Search

In this paper we propose a two-pass classification method and apply it to face recognitions. The method is obtained by integrating together two approaches, the hyper-ellipsoid neural networks (HENN's) and the SVM's with error correcting codes. This method realizes a classification operation in two passes: the first one is to get an intermediate classification result for an input sample by using the HENN's, and the second pass is followed by using the SVM's to re-classify the sample based on both the input data and the intermediate result. Simulations conducted in the paper for applications to face recognition showed that the two-pass method can maintain the advantages of both the HENN's and the SVM's while remedying their disadvantages. Compared with the HENN's and the SVM's, a significant improvement of recognition performance over them has been achieved by the new method. [ABSTRACT FROM AUTHOR]

This paper presents an approach for solving the driving load forecasting problem based on Cascade Neural Networks with node-decoupled extended Kalman Filtering (CNN-NDEKF). Because of the inherent advantages, hybrid electric vehicles (HEV) are being given more and more attention. The power control strategy of HEVs is the key technology which determines the HEV's efficiency and pollutive emission level. Since the extent of improvement involved with HEV power control strategies greatly depends on the future driving load forecasting, in this paper, we attempt to achieve driving load forecasting using CNN-NDEKF. Instead of forecasting the entire load sequence, we define 5 load levels by a fuzzy logic method and then we forecast the load level. Simulation study is given to illustrate the feasibility of the driving load forecasting approach. [ABSTRACT FROM AUTHOR]

Aiming at improving the convergence performance of conventional BP neural network, this paper presents an improved PSO algorithm instead of gradient descent method to optimize the weights and thresholds of BP network. The strategy of the algorithm is that in each iteration loop, on every dimension d of particle swarm containing n particles, choose the particle whose velocity decreases most quickly to mutate its velocity according to some probability. Simulation results show that the new algorithm is very effective. It is successful to apply the algorithm to gas turbine fault diagnosis. [ABSTRACT FROM AUTHOR]

The controlled system is an uncertain nonlinear differential-algebraic subsystem (DASs) in a large-scale system. The problem of robust stabilization for such class of uncertain nonlinear DASs is considered in this paper. The robust stabilization controller is proposed based on backstepping approach using two-layer Artificial Neural Networks (ANN) whose weights are updated on-line. The closed-loop error systems are uniformly ultimately bounded (UUB) and the error of convergence can be made arbitrarily small. Finally, using the design scheme proposed in this paper, a governor controller is designed for one synchronous generator in a multi-machine power systems. The simulation results demonstrate the effectiveness of the proposed scheme. [ABSTRACT FROM AUTHOR]

This paper develops a new blind channel identification method in frequency domain. Oversampled signal has the property of spectral redundancy in frequency domain which is corresponding to the cyclostationarity property in time domain. This method exploits the cyclostationarity of oversampled signals to identify possibly non-minimum phase FIR channels. Unlike many existing methods, this method doesn't need EVD or SVD of correlation matrix. Several polynomials are constructed and zeros of channels are identified through seeking for common zeros of those polynomials. It is in the similar spirit of Tong's frequency approach, but this new algorithm is much simpler and computationally more efficient. A sufficient and necessary condition for channel identification is also provided in this paper. This condition is quite similar to Tong's time domain theory but it is derived from a novel point of view. [ABSTRACT FROM AUTHOR]

Most of the algorithms proposed in the literature deal with the problem of digital modulation classification and consider classic probabilistic or decision tree classifiers. In this paper, we compare and analyze the performance of 2 neural network classifiers and 3 support vector machine classifiers (i.e. 1-v-r type, 1-v-1 type and DAG type multi-class classifier). This paper also deals with the modulation classification problems of classifying both analog and digital modulation signals in military and civilian communications applications. A total of 7 statistical signal features are extracted and used to classify 9 modulation signals. It is known that the existing technology is able to classify reliably (accuracy ≥ 90%) only at SNR above 10dB when a large range of modulation types including both digital and analog is being considered. Numerical simulations were conducted to compare performance of classifiers. Results indicated an overall success rate of over 95% at the SNR of 10dB in all classifiers. Especially, it was shown that 3 support vector machine classifiers can achieve the probabilities of correct classification (Pcc) of 96.0%, 97.3% and 97.8% at the SNR of 5dB, respectively. [ABSTRACT FROM AUTHOR]

