This research makes an attempt of using historical stock big data to predict near future trend of the stock. To fulfill this task, a novel approach based on fused intelligent computing is introduced and investigated. It is composited of four main parts, including data discretization, attribute reduction, classification and decision fusion. Further, one or two algorithms are adopted to realize specific function in each part, respectively. The given stock indexes are selected by the reduction algorithm of discernibility matrix, and the outputs of multiple classifiers are fused by the decision fusion algorithm. These processes and other ones are dedicated to enhancing the accuracy of stock trend prediction. To demonstrate the effectiveness of our approach, a variety of experimental simulations utilizing historical data of three stocks in NASDAQ are carried out, and the prediction accuracy of the proposed approach are compared as well. The experimental results prove that our approach could accomplish the prediction task with high accuracy.

In this paper, the specific framework and characteristics of positive approximation set are investigated. A novel attribute reduction algorithm is introduced for accelerating the existing reduction approaches for incomplete decision table. Consequently, a tolerance relation based algorithm (IFSPA) is proposed and employed to improve the performance of existing heuristic reduction algorithms. A series of experiments using real-life data sets are carried out to verify the effectiveness of the improved algorithms and compare their efficiency with that of the original algorithms. The simulation results prove the fact that the improved algorithms would output exactly the same reducts as the original ones, while the improved ones could accomplish the reduction task in a shorter time and even in a more stable method.

The study addresses the sensor control problem for multi‐target tracking via delta generalised labelled multi‐Bernoulli ( ‐GLMB) filter, and proposes two novel single‐sensor control schemes. One is that the Rényi divergence is used as the objective function to measure the information gain between the predicted and posterior densities of the ‐GLMB filter, and it is superior for the overall performance of the system. Since most of the sensor control schemes, including the scheme the authors proposed, are faced the curse of computation, thus the other novel scheme is proposed. This scheme, in which the sum of the statistical distances between the predicted states of targets and sensor is used as the objective function, evades the updated step of the multi‐target filter, when computing the objective function for each admissible action. Moreover, these two sensor control schemes are applied to a distributed multi‐sensor system, in which the proposed schemes are used for each sensor node and the generalised covariance intersection method is used to compute the fused multi‐target posterior density. Finally, they adopt the sequential Monte‐Carlo method in bearing and range multi‐target tracking scenarios to illustrate the effectiveness of the proposed methods.

This paper considers the sensor selection problem for target tracking in large-scale sensor networks. We propose a new sensor selection strategy based on dual-criterion optimization. Both the bias change detection and information gain maximization are considered as criteria in our proposed sensor selection strategy. This new approach extends the sensor selection problem from single criterion optimization to dual-criterion optimization. Therefore, our proposed approach can be used widely in many target tracking applications. Simulation results show the effectiveness of our proposed sensor selection approach.

This paper investigates the box-particle filter for multi-target tracking, and proposes a clustering based box-particle implementation of PHD filter. A subdivision step is added before the estimation of states. Each box is divided into several sub-box based on the estimated number of targets. An equivalent set of particles can be extracted from the set of subdivided boxes. Then, clustering technique is applied to get the states from the particle set. Compared to the conventional method, this operation avoids the hard division of which target one specific box-particle belongs to. The proposed algorithm can improve the performance of the tracking result, especially for the circumstances some of the targets are close to each other in the surveillance area. Simulation results verify the effectiveness of the algorithm.

Ensemble clustering consists in combining multiple clustering solutions into a single one, called the consensus, which can produce a more accurate and robust clustering of the data. In this paper, we attempt to implement ensemble clustering using Dempster-Shafer evidence theory. Individual clustering solutions are obtained using evidence theory and a novel diversity measure is proposed using the distance of evidence for selecting complementary individual solutions. After establishing the correspondence among different clustering solutions' labels, the consensus clustering solution can be obtained using evidence combination. Experimental results and related analyses show that our proposed approach can effectively implement the ensemble clustering.

