Computational Intelligence and Intelligent Systems

Particle Swarm Optimization (PSO) is slow but steady learner although it exhibits strong competence in solving complicated problems. However, during the course of searching process, the particles gradually gather into the vicinity of the best particle found so far. Furthermore, some evidences show that the unreasonable setting of its inertial term in the kinetic equations may lead to slow convergence of PSO. Thus, a differential opposition-based particle swarm optimization with adaptive elite mutation (DOPSO) is presented to overcome these drawbacks in this paper. There are two strategies are introduced into DOPSO to balance the contradiction between exploration and exploitation during its searching process: (1) Firstly a new particle’s position update rule in which differential term replaces the inertia term is designed to accelerate its convergence; (2) Secondly an adaptive elite mutation strategy (AEM) is included to avoid trapping into local optimum. Experimental results show that the proposed method has a significant improvement in performance compared with some state-of-art PSOs.

In this paper, a hierarchical cooperative algorithm based on the genetic algorithm and the particle swarm optimization is proposed that utilizes the global searching ability of genetic algorithm and the fast convergence speed of particle swarm optimization. The proposed algorithm starts from Individual organizational structure of subgroups and takes full advantage of the merits of the particle swarm optimization algorithm and the genetic algorithm (HCGA-PSO). The algorithm uses a layered structure with two layers. The bottom layer is composed of a series of genetic algorithm by subgroups that contributes to the global searching ability of the algorithm. The upper layer is an elite group consisting of the best individuals of each subgroup and the particle swarm algorithm is used to perform precise local search. The experimental results demonstrate that the HCGA-PSO algorithm has better convergence and stronger continuous search capability, which makes it suitable for solving complex optimization problems.

As an intelligent algorithm inspired by the foraging behavior in nature, particle swarm optimization (PSO) is famous for its few parameters, easy to implement and higher convergence accuracy. However, PSO also has a weakness over the local search, also called the prematurity, which resulted in the convergence accuracy reduced and the convergence speed slowed. For this, extremal optimization (EO), an excellent local search algorithm, has been introduced to be improved (CEO) and enhance the local search of PSO. Meanwhile, for improving its global search further, an improved opposition-based learning based on refraction principle (UOBL) has been chosen to enhance the global search of PSO, which is a better global optimization algorithm. In order to balance both of PSO to improve its optimization performance further, an adaptive hybrid PSO based on UOBL and CEO (AHOPSO-CEO) is proposed in this article. The large number of experiment results and analysis reveals that AHOPSO-CEO achieves better performance with other algorithms on the convergence speed and convergence accuracy for optimization problems.

Problem decomposition and subcomponent optimization play a key role in cooperative coevolution (CC) for large scale global optimization. In this paper, we firstly introduce a new variable interactions identification (VII) method to recognize the indirect decision variables. Then, we proposed a new reallocate computational resources method, aims to give more computational resources to the more important subcomponents. Hence, a novel ITÖ algorithm with cooperative coevolution (CCITÖ) strategy based on above two strategies is proposed. In order to understand the characteristics of CCITÖ, we have carried out extensive computational studies on the CEC’2010 benchmark function. Experimental results show that our algorithm achieves competitive results compared with other four state-of-the-art algorithms in the large scale global optimization problems.

During the last decade, multi-objective approaches to dealing with constraints in evolutionary algorithms have drawn more and more attention from researchers. In this paper, a conical area differential evolution algorithm (CADE) with dual populations is proposed for constrained optimization by borrowing the ideas of cone decomposition for bi-objective optimization. In CADE, a conical sub-population and a feasible one are designed to search the global feasible optimum along the Pareto front and the feasible segment, respectively. The conical sub-population aims to construct and utilize the Pareto front by a biased cone decomposition strategy in geometric proportion and a conical area indicator. Afterwards, neighbors in both sub-populations are adequately exploited to help each other. 13 benchmark test instances are used to assess the performance of CADE. The result reveals that CADE is capable of producing significantly competitive solutions for constraint optimization problems compared with the other popular approaches.

