Abstract:Data denoising is a classic issue in the field of signal and image processing which has been widely applied in various engineering practices. Due to the diversity of noise sources, denoising is a challenging and active research topic, and a variety of classical denoising methods have been developed. In recent years, with the development of compressed sensing theory, the methods for solving inverse problem based on sparse representation and regularization constraint have become important research directions and technical approaches in the field of image denoising. This paper firstly reviews and summarizes the sources and types of image noise, and then according to the different types of image noise, gives a comprehensive review focusing on the image denoising techniques based on sparse representation and regularization constraints. In addition, we analyze and describe the principle, advantages and disadvantages of several major denoising methods. Finally, the performance evaluation of denoising algorithm is summarized.

Abstract:Tunable Fabry-Perot (F-P) filter-based demodulation methods are more used in the wavelength demodulation of fiber Bragg grating (FBG). However, the hysteresis and creep of the piezoelectric ceramic (Pb(Zr1-xTix)O3，PZT) in the F-P filter make the output wavelength of the F-P filter nonlinearly change with time, resulting in the problems of large wavelength demodulation error and low sensing accuracy. In this paper, we start with the study of hysteresis and creep characteristics of PZT in F-P filters, and propose an F-P filter-based wavelength demodulation method based on thermal etalon and hysteresis and creep compensation control of the PZT. The driving voltage of PZT is controlled to change nonlinearly with time so that the wavelength of the broadband light source output through the F-P filter can be changed at equal intervals with time.The peak wavelength of the etalon transmission spectrum is thus obtained and also changes linearly with time. It acts as a "tick" for the wavelength scale and provides an accurate wavelength reference for demodulating the wavelength of the sensing grating. The experimental results of the FBG temperature measurement show that, compared with the method without hysteresis and creep compensation control, the proposed method improves the temperature measurement accuracy about 20 times and the temperature measurement error is less than 0.5 ℃.

Abstract:To cope with the problem that the traditional fine feature extraction methods for identifying communication transmitters suffer from the lack of the labeled samples in real complex electromagnetic environment, an efficient fine feature extraction method, called locally neighborhood preserving regularized semi-supervised discriminant analysis, is proposed for communication transmitter recognition. Based on the bispectrum estimation, manifold structure information is incorporated into the linear discriminant model by unlabeled samples, which extends the linear discriminant analysis to the semi-supervised learning. Extensive experiments on the real-world database sampled from different FM communication radios with the same model, manufacturer, manufacturing lot, and work pattern demonstrate that the proposed method can obtain better recognition performance.

Abstract:A parameter-estimation method based on subsampling for multi-component linear frequency modulated (LFM) signal is proposed. The received signal is subsampled by multiple analog to digital converters (ADCs) with identical sampling rate and different sampling time. The total sampling rate of all ADCs is lower than the Nyquist rate of original LFM signal. As the product ambiguity function of a single subsampling sequence is a single tone, the chirp rate is estimated by searching for the peak of the product ambiguity function. Then dechirp operation is conducted for all subsampling sequences to obtain multitone signals, from which the initial frequencies of LFM components are estimated by solving moment-preserving problem and over-determined equations. In the proposed method, parameters of the LFM signal are estimated from sub-Nyquist samples, and the estimation is easily realized with simple operations. Simulation results verify the effectiveness and accuracy of the proposed method.

Abstract:To improve the spectrum sensing efficiency of cognitive radio Ad hoc networks and to solve the problem of cognitive radio Ad hoc networks clustering, a clustering scheme based on spectrum sensing for cognitive radio Ad hoc networks clustering is proposed. By introducing the detection factor, considering multiple primary user signals overlapping and shadow fading effect, the clustering problem of secondary users and primary user channels is mapped to bipartitie model. Then the clustering problem can be simplified to constraint maximum-weight edge biclique(C-MWEB) decomposition problem and solved by greedy algorithm. Simulation result shows that the proposed clustering scheme is more effective and reliable than the traditional one in the case of multiple primary user signals overlapping and shadow fading effect.

