018; Accepted on March 5, 2018
References: 18

For the various sizes of random tasks, the possible cost is constrained in the process of cloud resources scheduling. The electricity price of the worldwide dynamic time zones is proposed based on the different electricity price of the world time zone characteristics, the network bandwidth and load levels. The optimization model of energy consumption of such system with execution cost as constraint condition is proposed, which optimizes the energy consumption of cloud system through the load level, electricity price and other factors in the resource scheduling process. In this model, the task hierarchical strategy is designed to realize the hierarchical task energy consumption. Thus, the scheduling algorithm of energy optimization with cost constraint is proposed. The results of experiments show that the algorithm can both optimize the energy consumption and reduce the service cost.

Submitted on December 25, 2017; Revised on February 2, 2018; Accepted on March 16, 2018
References: 11

In order to improve the optimization accuracy and convergence speed of gravitational search algorithm, the gravitational search algorithm with the mechanism of adaptive mixed random mutation is proposed. A mutation trigger function with adaptive property is introduced into the algorithm, so that every particle has the probability of mutation at any time, and the number of particles that change in the population tends to decrease with the increase of iteration times. At the same time, in the whole optimization process of the algorithm, the uniform mutation and Laplace-normal hybrid mutation cooperate together. The uniform mutation enables the algorithm to find the global optimal area quickly, and then continue deep search with hybrid mutation to improve the optimization performance. The simulation results show that in solving the problem of extremum optimization, the improved algorithm has significantly improved optimization performance, and has high convergence accuracy and faster convergence speed.

Submitted on January 10, 2018; Revised on February 24, 2018; Accepted on March 27, 2018
References: 18

A key problem in user relationship analysis is the identification and representation of user interest. The basis to tackle this issue is user similarity measures. In social tagging system, users collaboratively create and manage tags to annotate and categorize content for searching and recommending. Due to the contribution to reflect users’ opinions and interests, tags are metadata for user similarity measures. However, there are some issues about it such as data sparseness, the user none-distinguished interest areas and relatively little consider about user influence. This article argues a similarity measure method that based on user’s interest topic division. First, we construct tag clustering and divide the user community according to user interest areas. Second, we improve user similarity measurement model using social network analysis (SNA) and PageRank. Finally, the validity of the improved method about user similarity calculation is verified using del.icio.us data set. Experimental results show that the improved method gets the highest P@N and sorting accuracy compared with the traditional tag-based user similarity.

Submitted on December 22, 2017; Revised on January 30, 2018; Accepted on March 8, 2018
References: 16

For the problems in net primary productivity estimation of forest vegetation such as complex model, great difficulty in parameter acquisition, only appropriate for specific area and slow remote-sensing data processing platform computation speed, etc., the improved net vegetation primary productivity estimation model (Cloud-ICASA) is proposed by using the domestic GF-1 high resolution image based on the specific ecological environment of research region Mao’er Mountain forest farm. The Spark-based remote-sensing data processing platform is constructed to process the remote-sensing image in parallel environment. The research results show that the improved Cloud-ICASA model simplifies the parameters, improves the estimation accuracy and is appropriate for estimation of net primary productivity for the vegetation in research region. The Spark based remote-sensing data processing improves the node utilization rate, increase the computation speed and can satisfy the real-time dynamic evaluation requirements.

Submitted on January 11, 2018; Revised on February 18, 2018; Accepted on March 21, 2018
References: 17

In order to solve the problem that control and forwarding are closely coupled in traditional network, the network lack of innovation and programmability, and the network management and maintenance difficulty, an SDN load balancing method based on K-Dijkstra is proposed. By using SDN technology to achieve the separation of control and forwarding, the controller is responsible for the global scheduling, making the network flow and task scheduling more flexible than the traditional network. Through the integration of flow management, traffic monitoring, dynamic load balancing and load calculation in the SDN control layer, the K-Dijkstra algorithm and the HRRF algorithm are combined in the load balancing module to solve the problem of path selection. The traffic environment also has a better load balancing effect, optimizing the control layer structure, improving network management efficiency and achieving dynamic load balancing of network traffic. The simulation results on Mininet show that the method can significantly improve network delay, packet loss and throughput compared with traditional networks.