This article presents an information theory based sensor management method to be used for aerospace multi-target collaborative detection and tracking. The proposed sensor management method follows an information theoretic approach, in which PCRLB is used to calculate the tracking accuracy of multi-target. The detection particles are employed to determine the detection probability of incoming targets. Moreover, BPSO is adopted as well to find the optimal allocation scheme of sensors for maximizing the tracking accuracy and detection probability. The results of simulation experiments demonstrate that the proposed method succeeds in assigning sensors optimally in region surveillance task.

In this paper, a new entropy based uncertainty measure is introduced for evaluating the significance of subsets of attributes in incomplete decision tables. Some properties of rough conditional entropy are derived. And three attribute reduction algorithms are provided, including an algorithm using exhaustive search, an algorithm using heuristic search and an algorithm using probabilistic search for incomplete decision tables. Furthermore, several simulation experiments on real incomplete data sets are carried out to assess the efficiency of the proposed algorithms. The final simulation results indicate that two of above algorithms can give satisfying performances in the procedure of attribute reduction for incomplete decision tables.

This paper presents a compressed sensing based sensor selection algorithm for direction-of-arrival estimation, in a large scale randomly distributed sensor array. First, a target tracker is employed for the prior information of target position. Second, according to the prior information, we reduce rank of sensing matrix by narrowing the interesting area. Third, a linear independence combination of sensors is utilized to represent the whole sensor array through eigendecomposition. Finally, the statistic and dynamic simulations are carried out, which demonstrate the feasibility of the proposed algorithm.

This study proposes a sensor selection approach based on maximum entropy fuzzy clustering to address the target tracking problem in large‐scale sensor networks. The authors try to deal with this problem at two levels: (i) sensor‐level tracking: data association problem and sensor‐level tracking are carried out at the local level, and only the track outputs are transmitted to the fusion centre for data fusion; (ii) global‐level fusion: two sensor selection strategies are adopted at the fusion centre, which seek to only choose a subset of reliable sensors for track‐to‐track fusion and bias registration. In addition, an improved sensor selection approach is proposed for data fusion in both sparse and dense target environments, and a new fuzzy membership reconstruction strategy is introduced for data association in dense target environments. Furthermore, the proposed sensor selection strategy is also effective in the presence of the possible changing sensor biases. Simulation results are given to evaluate the performance of the proposed approaches.

In this paper we propose a new extension of the k-means algorithm by changing the model it uses to represent clusters. The objectives of traditional k-means and lots of other k-means-related clustering algorithms are all center-based. We suggest using an alternative way to represent the clusters while computing the similarity between an object and a certain cluster. The purpose is to preserve more information of the clusters and at the same time keep the simplicity of the algorithm. In this paper we use intervals to represent clusters and propose a new clustering algorithm k-intervals based on this model. Experimental results on both synthetic data sets and real data sets (several UCI data sets and the ORL face database) demonstrate the effectiveness and the advantages of the proposed algorithm.

The commonly used root‐mean‐square error for estimation performance evaluation is easily dominated by large error terms. So many new alternative absolute metrics have been provided in X. R. Li's work. However, each of these metrics only reflects one narrow aspect of estimation performance, respectively. A comprehensive measure, error spectrum, was presented aggregating all these incomprehensive measures. However, when being applied to dynamic systems, this measure will have three dimensions over the total time span, which is not intuitive and difficult to be analysed. To overcome its drawbacks, a new metric, dynamic error spectrum (DES), is proposed in this study to extend the error spectrum measure to dynamic systems. Three forms under different application backgrounds are given, one of which is balanced taking into account both good and bad behaviour of an estimator and so can provide more impartial evaluation results. It can be applied to a variety of dynamic systems directly. Then the challenge in performance evaluation of the interacting multiple model (IMM) algorithm is considered, and the IMM algorithm is chosen as the testing case to illustrate the superiority of the DES metric. The simulation results validate its utility and effectiveness.