Cluster analysis is one of the important research contents in data mining. The basic Particle Swarm Optimization algorithm (PSO) can be combined with the traditional clustering algorithm to achieve clustering analysis. Aiming at the disadvantages of the basic particle swarm optimization algorithm is easy to fall into local extremum, the search accuracy is not high, and the traditional K-means and FCM clustering algorithm are affected by the initial clustering center. This paper proposes a new particle swarm clustering algorithm based on tree structure and neighborhood (TPSO), which designs the structure of the particle group as a tree structure, uses the breadth of traversal, increases the global search ability of the particle, and joins the neighborhood operation to let the particle close to the neighborhood optimal particles and accelerate the convergence speed of the algorithm. Our experiments using Iris, Wine, Seed, Breast-w4 group of UCI public data sets show that the accuracy obtained by the TPSO algorithm implementing the proposed K-means and FCM is statistically significantly higher than the accuracy of the other clustering algorithms, such as K-means algorithm, fuzzy C-means algorithm, the basic particle swarm optimization combined with traditional clustering algorithm, etc., Comparison experiments also indicate that the TPSO algorithm can significantly improve the clustering performance of PSO.

In the design and optimization process of ultra-wideband antenna, the fitness function is unknown and the antenna modeling process is complex. To solve these problems, this paper introduces a method to construct evaluation function. The method is based on MATLAB and HFSS joint simulation platform. The antenna modeling and simulation analysis process is packaged as a ‘black box’ which acts an evaluation function of the optimization algorithm. Real-time data exchange is carried out through the joint simulation platform. The particle swarm optimization algorithm (PSO) is used to optimize the antenna structure parameters automatically. Simulation results show that this method can replace the mathematical fitness function and simplify the antenna modeling process. In addition, the proposed method can reduce the antenna return loss effectively and improve the overall performance of the antenna.

It is generally accepted that the clustering-based differential evolution (CDE) algorithm exhibits better performance in comparison with the standard differential evolution. However, such clustering method mechanism that is only based on input data may lead to some limitations such as premature convergence. In this study, we propose a divisive multi-level differential evolution algorithm (DMDE) to alleviate this drawback. The proposed divisive method is based not only input data but also the output fitness. In particular, DMDE becomes the conventional CDE when the output fitness in not considered in the process of clustering. Several benchmark functions are included to evaluate the performance of the proposed DMDE. Experimental results show that the proposed DMDE exhibits a promising performance when compared with CDE, especially in case of high-dimensional continuous optimization problems.

After MapReduce Parallel refactoring, there occurred an enormous difference between the latest version of MapReduce parallel and the old version on framework. At first, this paper introduces the framework of the old version of MapReduce parallel, the procedure of implementing tasks and how to schedule tasks and resources allocations, etc. Also, it points out the limitation of the old parallel. Then, it explained a framework of the new version of MapReduce parallel, task scheduling and resource allocation and so on. From the perspective of the framework, task scheduling, and resource allocation, it compared these two different versions, putting forward the advantages of the renewed framework. New-generation of MapReduce has a framework YARN, which is sharing model, in other words, it’s an application program that enables various compiling of computing structures to run on the same cluster. Meanwhile, it makes operation and maintenance more smooth and makes full use of cluster resource.

A novel meminductor circuit model is showed in this paper, and the mathematical model is also derived. By utilizing this circuit model, a simple series chaotic circuit is constructed by resistor, capacitor, and meminductor. Stability and dynamical behaviors of the third-order meminductor chaos circuit are studied by a variety of nonlinear analyzing tools. The result of the research shows that the system has complicated nonlinear phenomena; it is very suitable for confidential communication.

Urban traffic is an extremely complex dynamic system. Urban traffic modeling and forecasting is still a challenge, the main difficulty is how to determine supply and demand and how to parameterize the model. This paper tries to solve these problems with the help of a large number of floating taxi data. We describe the first solution to the challenge of finding a taxi destination. The tasks included at the beginning of its trajectory prediction of a taxi destination, it is expressed as the GPS point of variable length sequences, and related information, such as the departure time, the driver id and customer information. We use a neural network based approach that is almost completely automated. The architecture we are trying to use is a multi-layer perception, bidirectional recursive neural network, and a model inspired by the recently introduced memory network. Our approach can be easily adapted to other applications, with the goal of predicting the fixed-length output of a variable length sequence.

