Search

Non-negative Matrix Factorization (NMF) is a method to obtain a representation of data using non-negativity constraints. These constraints lead to a part-based representation in the vector space because they allow only additive, not subtractive, combinations of original data. This is how NMF learns a part- based representation. This paper introduces briefly the theory and algorithm of NMF. Then a NMF ANN framework is presented to classify spontaneous EEG in five metal tasks. Several comparisons and experiments were carried out. The results showed that NMF lead more localized and sparse features than power spectrum method and principal component analysis method did, and that the NMF-ANN structure preserved the spatio-temporal characteristics of EEG signals. Its best cognition rate of five mental task pairs can achieves better than 88.0%. It may be a promising classifier for Brain Computer Interface (BCI) scheme.

The event related potentials (ERPs) is an electrical change recorded from the brain in relation to an event that occurs either in the external world or within the brain itself. Here we are to design a system capable of learning a particular mapping between ERPs and different mental tasks is of great significance. ERPs is traditionally extracted by averaging. The paper proposed a new ERP extraction algorithm using artificial neural network. It discussed the configuration, learning and running of the designed network. The partial least square regression was introduced to train the neural network. Series experiments showed that the method is effective and is suitable for single-trail ERP estimation.

Based on the convergence and stability of the new mean shift fast algorithm, an adaptive watermark algorithm in contourlet domain based on mean shift texture features clustering is proposed in this paper. Through the texture recognition method based on gray co-occurrence matrix, watermark is embedded into the coefficients in contourlet domain, which makes the capability of the watermark more covert, anti-noise attack and robust. During the clustering, three texture features including energy, entropy and contrast were selected for mean shift fast clustering algorithm. Strong regional textures of host images are extracted directly, accurately and efficiently, and the embedding intensity can be gain automatically then. The experiment shows that this algorithm has the strong robustness to Gauss low pass filter, Wiener filter, median filtering, Salt and pepper noise, Gaussian noise, JPEG compression, shear attack etc. It is a blind detection, and adapt to various of images.

There are inherent speckle noises in the medical ultrasound image. Speckle is multiplicative noise and the main reason to cause ultrasonic image degenerate. A new method of speckle reduction based on wavelet transform and soft threshold is presented and discussed. We compare our proposed technique with two existing speckle reduction methods. The results show that our method performances well not only in reducing the speckle noise but also in preserving the resolvable details. Finally, the prospect of the nonlinear filtering method is discussed.

Statistical characteristics of stationary processes in the cyclic autocorrelation based PSK symbol rate estimation are detailedly studied in this letter. The analytic expressions of the means and the variances of those undesired items which are produced by stationary processes are deduced, and the simplified forms for the Gaussian white noise case are also provided. Simulation results are presented to confirm the theoretical conclusions.

The detection and parameter estimation of spread spectrum signals under non-cooperative conditions are key problems to which more attentions are paid in the field of signal processing. A cyclic-cumulant based DS-SS (direct sequence spread spectrum) signal detection scheme is developed in this paper by extending the asymptotically optimal chi-squared test for the presence of cyclostationarity to DS-SS signal detection. The explicit formula of the second-order cyclic-cumulant is derived, and the influence of time interval on the cycle-frequency estimation from the cyclic-cumulant is analyzed in detail. On the base of this, a new cyclic-cumulant based DS-SS signal parameter estimation method is proposed, which can perform blind detection and blind parameter estimation without a priori knowledge of the signal such as pseudorandom sequence and frequencies. Moreover, the method also avoids multi-dimensional searching in parameter estimation. Simulation results have proved the validity of the method.

The mean-shift algorithm is an efficient technique for tracking 2D target through an image. The classic mean-shift algorithm has a fixed spatial template. Therefore, the tracking result is affected by changes of the tracked object. This paper presents an adaptive template modification strategy which overcomes the influence. The proposed method adjusts adaptively template orientation during the mean shift procedure according to the analysis of the moments of the tracked target. Therefore, the tracker has the ability to handle target rotation. Experimental results demonstrate its effectiveness

Particle filtering shows great promise in addressing a wide variety of non-linear and /or non-Gaussian problem. A crucial issue in particle filtering is the selection of the importance proposal distribution. In this paper, the iterated extended Kalman filter (IEKF) is used to generate the proposal distribution. The proposal distribution integrates the latest measurements into system state transition density, so it can match the posteriori density well. The simulation results show that the new particle filter superiors to the standard particle filter and the other filters such as the unscented particle filter (UPF), the extended Kalman particle filter (PF-EKF), the EKF.

Maneuver detection is a key issue in multiple-model target tracking applications. In the Kalman filter, the conventional maneuver detection algorithm fails when the maneuver is almost buried in noise. The multiresolution approach behaves well for maneuver detection, but the time delay limits its application. We suggest a new algorithm for maneuver detection based on the third-order cumulant of the measurements. As cumulants are blind to Gaussian noise, the maneuver is easy to detect in the cumulant domain. Using sliding window processing, the maneuver detection algorithm is real-time without any time delay. Simulation results demonstrate the effectiveness of the algorithm over the conventional one, especially at low SNR.

With the good performance of higher-order spectra (HOS) techniques for non-Gaussian signal processing and Gaussian noise suppression capability, an AR model parametric bispectrum estimation is presented for conventional radar target return analysis. A reasonable selection of target bispectrum features is made with a formation of target feature vector for target classification. The classification experimental results-on-actual target returns are given, with satisfactory results.

As targets flying in groups may spread over several range cells and their Doppler frequencies are time-varying, a scheme of target number detection in groups using a conventional low-resolution radar is proposed based on a time-range-Doppler image cube which is constructed by substituting a time-frequency transform for the Fourier transform in the conventional range-Doppler imaging. The image cube provides not only a clear range-Doppler image with superior resolution at each time instant but also the time-varying history of the images, hence a good solution to the problem of range spread and image blunting. The experimental results of actual target echoes demonstrate the validity of the scheme with successful detection of four targets flying in groups.

It is well known that fuzzy c-means (FCM) algorithm is one of the most popular methods of cluster analysis. However, the traditional FCM algorithm does not work for the interval-valued data and fuzzy-valued data. To this end, a feature mapping method is proposed to preprocess these special type data, and then the traditional FCM algorithm can also be employed to analyze the interval-valued and fuzzy-valued data. Therefore, a novel FCM clustering algorithm is formed for interval-valued data and fuzzy-valued data. The experimental result demonstrates its effectiveness.

As the targets flying in groups may spread over several range cells and their Doppler frequencies are time-varying, a scheme of target number detection in groups on a conventional low-resolution radar is proposed based on a time-range-Doppler image cube which is constructed by substituting a time-frequency transform for the Fourier transform in the conventional range-Doppler imaging. The image cube provides not only a clear range-Doppler image with superior resolution at each time instant but also the time-varying history of the images, hence a good solution to the problem of range spread and image blurring. The experimental results of actual target echoes demonstrate the validity of the scheme with successful detection of four targets flying in groups.

Due to the drawback that a conventional radar can not directly resolve the targets flying in a group in both range and azimuth, a method of target number detection is proposed by using the fractional Fourier transform based on the fact that the targets may be resolved in the Doppler frequency domain. The combination of coarse and fine search steps adopted effectively resolves the contradiction between the computation time and the search accuracy. In addition, being a 1-D linear transform and without the influence of the cross-term as in other time-frequency transforms, the method performs well under either higher SNR or lower SNR. The experimental results of actual target echoes demonstrate the performance of the method.