With the development of uncertainty reasoning and information fusion, there have emerged several theories including fuzzy set theory, Dempster-Shafer evidence theory, probability theory and rough set theory, etc., for representing and dealing with the uncertain information. When the fusion of the uncertain information originated from different sources is needed, how to construct a general framework for different theories of uncertainty or how to establish the connection between different theoretical frameworks has become a crucial problem. Particularly, to combine two kinds of information represented respectively by the BPA and the FMF, this paper proposes two transformations of an FMF into a BPA by solving a constrained maximization or minimization optimization problem. The objective function is the uncertainty degree of the body of evidence (BOE) and the corresponding constraints are established based on the given FMF. In fact the transformation of an FMF into a BPA is the transformation of fuzzy sets into random sets, which is currently accepted as a unified framework for several theories of uncertainty. Our proposed approaches have no predefinition of focal elements and they can be used as the general transformations of fuzzy sets into random sets. Some examples and analyses are provided to illustrate and justify the rationality and effectiveness of the proposed approaches.

This paper presents a new method on generating realistic human animation of various styles with given step constraints and a specific skeleton. Given a set of normal walking data captured from different subjects, a hierarchical geostatistical model is automatically learned to encode variety of walking styles by representing human body as a hierarchy of joint groups. For each child hierarchy level, there is a learned geostatistical model, whose low-dimensional control parameters are automatically constructed from the synthesized motion of its parent hierarchy level. The top level is controlled by the given step constraints. Also a realistic transition is finished during each three sequential stances from two interpolated cycles satisfying input step constraints thanks to the local model. We show that the normal walking styles are flexibly controlled by a simple graphlike user interface representing the given skeleton and steps. Our results demonstrate that this method would be helpful to remove the motion clones in group or crowd animation.

Recently, many unified learning algorithms have been developed for principal component analysis and minor component analysis. These unified algorithms can be used to extract principal components and, if altered simply by the sign, can also serve as a minor component extractor. This is of practical significance in the implementations of algorithms. This paper proposes a unified self-stabilizing neural network learning algorithm for principal and minor components extraction, and studies the stability of the proposed unified algorithm via the fixed-point analysis method. The proposed unified self-stabilizing algorithm for principal and minor components extraction is extended for tracking the principal subspace (PS) and minor subspace (MS). The averaging differential equation and the energy function associated with the unified algorithm for tracking PS and MS are given. It is shown that the averaging differential equation will globally asymptotically converge to an invariance set, and the corresponding energy function exhibit a unique global minimum attained if and only if its state matrices span the PS or MS of the autocorrelation matrix of a vector data stream. It is concluded that the proposed unified algorithm for tracking PS and MS can efficiently track an orthonormal basis of the PS or MS. Simulations are carried out to further illustrate the theoretical results achieved.

We present a new segmentation method for the fully polarimetric synthetic aperture radar (PolSAR) data by coupling the cluster analysis in the tensor space and the Markov random field (MRF) framework. The PolSAR data are usually obtained as a set of 3 × 3 Hermitian positive definite polarimetric covariance matrices, which do not form a Euclidean space. If we regard each matrix as a tensor, the PolSAR data space can be represented as a Riemannian manifold. First, the mean shift algorithm is extended to the manifold to cluster such tensors. Then, under the MRF framework, the data energy term is defined by the memberships of all tensors in all the clusters, and the smoothness energy term is defined according to the cluster overlap rates. These parameters regarding the cluster analysis are computed under the Riemannian framework. The total energy is minimized using a graph-cut-based optimization to achieve the segmentation results. The effectiveness of the proposed method is verified using real fully PolSAR data and synthetic images.

The minor component analysis (MCA) deals with the recovery of the eigenvector associated to the smallest eigenvalue of the autocorrelation matrix of the input dada, and it is a very important tool for signal processing and data analysis. This brief analyzes the convergence and stability of a class of self-stabilizing MCA algorithms via a deterministic discrete-time (DDT) method. Some sufficient conditions are obtained to guarantee the convergence of these learning algorithms. Simulations are carried out to further illustrate the theoretical results achieved. It can be concluded that these self-stabilizing algorithms can efficiently extract the minor component (MC), and they outperform some existing MCA methods.

The Mellin transformation-based method is developed to estimate the parameters of non-Rayleigh radar cross section (RCS) models for synthetic aperture radar (SAR) images from the observed image. Models investigated include heavy-tailed Rayleigh and Weibull. For each model, we consider the three kinds of images: intensity, square-root of intensity, and multi-look averaged amplitude. Using the Mellin transformation, we derive the analytical expressions of the first two second-kind cumulants for speckle and RCS respectively, and obtain the estimators according to the multiplicative model of SAR images and the Mellin convolution. Results of parameter estimation from Monte Carlo simulation and real SAR images demonstrate that the proposed estimators, which are easy to implement in the form of closed expressions, are efficient in estimating the parameters of non-Rayleigh RCS models from the observed SAR images.