This paper introduces a time satisfaction function to build the emergency materials transshipment model, combining the traditional point to point transport model and hub-and-spoke distribution mode. The proposed model, emergency materials transshipment model with time window constraints, has two vital factors that are quantity and time of emergency material transportation. The quantity of materials is considered as the weight of time satisfaction. The total time satisfaction is the sum of product of the quantity and time satisfaction. A hybrid of colliding body’s optimization (CBO) and genetic algorithm (GA) imbedding with linear programming algorithm is proposed to solve the problem through analyzing the trait of the model. The hybrid algorithm improves the performance of CBO algorithm in the discrete field. Experimental results demonstrate the efficiency of the model and algorithm.

The Primal-dual interior-point methods with a filter are one of hot issues in optimization with both equality and inequality constraints. Extensive attention has been paid and great progress has been made for a long time. Interior-point methods not only have polynomial complexity, but are also highly efficient in practice.In this paper, we first generalize the dual interior filter algorithm for referred paper. Then a dual interior filter algorithm based on the orthogonal design is proposed. The orthogonal design is used to evenly sample the solution space, then the points with small constraint violations are chosen as the initial points of the algorithm. The dual interior point filter algorithm based on orthogonal design (DIPFA-OD) choosing the initial points is compared with the dual interior point filter algorithm based on random initial points (DIPFA-RND). The experimental results show that DIPFA-OD has a faster convergence speed and obtains better objective values than DIPFA-RND.

A beam search algorithm is presented to solve the 2D rectangular packing problem. The basic algorithms work according to 7 rule vectors of heuristic selection rules designed to select a corner-sticking action. Furthermore, the trade-off scheme of breadth first search (BFS) and the depth first search (DFS) increases the algorithm’s effectiveness and efficiency. The improved version of the algorithm adopts a rough phase to get a height for the stripe and a refine phase to obtain better solution for the problem. Computational experiments run on two sets of well-known benchmark instances and the computational results show that the algorithm outperforms the current best algorithms. Especially, for two benchmark instances ZDF6 and ZDF7, our algorithm finds the best packing configurations so far.

If the university management is going to keep up with the pace of development of the times. In this process, we should give full play to the dominant position of students. Because in the present social environment, college students on the application of advanced scientific and technological achievements have a high sensitivity. This reality is the interest of college students. Therefore, colleges and universities is to implement effective management of students. If it still use the traditional management model, it is difficult to play the timeliness of student management. We are now in the information age, multimedia technology in all areas of society are very popular. College students lead the forefront of network technology, so the use of traditional management model is a clear lag of management. Information environment, colleges and universities must use multimedia technology to carry out student management, in order to achieve student management innovation and improve the efficiency of student management. In this article, we first make a simple price introduction to the research background of university student management and bring out the key word of multimedia technology. Then, we will show the current mode of university student management and the research status quo. In view of this status quo, the main problems that exist in the management of college students are pointed out. Finally, according to the problems encountered in the actual process, put forward specific solutions to the management of colleges and universities.

With the popularity of social media and GPS equipment, a large number of the location-text data have been produced in the form of stream. The popularity leads to a variety of applications, such as based-location advice and location information transmission. Existing top-k location-text publish/subscribe schemes need subscriber to set a threshold, k value (the number of returned top results) and preference parameter $$ \delta $$ δ (the decision of which one is more important about location or text). The threshold brings lots of disadvantages to subscribers and publishers. Therefore in this paper, we propose an efficient top-k location-text publish/subscribe scheme without threshold which is named as TGT. Our scheme only needs subscriber to input k value and preference parameter $$ \delta $$ δ . Then TGT returns top-k results to the subscriber based on k and $$ \delta $$ δ without any threshold. Therefore, our scheme can reduce redundant computation, improve the recall ratio and facilitate the subscriber. Extensive experiments prove the efficiency and effectiveness of the proposed scheme.

