ComparisonofclassificationmethodsonEEGsignalsbasedonwaveletpacketdecomposition资源-CSDN文库

198 浏览量 2021-02-06 18:03:01 上传评论收藏 624KB PDF 举报

### 基于小波包分解的脑电图信号分类方法比较 #### 摘要与背景本文探讨了基于小波包分解的小波技术在脑电图（EEG）信号处理中的应用，并通过对比几种不同的分类算法评估了其性能。脑电信号作为一种重要的生理信号，在神经系统疾病的诊断和心理生理过程的研究中扮演着至关重要的角色。尽管近年来磁共振成像（MEG）、功能性磁共振成像（fMRI）和正电子发射断层扫描（PET）等其他脑功能分析方法得到了发展，但由于其成本较低且对患者来说较为便捷，EEG信号仍然是监测大脑活动的有效工具之一。 #### 主要内容本文的主要内容包括以下几个方面： 1. **特征提取方法**：本文采用了近似熵（Approximate Entropy, ApEn）和样本熵（Sample Entropy, SampEn）作为特征提取方法。这两种熵是衡量信号复杂性的常用指标，可以有效地提取出脑电信号中的有用信息。 2. **分类器**：支持向量机（Support Vector Machine, SVM）和极限学习机（Extreme Learning Machine, ELM）被选作分类器。SVM是一种基于统计学习理论的机器学习方法，特别适用于高维空间中的分类问题；ELM是一种单层前馈神经网络，训练速度快，泛化性能好。 3. **实验设计**：文中分别对癫痫脑电数据和五种不同心理任务的数据进行了实验。通过对这些数据集的分析，作者旨在验证所提出方法的有效性和适用性。 4. **结果分析**：实验结果显示，样本熵结合极限学习机的方法表现出了较高的分类准确率和较短的训练时间，表明该组合策略具有较好的性能。 #### 关键词解析 - **EEG**：脑电图信号，记录了大脑皮层神经元活动产生的电信号，常用于临床诊断和科学研究。 - **小波包分解**：一种信号处理技术，能够提供更精细的时频分辨率，有助于提高特征提取的效果。 - **分类**：将EEG信号根据其特性划分为不同的类别，如健康与疾病状态、不同心理任务等。 - **近似熵**：用于量化信号复杂性的度量标准，值越大表示信号越随机或复杂。 - **极限学习机**：一种快速的单层前馈神经网络模型，能够高效地解决分类和回归问题。 #### 讨论与结论本文提出的基于小波包分解的特征提取方法与样本熵及极限学习机相结合的分类策略，在处理EEG信号时表现出良好的性能。特别是在处理癫痫脑电数据和不同心理任务的数据时，该方法不仅分类准确率较高，而且训练时间较短，为构建有效的脑机接口（BCI）系统和改善临床诊断工具提供了有力的支持。此外，这种方法还可以为未来的研究提供一个有价值的参考框架，尤其是在需要快速准确处理大量EEG数据的应用场景中。

资源推荐

资源详情

资源评论

ORIGINAL ARTICLE

Comparison of classiﬁcation methods on EEG signals based

on wavelet packet decomposition

Yong Zhang

•

Yuting Zhang

•

Jianying Wang

•

Xiaowei Zheng

Received: 17 September 2014 / Accepted: 9 December 2014 / Published online: 24 December 2014

 The Natural Computing Applications Forum 2014

Abstract EEG signals play an important role in both the

diagnosis of neurological diseases and understanding the

psychophysiological processes. Classiﬁcation of EEG sig-

nals includes feature extraction and feature classiﬁcation.

This paper uses approximate entropy and sample entropy

based on wavelet package decomposition as the feature

exaction methods and employs support vector machine and

extreme learning machine as the classiﬁers. Experiments

are performed in epileptic EEG data and ﬁve mental tasks,

respectively. Experimental results show that the combina-

tion strategy of sample entropy and extreme learning

machine has shown great performance, which obtains good

classiﬁcation accuracy and low training time.