With the increase of the training set¿s size, the efficiency of support vector machine (SVM) classifier will be confined. To solve such a problem, a novel pre-extracting method for SVM classification is proposed in this paper. In SVM classification, only support vectors (SVs) have significant influence on the optimization result. We adopt a non-parametric k-NN rule called relative neighborhood graph (RNG) to extract the probable SVs from all the training samples. Experimental results verify that the approach proposed can effectively reduce training set¿s size and accelerate the learning speed. At the same time, the classification accuracies are still competitive.

In order to model the synthetic aperture radar (SAR) amplitude images as the heavy-tailed Rayleigh distribution, we focus our attention on two questions in this paper. First, based on the negative-order moments, we propose the logarithmic moment estimator and the iterative logarithmic moment estimator to accurately estimate the parameters of the heavy-tailed Rayleigh distribution. Second, we use the asymptotic series to evaluate the density function of heavytailed Rayleigh distribution, and propose an efficient threestep method suitable for real-time calculation based on interpolating polynomial fit. SAR image modeling experiments demonstrate that the heavy-tailed Rayleigh distribution reflects the high peak and heavy tail of SAR amplitude images, so it is an accurate tool for the modeling of SAR amplitude images.

We propose a three step method to extract buildings from the high-resolution PolSAR data, using both the region-based and edge-based information. Firstly, low-level detectors are employed to provide raw region and edge information of the scene. In the second step, improved region-based building detection results are achieved by fusion of label fields, meanwhile the building profile line segments are extracted under a line Markov random field framework. The last step gives the final building footprint estimates: initial rectangle buildings are defined from the building line segments; by optimizing a surface criterion, the final rectangles are retrieved to fit the region-based building detection results. The effectiveness of this method is demonstrated using the real full PolSAR data.

In this paper, a data fusion approach to the segmentation of SAR and optical images in Markov random field (MRF) framework is proposed. In the joint segmentation scheme based on an MRF model defined on a region adjacency graph (RAG), a fusion rule made on local features of source images is developed for appropriately measuring the feature saliency and incorporating the source reliability of each data source to weigh the source influence in the segmentation procedure. A specific scheme for segmentation of a set of Landsat Thematic Mapper (TM) images and a synthetic aperture radar (SAR) image is presented in detail. Comparative analysis of the proposed segmentation approach against several conventional segmentation approaches carried out on synthetic and real datasets confirms the effectiveness of the proposed approach.

In practical system, the sensor biases may jump abruptly. An adaptive on-line algorithm is presented in this paper for this situation. The algorithm can detect the jump onset time and estimate the jump level base on General Likelihood Ratio (GLR) test. The Monte Carlo results show, our algorithm can adaptively estimate the bias jump level well and the estimation error will not increase remarkably as other previous registration algorithms. The bias estimation error also converges to the Cramer-Rao lower bound (CRLB) after the jumping.

Based on the perfect properties of stochastic particle swarm optimization (SPSO), such as the property of robust and quick convergence, a new scheme is applied to estimate scaling factor for radar constant false alarm rate (CFAR) detectors. Owing to few constraints, it can estimate scaling factor for single radar as well as radar netting system. The numerical results indicate that the particle swarm optimizer has been found to be accuracy and fast in searching the threshold factor T of CFAR detector under any designed probability of false alarm.

The paper develops an adaptive particle swarm optimization(PSO) based on its simplification. With the hope to prompt its global optimization performance, the improved menthod introduces a PD controller into the architecture of the standard PSO(SPSO-PD). The PD controller can control the particle dynamics, and prompt the particles to respond to the change of their exemplars correctly and rapidly, further to surpass the limit of their exemplars with more chances, which greatly contributes to a successful global search. The proposed SPSO-PD was applied to some well-known benchmarks and compared with the standard PSO. The relative experimental results show SPSO-PD performs better than SPSO on the complex optimization functions.