Recently, the distribution and using of the digital multimedia contents are easy by developing the internet application program and related technology. However, the digital signal is easily duplicated and the duplicates have the same quality compare with original digital signal. To solve this problem, there is the multimedia fingerprint which is studied for the protection of copyright. Fingerprinting scheme is a technique which supports copyright protection to track redistributors of electronic information using cryptographic techniques. Only regular user can know the inserted fingerprint data in fingerprinting schemes differ from a symmetric/asymmetric scheme and the scheme guarantee an anonymous before re-contributed data. In this paper, we present a new scheme which is the detection of colluded multimedia fingerprint by neural network. This proposed scheme is consists of the anti-collusion code generation and the neural network for the error correction. Anti-collusion code based on BIBD(Balanced Incomplete Block Design) was made 100% collusion code detection rate about the average linear collusion attack, and the Hopfield neural network using (n,k) code designing for the error bits correction confirmed that can correct error within 2bits. [ABSTRACT FROM AUTHOR]

Currently, there is little quantitative research on macroscopic real estate finance at home and abroad. Seen from the whole system of real estate finance, this paper chooses 14 main indexes to compose an evaluation index system. Based on the evaluation index system, an error -back-propagation BP network model is built to evaluate the growth of real estate finance. Data of real estate financial system from 1997-2005 are used as train and test samples of BP neural network. After training, the BP neural network is used to evaluate and forecast by simulation. Through the good accuracy of evaluation and forecasting, the model is proved to be very efficient. By comparing the growing difference of two adjacent years and analyzing the related macro financial policies in related years, the running effect of related real estate financial policies in related years is gained. So by using the evaluation model of this paper, decision makers can decide to use what kind of macro adjusting and controlling policies to gain anticipated aim of real estate finance in the future. [ABSTRACT FROM AUTHOR]

Aiming at the existence of relativity between repeat or similar samples and character parameters during diagnosis of character data, this paper presents an effective data analysis approach for character data compression from bi-direction, which can reduce the burden of learning machine without losing the connotative character knowledge of character data. At the first step of the algorithm, basing on the theory of component analysis, the paper adopt a principal component analysis approach to reduce the dimension of data horizontally, then after comparison of existing clustering algorithms, put forward an immune clustering algorithm based on similarity measurement of principle component core for vertical reduction by using related mechanism of clone selection as well as immune network self-stabilization in organism natural immune system for reference. Finally, to analyze machine behavior quantitatively, a pattern discrimination model based on a cerebellar model articulation controller neural network (NN) was developed. Simulation experiments proved the effectiveness of this algorithm. [ABSTRACT FROM AUTHOR]

Aiming at the problem of parameter estimation in radar detection, a modified RBF neural network is proposed to estimate parameter accurately because of its good approximation ability to random nonlinear function and quick convergence speed. Two classical detection methods, which widely used in radar field, are listed in this paper, and their corresponding parameters are estimated with modified RBF neural network. Theoretical analysis and numerical results both show that the proposed method has good parameter estimation accuracy and quick convergence speed. [ABSTRACT FROM AUTHOR]

Image segmentation is of great importance in the field of image processing. A wide variety of approaches have been proposed for image segmentation. However, SAR image segmentation poses a difficult challenge owing to the high levels of speckle noise. In this paper, we proposed a SAR image segmentation method based on one-class support vector machines (SVM) to solve this problem. One-class SVM and two-class SVM for segmentation is discussed. One-class way is a kind of unsupervised learning, and one-class SVM based segmentation method reduces greatly human interactions, while yielding good segmentation results compared to two-class SVM based segmentation method. The segmentation results based on SVM are also compared to threshold method and adaptive threshold method. Experimental results demonstrate that the proposed method works well for image segmentation while reducing the speckle noise. [ABSTRACT FROM AUTHOR]

Visual tracking has been an active area of research in computer vision. However, robust tracking is still a challenging task due to cluttered backgrounds, occlusions and pose variations in the real world. To improve the tracking robustness, this paper proposes a tracking method based on multi-cue adaptive fusion. In this method, multiple cues, such as color and shape, are fused to represent the target observation. When fusing multiple cues, fuzzy logic is adopted to dynamically adjust each cue weight in the observation according to its associated reliability in the past frame. In searching and tracking object, neural network algorithm is applied, which improves the searching efficiency. Experimental results show that the proposed method is robust to illumination changes, pose variations, partial occlusions, cluttered backgrounds and camera motion. [ABSTRACT FROM AUTHOR]

