Hilbert-Huang Transform And Its Application In EEG Signals Processing

As an important part of the subject,"Research on the coherent collaborative learning algorithm of audio-visual cross-modal"(No.60873139) supported by National Natural Science Foundation of China, this thesis aims at exploring the visual and auditory modal coherence in human cognition session. Over eighty percent of human perception on environment is achieved by vision. Hearing also plays an important role in the process. What are the relationships between visual mode and auditory mode if they exist? At present, many subjects, including psychology, artificial intelligence and life science, are focusing on this problem. In order to reveal solid evidence of the con-elation between these two modes, this thesis, mainly by means of HHT method, analyzes the relationship during visual evoked potentials, auditory evoked potential and audio-visual cross-modal evoked potentials. The correlation coefficients gained from experiments indicate the coherence of the two modes. In addition, a series of technical problems on the EEG signal processing with HHT has been addressed. Overall, the contributions of this thesis lie in the following four aspects.(1)EEG signal is typically nonlinear and non-stationary. Wavelet transform (WT) and Hilbert-Huang transform (HHT) are commonly used to process and analyze EEG signals. However, which method is more efficient is also a key point. This thesis gives a detailed comparison of these two methods from various aspects. Corresponding simulation results using real EEG data show that HHT method is more efficient than wavelet transform.(2)The end effect usually appears in the EEG signal processing with HHT method. To account for it, a new method based on weighted matching similar waveform is proposed in this thesis, after analyzing some existing improved methods and comparing relevant results.â‘ Based on the measurement and evaluation of those existing improved methods, a new general evaluation system is put forward, which accommodates both the decomposition result and decomposition efficiency. The decomposition results can be examined from two aspects, namely the correlation coefficient of effective components after decomposition and the original signal, and the energy difference before and after decomposition. Besides, the decomposition efficiency can be evaluated through computing time of those improved methods.â‘¡ccording to the analysis of those existing manners to restrain the end effects, this thesis proposes a new method of the weighted matching similar waveform. In order to make the extension signal conform to the trend of the original signal more, it utilizes a weighted wavelet of a number of similar wavelets to extend the endpoints. The experimental results of simulations, which use real EEG data, indicate that this new method can generate more reasonable results and restrain the end effects effectively.(3)In order to address the mode mixing problem caused by traditional EMD method in EEG signal processing, ensemble empirical mode decomposition (EEMD) method is proposed in this thesis, aiming at improving the situation. Furthermore, CUDA computation pattern is adopted to increase the decomposition efficiency.â‘ As a noise-assisted analysis technique, EEMD maintains the continuity of the signal in time domain by adding a series of white noise signals to the original signal. Due to the noise characteristics of zero mean, the noise will be offset to improve the mode mixing after several averaging. â‘¡The increased time complexity of the decomposition algorithm makes the EEMD method does not work well for real-time EEG signal analysis. The reason is that it causes addition of many white noise signals. To this end, an improved EEMD method is put forward in this thesis to enhance the computational efficiency, which is based on the calculation mode of GPU and CPU.(4)In order to investigate the correlation between auditory mode and visual mode, an experiment is designed in this thesis. After the EEG data are collected and analyzed, the signal correlation of the two modes are examined from the following two aspects.â‘ The correlations between visual modal signals, auditory modal signals and cross-modal signals are studied. Then, by comparing the two correlation coefficients, it concludes that vision is related to audition.â‘¡The main components of visual single-modal signals and auditory single-modal signals with I-EEMD method are obtained. The correlation coefficients of the main components and audio-visual consistent signals, and those of the main components and audio-visual inconsistent signals are derived respectively. Those correlation coefficients are then compared. The comparison results also show that vision is linked to audition.