We propose a hybrid method of seeded region growing and region hue-area information fusion for object segmentation under patterned background. At first, image is segmented into many small regions according to hue homogeneity by seeded region growing algorithm, then background texture mode is discovered by the regions' hue-area information fusion, finally, the background texture is removed according to the discovered background mode and seeded region growing, the residual region is the object segmentation result. Experimental results show that the method can be used in object segmentation under patterned background.

The idea of particle filter is to represent probability density function (PDF) of nonlinear/non- Gaussian system by a set of random samples. One of the key issue of particle filter is the proposal distribution. In this paper, the iterated unscented Kalman filter (IUKF) is used to generate the proposal distribution for particle filter. The proposal distributions integrate the current observation, thus greatly improving the filter performance. To evaluate the efficacy of the new algorithm, we apply it in a real-world estimation problem. The simulations results are compared against those of the widely used unscented particle filter (UPF), the extended Kalman particle filter (PF-EKF) and have demonstrated superior estimating performance.

A novel method involved the time-varying tracking model under the nonlinear state-space evolved system is presented, in which the expectationmaximization (EM) algorithm is used to identify the state transition matrix F and the process noise covariance Q online. The typical maneuvering models, as described, essentially, are prior models and use fixed and constant evolved matrix and designed noise level for whole filtering procedure. Actually, the motion of target is always too complicated to be prior modeled as a fixed form, meanwhile, Q used to reflect the mis-match error between the mathematic model and the actual maneuvering mode, thus is time-varying and severely influenced by the environment around the target. Therefore, the prior F and Q can not characterize the maneuvering mode exactly, hence, by assumption that the state evolution and the likelihood of measurements data can be represented by Gaussian distribution, the method of identifying F and Q online is developed. Comparing with the standard IMM filtering, Monte Carlo simulations show that the proposed algorithm is efficient and filtering precision can be improved to some extent.

This paper present a new distributed CFAR (constant false alarm rate) detector based on the adaptive censored cell-averaging CFAR technique. In the scheme, every local decision of individual detector, resulting from the comparison between its sample in test cell and the estimate of clutter power level of its reference samples, takes the value zero or one. In local processor, the CCA-CFAR (censored cell-averaging) technique is utilized to get the local decision, Then, the fusion center makes the global decision based on the total local decisions, which are transmitted from each local sensor. The overall decision, which is zero or one, is obtained at the data fusion center grounded on “k/N” fusion rule. The results show that for the nonhomogeneous background caused by multiple interfering targets, this approach is more reality. Particularly in multiple target situations, it exhibits robustness than MOS (maximum order statistic), mOS(minimum order statistic), and OSOR(ordered statistics), ORAND in distributed sensor networks. Under Swerling 2 assumption, the analytic expression of false alarm and detection probability are derived.

In the sense of likelihood ratio test (LRT), a new type of distributed constant false alarm rate (CFAR) scheme-CCAWCA(censored cell-averaging -R-weighted cell averaging) CFAR detector is presented. Its characteristic is that censored cell-averaging (CCA) CFAR algorithms are used in local processors to form the estimation of SNR of local observations, and then the estimation transmitted to the data fusion center (DFC). Finally, the fusion center makes the final decision based on the weighted cell averaging (WCA). Since the weights are adjusted according to different SNR adaptively, the proposed detector can be available in the case that the target echo and noise/clutter have different level for every sensor. In addition, it does not need a priori knowledge about the interference in order to perform well. Furthermore, unlike the OS-CFAF the tolerance of interfering targets is restricted in appointed k value’s. Under Swerling 2 assumption, the analytic expression of detection probability and false alarm probability are derived.

The paper presents a spatial-color model of object and develops an efficient visual tracking algorithm based on particle filter. This spatial-color model captures richer information than the general color histogram because it incorporates spatial distribution of pixels in addition to color. In order to fast compute the weight of each particle, the Integral Images for computation of histogram, mean and variance are proposed, with which the parameters of spatial-color histograms can be computed by using only simple arithmetic operations and thus the computation load is reduce to linear complexity. Experiments on real image sequences show our method results in more accurate tracking than a standard color particle filter.