To elucidate the relationship between agronomic traits and Se/Cd contents of rapeseeds, we conducted randomized block tests with 6 varieties of rapeseed (30 g/hm2 foliar Se fertilizer). With the response relationship between agronomic traits and Se fertilizer of rapeseed as a template, the effects of foliar Se fertilization on the primary branch number, secondary branch number, rapeseed yield, rapeseed Se content, and grain Cd content were analyzed to find out the differences among cultivars in response to Se application. This study was aimed to actively promote the search for Se-sensitive rape, screen Se advantage in the future, and provide reference for rational and effective use of Se resources into rapeseed varieties. Least significant difference analysis of variance model was used to scrape out the average Se content of 518 seeds of Deye Oil 0.248 mg/kg, The potential for Se enrichment was strong, and it was expected to become a typical variety of Cd-tolerant rapeseeds. Seeds of Huayouza 9 made maximum use of exogenous Se under relatively low Se content (0.298 mg/kg). The soil Se content was lower for the main varieties of planted areas. The Cd accumulation of Huiyouzaoxian 6815 was weak and did not change much when the leaves were sprayed with 30 g/hm2 Se, which could be used as a favorable material for Se-rich and low-Cd production.

Because of the insufficiency of such traceability systems as conventional bar code, and two-dimensional code, fresh-water fish traceability system based on NFC technology is designed, which consists of NFC cellphone, and Mifare electronic tag. As for Mifare electronic tag, disposable wristband Mifare tag is adopted due to its features of usage convenience, water and high temperature resistance and anti-copy by special modes. Three identities are contained in the system, providing different users with different operation authorities. Encrypted key traceability information is written into Mifare sectors through APP of NFC cellphone, which, by consumers, can be placed near tags, for the system to automatically analyze and read the traceability information, thus achieving the purpose of traceability.

Thorough analysis of the traditional linear model of information filtering, an improved model is proposed based on neural network, which reflects the user’s expectation. Taking 200 Email as the test object, the advantages and disadvantages of the linear model and the improved model are compared. The improved information filtering model has strong self-learning ability and adaptive ability, and improves the recognition rate.

As the computer technology is advancing endlessly and the information quantity is increasing exponentially, people have raised higher and higher requirements on retrieval technique, especially with the appearance of network technique and multimedia technology. The software and hardware environment of information retrieval technique is remarkably improved, making information retrieval technique develop from traditional linear retrieval to non-linear retrieval of hypertext support, and the traditional Boolean logic retrieval model no longer dominates in the information retrieval. It is hard to predict the new changes, new technologies and new ideas induced by technological advancement, yet we can grasp the correct direction for future development of information retrieval technique via comprehensive research, comparison, and analysis.

The development of logistics in developed countries should be earlier than us, and it is more mature in theory and practical application. Domestic logistics development is relatively late, but logistics research has also become a hotspot in recent years. The distribution center of logistics is also becoming a hot spot. The article combines the existing research results and summarizes the selection path, location layout and storage strategy, which can be used as a reference for future research.

The article puts forward a 3-dimensional searching approach that can factorize odd composite integers. The article first proves that, an odd composite number can be expressed by a trivariate function, then demonstrates that factorization of an odd integer can be turned into a problem of searching a point in a 3-dimensional cube whose points can be searched rapidly via octree search algorithm or other 3-dimensional searching algorithm. Mathematical principles with their proofs are presented in detail, and an algorithm that reaches square of logarithm time-complexity is proposed with numerical examples. The proposed algorithm can be applied both in sequential computation and parallel computation.

Taking Chinese as the language object, after analyzing the current Chinese word segmentation algorithm and Lucene relevance ranking algorithm, an improved word segmentation algorithm and an improved relevance ranking algorithm based on Lucene full-text search toolkit were proposed. This paper also uses distributed storage, parallel computing, inverted indexing and retrieval techniques to analyze and design a search engine for digital information in the network to provide users with fast and accurate search service for massive digital information. The experimental analysis compares the speed of word segmentation and word segmentation by comparing various word segmentation algorithms and compares their response time, the number of hits, the accuracy and the recall rate of the keyword search results. The experimental results show that the system greatly improves the information Search speed to ensure the accuracy of search results.