Keywords EEG  Wavelet packet  Classiﬁcation 

Approximate entropy  Extreme learning machine

1 Introduction

The electroencephalogram (EEG) signals play an important

role in both the diagnosis of neurological diseases and under-

standing the psychophysiological processes. Although several

methods for the brain function analysis such as

magnetoencephalography (MEG), functional magnetic reso-

nance imaging (fMRI), and positron emission tomography

(PET) have been introduced, the EEG signal is still a valuable

tool for monitoring the brain activity due to its relatively low

cost and being convenient for the patient [1]. Effectively

classifying EEG is the foundation for building usable Brain–

Computer Interfaces (BCI) and improving the performance of

EEG analysis software used in clinical and research settings.

Feature extraction and classiﬁcation of EEG signals is a

challenge for engineers and scientists. In recent years, there

has been an increasing interest in the application of

machine learning methods for EEG signal. Many methods

have been proposed to classify and analyze EEG signal.

Among these methods, support vector machine (SVM) and

artiﬁcial neural network [2, 3] are two widely used clas-

siﬁcation techniques. Most of EEG signal classiﬁcations

are built by time–frequenc y domain-based feature extrac-

tion followed by a variety of classiﬁcation models. So, the

classiﬁcation performance of these methods depends on the

feature extraction of the EEG signal.

Wavelet has played an important role in biomedical

signal proce ssing for its ability to capture localized spatial-

frequency information of EEG signals [4]. In the feature

extraction phase, this paper uses approximate entropy

(ApEn) and sample entropy (SampEn) based on wavelet

packet decomposition (WPD). Obtained feature values are

employed as the inputs of a classiﬁer, such as extreme

learning machine (ELM) and SVM. Experiments are per-

formed in four different combinations of feature extrac-

tions and classiﬁers. Experimental results in epileptic EEG

data and ﬁve mental tasks indicate that the combination of

sample entropy and ELM has shown great performance.

The remaining of the paper is organized as follows. In

Sect. 2, we brieﬂy presented the literature survey that has

been performed. In Sect. 3, we described about the

Y. Zhang  Y. Zhang  X. Zheng (&)

School of Computer and Information Technology, Liaoning

Normal University, No. 1, Liushu South Street, Ganjingzi

District, Dalian, Liaoning Province 116081, China

e-mail: zhengxw_ln@163.com

Y. Zhang

e-mail: zhyong@lnnu.edu.cn

J. Wang

School of Psychology, Liaoning Normal University,

Dalian 116029, China

123

Neural Comput & Applic (2015) 26:1217–1225

DOI 10.1007/s00521-014-1786-7

materials and methods. In Sect. 4, we compared the clas-

siﬁcation results of four different combinations in epileptic

EEG data and ﬁve mental tasks, respectively. And Sect. 5

concluded the paper.

2 Related wor k

In the feature extraction for EEG signals, Burke et al. [5]

extend the usual autoregressive (AR) models for EEG

feature extraction. The extensi on model is an AR with

exogenous input (ARX) model for combined ﬁltering and

feature extraction. Polat and Gu

nes [6] use fast Fourier

transform as feature extraction method. Wavelet transform

is also widely used in extracting EEG features [7, 8], in

which EEG signals are decomposed by wavelet transform

to calculate approximation and detail coefﬁcients. ApEn is

a statistical parameter that measures the predictability of

the current amplitude values of a physiological signal

based on its previous amplitude values. ApEn is used for

the ﬁrst time in the proposed system for the detection of

epilepsy using neural networks [2]. A novel feature

extraction method based on multi-wavelet transform and

ApEn is proposed by Guo et al. [9]. The proposed method

uses approximate entropy features derived from multi-

wavelet transform and combines with an artiﬁcial neural

network to classify the EEG signals. Song et al. [10, 11]

use sample entropy (SampEn) as a feature extraction

method for detecting epileptic seizures. In Ref. [10],

SampEn is selected as a feature extraction method to

classify the task of three different kinds of EEG signals and

detecting epileptic seizures. Based on Ref. [10], the authors

propose an optimized sample entropy algorithm to identify

the EEG signals in [11].