Submitted on January 5, 2018; Revised on February 23, 2018; Accepted on March 27, 2018
References: 16

The streamline visualization method can intuitively reveal the nature and change law of flow field. However, the streamline visualization method easily causes the disorder of visualization effect, which is especially serious in the 3D flow field. With the improvement of calculation accuracy, the time consumed by calculation in the streamline generation process also increases accordingly, which impacts the visualization efficiency. In view of the above problems existing in the streamline visualization method, this paper researches the parallel visualization method of flow field based on streamline similarity. First, this paper judges the streamline shape similarity through the feature points and designs the method for judging 3D streamline similarity by combing the methods for judging the distanced-based similarity between streamlines. This reduces the streamline disorder phenomena in visualization results. Second, this paper utilizes the parallel computing technology to improve the computational efficiency. On the basis of seed point parallel strategy, this paper proposes the parallel task partition method that combines the equal partition of tasks with repartition of redundant tasks. In view of the blocking wait that exists in parallel judgment of similarity, this paper proposes the cyclic check method based on a double-caching line to avoid the blocking wait problem, obviously reducing the flow field visualization time and effectively improving the flow field visualization efficiency.

Submitted on December 25, 2017; Revised on February 2, 2018; Accepted on March 10, 2018
References: 14

Multi-node and multi-feature fusion is an important approach for digital modulations signal recognition in modern communication field. Information obtained from multi-node and multi-feature needs to be fused because of incompleteness of single feature and uncertainty of single node. As a powerful method for data fusion and conclusion reasoning in uncertain environments, evidence theory is widely used. Establishing reliable BPA is the prerequisite for evidence fusion. In this paper, in order to improve the coincidence of basic probability assignment (BPA) with real probability, the notion of bell function (Bell-F) based similarity evaluation model (SEM) is introduced. Through comparative experiments, it is proved that the new method based on Bell-F is effective for BPA acquisition. Furthermore, a new information fused based digital signal modulation recognition scheme is described. Finally, a case study is given to illustrate the performance of the proposed model. Through test and calculations under the digital modulation signal data set, under , the recognition rate based on the Bell-F fusion method is above 90%, which is 20% higher than methods without fusion. Under or less, the integrated recognition rate of the Bell-F is 40% higher than the Gray relation method.

Submitted on January 4, 2018; Revised on February 17, 2018; Accepted on March 26, 2018
References: 25

In the transmission process of 1090ES ADS-B signal, it is easy to see the relationship between signal overlap and signal interference. The signal noise under interference easily causes the error of ADS-B signal decoding. In this paper, a 1090ES ADS-B signal separation algorithm based on PCA and FastICA is designed to achieve the separation of ADS-B signals and the separation of overlapped signals. Through simulation and verification, the 1090ES ADS-B signal separation algorithm based on PCA and ICA can realize the denoising and separation of ADS-B signal and reduce the computation and speed of separation. The signal reduction degree after separation is relatively high.

Submitted on January 15, 2018; Revised on February 18, 2018; Accepted on March 29, 2018
References: 13

Gas turbine engines always run during poor working conditions that have high temperatures, high pressures, and high mechanical and thermal stress. Thus, the performance of the gas path components gradually degrades, leading to serious faults. So, the health status of the engine gas path components provides essential information for users and operators. Here, a new gas path analysis approach has been developed to predict gas turbine engine health status by using gas path measurements. The developed approach has been tested in seven test cases where the degradation of a model gas turbine engine similar to a three-shaft marine engine has been analyzed. The case studies have shown that the approach can accurately and quickly detect, isolate, and quantify the degradation of major engine gas path components with the existence of measurement noise. The test cases have also shown that the time cost by the approach is short enough for its potential application of online health monitoring.

Submitted on December 20, 2017; Revised on February 2, 2018; Accepted on March 26, 2018
References: 17

It is difficult to apply the bitcoin transaction system’s blockchain technology outside the electronic currency transaction. This paper proposes a Multi-Transaction Mode Consortium Blockchain (MTMCB), which generalizes the traditional buying and selling transaction into event processing of the business system. It prevents all kinds of high value event processing results from repudiation and tampering. MTMCB is low-cost and obtains credible results without accumulating technology. It extends the user type (not only the PC user), the storage type (not only local storage) and the storage content (not only the block data). MTMCB optimizes transaction verification, consensus decision, block generation, block storage, comparative verification and block linking mechanism. It provides visual audit services, including event proof, abnormal transaction early warning, tampering discovery and reconstruction, etc. Meanwhile, example simulation, security analysis and performance analysis are processed. Its results show that MTMCB has the same security level as the bitcoin core transaction system, and it has good adaptability in data resource protection of high value event processing.

