Research On The Analysis Of Sleep EEG Signal Based On Wavelet Transform

Sleep activity is one of most important physiological activities, which affects people’s health and life directly. However, the pace of life of modern is becoming more and more quickly, leading to a high incidence of many kinds of sleep diseases. The significance of sleep research is highlighting gradually. Since the electroencephalogram(EEG) is the biological signal which can reflects the brain activity essentially, sleep researches based on EEG signals are attracting more and more attention.Sleep EEG signal is in essence a kind of typical nonlinear and non-stationary random signal, so lots of typical signal analysis methods are inappropriate to analyze the EEG signal. The wavelet transform has an ability to well capture the local characteristic of the raw signal with the high resolution in time domain and frequency domain, therefore, it became the powerful tool to analyze the sleep EEG signal.In this paper, we performed a time-frequency analysis of sleep EEG signals based on the wavelet analysis, and employed the wavelet transform to achieve an automatic sleep staging approach and a sleep spindles detection algorithm. According to the definitions, sleep staging is to divide the sleep activity into several different stages on the basis of the characteristics of different sleep stages. It can be used for analyzing the structure of sleep and evaluating the quality of sleep. Sleep spindles detection is to find out the randomly appear transient spindles from a long sleep period, providing theoretical guidance for the prevention of some sleep diseases in clinical. To solve the problem of sleep staging and spindles detection, foreign researchers have achieved some progress, but some improvements on the accuracy and the complexity of algorithm are desired.Based on the Morlet wavelet transform, a sleep staging algorithm was proposed in this paper. In this approach, the discrimination of the features on each frequency was evaluated for selecting the features with great differentiation, and then a sparse multinomial logistic regression method was employed to achieve the automatic sleep staging. To verify the feasibility of the algorithm, a public available dataset and the EEG recordings from the members of our laboratory were used in this paper.Based on the Mexh wavelet transform, a spindles detection algorithm was developed. The Mexican Hat wavelet features of EEG signal was extracted at first. Subsequently, a local evaluation of each sample point with its adjacent sample points was performed, converting the original signal to a binary signal based on the evaluation results. Finally, a sliding window was applied to estimate the probability of each sample point on the binary signals. Additionally, a more scientific evaluation method was suggested to compared the performance of the proposed algorithm and the other four algorithms on a public available dataset.In a conclusion, the result showed that the performances of the sleep staging and sleep spindle detection proposed in this paper were well and the work of this study have a significant research value.