New retail is the trend of future development, currently, whether it is e-commerce channel or physical retail channel, all have some problems, are badly in need of transformation. In the “Internet +”, online and offline integration, realize the difference operation; with each other online and offline diversion, increasing flow; With the help of online and offline mutual diversion, the flow is increased continuously; Building a smart logistics distribution system, serving the integration of e-commerce channels and physical retail channels is a common choice for many enterprises, it is the main content of the new retail too. In this regard, this paper analyzed several modes of integrated development on current e-commerce channel and physical retail channel, and put forward the relevant suggestions on the integration development.

To assess the controlling potential of Crematogaster rogenhoferi against Parasaissetia nigra Nietner 1st instar larva, we determined the functional response and searching efficiency of C.rogenhoferi preying1st instar larva of P.nigra and analyzed its predatory behaviors. The results showed that predation functional response of C.rogenhoferi conformed to Holling’s type II model, and the feeding quantity rose with the increase of density. However, as the 1st instar larva of P.nigra increased, the searching efficiency of C.rogenhoferi decreased. A mutual interference effect existed among different individuals, and the searching efficiency of C.rogenhoferi accosted with a Hassell-Varley model. In the predation, although the feeding and espionage proportion of C.rogenhoferi were not high, the proportions of casting and searching behavior were large. Observations of comprehensive feeding behaviors of C.rogenhoferi showed C.rogenhoferi had certain control function and control potential on the 1st instar larva of P.nigra.

In the e-commerce environment, the quantity of order in e-commerce enterprises is increasing, the number of single order is small, and there are many kinds of them, which puts forward higher requirements for picking operation. The distribution of the cargo location in the e-commerce warehouse is the key factor affecting the selection efficiency, which directly affects the response speed of the customer order. The most effective way to improve the efficiency of storage operation is to optimize the storage location of goods.

The DE algorithm has strong global search ability and robustness, but also has the shortcoming of slow convergence speed and local search ability is insufficient, and TLBO algorithm has the advantage of strong local search ability and faster convergence speed, but will be fall into the local optimum when dealing with complex problems. In this paper, the DE algorithm and TLBO algorithm are combined to construct a two-population co-evolutionary algorithm based on the DE and TLBO algorithm (DPCEDT). By theory analysis, the proposed DPCEDT algorithm can be used to improve the SOC estimation algorithm of power battery which is an extremely complex problem.

The development from C2M model to C2B model is the main trend in the development of current e-commerce platform, C2M e-commerce model eliminated the intermediate circulation of commodities, to make consumers and manufacturers realized the docking and communication directly, manufacturers can make production according to the personalized needs of consumers. This model has been obtained a very good development in the e-commerce platform, such as clothing industry, automobile industry, necessary malls and so on, also exposed some problems, need other industries to continue to improve, to ensure the stability and maturity of the development for C2M e-commerce model.

Today, fashion design has basically matured. Based on the interaction and internal connection between computer clothing design and fashion design, this paper also put forward a series of suggestions for the computer clothing design in the apparel industry played a reference role. In addition, this article also studied how to develop fully automated garment manufacturing, clothing design provides a certain way of development. Clothing CAD/CAM has three meanings, namely, intelligent design, graphic design and virtual simulation. Major issues in apparel design technology include design database construction, 3D modeling, mesh generation, fabric drape modeling and crash testing. In the process of researching and analyzing garment manufacturing automation system, the author also analyzes these technical parameters emphatically. All in all, when the garment is fully automated, the era of mass customization of apparel is imminent.

