Feature Extraction And Brain Network Construction Based On Athletes' EEG Signals

As a complex dynamic system,the brain's structure and function will constantly undergo plastic changes under the influence of the external environment and experience.Athletes and patients with motor dysfunction can improve professional skills and repair nerve damage through long-term sports training based on this feature.However,very few studies are available on the effects of different exercise methods on brain function.To this end,this article uses Electroencephalogram(EEG)technology to analyze the EEG characteristics of professional athletes in the area of martial arts and track and field from the perspective of time-frequency domain and brain network.This provides necessary reference for exploring the influence of different sports training methods on the function of the brain.The thesis have the following objectives:(1)Based on multi-modal data requirements,a joint experimental paradigm of resting state and visual-auditory intersection is designed,which improves the sensitivity of the induced P300 component during the experiment.Athletes from different professions and ordinary college students are chosen as study subjects for the data collection.A set of preprocessing procedures are designed to remove various artifacts mixed in the data acquisition process.An automatic artifact recognition and correction algorithm that combines independent component analysis and correlation analysis is proposed for the electrooculogram artifacts.It improves the quality of acquired signal.(2)The paper uses a linear analysis method for time-frequency analysis.The time domain characteristics of the P300 component of the task state data and the frequency domain energy characteristics of the resting state data are extracted.But traditional time-frequency analysis method faces problems and shortcomings stemming from wavelet basis selection,poor resolution in the high frequency part of wavelet transform.To solve such issues,this paper uses the signal-to-noise ratio and correlation coefficient as indicators to judge the denoising effect of different wavelets for determining "Bior3.9" as the optimal wavelet basis and extracts energy entropy as a feature to simultaneously characterize the rhythm and nonlinearity of the signal by wavelet packet decomposition.(3)Independent channel signal analysis methods lacks comprehensiveness and spatial distribution.To solve this,this paper uses correlation analysis and phase slope index based on complex network theory to construct functional brain networks and effective bra in networks for resting state and task state EEG data.In addition,a range threshold selection method based on the sparsity interval is used to perform multi-network combination analysis for the lack of unified standards for threshold selection.(4)The complexity of the brain network topology and unclear qualitative results are taken into account of.For this,further analysis is made on its statistical characteristics by using topologies such as node degree,clustering coefficient,characteristic path length/global efficiency,and verifies its small-world characteristics.The study found that martial arts athletes and track and field athletes show obvious differences from ordinary people in terms of time-frequency characteristics and brain network characteristics.In order to verify the effectiveness of feature extraction,four classification algorithms such as support vector machines are used to distinguish different athletes and ordinary people,with average accuracy rate of 90%.