This paper illustrates the potentials of the PCNN for image processing. A description of three schemes for image processing using the PCNN is presented in this paper. The first scheme is related to image segmentation, the second to automatic target location, ATL, and the third to face recognition. The first scheme was developed in order to obtain an insight of the behavior of the PCNN as a preprocessor element, the second one is an application to test the performance of the PCNN in an ATL problem. The third is a feature extraction method for face recogiton. The segmentation scheme showed great potentials to perform pixel grouping. The second scheme turned into a system with an ATL performance as good as other systems reported in the literature. And the third scheme seems to improve the performance of a face recognition system. [ABSTRACT FROM AUTHOR]

In this paper, a texture image segmentation algorithm based on improved wavelet neural network is proposed.This algorithm can overcome shortcomings of traditional threshold segmentation techonologies. By using texture features of images, a series of fractal texture feature parameters which will be taken as input layer factors of wavelet network are created by this algorithm. Then, the wavelet neural network is trained with self-adaptive pheromone volatilization mechanism and dynamic heuristic search strategy of improved ant colony algorithm. Finally, the trained wavelet neural network is taken as the classifier of image pixel to realize segmentation of texture images. Simulation experiment shows that, improved algorithm could realize self-adaptive segmentation based on different texture features of images and it is robuster. However, further researches on methods of improving convergence speed of this algorithm and objective criteria for assessing whether texture images have been segmented successfully or not are needed. [ABSTRACT FROM AUTHOR]

During preceding theory study and engineering application, we dealed with the parameter estimation of one-dimension long memory process actually, and rarely take into account high dimensions. There are few papers about it. In this paper, using the decorrelation property of discrete wavelet transform, high dimension situation (mainly 2D) is simplified to 1D and corresponding referrers are improved according to new idea, combining with matrix transform. So the computation complexity is reduced effectively and estimation precision is satisfied. Some experiment results show that this algorithm has a better general performance. [ABSTRACT FROM AUTHOR]

A novel Support Vector Regression(SVR) algorithm has been proposed recently by us. This approach, called Lagrangian Support Vector Regression(LSVR), is an reformulation on the standard linear support vector regression, which leads to the minimization problem of an unconstrained differentiable convex function. During the process of computing, the inversion of matrix after incremented is solved based on the previous results, therefore it is not necessary to relearn the whole training set to reduce the computation process. In this paper, we implemented the LSVR and tested it on Mackey-Glass time series to compare the performances of different algorithms. According to the experiment results, we achieve a high-quality prediction about time series. [ABSTRACT FROM AUTHOR]

As high-voltage electric equipment has complex structure and works in harsh environment, this paper is aimed at applying Optical Fiber Sensors to a temperature-variation fault diagnosis system of high-voltage electric equipment based on the combination of neural network and expert system. Neural network has the characteristics of self-adapted, distributed storage and associative memory. Using BP neural network, we can make on-line diagnosis of temperature-variation fault of high-voltage electric equipment. All the uses above can increase the speed of diagnosis and make results be more exact. [ABSTRACT FROM AUTHOR]

This paper presents an adaptive threshold neural-network scheme for Rotorcraft Unmanned Aerial Vehicle (RUAV) sensor failure diagnosis. The approach based on adaptive threshold has the advantages of better detection and identification ability compared with traditional neural-network-based scheme. In this paper, the proposed scheme is demonstrated using the model of a RUAV and the results show that the adaptive threshold neural-network method is an effective tool for sensor fault detection of a RUAV. [ABSTRACT FROM AUTHOR]

This paper presents an efficient approach based on Hopfield network for solving nonlinear optimization problems, with polynomial objective function, polynomial equality constraints and polynomial inequality constraints. A modified Hopfield network is developed and its stability and convergence is analyzed in the paper. Then a mapping of nonlinear optimization problems is formulated using the modified Hopfield network. Simulation results are provided to demonstrate the performance of the proposed neural network. [ABSTRACT FROM AUTHOR]

