Abstract:With the increasing requirement for audio visual experience, three dimension (3D) multimedia technology, especially 3D audio technology that is less developed than 3D video, has received widespread attentions in recent years. Current researches of 3D audio can be divided into two kinds of technologies based on physical reconstruction of acoustic field and perceptual reconstruction of sound scene respectively. The representatives of the former are ambisonics, which is a full sphere surround sound reproduction technique based on spherical harmonic decomposition, and wave field synthesis (WFS), which is a spatial audio rendering technique based on wave fronts synthesis. Perceptual reconstruction techniques of sound scene mainly include the amplitude panning （AP） technique and the binaural sound synthesis technique based on head related transfer function （HRTF）. The above four kinds of 3D audio technologies and the corresponding typical systems are introduced and compared in this paper. The current hot researches of 3D audio techniques are also presented mainly from three aspects: spatial hearing perceptual mechanism, 3D audio coding and multichannel simplification of 3D sound system.

Abstract:In compressed sensing, measurement matrix plays an important role in signal acquisition and reconstruction. The traditional random measurement matrices can achieve good performance on the condition that the sampling rate is high enough, whereas the reconstructions are not satisfactory at low sampling rates. Compared with these random measurement matrices, the deterministic measurement matrices possess their own constraints, which lead to worse performance. Based on the generalized rotation (GR) matrix, two kinds of structured random matrices are proposed as the generalized binary rotation (GBR) matrix and the pseudo random generalized binary rotation (PGBR) matrix. Simulation results for two dimensional signals show that the two series of new matrices perform better than the traditional measurement matrices. The amount of time required by the traditional and the new approaches is about the same. Moreover, they can obtain more accurate reconstructions at low sampling rates.

Abstract:Non stationary, non-linear, and weak signals are difficult to analyze and process. A novel signal processing method based on empirical mode decomposition (EMD) and learning vector quantization (LVQ) neural network is proposed and applied in the field of biological signal processing (left and right hands move imagery electroencephalogram (EEG) signal). Firstly, EMD is used to decompose EEG signal. Secondly, the major intrinsic mode function components are extracted and their mean absolute values are calculated as the features. Finally, LVQ is used to finish the classification task. Then the results are compared with the support vector machine and error back propagation neural network classification algorithms. The experimental results show that the classification accuracy rate of the proposed algorithm reaches 87%. Compared to the other two contrast algorithms, the new algorithm has better performance in the specific signal processing field and thus has high reference and research value.

Abstract:The poor channel environment of cognitive radio networks often results in the decline of local spectrum detection performance during cooperative spectrum sensing. Sometimes, fault nodes or malicious nodes deliberately send some wrong information to disturb the fusion center to make global decision. According to the historical sensing information of sensing nodes, the sensing nodes are classified intoreliable nodes, fault or malicious nodes and unreliable nodes, and a cooperative spectrum detection algorithm is proposed based on node recognition. Therefore, the fault or malicious nodes are discarded and prevented from data fusion in fusion center. At the same time, the current local detection results of unreliable nodes are not considered in current global decision. Thus, not only the effect of fault or malicious nodes on the global decision is eliminated, but also the calculation complexity of fusion center is reduced. The simulation results show that the proposed algorithm can avoid the interference of fault or malicious nodes and improve spectrum sensing performance effectively.

Abstract:Traditional fire monitoring systems usually adopt single sensor to detect fire by wire transmission systems, and they have inconvenient wiring, low environmental adaptability and poor anti interference ability. A fire detection system based on ZigBee wireless multi-sensor network is proposed with sensors including smoke, temperature and carbon monoxide gas sensor to detect combustion situation. The trust functions for smoke, temperature and carbon monoxide caused by fire are assigned according to fire situation, and the D-S evidence theory is utilized to integrate the three kinds of sensor information to determine whether fire happens or not. Multi-sensor combustion response experiments are conducted. Theoretical analysis and experimental results show that the fire detection system based on wireless multi sensor can detect fire more accurately, reduce the false alarm rate and improve the credibility of thesystem.

Abstract:A new banknote classification method is proposed by using phase concept of reduced quaternion wavelet transform (RQWT) to improve the banknote recognition rate and feature extraction. Banknote is preprocessed including edge detection and slant correction. And image is decomposed by reduced quaternion wavelet. The statistical characteristics of the decomposition coefficients are used as features of the banknote image for classification. Finally, the support vector machine is applied as classifier in the banknote classification system. The experimental results show that the proposed method can obtain better results compared with other conventional methods and satisfy the real time requirements.