Based on two-domain DNA strand displacement, a computing model is proposed. The model is used as a “referee” for two players in a well-known Rock-Paper-Scissors game, which can be utilized as an example of the study of game theory and artificial intelligence (AI). A molecular model based on Two-domain strand displacement is applied to emulate the process of the game. The output of the circuit shows the final result of the underlying game, that is, each player’s win, lose and draw. The two players hold a win of one inning and two win of three innings which are simulated by employing Visual DSD software. The simulation results show that the molecular model is correct and feasible. The establishment of the computing model is hoped to provide some new insights for the AI in the field of nanotechnology.

On the precondition of analyzing the algorithm of GBR and PDB, a business resource scheduled algorithm of TD-LTE trunking system based on QoS is presented in this paper. The simulation suggests that this algorithm can serve mixed styles of business better and meet different QoS requirements, especially to the throughput of high rate GBR business. It can make the system gain a higher throughput as well as decreasing the delay of the business.

Many prevail applications, such as data cleaning, sensor networks, tracking moving objects, emerge an increasing demand for managing uncertain data. Probabilistic relational databases support uncertain data management. Informally, a probabilistic database is a probability distribution over a set of deterministic databases (namely, possible worlds). Assumption queries in probabilistic relational databases have natural and important applications. To avoid unnecessary updates of probabilistic relational databases in existing general methods of assumption queries processing, an optimization method by computing conditional probability is proposed to handle assumption queries. The effectiveness of the optimization strategies for assumption queries is demonstrated in the experiment.

The two-wheeled self-balancing robot has a simple structure, low cost and high flexibility, which is very suitable for indoor space. In this paper, we designed a self-balancing robot, by using the STM32 microprocessor as the main controller, and the attitude sensor MPU6050 is used to collect the obliquity and angular velocity. However, the gyroscope and the accelerometer make noise interference and drift error, the Kalman filter algorithm, therefore, is used to fuse the obliquity and angular velocity, in order to obtain the optimal obliquity. The PID control algorithm will combine the optimal obliquity and the real-time speed obtained by the high-precision encoder of the coaxial motor, to output the stable and reliable PWM signal, which can be sent to the motor drive chip. The motor drive chip can drive the operation of the two motors, to obtain the more ideal operation control effect. The results showed that the self-balancing robot could achieve stable self-balancing control.

Many applications of road networks are emerging nowadays and skyline routes in road networks are important for users. In this paper, we design and implement a route skyline query system based on weighted Voronoi diagrams. After introducing related techniques such as route skyline, weighted Voronoi diagram and Dijkstra’s algorithm, we provide the design of our route skyline query system. The implementation of our system mainly focuses on user interfaces and integration of the weighted Voronoi diagram constructing algorithms as well as skyline route query processing algorithms. In addition, our system is able to display executing strategy of Dijkstra’s algorithm and parallelize network weighted Voronoi diagram algorithm by using thread mechanism.

With the wide application of Radio Frequency Identification (RFID) technology in many fields, anti-collision algorithm to solve the problem of multi-tag identification becomes more and more important. The current RFID anti-collision algorithm is mainly divided into two categories: ALOHA based algorithm and tree based algorithm. The traditional tree based anti-collision algorithm has a long time and low efficiency. Based on this, this paper proposed an improved RFID anti-collision algorithm based on quad-tree. It can eliminate idle timeslots of the identification process by grouping and re-encoding the original ID code of electronic tag. The mathematical analysis and simulation results show that the identification performance of the proposed algorithm is greatly improved compared with other traditional tree based algorithms.

Nowadays, with the quick development of the whole society, the development of computer technology is also rapid. At the same time, computer aided technology plays an more and more important role of CAM (Computer aided manufacturing) in human production and life in the process of technology, which uses the computer aided manufacturing to complete the whole process from the production activities to help improve the efficiency.

In this paper, an antenna with a miniature structure and wide-band is presented. We designed a two-arm conical spiral antenna according to the structure features of the Archimedes spiral and the conical helical antenna, and proposed an exponential asymptote balun to match the impedance. Unlike traditional antenna designs which optimize antenna and matching module separately, we adopted the differential evolution (DE) algorithm to optimize both the antenna and balun simultaneously. In addition, the peak radiation direction of the antenna was added as a constraint when evolving the antenna, which is usually ignored in normal evolutionary antenna designs. Simulation results indicate that the evolved antenna can basically fulfills the requirements. And the evolved antenna with the additional constraint has smaller deviation angle between the peak radiation direction and the antenna’s axis than that without the constraint.