Collaborative optimization (CO), one of the multidisciplinary design optimization (MDO) approaches, is a two-level optimization method for large-scale and distributed-analysis engineering design problem. In practical application, CO exists some known weaknesses, such as slow convergence, complex numerical computation, which result in further difficulties when modeling the satisfaction degree in CO. This paper proposes the use of approximation response model in place of discipline-level optimization in order to relieve the aforementioned difficulties. In addition, a satisficing back propagation neural network based on multiple-quality and multiple-satisfaction mapping criterion is applied to the design of the satisfaction degree approximation for disciplinary objective. An example of electronic packaging problem is provided to demonstrate the feasibility of the proposed method. [ABSTRACT FROM AUTHOR]

In this paper, in terms of a linear matrix inequality (LMI), using a delayed Lagrangian network to solve quadratic programming problems, sufficient conditions on delay-dependent and delay-independent are given to guarantee the globally exponential stability of the delayed neural network at the optimal solution. In addition, exponential convergence rate is estimated by the equation in the paper. Furthermore, the results in this paper improved the ones reported in the existing literatures and the proposed sufficient condition can be checked easily by solving LMI. Two simulation examples are provided to show the effectiveness of the approach and applicability of the proposed criteria. [ABSTRACT FROM AUTHOR]

In this paper, we present a neural network based optimization method for solving a kind of quadratic programming problems (QPPs) with equality and inequality constraints. The proposed method is appropriate for distributed implementation and can be used as a basic optimization module for managing optimization problems of large distributed systems. We test the proposed method in a real PC-Network for power system state estimation problem. Several cases are considered and obtain some successfully results. [ABSTRACT FROM AUTHOR]

This paper proposes an effective particle swarm optimization (PSO) based memetic algorithm (MA) for designing artificial neural network. In the proposed PSO-based MA (PSOMA), not only the evolutionary searching mechanism of PSO characterized by individual improvement plus population cooperation and competition is applied to perform the global search, but also several adaptive high-performance faster training algorithms are employed to enhance the local search, so that the exploration and exploitation abilities of PSOMA can be well balanced. Moreover, an effective adaptive Meta-Lamarckian learning strategy is employed to decide which local search method to be used so as to prevent the premature convergence and concentrate computing effort on promising neighbor solutions. Simulation results and comparisons demonstrate the effectiveness and efficiency of the proposed PSOMA. [ABSTRACT FROM AUTHOR]

In this paper, we present the general analysis of global convergence for the recurrent neural networks (RNNs) with projection mappings in the critical case that M(L,Γ), a matrix related with the weight matrix W and the activation mapping of the networks, is nonnegative for a positive diagonal matrix Γ. In contrast to the existing conclusion such as in [1], the present critical stability results do not require the condition that ΓW must be symmetric and can be applied to the general projection mappings other than nearest point projection mappings. An example has also been shown that the theoretical results obtained in the present paper have explicitly practical application. [ABSTRACT FROM AUTHOR]

In this paper, a class of reaction-diffusion recurrent neural networks with time-varying delays and Dirichlet boundary conditions are considered by using an approach based on the delay differential inequality and the fixed-point theorem. Some sufficient conditions are obtained to guarantee that the reaction-diffusion recurrent neural networks have a periodic orbit and this periodic orbit is globally attractive. The results presented in this paper are the improvement and extension of the existed ones in some existing works. [ABSTRACT FROM AUTHOR]

A novel equalization scheme for a wireless ATM communication channel using a recurrent neural network is proposed in this paper. The recurrent neural network used in this scheme is the Complex Bilinear Recurrent Neural Network (CBLRNN). A reduced version of the CBLRNN for faster and stable convergence is first proposed in this paper. The R-CBLRNN is then applied to equalization of a wireless ATM channel for 8PSK, which has severe nonlinearity and intersymbol interference due to multiple propagation paths. The experiments and results show that the proposed R-CBLRNN gives very favorable results in the Mean Square Error (MSE) criterion over conventional equalizers. [ABSTRACT FROM AUTHOR]

Parallel process neural network (PPNN) is a novel spatio-temporal artificial neural network. The approximation capability analysis is very important for the PPNN to enhance its adaptability to time series prediction. The approximation capability of the PPNN is analyzed in this paper, and it can be proved that the PPNN can approximate any continuous functional to any degree of accuracy. Finally, the PPNN is utilized to predict the iron concentration of the lubricating oil in the aircraft engine health condition monitoring to highlight the approximation capability of the PPNN, and the application test results also indicate that the PPNN can be used as a well predictive maintenance tool in the aircraft engine condition monitoring. [ABSTRACT FROM AUTHOR]

