多任务状态下的脑纹识别研究
发布时间:2018-04-27 17:38
本文选题:脑电信号 + 脑纹识别 ; 参考:《杭州电子科技大学》2017年硕士论文
【摘要】:在个人信息安全愈加重要的当今社会,如何安全有效地进行身份识别已经成为一个重要话题。基于脑电信号的身份识别(脑纹识别)因此受到了越来越多的关注。区别于传统身份识别特征所存在的各种各样的缺陷,脑纹具有高隐蔽性、不可窃取性、不可仿制性以及必须活体等方面的独特优势。目前基于脑电信号的身份识别研究大多要求执行某种特定的任务或者需要在固定环境中采集脑电信号进行分析。这些限制条件对脑纹识别的现实应用发展有很大的局限性。本文着重研究了多任务状态下的脑纹识别,对处于不同环境和执行不同任务所采集到的脑电信号,通过研究脑电信号的相位同步特征,利用脑功能网络和深度信念网络来研究多任务状态下的脑纹识别。本文基于多个不同任务状态的脑电数据集进行脑纹识别研究,主要做了以下三方面的工作。1)本文提出将相位同步作为脑纹识别的公共特征提取方法,区别于传统脑电信号特征提取针对一类任务使用一种特征提取方法,对于多类任务进行脑纹识别时均使用相位同步进行脑纹特征提取。同时,对脑电信号相位同步进行了研究分析并提出均值相锁值对相位同步进行测量。2)本文提出利用功能脑网络进行了二次脑纹特征提取。为了便于进行脑纹识别,利用均值相锁值构建脑功能网络,将网络属性:节点的度,全局有效性,聚类系数进行组合作为二次脑纹特征用于脑纹识别,并绘制脑地形图观察比较各项属性在脑纹识别时所表现出的类内的一致性和类间的差异性。同时研究发现,脑网络的全局有效性和聚类系数始终呈近似正相关趋势。3)本文运用深度信念网络进行多任务状态下的脑纹识别。对不同任务和不同环境下的脑电信号利用公共特征提取方法相位同步进行特征提取,训练网络模型,并对网络的各项参数进行学习。研究发现,对不同数据集进行训练时,一般要求不同的网络层数和神经元个数,利用学习后的特征进行分类,能够进行有效识别。同时,在该部分,我们对一个含有两种不同任务的混合型数据集进行了脑纹识别研究,该混合数据集中的被试执行两种不同类型的任务。结果显示,该方法针对不同任务的混合数据集也能进行有效识别,对32名被试的脑纹识别准确率超过96%。与以往基于脑电信号幅值研究的不同之处在于本文从信号的相位角度对基于脑纹的身份识别进行了研究分析。本文创新性地将相位同步作为多任务状态下脑纹识别的公共特征提取方法,并在五个数据集上进行了脑纹识别实验,对于9类,12类,14类,20类和32类数据集进行识别均取得了比较好的脑纹识别效果,准确率分别为99%,98%,99%,98%和96%。以上结果说明了相位同步作为多任务脑电信号的公共特征提取方法是有效的。本文工作在一定程度上突破了脑纹识别中任务与环境的局限,使得基于脑纹的识别方法更具有泛化性,这对于如何改进现有脑电身份识别方法大多限定于特定的任务或环境的局限性有一定的参考价值。
[Abstract]:In today's society, the security of personal information is becoming more and more important, how to identify the identity security and effectively has become an important topic. The identification of EEG based identification (brain pattern recognition) has attracted more and more attention. At present, most of the research on identification based on EEG requires the implementation of certain specific tasks or the need to collect EEG signals in a fixed environment. These limitations have great limitations on the development of the practical application of brain pattern recognition. This paper focuses on the study of brain pattern recognition in multi task state, the EEG signals collected in different environments and different tasks. By studying the phase synchronization characteristics of EEG signals, the brain pattern recognition under multi task state is studied by using brain function network and depth belief network. This paper is based on the number of EEG numbers in different task states. According to the research of brain pattern recognition, the main work has been done in the following three aspects. In this paper, the phase synchronization is used as the common feature extraction method of the brain pattern recognition, which is different from the traditional EEG feature extraction. A feature extraction method is used for a class of tasks, and the phase synchronization is used for all kinds of tasks in the brain pattern recognition. At the same time, the phase synchronization of the EEG signal is studied and analyzed and the mean phase lock value is proposed to measure phase synchronization.2. This paper proposes two brain pattern features extraction using functional brain network. In order to facilitate the brain pattern recognition, the mean phase lock value is used to construct the brain function network, and the network attribute: node degree, The global validity and clustering coefficient are used as the two pattern of brain pattern to identify the brain pattern, and draw a brain topographic map to observe the conformance and the difference between classes in the brain pattern recognition. At the same time, it is found that the global effectiveness and clustering coefficient of the brain network always have an approximate positive correlation trend.3). In this paper, the depth belief network is used to identify the brain pattern under multi task state. The EEG signals in different tasks and different environments are extracted by using the phase synchronization of the common feature extraction method. The network model is trained and the parameters of the network are studied. It is found that the training of different data sets is generally not required. The number of the same network layer and the number of neurons, using the characteristics of the learning to be classified, can be effectively identified. At the same time, in this part, we have carried out a brain pattern recognition study on a mixed data set containing two different tasks. The subjects of the mixed data set perform two different types of tasks. The results show that the method is directed against this method. The mixed data set of different tasks can also be effectively identified. The difference between the accuracy rate of the 32 subjects' brain pattern recognition is more than 96%. and the previous research on the amplitude of EEG signal based on the phase angle of the signal. This paper analyzes the identification of the brain based identification from the phase angle of the signal. This paper innovatively uses phase synchronization as a multi task state. The common feature extraction method of the lower brain pattern recognition, and the brain pattern recognition experiment on five data sets. The recognition results for the 9, 12, 14, 20 and 32 types of data sets are better. The accuracy rate is 99%, 98%, 99%, 98% and 96%., respectively, indicating the phase synchronization as the multi task EEG signal. The common feature extraction method is effective. This paper breaks through the limitations of the task and environment in the brain pattern recognition to a certain extent, making the recognition method based on the brain pattern more generalization, which is of certain reference value for how to improve the existing EEG identification methods to limit the limitations of specific tasks or environment.
【学位授予单位】:杭州电子科技大学
【学位级别】:硕士
【学位授予年份】:2017
【分类号】:TN911.7;TP391.41
【参考文献】
相关期刊论文 前5条
1 张宁;臧亚丽;田捷;;生物特征与密码技术的融合——一种新的安全身份认证方案[J];密码学报;2015年02期
2 刘晶;白艳茹;许敏鹏;殷涛;何峰;周鹏;綦宏志;明东;;基于Farwell范式诱发ERP的身份识别研究[J];电子测量与仪器学报;2015年02期
3 谢松云;张振中;杨金孝;张坤;;脑电信号的若干处理方法研究与评价[J];计算机仿真;2007年02期
4 崔建国,王旭,訾学博,张大千;脑电信号的最新研究方法[J];沈阳航空工业学院学报;2004年02期
5 孙冬梅,裘正定;生物特征识别技术综述[J];电子学报;2001年S1期
相关硕士学位论文 前2条
1 李鹏宇;基于脑电信号的身份认证[D];北京邮电大学;2015年
2 闫彤;基于脑电的静息态功能连接分析[D];北京工业大学;2014年
,本文编号:1811673
本文链接:https://www.wllwen.com/kejilunwen/xinxigongchenglunwen/1811673.html