Differential evolution (DE) is an efficient optimization technique, which has been applied to solve various engineering optimization problems. In this paper, DE is used to optimize the element spacing and lengths of Yagi-Uda antennas. An internal system with interactive simulation is developed based on C++ and CST Microwave Studio. To verify the performance our approach, the Yagi-Uda antenna for 60 GHz communications is designed in the experiments. Simulation results show the effectiveness of our approach.

Under the New Retailing mode, the sales model of fresh agricultural produce is significantly different from traditional model and O2O; therefore, it is necessary to study the coordination of fresh supply chain. This paper analyzed the fresh supply chain under the New Retailing, analyzed the supply chain coordination in detail based on the centralized decision-making mode, put forward some strategies that we should increase the proportion of online business, reduce the price of online sales, and build the information sharing platform under New Retailing mode.

Visual object tracking remains an active and challenging topic in computer vision due to a great variety of intricate factors such as illumination variation, object deformation and background clutter. Recent research efforts have achieved impressive success in object tracking, but they commonly have to utilize complicated models requiring high computation cost, which renders these methods hardly suitable for many applications. Considering depth information of the scene can provide effective complement to color images, in this paper, we propose a novel and efficient method for tracking an object in RGBD videos by using collaborative appearance and motion models. Experimental results demonstrate that our method achieves superior tracking performance over several state-of-the-methods while running efficiently.

Most recently, the lip password that embeds the password content into lip motion has been proposed for visual speaker verification (Liu and Cheung 2014). One merit of lip password is that it provides double security on the speaker verification, where only the target speaker saying the correct password can be accepted. Nevertheless, the previous work of lip password is based on identifying the distinguishing subunits of purely-digit password contents, thus limiting the application domain of lip password. To tackle this problem, we propose a novel visual speaker verification approach based on lip password without a priori knowledge of speech language, i.e. unknown language alphabet. We take advantage of the diagonal structure of sparse representation to preserve the temporal order of lip sequences by employ a diagonal-like mask in pooling stage and build a pyramid spatiotemporal features containing the structural characteristic under lip password. Experiments show the efficacy of the proposed approach comparing with the state-of-the-art ones.

In this paper, we propose a novel method based on Gene Expression Programming (GEP) to construct human motion model. Our approach better describes human motion features, which can be applied to improve the accuracy of human behavior recognition. On one hand, this method combines Genetic Algorithm (GA) and Genetic Programming (GP), and overcomes the limitation of traditional high-dimension function approaching method, realizing the generalization of Gene Expression Programming (GEP) on Function Mining. On the other hand, it implements the human motion capture technique of Kinect sensor, interpolates data and increases the training data accuracy. In the experiments result, we use GEP to develop human trajectory dynamics model, which has characteristics like encoding and gene structure flexibility that can lead the trajectory simulation error much decline. Given that the result is better than traditional methods and able to maintain most of the human motion features, our human motion model can be applied to human behavior analysis area and other similar domains.

Taking the characteristic value as the core, a population abnormality detection algorithm is used to process the crowd surveillance video. Using density detection, the density of the population is first obtained. Object-based feature extraction is used in low-density scenes, and pixel-based feature extraction in high-density scenes. So as to obtain the crowd of exercise intensity, trajectory gradient, entropy and local density and other characteristic value. Finally identify the abnormal behavior of the population based on characteristic value. The experimental results show that the characteristic value is obvious when the abnormality occurs. The algorithm’s performance index is superior to the traditional crowd behavior recognition algorithm with high recognition rate.