In this paper, the control approach to irregular structures excited by multi-dimensional ground motions is presented by using semi-active tuned liquid column damper (TLCD). A back propagation Artificial Neural Network (ANN) is used to predict the responses of structure due to two-dimensional seismic inputs. The semi-active control strategy is established and implemented based on ANN. The numerical examples have shown that it is an effective method presented for controlling the translational and rotational responses of irregular structures. [ABSTRACT FROM AUTHOR]

The paper presents a wavelet packet neural network (WPNN) approach for solving the waveform distortion problem of protective relaying testing instrument. With its excellent time-frequency localization property and approximation ability, WPNN is used to establish an identification model of the non-linear amplifier of the protective relaying testing instrument. The fault data to be put into the instrument is compensated by an adjusting function getting from the identification model, which makes the whole instrumentation system show linear performance so that the distortion of the output waveform is constrained greatly. Simulation results indicate the feasibility and validity of the proposed approach, and a prototype has been put into practical operation. [ABSTRACT FROM AUTHOR]

The paper describes the application of an artificial neural network in natural language text reasoning. The task of knowledge discovery in text from a database, represented with a database file consisting of sentences with similar meanings but different lexico-grammatical patterns, was solved with the application of neural networks which recognize the meaning of the text using designed training files. We propose a new method for natural language text reasoning that utilizes three-layer neural networks. The paper deals with recognition algorithms of text meaning from a selected source using an artificial neural network. In this paper we present that new method for natural language text reasoning and also describe our research and tests performed on the neural network. [ABSTRACT FROM AUTHOR]

In this paper, we address the problem of detection and correspondence of moving persons in a multi-camera set. A novel algorithm is proposed based on object occupancy random field (abbreviated by OORF), which has a robust performance even under severe occlusions and a fast implementation speed. The core of our algorithm is OORF and object window structure. The latter is essential to compute OORF and provides a scheme to detect and correspond objects simultaneously. [ABSTRACT FROM AUTHOR]

This paper presents a method for predicting nonlinear time series. It is based on the multiscale filtering, fundamental and technical model and artificial neural networks. In the technical model we used wavelet transform for disjoin the time series trends then to smooth the economic time series by multiscale filtering. We used too the fundamental analysis, that is, financial and macroeconomics variables to improve the network forecasting. The results were compared with the technical analysis showing that the multiscale filtering and addition of the fundamental variables increase the network forecasting ability. [ABSTRACT FROM AUTHOR]

Stock market prediction has always been one of the hottest topics in research, as well as a great challenge due to its complex and volatile nature. However, most of the existing methods neglect the impact from mass media that will greatly affect the behavior of investors. In this paper we present a system that combines the information from both related news releases and technical indicators to enhance the predictability of the daily stock price trends. The performance shows that this system can achieve higher accuracy and return than a single source system. [ABSTRACT FROM AUTHOR]

This paper proposes a novel off-line signature verification method based on adaptive multi-resolution wavelet zero-crossing and one-class-one-network classification. First, the horizontal, vertical, 45 degree direction and the 135 degree direction projections of the binarizated signature images are calculated, respectively. The curvature data of the projections are decomposed into multi-resolution signals using wavelet transforms. Then the zero-crossings corresponding to the curvature data are extracted as features for verification. At last, one-class-one-network classifier is used to verify the signatures. The signature verification system was experimented on real data sets and the results show the system is very effective. [ABSTRACT FROM AUTHOR]

This paper addresses the problem of designing a Markovian H ∞  filter for a class of nonlinear stochastic Markovian jump systems. Firstly, neural networks are employed to approximate the nonlinearities in the different jump modes. Secondly, the overall system is represented by the mode-dependent linear difference inclusion (LDI). Then, a neural network-based Markovian H ∞  filter is developed using the stochastic Lyapunov-Krasovskii stability theory under some linear matrix inequality (LMI) constraints. Finally, a numerical example is worked out to show the usefulness of the theoretical results. [ABSTRACT FROM AUTHOR]