Artiﬁcial neural networks (ANNs) have been widely

applied to classify EEG signals [2, 9]. Kumar et al. [7] use

discrete wavelet transform (DWT)-based ApEn and ANN to

detect epileptic seizures. SVM is also one of the classiﬁcation

methods for EEG signals [3, 4]. Zhang et al. [12] present an

improved method to calculate phase locking value (PIV)

based on Hilbert–Huang transform, and the PLVs are used as

features for a least squares support vector machine (LS-SVM)

to recognize normal and hypoxia EEG. Bajaj et al. [13]

present a new method for classiﬁcation of EEG signals using

empirical mode decomposition (EMD) method. The pro-

posed method for classiﬁcation of EEG signals is based on the

bandwidth features and employs LS-SVM for classifying

seizure and non-seizure EEG signals. Wu et al. [14] propose a

multiscale cross-approximate entropy method to quantify the

complex ﬂuctuation between R–R intervals series and pulse

transit time series. Ahangi et al. [15] use a multiple classiﬁer

system for classiﬁcation of EEG signals. The proposed

method uses DWT decomposition, and a variety of classiﬁer

combination methods along with genetic algorithm feature

selection is evaluated.

3 Materials and methods

In this section, we give a description of proposed method

for EEG signal classiﬁcation problem, present feature

extraction based on the WPD, and give a brieﬂy review of

two classiﬁers, ELM and SVM.

Figure 1 shows the block diagram of the propos ed

method based on WPD. In our method, we ﬁrst use the

WPD to decompose EEG data into sub-band signals. Then,

ApEn values or SampEn values are calculated by approx-

imation and detail coefﬁcients. Obtained feature vectors

are used as the inputs of classiﬁers, such as SVM and ELM.

Last, we evaluate the classiﬁcation accuracy.

3.1 Feature extraction based on wavelet packet

decomposition

Feature extraction of EEG signals includes two phases. In

the ﬁrst phase, EEG signals are decomposed by the WPD to

calculate approximation and detail coefﬁcients. In the

second phase, approximate entropy values or sample

entropy values of the approximation and detail coefﬁcients

are calculated, which form feature vectors. These feature

vectors are used as the inputs of a classiﬁer.

In this subsection, we ﬁrst simply analyze the charac-

teristic of the WPD, and then, give a brieﬂy review of

approximate entropy and sample entropy.

3.1.1 Wavelet packet decomposition

Feature extraction is very important to EEG signals ana-

lysis and processing. WPD is a wavelet transform where

Initial EEG data

Pre-processing

Feature extraction

WPD+ApEn/SampEn

Training data

Constructing

ELM/SVM classifier

Testing and Evaluation

Fig. 1 Block diagram of the

proposed method

1218 Neural Comput & Applic (2015) 26:1217–1225

123

剩余8页未读，继续阅读

评论收藏

内容反馈

weixin_38631729

粉丝: 8
资源: 905

Comparison of classification methods on EEG signals based on wav...

最新资源

Comparison of classification methods on EEG signals based on wav...

Comparison of simultaneous signals obtained from a dual-field-of-view lidar and its application to noise reduction based on empirical mode decomposition

Fuzzy support vector machine for classification of EEG signals using wavelet-based features

wavelet packet

Mode decomposition of signals

一种新的基于小波包分解的EEG特征抽取与识别方法研究

EEG Signal processing

EEG.rar_EEG_EEG paper

An Experimental Comparison of Min-Cut/Max-Flow Algorithms

Comparison of chip for developing low noise EEG amplifier.pdf

主元素分析方法A comparison of three methods for principal component

image-classification-different-methods

室内定位文献 Classification and Comparison of Range-Based

Digital Change Detection by Post-Classification Comparison of RS Data in Land Use of Guangzhou

A Survey and Comparison of Time-Delay Estimation Methods in Linear Systems

Comprehensive Comparison of Region-Based Image Similarity Models

A comparison of single-cell trajectory inference methods.pdf

Comparison of Various Learning Rate Scheduling Techniques on Con

Performance comparison of two on-demand routing protocols for ad hoc networks.pdf

Comparison of sEMG-Based Feature Extraction and Hand Motion Classification Methods

A comparison of 3D file formats.pdf

多目标优化（Comparison of Multiobjective Evolutionary Algorithms: Empirical Results）

comparison of db2 vs oracle

A Comparison of Dictionary Implementations

The Algorithms currently implemented for classification of EMG

Performance-Comparison-of-VSC-Based-Shunt-and-SER_CONVERTER VSC_

PCIE Transmitter PLL Testing A Comparison of Methods. Primer.pdf

comparison of nonlinear filter

最新资源