Abstract:Two-way product-and-forward (PF) relay networks can provide lower asymptotic symbol-error rate (SER) than amplify-and-forward (AF) relay networks. In order to study the performance of PF relay networks over Rayleigh fading channels, the asymptotic SER expression of the PF relay networks is derived and analytical results show that PF relay networks have better SER performance than AF relay networks. Experimental results agree well with the analytical results. In order to enhance the SER performance of PF relay networks, a power allocation algorithm to minimize SER under total transmit power constraint is proposed. Simulations show that proposed algorithm can significantly enhance the SER performance.

Abstract:In order to implement the frequency offset estimation in direct sequency spread spectrum (DSSS) system based on the standard of IEEE 802.15.4k under low SNR circumstance, an improved frequency offset estimation algorithm with high accuracy and stability is proposed in this paper. The proposed algorithm firstly makes biased auto-correlation and iteration to the signals by using the phase and amplitude information at the same time under the premise of ensuring the frequency offset estimation range, which improves the SNR of sequences including the phase information. And then it calculates the phase angles for each order difference of biased auto-correlation signals and makes a weighted mean, which overcomes the difficulty in meeting approximate conditions for Fitz method under low SNR circumstance. The simulation results show that the proposed method can achieve high accuracy estimation without reducing the estimation range under extremely low SNR circumstance, and the estimation precision meets the tolerance to residual frequency offset for the receiving system under the standard of IEEE 802.15.4k.

Abstract:How to find the good representation from raw data is a key and very important issue in machine learning. Most traditional approaches are based on the relationship among data or utilize simple linear combination, in which deep learning algorithm can perform very well in various machine learning tasks and achieve very good representations. However, most existing algorithms are implemented in serial, which cannot handle large-scale data. This paper proposes an effective parallel auto-encoder (PAE) based on Spark. The proposed PAE not only can learn satisfying representation, but also can speed up the executing time based on Spark. And then the paper adapts PAE to deal with the sparse data. Experiments conducted on two tasks, i.e., classification and collaborative filtering, demonstrate the effectiveness and efficiency of the proposed PAE.

Abstract:In the imbalanced internet protocol television(IPTV) dataset, the traditional algorithm performs not well in terms of predicting the user′s complaint. For this problem, this paper combines traditional network parameters that influence the network quality of service (QoS) with MOS score that objectively reflects the quality of experience (QoE) to predict user′s complaint. And then we propose an improved algorithm based on the existing ODR-BSMOTE-SVM algorithm for the defects that the over-sampling algorithm will produce noise and there is not any optimization for kernel parameters. In the improved algorithm, under-sampling algorithm, over-sampling algorithm and data cleaning algorithm are firstly used to process the original imbalanced dataset. Then, through searching for the approximate optimal value by adaptive variable kernel parameters, the classification effect is ultimately improved. Experimental results show that the improved algorithm performs better than the traditional standard support vector machine (SVM) and the ODR-BSMOTE-SVM algorithm in predicting user′s complaint.

Abstract:A joint sparse reconstruction method is proposed for the estimation of the direction-of-arrival (DOA) of narrowband or wideband signals by using a co-centered orthogonal loop and dipole (COLD) vector antenna array. This method is based on the sparse representation of the first column vector of the polarization-space covariance matrix constructed by the output of the COLD array, which can be regarded as two perpendicular loops and dipole separated antenna arrays. The DOA estimates are then obtained via the convex optimation (l1-norm based) joint sparse reconstruction technique under the l2-norm constraint. Simulation results show the superior performance of the presented method over the traditional methods in terms of resolution and DOA estimation accuracy.

Abstract:The polar format algorithm (PFA) for spotlight synthetic aperture radar (SAR) is influenced by wavefront curvature error. The error introduced will cause the missile borne SAR image edge blur and geometric distortion seriously. The previous wavefront curvature compensation methods are all assumed that the height of the radar is constant. These compensation methods cannot be directly applied to the case of missile borne SAR where the platform always has large dive angle. The application range of wavefront curvature error compensation method is restricted greatly. In this paper, the precise expression of wavefront curvature phase error in the frequency domain of missile borne SAR is derived. After image domain post-processing steps, the problems of distortion and defocus of PFA image in the missile borne SAR are solved effectively. This method further improves the precision of wavefront curvature compensation. At last, the correctness of formula and method is verified by simulation.