On the freeway, the serious fog and haze weather frequently appears on some road sections due to the geographical factors. The haze seriously damages the image quality of the road monitoring system. In this paper, we proposed a novel monitor image de-hazing algorithm (IDHA) for the heavy haze on the freeway. IDHA can accurately segment the haze monitor image into two regions (road region and non-road region), according to the prior knowledge of edges learned by the other fine weather monitor image of the same camera. An improved guided filtering method with dark channel prior and an improved adaptive histogram equalization algorithm is used on these two regions, respectively. Experiments show that the proposed algorithm IDHA can significantly outperform the dark channel prior algorithm and histogram algorithm on the running time and the de-haze effect on the heavy haze monitor image.

Recently, visual object tracking based on kernel correlation filtering has achieved great success. Application of robust feature, such as the Histogram of Oriented Gradients, is an important reason for the success of the kernel correlation filtering. However, the extraction of the HOG feature may bias the estimation of the target. To overcoming such kind of deviation, this paper proposes a real-time tracker with a multi-center strategy based on the kernel correlation filtering. Finally, abundant experimental results show that the multi-center kernel correlation filtering tracker of this paper has been made great progress relative the kernel correlation filtering tracker.

Due to the coming of the era of big data, and the computer processing power has been greatly improved. This will provide favorable conditions for the development of the convolutional neural network and then it will become an important object in the field of computer vision. Firstly, it summarized the development of the convolutional neural network and enumerated some successful models of convolutional neural network. Secondly, it introduced the working principle of convolutional neural network in detail, and also analyzed the operation mode of convolutional layer and sampling layer. Finally, it realize the recognition of handwritten figures Based on Convolutional Neural Network, and the experimental result shows, 196 were correct and 4 were wrong when the samples are 200, the recognition rate was 98%.

This paper adopts 10 machine learning algorithms to present the classification results of handwritten digit recognition on Minist dataset. These algorithms include k-nearest neighbors, support vector machine (SVM), decision trees (DT), random forest (RF), naive bayes, multilayer perception (MLP), logistic regression with neural network, artificial neural network (ANN), back-propagation (BP), convolutional neural network (CNN) and so on. We execute the experiments through matlab2015b and anaconda (python 3.6), and the result (accuracy and run-time) shows that SVM and RF achieve better performance. They has the accuracy of 98.08% and 97% separately, less running-time is taken compared with other methods. All the experiment are executed in CPU environment, without GPU. We also execute CNN algorithm for handwritten digit recognition in GPU (Nvidia GeForce GTX 1060), finally find that this algorithm achieves the best performance and the best classification result, the accuracy is up to 99%.

With the rapid growth of LTE user, the overload problem of high-speed rail network is more and more serious, and identifying the public network users who access to the rail network is the key and difficult point to solve the overload problem of high-speed rail network, especially for the low speed scenario where the existing speed detection algorithm effect is poorer. To this end, this paper proposes a high-speed rail network user identification algorithm based on the switching link, which realized the user’s identification by extracting cell level switch link, and simulation results show that the algorithm can effectively intercept 96% of public network users who access to the high-speed rail network, and there was no significant difference in both high speed and low speed scene of the performance of the algorithm.

Chinese has a history of literature more than three thousand years, it is very valuable to observe the evolution process of language. However, until now there is no such a computation model to describe the evolution process of Chinese, particular for Chinese functional words. In this paper, we propose a new model (NSkip-gram) to describe representations of Chinese functional words. By training Chinese data from early Chinese to Mandarine Chinese, we get the vector space representation of words. The experimental results of the statistical analysis on the preposition reveal that, the preposition has lost the ability of co-occurrence with the location argument gradually. And this implies the proposed NSkip-gram model can describe the situations of in Chinese evolution process well, and it can also be applied to other Chinese functional words.

The localization process of sensor networks using mobile reference nodes is discussed in detail, and errors in each phase are analyzed. Whether the selection of reference node path is appropriate, whether the unknown nodes can finish localization in closely related monitoring area, the reference node moving distance, and the time and energy consumed, concerning the above problem this paper proposed a covering algorithm to set the location of the radio beacon in advance, and the path traversal of these points planning. Simulation results show that the proposed mechanism can make the mobile reference node traverse the whole monitoring area as much as possible and avoid the obstacles effectively.