Optimal message scheduling is one of the key issues in the field of controller area network (CAN) bus system. There are numerous approaches related to this issue. Most of them are essentially based on priority-based strategies. In 1, we utilized Radial Basic Function (RBF) network 2 as a message scheduling controller to dynamically schedule messages. Furthermore, an online Backward-Through-Time (BTT) algorithm is presented for parameter optimization under a priori fixed network structure. Intuitively, an inappropriate RBF network structure leads to performance degradation. In the worst case, the CAN system diverges. In this paper, we extend our previous works by including Minimal Resource Allocation (MRA) algorithm for structure determination. In this way, both problems of parameter optimization and structure determination can be resolved at the same time. Simulation results demonstrated that the proposed BTT with MRA methods outperform our previous results in terms of convergence time, stability, and the number of required hidden neurons (or radial basis functions). [ABSTRACT FROM AUTHOR]

In the new H.264 video coding standard, motion estimation takes up a significant encoding time especially when using the straightforward full search algorithm (FS). We present an efficient scheme based on BP neural network algorithm which we believe can overcome to a significant degree this shortcoming. The mean squared error (MSE) between the current block and the same position in the reference frame is an often used matching criteria in block matching process. The scheme presented is very well suited to neural network training where the performance index is the mean squared error. The experimental results in Table 1 and Table 2 in the full paper compare our method with the full search algorithm. These comparisons show preliminarily but clearly that our method dose overcome to a significant degree the shortcoming of FS mentioned at the beginning of this abstract with neglectable coding efficiency loss. Keywords:Motion Estimation; mean squared error; BP neural network. [ABSTRACT FROM AUTHOR]

The effect of image quality on the performance of multimodal biometric verification is studied. A biometric system based solely on single modality is often not able to meet the system performance requirements for poor image quality. Prior studies of multimodal biometric authentication have shown that it can improve performance over use of a single unimodal biometric. The well-known multimodal methods do not consider the quality information of the data used when combining the results from different matchers. In the paper, a novel SVM-based multimodal biometric authentication system is presented. It is based on SVM classifiers and quality measures of the input biometric signals. Experimental results on a prototype application based on fingerprint and face are reported. The proposed scheme is shown to outperform significantly multimodal systems without considering quality signals and unimodal systems over a wide range of image quality. [ABSTRACT FROM AUTHOR]

Image magnification is among the basic image processing operations. The most commonly used techniques for image magnification are based on interpolation method. However, the magnified images produced by the techniques, such as nearest neighbor, bilinear and cubic method, often appear a variety of undesirable image artifacts such as 'blocking' and 'blurring' into the several processing for image magnification. In this paper, we propose image magnification method by properties of human visual system which reduce information during transforming from receptors to ganglion cells in retina and magnify information at visual cortex. Our method uses the whole image to exactly detect the edge information of the image and then emphasizes edge information. Experiment results show that the proposed method solves the drawbacks of the image magnification, such as blocking and blurring, and has a higher PSNR and Correlation than the traditional methods. [ABSTRACT FROM AUTHOR]

The enlargement of the digital image implies the improvement of the image resolution, where the high frequency components lost in sampling must be estimated. In this paper, an image enlargement method using a high resolution neural network is proposed, corresponding to the region with rapid change (high local variance) and the region requiring a smooth interpolation (low local variance). It is shown that the high resolution NN has high potential ability. [ABSTRACT FROM AUTHOR]

A novel image quality metric based on the characteristics of wavelet coefficients of images is proposed in this paper. An image is decomposed into several levels by means of wavelet transform. The standard deviations of the diagonal details (HH coefficients) at each level increase with the noise standard deviation increasing and decrease with the blurring radius increasing. According to that, an image quality can be measured by analyzing the characteristics of its wavelet coefficients. Neural network is used to realize the algorithm of image quality assessment. The results of experiments demonstrate that the image quality metric is reasonable and the algorithm realization using neural network is feasible and performs well. [ABSTRACT FROM AUTHOR]

A novel edge enhancement based on Hopfield neural net is presented in this paper, which is a post-processing complement for a pre-existing edge detector. This term is added to the output of the edge detector. Firstly, the energy function which is used to find the final stable edges is provided in the Hopfield neural net, and then, based on the window iteration, it improves the performance of the edge detector by recovering missing edges and eliminating false edges. In experiments conducted on various images, we demonstrate the performance of the algorithm on them. [ABSTRACT FROM AUTHOR]