Based on the concept of sequential importance sampling (SIS) and the use of Bayesian theory, particle filter is particularly useful in dealing with nonlinear and non-Gaussian problems. In this paper, a new particle filter is proposed that uses a divided difference filter to generate the importance proposal distribution is proposed. The proposal distribution integrates the latest measurements into system state transition density so it can match the posterior density well. The simulation results show that the new particle filter performs superior to the generic particle filter and other particle filters such as the extended Kalman particle filter and the unscented particle filter.

Multiplicative noise makes the interpretation of image extremely difficult, and the fixed-size window filters cannot achieve good trade-off between noise suppression and edge keeping. Based on adaptive windowing and local structure detection, a new filtering algorithm of multiplicative noise is developed in this paper. The sliding window size is automatically adjusted by adaptive windowing, and the local structure detection is required for each window determined because appearance of point target and edge feature cannot satisfy the basic premise of stationary in increment for all statistical filters. Point target is preserved to keep edges and fine details, and the most homogeneous semiwindow on which the central pixel lies is chosen by gradient masks to enhance noise reduction in edge areas. The denoising experiments demonstrate that the proposed filter is superior both in noise suppression and in fine detail preserving.

In the condition of sensor network (SN), to associate local tracks front multiple sensors is a complex task, due to the combination explosion caused by the increasing number of sensors and targets. A new graphical models-based technique for track association is proposed in this paper to deal with the problem. Firstly, by means of the sparse structure inherent in multisensor multitarget tracking scenario, the graphical structure for track association is established; Secondly, the compatibility function of nodes and edges in graph is properly defined to describe the objective function with constrains about track association problem; finally, max-product message passing scheme is employed to obtain the optimal association result. Simulation results demonstrate the efficiency of the presented method.

In natural complex terrain surfaces, scattering targets with random orientations produce random fluctuating echoes which lead to confused classifications by directly using target decomposition on polarimetric SAR (PolSAR) image. In order to reduce the influence, the target vector is transformed into the state with minimization of cross-polarization. Then a set of new parameters u/v/w are used to characterize scattering mechanisms under the deorientation theory, and the fuzzy membership is adopted instead of "hard" division of parameter plan. Characterizing the sample coherency matrices as complex Wishart distribution, the PolSAR image is clustered based on Bayes maximum likelihood (ML) criteria. Experiment is carried out on an L-band NASA/JPL SIR-C PolSAR image over Danshui town, Guangdong, China. Comparison results with the popular used methods show that the proposed method provides a significant improvement in classification and the associated scattering mechanism of class is more accurate and beneficial for automatic terrain recognition.

Positive alpha-stable distribution is used to model the nonnegative quantity with impulsive property. Based on negative-order moments, three methods are used to estimate the parameters of positive alpha-stable distribution in this paper. First, a ratio estimator based on the ratio of negative-order moments is presented whose performance is significantly determined by the choice of order. Second, a new estimator with explicit closed form is presented and it is shown to be robust compared to the ratio estimator. Last, an iterative estimator is proposed and it achieves better performance only using fewer samples in each step computation. Monte Carlo simulation results demonstrate that the proposed iterative estimator is high efficient for the positive alpha-stable distribution.

In recent years, there has been considerable interest within the tracking community in the approach to data association based on the assignment algorithm. The focus of this paper is to describe a new dynamic S-D assignment algorithm. In the presence of clutter, missed detections, the new method incorporates the image intensity information into the dynamic assignment cost coefficient. To reasonably utilize the special feature of IR sensors, the association becomes more accurate and, at mean time, the complexity of the algorithm has not obviously increased. The simulation results show that the proposed approach has better performance than traditional approaches.

