基于EEG的信号处理和睡眠分期研究

发布时间：2018-02-16 04:55

本文关键词： 脑电信号小波包非线性动力学粒子群算法最小二乘支持向量机　出处：《中国矿业大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着现代社会的快速发展,人们的生活节奏在不断的加快,各种生活和心理上的压力使得现代人的身心健康状态备受威胁。由压力造成的影响主要表现在现代人的睡眠状态上,越来越多的人饱受失眠或者睡眠质量低下的痛苦,而由此衍生出的生理或心理疾病也是数不胜数。因此对于睡眠脑电的研究越来越受到人们的关注。本文对睡眠脑电信号的基本特征和基本波形进行了介绍,由于不同的睡眠状态,其对应的脑电信号均呈现不同的特点,为接下来的睡眠分期提供了理论上的依据。然后根据麻省理工学院提供的研究心律失常的数据库来获取本课题中所需要的睡眠脑电测试数据。本文主要从信号预处理、睡眠特征提取和特征参数分类三个方面实现算法设计。首先,由于脑电信号中存在着大量的背景噪声,因此需要选取合适的去噪方法。通过实验对比了小波包去噪结果,逐步实现了小波基函数、小波分解层数和阈值函数的选取。其次,运用非线性动力学对滤波后的睡眠脑电信号进行特征分析,验证了相关维数、复杂度、Lyapunov指数和近似熵等特征参数均可以用来表征睡眠状态。Lyapunov指数证明了大脑的混沌特性,相关维数、复杂度和近似熵则在不同的睡眠状态中展现出了规律性的变化。最后,在对不同的分类器的原理和算法进行了分析对比的基础上提出了使用粒子群算法对参数进行优化。通过对比人工神经网络、最小二乘向量机和经过粒子群算法优化后的最小二乘向量机的睡眠分期结果,验证了所提出的优化算法的有效性。实验结果表明,通过本文设计的分期系统,在对7*400组特征值进行的识别实验中,清醒期识别正确的有338组,错误的有31组,准确率为91.60%;浅睡期识别正确的有411组,错误的有47组,准确率为89.74%;深睡期识别正确的有242组,错误的有28组,准确率为87.68%;快速眼动期识别正确的有269组,错误的有28组,准确率为90.57%。最终得到的平均准确率可以达到90.00%,由此证明了本课题中设计的自动分期系统是可行的。
[Abstract]:With the rapid development of modern society, people's rhythm of life is speeding up, and various kinds of life and psychological stress make the physical and mental health of modern people under threat. The influence of stress is mainly reflected in the sleep state of modern people. More and more people suffer from insomnia or poor sleep quality. Therefore, more and more attention has been paid to the study of sleep EEG. The basic characteristics and waveforms of sleep EEG are introduced in this paper. Because of different sleep states, their corresponding EEG signals show different characteristics. It provides a theoretical basis for the following sleep stages. Then according to the database provided by the Massachusetts Institute of Technology to study arrhythmia to obtain the sleep EEG test data needed in this study. The algorithm is designed in three aspects: sleep feature extraction and feature parameter classification. Firstly, due to the large amount of background noise in EEG signal, it is necessary to select the appropriate denoising method. The results of wavelet packet denoising are compared by experiments. The selection of wavelet basis function, wavelet decomposition layer number and threshold function is realized step by step. Secondly, nonlinear dynamics is used to analyze the characteristics of the filtered sleep EEG signal, and the correlation dimension is verified. The characteristic parameters such as complexity Lyapunov exponent and approximate entropy can be used to characterize the sleep state. Lyapunov exponent proves the chaotic characteristics of the brain. The correlation dimension, complexity and approximate entropy show regular changes in different sleep states. On the basis of analyzing and comparing the principles and algorithms of different classifiers, the particle swarm optimization (PSO) algorithm is proposed to optimize the parameters. The sleep staging results of the least square vector machine and the least squares vector machine optimized by particle swarm optimization verify the effectiveness of the proposed optimization algorithm. In the experiment of recognizing the characteristic values of 7C400 groups, 338 groups were correctly identified during waking period, 31 groups were wrong, the accuracy rate was 91.60%, 411 groups were correct in shallow sleep period, 47 groups were wrong, and the accuracy rate was 89.7474%, while 242 groups were correct in deep sleep. There are 28 groups of errors with the accuracy of 87.68, 269 groups with correct recognition of rapid eye movement period, and 28 groups with accuracy of 90.57. The average accuracy can reach 90.00g, which proves that the automatic staging system designed in this paper is feasible.
【学位授予单位】：中国矿业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R740;TN911.7

【相似文献】