Abstract:Cross-range scaling of the inverse synthetic aperture radar (ISAR) determines the real size of one cross-range unit, by which the geometry character of the target can be obtained. Considering the linear frequency modulation of the echo signal in slow-time domain, a new approach based on fractional Fourier transform (FrFT) is proposed for cross-range scaling. The approach estimates the chirp rate of the Doppler history and the target rotation vector according to the energy concentration property of the signal in the different fractional Fourier domain. The simulation results show that the algorithm is effective and accurate.

Abstract:Locally linear embedding (LLE) algorithm has no direct relationship with the classification. Meanwhile, the recognition effect is decreased when the LLE algorithm is affected by different facial expressions, illumination and pose, etc.,and the distribution of the original sample is usually nonlinear and complex. Therefore, an efficient dimensional reduction and classification algorithm is presented, that is fuzzy difference embedding projection (FDEP) algorithm. The FDEP algorithm constructs different radiograms to characterize the local and the global structure information using fuzzy membership degree (fuzzy sets) under fuzzy thinking, and then uses the maximum margin criterion (MMC) to construct the objective function for avoiding the ″small-size sample″problem. Finally, the algorithm solves the constrained optimization by Lagrange operators. The FDEP algorithm maintains the original neighbor relations for neighboring data points of the same class and is also crucial to keep away neighboring data points of different classes. The results of face recognition experiments on ORL, Yale and AR face databases demonstrate the effectiveness of the FDEP algorithm.

Abstract:The traditional principal curve algorithm is widely used in many fields, but it is ineffective in extracting the principal curves for complex data. To solve the kind of the problem, one of most effective ways is to combine the granular computing with the principal curve algorithm. Therefore, a new multi-granularity principal curve extraction algorithm for complex data based on granular computing is proposed. Firstly, we use the spectral clustering algorithm based on t-nearest neighbor (TNN) to granulate the data and propose the inflexion point estimation to automatically determine the number of granules. Then the local principal curve extraction for each granule is carried out by using soft K-segments principal curve algorithm and optimized by removing the false edges. Finally, a local-to-global strategy is adopted to extract the multi-granularity principal curves to optimize overfitting curves and a principal curve which can describe the original data distribution pattern can be obtained. Experimental results demonstrate the excellent feasibility of the proposed principal curve extraction algorithm.

Abstract:Hyperspectral images have been widely used in target dectection terrain classification and so on owing to its rich spectral information. Classification, being the fundamental step to further explore the hyperspectral images, attracts wider concern. The spatial information describes the connections between pixels with its spatial neighbors which can help to solve the problems like metameric substance of same spectrum, metameric spectrum of same substance and insufficient labeled samples with a high dimension while the spectral information cannot handle well. The traditional preprocessing uses a structure element to obtain the spatial neighbors and assist the last classification with the extracted spatial features. It is obvious that the structure element matters, however one cannot find a suitable size to meet all demands. For dealing with this, a method combing watershed segmentation with composite-kernels support vector machine (SVM) is prposed. It is the characteristics of over segmentation that we use to get a self-adapting spatial neighbors, containing less dissimilar pixels and being more discriminant for every pixel, then we fuse the spatial features and the spectral through the composite-kernels SVM and give a reliable judgement. Experiments show that the proposed method can make a better use of the spatial imformation and achieve a high accuracy with limited training samples.

Abstract:As a basic and fundamental issue in computer vision area, many algorithms have been proposed to address the issue of circle detection, such as Hough transform, randomized Hough transform, randomized circle detection and so on. However, the low efficiency of these methods makes them hard to be used in complicated situations or conditions that require much higher circle detection speed. To improve the efficiency of circle detection, this paper analyzes three stages, including the selection of sampling points, the determination of candidate circle and the confirmation of true circle. Combined with the optimization of these three stages, a circle detection algorithm with multi-stage optimization is proposed. Experimental results of synthetic images and real images indicate that the proposed algorithm has faster detection speed compared with other methods, and has a high detection accuracy and strong robustness.

