脑电信号的非平衡数据处理及特征提取方法的研究

发布时间：2018-01-30 17:24

本文关键词： 癫痫脑电图非平衡数据集特征提取中心核半径超限学习机(ELM) 支撑向量机(SVM)　出处：《西北大学》2017年硕士论文　论文类型：学位论文

【摘要】：癫痫是一种常见的神经系统紊乱疾病,以反复、突发的癫痫性发作为主要特点,严重危害着人类的日常生活乃至生命。传统的癫痫检测方法是医生根据自身的经验结合病人的脑电图进行视觉诊断完成的,但由于监测过程中会产生海量的非平衡数据,从而会造成传统方法十分耗时、主观性强等缺点。因此,利用计算机技术进行癫痫性发作的自动检测成为近几年来的研究热点。作为一种典型的非平衡数据,如何对获取的长时程脑电数据进行恰当的平衡预处理,以及如何设计合适的脑电特征提取方法是完成这一过程的两个关键环节。基于此,本文分别从脑电信号的平衡处理以及特征提取两个角度出发,设计了一种恰当有效的用于癫痫性发作自动检测的非平衡数据处理方法,同时提出了一种新的癫痫脑电特征提取方法。主要工作内容如下:第一章系统地阐述了癫痫性发作自动检测的研究背景及现状;第二章简要地阐述癫痫及脑电图,并对癫痫性发作自动检测中常用的数据库以及分类器做了简单介绍。第三章基于K-means、Silhouette指标以及M-近邻下采样提出了一种新的数据平衡方法K-M-S,并通过数值实验对该方法的有效性进行了验证。第四章基于非线性动力系统提出一种新的脑电特征提取方法中心核半径(Kernel Radius-based Feature,KRF),并通过数值试验验证了该特征的有效性。
[Abstract]:Epilepsy is a common nervous system disorder disease, with repeated, sudden epileptic hair as the main characteristics. The traditional epilepsy detection method is based on the doctor's own experience combined with the patient's electroencephalogram (EEG) for visual diagnosis. However, due to the monitoring process will produce a large amount of unbalanced data, which will lead to the traditional method is very time-consuming, strong subjectivity and other shortcomings. Automatic detection of epileptic seizures using computer technology has become a hot topic in recent years. As a typical non-equilibrium data, how to do the proper balanced preprocessing of the obtained long-term EEG data. And how to design appropriate EEG feature extraction method is the two key links to complete the process. Based on this, this paper starts from the two angles of EEG balanced processing and feature extraction. A suitable and effective non-equilibrium data processing method for the automatic detection of epileptic seizures was designed. At the same time, a new method of feature extraction of epileptic EEG is proposed. The main work is as follows: the first chapter systematically describes the background and present situation of the automatic detection of epileptic seizures; The second chapter briefly describes epilepsy and EEG, and makes a brief introduction to the database and classifier commonly used in the automatic detection of epileptic seizures. Chapter 3 is based on K-means. A new data balance method, K-M-S, is proposed for Silhouette index and M-nearest neighbor down-sampling. The effectiveness of the method is verified by numerical experiments. In Chapter 4th, a new method for extracting EEG features is proposed based on nonlinear dynamic systems. Kernel Radius-based Feature. The effectiveness of this feature is verified by numerical experiments.
【学位授予单位】：西北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R742.1;TN911.7

【相似文献】