Abstract:In the traditional variational level set method for image segmentation, the evolving level set function needs periodical re-initialization to keep it close to a signed distance function during the evolution. It remains many serious problems such as when and how to apply the re-initialization. Li presented a new variational formulation that forces the level set function to be close to a signed distance function by adding an internal energy into the energy functional, and therefore completely eliminates the need of the expensive re-initialization procedure. We present a new image segmentation model based on variational level set method. It also completely eliminates the need of the re-initialization by adding a new and simple internal energy into the energy functional. In addition, a new external energy by redefining the edge stopping function is introduced, which makes the proposed model more robust to noisy image segmentation. The experimental results show that, compared with Li model, our model has some superiority in the convergence speed andsegmentation quality for noisy image.

Abstract:A script identification method is proposed based on Gaussian derivative filter bank. The texture characteristic of document images is analyzed. Compared with traditional wavelet transform, the proposed algorithm can extract edge and ridge features with more orientations. The support vector machine (SVM) is applied for training and classifying the extracted features to identify scrip ts in different languages. Experiments are performed upon document images with ten kinds of languages (including Chinese, Russian, English, Japanese, Korean, Arabic, etc). The effects of different Gaussian derivative filter parameters on the identification performance are tested, and other three script identification methods based on texture are selected for comparing. Experimental results show that the proposed algorithm can improve the speed and the correct rate of script identification.

Abstract:Topic summarization is a natural language processing for creating summaries of topic information. Previous work focused on summaries of news, web documents and blogs, while seldom on microblog topic summaries. A microblog topic summarization (MTS) method is proposedbased on topology structures for microblog retweets. First, representative terms are selected according to structural relationships between retweeting tweets. Second, topic areas are identified after topic nodes are merged by using depth-first and breath-frist methods. Third, topic-oriented summaries with topology structure are generated through measuring adjacent topic nodes on the retweeting graph. Finally, experiments on the real world event datasets show the effectiveness of the proposed methods. Visual topic summary trees are also produced for remarkably emphasizing the insight behind the evolving topics.

Abstract:An improved algorithm of speech emotion recognition is proposed based on modified multiple kernel learning. The algorithm based on Gaussian radial basis function improves classification performance by gathering different samples,utilizing different evaluation criteria and acquiring different parameters .In addition, with the help of multiple kernel technology, the trained Gauss kernel functions are used to construct the basis of multiple kernel learning and the efficiency of learning are greatly improved by utilizing the relaxation factor to construct objective function of soft margin .Experimental results show the presented algorithm can effectively improve the performance of speech emotion recognition.

Abstract:A novel method for multi-target angle estimation and mutual coupling self-calibration of bistatic multiple-input multiple-output (MIMO) radar in spatial colored noise is presented. The receiving data are preprocessed by the fourth-order cumulants, which can suppress the spatial colored noise. The rotation invariant matrices of the transmit arrays and receive arrays are exploited by using the character of Toeplitz matrix and the ESPRIT method. Further, the transmit and receive angles can be estimated without pairing algorithm. Moreover, mutual coupling self-calibration is realized by using the estimated transmit and receive angles and the method of Lagrange multipliers. The effectiveness of the proposed method is verified with computer simulation.

Abstract:Savitzky Golay (SG) filtering algorithm is an effective algorithm to remove the white noise in digital signal. The algorithm is based on the least squares algorithm, whose essential concern is how to ensure the filtering effect while remain the information of signals as much as possible. A SG threshold value filtering algorithm is put forward through the theoretical analysis to take different SG filtering strategy in different signal intervals with different characteristics. Then, based on the white noise singular point detection and iterative algorithm, a threshold value determination algorithm is proposed to enhance the practicability and the convenience of the SG threshold value filtering algorithm.

Abstract:A rate unknown image steganalysis scheme is proposed based on multiple classifier fusion. Firstly, various classified results are acquired by using the multi-rate classifiers established in the training phase. Secondly, these classified results are converted to evidence and enhanced through introducing weighted coefficients which are acquired according to the missed detection rates and the false alarm rates of different classifiers. Finally, the decision is obtained by Dempster-Shafer(D-S) evidence theory based on weighted coefficients. The detection work is presented to attack LSB matching. Experimental results show that the proposed method improves detection accuracy.

Abstract:A compound anisotropic diffusion filter algorithm is presented to apply edge sensitive instantaneuos coefficient of variation (ICOV) operator of the speckle reducing anisotropic diffusion(SRAD) model into the NCD model. The correlation coefficient matrix of ICOV operator which can measure the correlation of the image is proposed. The value of the coefficient matrix gets the maximum value in the vicinity of the edge. According to the correlation coefficient of the ICOV operator, This paper modifies the intensity of the diffusion and obtains effective nonlinear denoising. Compared with other algorithms, the experiment results show that the proposed algorithm has better effect in smoothing and edge preservation.