Due to the presence of speckle in synthetic aperture radar (SAR) image, the existing registration algorithms, which are successfully used in optical remote sensing image, are usually not applicable to it directly. An automatic SAR image registration algorithm is proposed in this paper. Firstly, the element triangles are constructed from the point targets detected from the SAR images; then they are matched by integrating triangle moment invariance proposed in this paper and region invariant moments; finally, the LMSE algorithm is used to estimate the affine transformation parameters, thus the SAR images can be registered automatically. The proposed algorithm is evaluated and compared with the existing methods by means of invariant moments (IM) and affine moment invariants (AMI). It is shown from Monte-Carlo simulations that the proposed algorithm is robust to detection error and partial correspondence of control points (CPs), and has higher ratio of correct matching than the methods using IM or AMI. Experimental results show that the proposed new algorithm is not only valid in the automatic registration of SAR images, but also can avoid the influence caused by speckle in feature detection and feature matching process

Data association is one of the key techniques on bearing-only tracking with infrared search and track (IRST) system. A new data association algorithm based on information fusion is proposed in this paper. Firstly, by considering the special feature of IRST system, the new method constructs several kinds of evidences that are based on the multi-type information, such as angular measurement, intensity level and so on. And then all results from different aspects are fused by utilizing Dempster combination rule. Finally, the association decision is obtained by maximizing the final mass function. The simulation results show that the proposed approach has better performance than the traditional ones. As observation conditions deteriorate, the advantage of new method becomes obvious. For uncertain data, this new data association algorithm based on evidence theory is excellent by utilizing reasonably the measurements

A novel knowledge discovery method to multiple classifier fusion is proposed. In the new method, all base classifiers are viewed as predictors relating to domain knowledge, and they may be allowed to operate in different feature spaces. Then the beliefs assigned to each base classifier are generated automatically from the established decision tables (DTs). For this purpose, two types of belief structures on DT are investigated based on generalized rough set model and Dempster-Shafer theory (DST). Correspondingly, two fusion approaches are designed based on the belief structures and the heuristic fusion function. Compared with plurality voting, the vegetation classification experiment on hyperspectral remote sensing images shows that the performance of the classification can be improved further by using the proposed method

Based on assumption that background would not be the parts which appear in the sequence for a short time, a background reconstruction algorithm based on online clustering was proposed in this paper. Firstly, pixels intensities are classified based on online clustering. Secondly, cluster centers and appearance probabilities of each cluster are calculated. Finally, a single or multi intensities clusters with the appearance probability greater than threshold are selected as the background pixel intensity value. Simulation results show that the algorithm can represent situation where the background contains bi-model or multi-model distribution, and motion segmentation can be performed correctly. The algorithm with inexpensive computation and low memory can accommodate the real-time need

The presence of speckle, which modeled as a strong multiplicative noise, makes the edge detection of synthetic aperture radar (SAR) images very difficult. The usual edge detectors using gradient yield poor results. By analyzing Harris' combined corner and edge detector, to solve the speckle-sensitivity and anisotropic response, an edge detection method used for SAR image is presented. Horizontal, vertical and diagonal ratios of averages of nonoverlapped neighboring pixel values instead of gradients are used to calculate edge strength maps (ESM); then they are fused to obtain the final ESM; finally, a threshold is presented to obtain edge detection result. Analysis and experiment results illustrate that the proposed method can be used for edge detection in SAR image effectively.

A novel methodology utilizing the direct data domain approach to space-time adaptive processing (STAP) in airborne radar environments is presented in this paper. The deterministic least squares adaptive signal processing technique operates on a snapshot-by-snapshot basis to determine the adaptive weights for nulling interferences and estimating signal of interest (SOI), and eliminates the requirement for the secondary data from neighboring range cell. This is in contrast to conventional adaptive techniques where processing is done by taking the time averages as opposed to spatial averages. Simulation results illustrate the efficiency of interference suppression in nonhomogeneous environment.

This paper deals with data (or information) fusion for the purpose of estimation. Three estimation fusion architectures are considered: centralized, distributed, and hybrid. A unified linear model and a general framework for these three architectures are established. Optimal fusion rules based on the best linear unbiased estimation (BLUE), the weighted least squares (WLS), and their generalized versions are presented for cases with complete, incomplete, or no prior information. These rules are more general and flexible, and have wider applicability than previous results. For example, they are in a unified form that is optimal for all of the three fusion architectures with arbitrary correlation of local estimates or observation errors across sensors or across time. They are also in explicit forms convenient for implementation. The optimal fusion rules presented are not limited to linear data models. Illustrative numerical results are provided to verify the fusion rules and demonstrate how these fusion rules can be used in cases with complete, incomplete, or no prior information.