基于双频SSVEP的颜色和朝向特征提取

发布时间：2018-11-27 14:40

【摘要】：对脑信号的模式分类一直是认知神经科学中的研究热点,不论是对功能磁共振信号和脑电图、脑磁图信号的学术研究方面,还是在实际应用中。通过模式分类可以找出不同信号之间难以用常规分析方法得到的信息。除了分类以外,建立线性模型也是越来越常用的一种方法,近年来有越来越多的研究根据已有的研究成果对人脑的视觉区域尤其是初级视皮层进行建模,并用于重建视觉刺激特征。本文的研究主要针对的是脑电信号,研究目的是对两个同时呈现的视觉刺激进行单独的信号特征提取和多元模式分析。本文的主要工作内容及成果有:1,设计了双刺激频率的稳态视觉诱发电位(SSVEP)的刺激呈现范式,使用两个刺激频率,且每种刺激频率由交替闪烁刺激频率叠加组成,两个刺激频率分别呈现两种视觉刺激,即有六种方向的光栅和有六种颜色的棋盘格。实验分为注意方向的条件和注意颜色的条件。2,对实验中出现的两种视觉刺激分别计算了正向编码模型,并在模型效果上得到了注意条件与非注意条件之间的显著主效应,并通过模型的预测结果调节正向编码模型的输入特征以得到最优的信号特征。3,使用多种方法来提取信号中与视觉刺激相关的特征,除了使用功率谱特征以外,提出了两种信号特征提取方法,一是通过注意对稳态视觉诱发电位的功率幅度调制机制,在信号的时频图中分别提取不同的信号特征,二是通过正向编码模型来提取最优的信噪比特征。4,使用三种信号特征,和三种不同的分类器,来进行多元模式分析,分类器包括栈式自编码神经网络、线性判别分析、以及支持向量机三种分类器。首先进行的是不同颜色之间的六分类和不同方向之间的六分类,然后进行了注意方向与注意颜色两种实验条件之间的二分类。线性判别分析的六分类结果表明,在注意条件下的分类准确率显著高于随机水平,并显著性高于非注意条件下的分类准确率。栈式自编码神经网络在注意与非注意条件下的准确率都高于随机水平,但是未发现注意的主效应。信噪比特征在该问题的二分类效果上不如功率谱特征,但在六分类的效果上更好。论文通过设计双视觉刺激的实验进行多元模式分析研究,提出了针对该问题的两种特征提取方法,比较了多种分类器和信号特征的效果,可以为复杂视觉特征的多元模式分析提供一定的参考。
[Abstract]:Pattern classification of brain signals has always been a hot topic in cognitive neuroscience, both in the academic research of functional magnetic resonance signals and EEG, and in practical applications. Through pattern classification, we can find out the information between different signals which is difficult to get by conventional analysis method. In addition to classification, linear modeling is also an increasingly common method. In recent years, more and more research has been done to model the visual regions of the human brain, especially the primary visual cortex. And used to reconstruct visual stimulation features. The purpose of this study is to extract the features of two simultaneous visual stimuli and analyze the multiple patterns. The main contents and achievements of this paper are as follows: 1. The stimulus paradigm of the steady-state visual evoked potential (SSVEP) with double stimulus frequency is designed. Two stimulus frequencies are used, and each stimulus frequency is composed of alternating flicker stimulation frequency. The two stimuli present two kinds of visual stimuli, that is, grating with six directions and checkerboard with six colors. The experiment was divided into two aspects: the condition of attention direction and the condition of attention color. 2. The positive coding model was calculated for the two kinds of visual stimuli in the experiment, and the significant main effect between the attention condition and the non-attention condition was obtained in the effect of the model. The input features of the forward coding model are adjusted by the prediction results of the model to obtain the optimal signal characteristics. 3. Various methods are used to extract the visual stimulus-related features, except for the power spectrum features. Two signal feature extraction methods are proposed. One is to extract different signal features in the time-frequency map of the signal by paying attention to the power amplitude modulation mechanism of the steady-state visual evoked potential. The second is to extract the optimal signal-to-noise ratio (SNR) features by the forward coding model. 4. Three signal features and three different classifiers are used for multivariate pattern analysis. The classifiers include stack self-coding neural networks and linear discriminant analysis. And support vector machine three classifiers. First there are six classifications between different colors and six classifications between different directions. Then two classifications between the two experimental conditions of attention direction and attention color are carried out. The six classification results of linear discriminant analysis show that the classification accuracy under attention condition is significantly higher than that at random level and is significantly higher than that under non-attention condition. The accuracy of stack self-coding neural network is higher than that of random level under both attention and unattention conditions, but the main effect of attention is not found. The signal-to-noise ratio (SNR) feature is not as good as the power spectrum feature in the two-classification of the problem, but it is better in the six-classification. In this paper, by designing the experiment of dual visual stimulation, we study the multi-element pattern analysis, propose two feature extraction methods to solve this problem, and compare the effects of various classifiers and signal features. It can provide some reference for multiple pattern analysis of complex visual features.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318;TN911.6

【相似文献】