Abstract:An image multiple description coding method is proposed based on compressive sensing theory. It measures the image by block compressive sensing technology on the encoding side, and then divides the result matrix into many descriptions by grouping equally in the rows. In the decoding side, it reconstructs the image by received descriptions. The more the descriptions are received, the better the quality of the reconstructed image is. Dividing the result matrix into groups directly can generate many descriptions easily and reduce computational complexity on the encoding side. Experimental results show that, in the same experimental conditions, the proposed method exhibits faster encoding speed and higher reconstruction image quality than other methods.

Abstract:The technologies of traffic information collection using floating car equipped GPS have been become one of the main important means for real-time collecting traffic information in intelligent transportation system. The intervals of traditional traffic information collection technologies using floating car equipped GPS are simplex and equivalent at present. The sampling interval cannot be obtained according to geometric condition of load network and diversity of traffic status. Aiming at the ineffectiveness of the existing sampling interval algorithms, a real road network oriented optimization method of floating car sampling interval is proposed. Firstly, the urban road network is divided via quad-tree model. Thereby, the spatial sampling resolution can be acquired; Secondly, the short-term speeds of floating car are predicted according to the history track; Finally, the optimal sampling intervals are obtained, simultaneously, the spatial sampling resolution cannot be influenced. The results of simulation and experiment show that the sampling interval can be dynamically determined under the circumstances of different complexities of road network. The sampling result can notonly ensure sampling data precision, but also reduce data capacity.

Abstract:There is a changing trend of attenuation exponential summation in a great capacity of electromagnetic response data in the research of time domain aviation electromagnetic exploration. According to the changing situation, characteristic parameters as poles are extracted from response data by matrix pencil method, and then the response data are dealt with reducing rank and reconstruction. Algorithm analysis and practical testing result prove that better reconstruction precision by matrix pencil method is presented in the condition of non-noise interference, and adding noise interference,a little fluctuaton of precision emerges with in a big variable extent of signal to noise rate which reflect the abilits of resisting noise and stable performance. Therefore, the proposed matrix pencil method is applicable to the analysis of aviation electromagnetic response data.

Abstract:A novel approach of blind separation of communication signal and interference is proposed for only one single-channel observation signal obtained and low signal noise ratio (SNR). The proposed algorithm aims to obtain the maximum a posterior (MAP) estimates of communication code and the unknown parameters using particle filtering by establishing the state space model for the observation signal, Specially， in order to overcome the sample impoverishment problem and estimate iteratively the particles that have more important weigh, genetic operation is introduced to the re-sampling process in particle filtering. In such a way, the number of needed particles is reduced and the variety of particles is retained during the sequential estimation process. Simulation results show that the proposed algorithm has superior performance than the classical particle filtering, and the method can effectivly separate communication signal and interference when the interference signal ratio (ISR) is less than 15 dB and SNR is more than 10 dB.

Abstract:In the continuous wave measurement while drilling (MWD) system, the accuracy of error rate prediction is low and the data transfer process is affected by interference signals.A model for error rate prediction of continuous-wave data transmission is proposed by using the improved least squares support vector machines (LS-SVM), and the genetic algorithm is used to search the optimized parameter to improve the prediction accuracy of the model. During establishing the model, Dixon criteria is used to screen the data and improve the error rate prediction accuracy. With small samples, mud continuous-wave data transmission model is established by using Matlabbased on the improved LS-SVM. The simulation results show that the model can avoid falling into local optimization problem effectively, and has strong generalization and prediction ability. Compared with back propagation(BP) and Elman neural network, the model has higher prediction accuracy, so it can be used to predict the error rate of mud continuous-wave data.

Abstract:An algorithm is proposed based on the affine projection (AP) adaptive filter theory for digital predistortion (DPD) system. A method of look-up table (LUT) with complex gain is applied to the system. The effects of step size parameter and the number of multiple constraints on the affine projection algorithm convergence rate are analyzed. The performance of AP algorithm and the normalized least mean square (NLMS) algorithm is compared. Simulation result shows the AP algorithm has a faster convergence rate when the numbers of step size parameter and multiple constraints are greater. The adjacent channel power ratio (ACPR) of AP algorithm is -59.3 dB while the one of NLMS algorithm is -44.2 dB at offset normalized frequency of 5 MHz with same iteration times, step size parameter and regularization coefficient. The performance of AP algorithm is better than that of NLMS.