Abstract:In the framework of block compressed sensing (BCS), the reconstruction algorithm based on the smoothed-projected Landweber iteration can achieve better performance of rate-distortion with a low computational complexity, especially for the case using the principle component analysis (PCA) to conduct adaptive hard-thresholding shrinkage. However, during learning PCA matrix, the reconstruction performance of Landweber iteration is affected because of neglecting the stationary local structural characteristic of image. To solve the above problems, the granular computing (GrC) is adopted to decompose an image into several granules depending on the structural features of patches, and then PCA is performed to learn the sparse representation basis corresponding to each granule. Finally, the hard-thresholding shrinkage is used to remove the noises in patches. The patches in granule have the stationary local structural characteristic, and the proposed method can thus effectively improve the performance of hard-thresholding shrinkage. Experimental results indicate that the reconstructed image by the proposed algorithm has a better objective quality when comparing with several traditional ones, and its edge and texture details are better preserved, which guarantees the better subjective visual quality. Besides, the method has a low computational complexity of reconstruction.

Abstract:In order to cut subscriber identification module （SIM）card slot via laser cutting with coaxial visual sensing system, this paper presents an algorithm for detecting the specific straight lines. This algorithm makes full use of the gradient magnitude and direction information, and selects the seed points based on the difference of gradient modulus values between adjacent pixels. Then the line-support regions are found via growing a region from the seed points using gradient angle measure. The line segment can be determined based on the gradient modulus values of pixels in the line-support regions. The numerical experiments demonstrate that the proposed algorithm has super performance of high detection precision, good robustness and low calculation cost, thus meeting the application requirements of industry detection.

Abstract:On the basis of analyzing causes of Internet public opinion emergency (IPOE) and features of emergency decision, the indicators of IPOE are constructed from three aspects of the state of Internet public opinion, the emergency itself and the public attention. Due to the uncertainty and diversity of indicators for IPOE and the difference of linguistic judgment matrices on crisis evaluation from experts, we propose a fuzzy multi-indicator group decision making of IPOE based on relative entropy and linguistic weighted operator. Firstly, according to judgment matrix of evaluating importance of indicators, we calculate the weights of indicators of each expert, and then use the multi-indicator combination weighting method based on relative entropy to get the experts′weights, so we can calculate comprehensive weights of indicators of IPOE. Secondly, based on linguistic indicator evaluation of IPOE from experts, we use the linguistic weighted arithmetic average operator to gain the comprehensive evaluation of the value of each IPOE, and quickly sort out IPOEs according to crisis levels. Finally, an example is carried out to show the practicality of the proposed method.

Abstract:Using context information to improve the accuracy of recommendation systems and enhance user experience is one of the hottest topics in the domain of recommend systems. However the issue of data sparse still challenges the existing context-aware recommender system. To better alleviate the data sparse problem, this paper proposes a rating prediction method, i.e., joint matrix factorization with user category prefernce(JMF-UCP). Based on the joint matrix factorization, the method addresses the data sparse problem by combining user′s rating information and user category preference to predict the rating score with higher accuracy. The time complexity of the proposed method linearly increases with the number of amount of dataset and is scalable to very large datasets. Experimental results on real world rating dataset MovieLens demonstrate that the proposed method can achieve better accuracy.

Abstract:As an important information platform, micro-blog has a large number of user visits every day, and important public opinion events will form a hot topic on micro-blog. In this study, we propose a novel micro-blog topic detection method, named TDFWN (Topic detection in frequent word networks),to excavate hot topics in micro-blog corpus. First, frequent k-item sets (k≥3) in Microblog text data are mined. Second, a word co-occurrence network is build based on these mined frequent k-item sets. Third, the network is partitioned into different communities by using a community detection method, where each community represents a micro-blog hot topic. At last, the micro-blog text data are clustered into different groups by computing similarity of each micro-blog text with the found topics. The empirical study shows that the TDFWN method is able to find hot topics in micro-blog text data and cluster the micro-blog text data by the found topics simultaneously.