Minfen Shen

A method for estimating the instantaneous frequency of non-stationary heart sound signals

Practical signals such as speech, biomedical measurement and communications turn out to be extremely non-stationary and nonlinear time series. Traditional FFT-based power spectral analysis fails to deal with these transient signals. To provide more efficient way for investigating non-stationary and nonlinear signals with high time-frequency resolution and extract more information regarding the transient frequency features involved in the signals, a novel method based on the instantaneous frequency distribution is developed in this paper to provide the time-frequency distribution of the practical signals. The aim of this contribution is to explore the role that both empirical mode decomposition and Hilbert transform can be used to play in such practical signals. Both simulation and experimental results were presented and analyzed to demonstrate the power and effectiveness of the proposed new time-frequency distribution.

Application of ICA to the digital image watermarking

Digital watermark processing technology has developed very quickly during the recent few years, and been widely applied to protect the copyright of digital image, audio, video and multimedia production. In this paper, a method based on the independent component analysis (ICA) technique for detection and extraction of digital image watermark is proposed. With ICA techniques, it is ensured that a better-extracted watermark can be obtained. Several results of experiments indicate the proposed method is remarkably effective.

Analysis of time-varying synchronization of multi-channel EEG signals using wavelet coherence

This paper proposes a novel method based on the time-frequency coherent representation for quantifying synchronization of multi-channel signals with high resolution. The presented wavelet-coherent technique provides the information regarding both the degree of coherence and the relation of phase difference. The wavelet coherence enables to provide the synchronization and the direction of information flow between two-channel signals. In addition, real EEG recordings are collected based on the cognitive targets during sentences identification and the wavelet coherence is employed for the analysis of the multi-channel EEG signals. It is observed from both the magnitude spectra and phase of the wavelet coherence that there are obvious differences between two kinds of cognitive activities. Finally, some results are illustrated with both simulation and real EEG lime series to show the effectiveness of the method.

Analysis of Time-Varying Coherence of EEG during Face Recognition Based on Harmonic Transform

Distinct cortical activity during face recognition was reported by a number of studies. Classical coherence analysis reflects the synchronization between two random processes in certain frequency under the assumption that the signals are stationary. In EEG study, the coherence analysis is mainly used to analyze the coupling and drive-response relation about the brain activities in different regions. Classical coherence analysis can extract the rhythmic consistency of the EEG activities in different regains. However, classical coherence can not extract the transient synchrony characteristics of brain activities. In order to track the spatial-temporal characteristics of EEG activities during cognitive conception, the time-varying coherence analysis is proposed. The EEG of 10 participants was recorded during recognition of familiar and unfamiliar faces, experiment results show that there is obvious difference about the model of information communication between temporal region and other brain regions in alpha rhythm during cognitive processing

Novel Approach for Time-Varying Bispectral Analysis of Non-Stationary EEG Signals

A novel parametric method, based on the non-Gaussian AR model, is proposed for the partition of non-stationary EEG data into a finite set of third-order stationary segments. With the assumption of piecewise third-order stationarity of the signal, a series of parametric bispectral estimations of the non-stationary EEG data can be performed so as to describe the time-varying non-Gaussian nonlinear characteristics of the observed EEG signals. A practical method based on the fitness of third-order statistics of the signal by using the non-Gaussian AR model, together with an algorithm with CMI is presented. The experimental results with several simulations and clinical EEG signals have also been investigated and discussed. The results show successful performance of the proposed method in estimating the time-varying bispectral structures of the EEG signals

Analysis of time-varying synchronization of EEG during sentences identification

The study of the synchronization of EEG signals can help us to understand the underlying cognitive processes and detect the learning deficiencies since the oscillatory states in the EEG reveal the rhythmic synchronous activity in large networks of neurons. As the changes of the physiological states and the relative environment exist when cognitive and information processing take place in different brain regions at different times, the practical EEGs therefore turn out to be extremely non-stationary processes. To investigate how these distributed brain regions are linked together and the information is exchanged with time, this paper proposes a modern time-frequency coherent analysis method that employs an alternative way for quantifying synchronization with both temporal and spatial resolution. Wavelet coherent spectrum is defined such that the degree of synchronization and information flow between different brain regions can be described. Several real EEG data are analysed under the cognitive tasks of sentences identification in both English and Chinese. The time-varying synchronization between the brain regions involved in the processing of sentences exhibited that a common neural network is activated by both English and Chinese sentences. The results of the presented method are helpful for studying English and Chinese learning for Chinese students.

Analysis of Time-Varying EEG Based on Wavelet Packet Entropy

To investigate the time-varying characteristics of the multi-channels electroencephalogram (EEG) signals with 4 rhythms, a useful approach is developed to obtain the EEGs rhythms based on the multi-resolution decomposition of wavelet transformation. Four specified rhythms can be decomposed from EEG signal in terms of wavelet packet analysis. A novel method for time-varying brain electrical activity mapping (BEAM) is also proposed using the time-varying rhythm for visualizing the dynamic EEG topography to help studying the changes of brain activities for one rhythm. Further more, in order to detect the changes of the nonlinear features of the EEG signal, wavelet packet entropy is proposed for this purpose. Both relative wavelet packet energy and wavelet packet entropy are regarded as the quantitative parameter for computing the complexity of the EEG rhythm. Some simulations and experiments using real EEG signals are carried out to show the effectiveness of the presented procedure for clinical use.

Analysis of Time-Varying VEP Signal Using Rational RBF Network

This contribution studies the problem of denoising single-trial visual evoked potentials (VEP) signal. The main objective for VEP detection is to extract the change of the response and the corresponding latency without losing the individual properties of each trial of the signals, which is meaningful to clinic and practical application. Based on the radial basis function neural network (RBFNN), we proposed normalized RBFNN to obtain preferable results against other nonlinear methods: adaptive noise canceling (ANC) with RBFNN prefilter and RBFNN alone. These three approaches are compared with MSE and the ability of tracking peaks. The experimental results provide convergent evidence for the view that NRBFNN significantly attenuates noise and can successfully identify variance between trials

Effective Approach for Detecting the Single Trial ERP Using Rational Gaussian Neural Network

A novel approach is proposed to deal with the problem of detecting the single trial ERP using a modified RBF neural network, rational Gaussian network. The Gaussian RBF is normalized to obtain optimal behavior of noise suppression even at low SNR. The performance of the proposed scheme is also evaluated with both MSE and the tracking ability. Several experimental results with real ERP signals provide the convergent evidence to show that the presented method can significantly enhance the SNR and successfully track the variation of the signal such as the specified ERP in the applications of biomedical signal processing

Nonextensive entropy analysis of non-stationary ERP signals

The application of a measurement of entropy defined from the nonextensive entropy, the Tsallis-like time-dependent entropy (TDE), is proposed to investigate the event-related potential (ERPs). The TDE carries information about the degree of order or disorder associated with a multi-frequency signal response. The statistical characteristics of the TDE for different signal distributions are studied. TDE was estimated for the ERP signals recorded from several healthy subjects with a specified cognitive task. A significant decrease of entropy was correlated with the responses to target stimulus (P300). The experimental results indicate that the TDE can be employed as a quantitative measurement for monitoring the ERPs activities and other signal processing.