Abstract:The saliency detection methods have been widely used in the field of image processing and computer vision. However, the saliency detection algorithms via global feature and local feature extraction have shortcomings. Therefore, a significant saliency detection algorithm is proposed based on fusion of global and local features. Firstly, an image is partitioned to non-overlapped blocks. When each image block is mapped to high dimensional space by principle component analysis(PCA) method, according to the law that the isolated feature points correspond to the salient regions, the saliency map based on the global features is obtained; Secondly, based on the color dissimilarities between center block and its neighborhoods, the saliency map via the local features is obtained; Lastly, based on the Bayes theory, the two obtained saliency maps are fused to the final saliency map. The simulation results on three public image database verify that the proposed algorithm can combine the significant advantages of the global and the local saliency detection algorithms, and it is more effective on saliency detection and object segmentation compared with other state of art algorithms.

Abstract:Image mosaic in image processing is widely used. However, its key issue is image registration, and clear mosaic from several relevant images with no splicing trace. Therefore,an algorithm is put forward: firstly, gray correlation method is utilized for matching search with image information in wavelet domain, and random sample consensus (RANSAC) algorithm is used to work on match point; Secondly, adaptive filtering is established in the wavelet domain for image flat fell,and the wavelet coefficient matrix is concluded;Finally, wavelet inverse transform on the wavelet coefficient matrix. Experimental results show that the proposed algorithm can finely smooth the seam line on the mosaicking boundary thanks to image mosaic in wavelet domain and adaptive filtering (instant changing filter's form)according to the adjacent two images' own information.

Abstract:Railway signal cable is long and under ground, which leads to difficulties in cable fault detection. The second correlation method is improved. The improved computation model based on FFT is designed, and its real time computation ability is analyzed. Based on the improved second correlation method,a signal cable sequence time domain reflectometry／spread sequence time domain reflectometry (STDR/SSTDR) detection model established. Moreover, the open and short circuit fault detection performance of the model is tested at various SNRs. The result shows that SSTDR based on the improved second correlation can be used to detect the signal sectioncable fault.

Abstract:Hyponymy, a kind of basic semantic relation between words， is widly used in areas, including text classification and information retrieval. Automatic extraction of such relation is an important issue in natural language processing. Two kinds of hyponymy extraction strategy, i.e., dictionary based strategy and encyclopedia basedstrategy are proposed to build a sophisticated hyponymy knowledge base. Chinese Concept Dictionary and Chinese Classied Subject Thesaurus are used as dictionary resources. Manual regex is introduced to extract hyponym from wikipedia,baidubaike and hudongbaike based on addresses of web pages. Extensive experimental evaluationdemonstrates that the proposed strategies outperform the NLP&CC 2012 evaluation results.

Abstract:Aiming at the requirements of precise prediction and health management of spacecraft, a method for combinational prediction of parameter values called particle swarm optimization grey neural network is promoted. The method enables particle swarm optimization algorithms, grey theory and neural network to complement each other. Firstly, a prognosis for output current values of southern sailboard of a certain satellite is taken as an example. Then, three evaluation indexes of prediction, including mean absolute error, mean absolute percentage error and root mean square error, are chosen to evaluate the results of different step length prediction is of particle swarm optimization fuzzy neural network. The results show that the particle swarm optimization fuzzy neural network is effective. Secondly, the mean absolute percentage errors of particle swarm optimization fuzzy neural network, grey model particle swarm optimization neural network, particle swarmoptimization neural network and grey model are calculated. The results show that the model of particle swarm optimization fuzzy neural network is the most precise one and more efficient in prediction than others. It has vast application prospects in the field of prediction of spacecraft parameter values.

Abstract:Most community detection methods are aiming at solving undirected and unweighted datasets. However, datasets are often directed and weighted with noisein real world. In order to process noisy and directed weighted community detection, a method based on nonnegative matrix factorization (NMF) is proposed. In the algorithm, wavelet threshold denoising is used to denoise the social network datasets. And the community structure is abtained by community detection through NMF. Simulations show the proposed method is more effective,i.e. for Lesmis dataset when SNR is 15, the accuracy of dividing community is 96% and the modularity of the method is improved by 29%. The proposed method is more applicable than other community detection methods for directed weighted datasets with noise.

Abstract:The development of technology and multiple requirements produce huge data.It is critical to mine patterns behind the data for farther in depth study. To mine period frequent patterns of moving objects, an algrithm of non-redundant period patterns(NRPP) is proposed. The algorithm set different limited conditions to solve the combinatorial explosion problem and the rare item problem, and it is more efficient. To prevent noise and other uncertainties, similarity-based pattern matching method is introduced. By facilitating existing methods, the proposed method can be more concise and accurate. The experimental results using open data show that the method can efficiently excavate the periodic frequent patterns.