Submitted on January 13, 2018; Revised on February 12, 2018; Accepted on March 23, 2018
References: 17

Differential Evolution (DE) is a heuristic stochastic search algorithm based on population differences, which has advantages of simple parameters and fast convergence rate. However, it has weak robustness, especially for multimodal problems. Therefore, this paper proposes a Cooperative Differential Evolution with Dynamical population (DynCDE). In the proposed algorithm, the K-means method is employed to partition the whole population. For the high convergence rate of DE/current-to-best/ 1/bin, the neighbor-based mutation strategy is applied and the dynamic population size method based on aging mechanism and lifecycle mechanism is designed to keep the balance between exploration and exploitation. This modified DE has the potential to improve prediction accuracy of neural networks. Finally, this DynCDE-based neural network model is applied to solving the short-term traffic flow prediction problem, which offers very excellent results.

Submitted on January 13, 2018; Revised on February 16, 2018; Accepted on March 24, 2018
References: 9

With the development of information technology, there exists a large amount of multi-media data in our lives; the data is heterogeneous with low-level features while consistent with semantic information. Traditional mono-media retrieval can’t cross the heterogeneous gap of multi-media data, and cross-media retrieval is arousing many researchers’ interests. In this paper, we propose a two-stage semantic matching for cross-media retrieval based on support vector machines (called TSMCR). Our approach uses a combination of testing images’ predictive labels and testing texts’ predictive labels as the next training labels. It makes full use of semantic information of both training samples and testing samples, and the experimental results on four state-of-the-art datasets show that the TSMCR algorithm is effective.

Submitted on December 29, 2017; Revised on February 2, 2018; Accepted on March 20, 2018
References: 9

Based on the evaluation model of the entropy evaluation method and mutation series method, this paper evaluates the independent innovation ability of high-tech industry areas of China during the period of 2005 to 2011. It also emphatically analyzes the independent innovation of an underdeveloped area of China, the Heilongjiang Province. The result of this evaluation shows that the independent innovation ability of the high-tech industry in Heilongjiang Province has gradually reduced, expanding the gap between the high-tech industry development of Heilongjiang and that of the national average. In recent years, the high-tech industry development of Heilongjiang has already fallen further behind the Eastern areas of China.

Submitted on January 3, 2018; Revised on February 22, 2018; Accepted on March 23, 2018
References: 12

On account of the fact that a fractional Brownian motion (fBm) with the Hurst parameter H ∈(0,1/2)∪(1/2,1) cannot follow the laws of the semimartingale and the Markov process, little work is presented about the T-stability for stochastic differential equations (SDEs) with fBm. Here, three results are obtained for the generalized Black-Scholes model (SDE) with H∈(1/3,1/2). Firstly, the sufficient conditions of the stochastical and asymptotical stability in the large for such equation are presented by the aid of the Lyapunov exponent. Secondly, the Euler-Maruyama (EM) numerical algorithm with a given step-size for such model is constructed. Lastly, by taking advantage of the stable average function, the sufficient conditions of the T-stability that originated from the EM algorithm are presented. All the results show that on the basis of the stability of such equation, the T-stable region produced by the EM algorithm can be found. Moreover, one numerical example is afforded to the main conclusions.

Submitted on December 21, 2017; Revised on January 29, 2018; Accepted on March 5, 2018
References: 32

After analyzing the disadvantages of the centralized multidimensional localization algorithms MDS-MAP (Multi-Dimensional Scaling-MAP) in positioning accuracy and computational complexity, we present a new localization algorithm based on a set of statistical vectors (SV). The solving equation of the double center matrix can be simplified by node coordinate transformation. In order to reduce the noise disturbance and decrease the effect of ranging error on the followed location accuracy, a new coordinate inner product matrix can be reconstructed by using a set of statistical vectors, which can be used to calculate the node coordinates directly. This algorithm can realize centralized localization, distributed localization and incremental localization of nodes.

Submitted on January 3, 2018; Revised on February 29, 2018; Accepted on March 25, 2018
References: 14