A new approach for maximum posterior probability (MAP) image restoration based on cellular neural network (CNN) is proposed in this paper, and hardware realization is also discussed. According to analysis of MAP image restoration, a new template is proposed for CNN image restoration. The computer simulation result proves the approach is reasonable, then a hardware system based on CNN processor is setup for the restoration algorithm, and the effectiveness of the CNN processor is also confirmed in this system. [ABSTRACT FROM AUTHOR]

A neural-based method for source separation in nonlinear mixture is proposed in this paper. A cost function, which consists of the mutual information and partial moments of the outputs of the separation system, is defined to extract the independent signals from their nonlinear mixtures. A learning algorithm for the parametric RBF network is established by using the stochastic gradient descent method. This approach is characterized by high learning convergence rate of weights, modular structure, as well as feasible hardware implementation. Successful experimental results are given at the end of this paper. [ABSTRACT FROM AUTHOR]

Extracting independent source signals from their nonlinear mixtures is a very important issue in many realistic models. This paper proposes a new method for solving nonlinear blind source separation (NBSS) problems by exploiting particle swarm optimization (PSO) algorithm and natural gradient descent. First, we address the problem of separation of mutually independent sources in post-nonlinear mixtures. The natural gradient descent is used to estimate the separation matrix. Then we define the mutual information between output signals as the fitness function of PSO. The mutual information is used to measure the statistical dependence of the outputs of the demixing system. PSO can rapidly obtain the globally optimal coefficients of the higher order polynomial functions. Compared to conventional NBSS approaches, the main characteristics of this method are its simplicity, the rapid convergence and high accuracy. In particular, it is robust against local minima in search for inverse functions. Experiments are discussed to demonstrate these results. [ABSTRACT FROM AUTHOR]

Using the neural network model for oriented principal component analysis (OPCA), we propose a solution to the data least squares (DLS) problem, in which the error is assumed to lie in the data matrix only. In this paper, We applied this neural network model to channel equalization. Simulations show that DLS outperforms ordinary least square in channel equalization problems. [ABSTRACT FROM AUTHOR]

Scarcity of the spectrum resource and mobility of users make Quality-of-Service (QoS) provision a critical issue in wireless multimedia networks. This paper uses neural network as call admission controller to perform call admission decision. A performance measurement is formed as a weighted linear function of new call and handoff call blocking probabilities of each service class. Simulation compares the neural network with complete sharing policy. Simulation results show that neural network has a better performance in terms of average blocking criterion. [ABSTRACT FROM AUTHOR]

This paper proposed a new fault-detection method based on the combination of Rough Set (RS) and Artificial Neural Network (ANN), called hybrid fault-detection method based on RS and ANN (HFDMRSNN), which uses RS to reduce parameters of a pipeline system and then uses ANN (three-layer neural network) to form a detection model. This method could detect fault of pipeline not only in stationary status but also in non-stationary status. The efficiency of the HFDMRSNN in detecting fault in real pipeline system is evaluated by an experiment in a long product oil pipeline in Shandong China. From the results, it is observed that the proposed HFDMRSNN is able to identify the status of complex pipeline effectively. [ABSTRACT FROM AUTHOR]

This paper develops a fast and accurate algorithm for training transductive SVMs classifiers, which utilizes the classification information of unlabeled data in a progressive way. For improving the generalization accuracy further, we employ three important criteria to enhance the algorithm, i.e. confidence evaluation, suppression of labeled data, stopping with stabilization. Experimental results on several real world datasets confirm the effectiveness of these criteria and show that the new algorithm can reach to comparable accuracy as several state-of-the-art approaches for training transductive SVMs in much less training time. [ABSTRACT FROM AUTHOR]

A novel method for the multiuser detection in CDMA communication systems based on a stochastic cellular neural network (SCNN) is proposed in this paper. The cellular neural network (CNN) can be used in multiuser detection, but it may get stuck in a local minimum resulting in a bad steady state. The annealing CNN detector has been proposed to avoid local minima; however, the near-far effect resistant performance of it is poor. So, the SCNN detector is proposed here through adding a stochastic term in a CNN. The performance of the proposed SCNN detector is evaluated via computer simulations and compared to that of the conventional detector, the stochastic Hopfield network detector, and the Annealing CNN detector. It is shown that the SCNN detector can avoid local minima and has a much better performance in reducing the near-far effect than these detectors, as well as a superior performance in bit-error rate. [ABSTRACT FROM AUTHOR]