基于特征融合神经网络的运动想象脑电分类算法

运动想象-脑机接口(MI-BCI)技术为运动障碍患者提供了一种新的与外界交流的能力。应用卷积神经网络(CNN)处理运动想象(MI)脑电分类问题时,多提取最后卷积层的特征,忽视了中间层大量可用信息,导致MI-BCI的分类性能较差。针对这一问题,提出模型内层融合(WMFF)和模型间层融合(CMFF)两种特征融合策略。WMFF策略提取CNN每一层特征进行融合;CMFF策略融合CNN和长短时记忆网络并提取每一层特征。本研究用BCI竞赛IV Datasets2a数据集对所提方法进行验证,WMFF和CMFF MI脑电信号四分类平均正确率分别达到76.19%和80.46%。结果表明,所提方法可有效提高MI脑电信号分类正确率,为MI脑电信号分类提供了新的思路。

基于1DCNN-XGBoost混合模型的癫痫脑电分类

在癫痫脑电自动分类研究中,针对传统方法需手动提取特征、特征提取不充分、适应性不强等问题,提出一种一维卷积神经网络结合XGBoost算法的单通道癫痫脑电分类模型1DCNN-XGBoost。该模型首先在原始脑电信号上训过练对一波维恩卷大积学模癫型痫,脑利电用数训据练集好进的行一分维类卷实积验模,型结实果现得到特了征9的8.自83动%提的取分,最类后准运确用率,X说GB明oos提t出算的法模进型行能训有练效和提分取类。癫通痫脑电信号的特征信息,并能进行准确的分类,分类精度和稳定性都有所提高。

基于皮尔逊最优电极选择的ADHD患者脑电特征提取及分类研究

事件相关电位(ERP)可用于注意缺陷多动障碍儿童(ADHD)和正常儿童的脑电特征提取与分类。首先,采用赌博任务范式,采集2类儿童的脑电信号;其次,基于皮尔逊相关系数算法选择最优电极,并预处理最优电极脑电信号;然后,提取预处理脑电信号的时域特征(均值、方差、峰值)和频域特征(Theta波段功率、Alpha波段功率);最后,利用传统分类方法支持向量机(SVM)、自适应增强(AdaBoost)、自举汇聚法(Bagging)、线性判别式分析(LDA)、反向传播(BP)和组合分类器的分类方法(LDA-SVM,BP-SVM)完成对2种脑电信号的分类。研究结果表明,传统方法 BP分类器的分类准确率可达80.52%,组合分类器BP-SVM的分类准确率可达88.88%。组合分类方法能提高ADHD儿童的分类准确率,为基于脑机接口技术的ADHD神经反馈康复治疗提供技术支持。

基于脑电功能连接拓扑表征的心算任务分类

使用脑网络图的方法分析脑电功能连接存在阈值选择、忽略了脑电图动力学特性的问题。针对这一问题,提出了一种使用拓扑动态建模的方法来分析脑电功能连接矩阵,以提高心算任务分类识别正确率。该方法首先将功能连接矩阵转换为无向加权图,然后使用持续同调工具来构建不同的复形,记录拓扑动态过程中形成的不同阶的同调特征,形成持续图,最后使用持续景观图特征作为分类特征,输入到随机森林分类器进行心算状态识别。在心算状态识别和心算质量分类两个任务中分别获得了最高99.26%、99.20%的识别准确率,97.87%、99.80%的敏感性,以及99.78%、97.64%的特异性,并且在跨个体验证实验中分别获得了66.81%、66.85%的准确率。实验结果表明,该方法能充分考虑所有可能的阈值,有效提取脑电功能连接的分类信息,实现脑电心算状态自动识别。

基于深度卷积网络的运动想象脑电信号分类方法

为了提高多分类运动想象脑电信号的解码精度,以此促进脑机接口系统在生产生活中的应用。采用基于深度卷积网络的LeNet和AlexNet模型分析四分类运动想象脑电特性。将脑电信号通过预处理、数据归一化和数据增强,然后分别输入两个模型中进行分类。通过与现有不同的特征提取和分类方法对比,实验结果表明,在多分类运动想象脑电解码研究领域中,深度卷积网络模型取得的分类效果较好。

基于多特征融合的多分类运动想象脑电信号识别研究

针对目前多分类运动想象脑电识别存在特征提取单一、分类准确率低等问题,提出一种多特征融合的四分类运动想象脑电识别方法来提高识别率。对预处理后的脑电信号分别使用希尔伯特-黄变换、一对多共空间模式、近似熵、模糊熵、样本熵提取结合时频—空域—非线性动力学的初始特征向量,用主成分分析降维,最后使用粒子群优化支持向量机分类。该算法通过对国际标准数据集BCI2005 Data set IIIa中的k3b受试者数据经MATLAB仿真处理后获得93.30%的识别率,均高于单一特征和其它组合特征下的识别率。分别对四名实验者实验采集运动想象脑电数据,使用本研究提出的方法处理获得了72.96%的平均识别率。结果表明多特征融合的特征提取方法能更好的表征运动想象脑电信号,使用粒子群支持向量机可取得较高的识别准确率,为人脑的认知活动提供了一种新的识别方法。

空间模型对单次运动相关脑电的分析

本研究提出一种从单次试验的多导EEG信号中提取运动相关去同步化和同步化电位特征的空间模型 ,区分左右手想象运动 ,作为一种新的通讯手段对外界设备进行控制。此模型根据各电极对分类的重要性自动获得其权值 ,并将EEG信号沿最适合分类的几个方向投影 ,沿投影方向计算一连续时间段内的方差 ,作为线性分类器的特征输入。对 8名被试者左右手想象运动时 5 9导EEG进行分类 ,正确率均在 70 %以上 ,与用多通道AR模型提取特征、神经网络做分类器的方法相比 ,效果好、速度快。

脑电时空多特征融合的数字图形界面认知负荷评价方法

准确地评价数字图形界面下操作员的认知负荷(cognitive load,CL),有助于实现认知反馈机制并最终提高人机工效.为了进一步提高评价方法的鲁棒性与泛化能力,结合EEG实验将Att-BLSTM应用于CL评价问题中.该方法首先利用Multi-CNN提取EEG的频域与空间特征,然后利用Att-BLSTM提取EEG的时域特征,最后通过多特征融合构建CL评价方法.通过招募12名被试,采集了2种CL条件下的EEG数据进行了实验.实验结果表明,文中方法在该数据集上的平均准确率为82%,比传统机器学习的方法具有更强的EEG信号表征能力;与其他深度学习方法相比,也能更准确地提取EEG的时域特征,且具有更强的鲁棒性.

卷积神经网络及其分析在抑郁症判别中的应用

在线脑电分类能准确评估严重抑郁症患者的脑状态并及时跟踪其发展状态可以将其陷入危险和自杀的风险降为最低。然而,在无经验监督条件下,在线脑电分类应用面临更大的挑战:脑电数据往往具有弱信号、高噪声与非平稳特性;缺乏有效解耦脑疾病发作时脑区与神经网路的复杂关系。为此,设计一个以卷积神经网络为核心的云辅助在线脑电分类系统,该系统直接应用于原始脑电信号,无需进行预处理和特征提取,能精准、快速判别抑郁症状态。在公开数据集上进行抑郁症评估实验,对健康控制组和抑郁症对照组分类的准确率、敏感度和特异度分别为99.08%、98.77%和99.42%。另外,通过对神经网络进行定量解释,表明抑郁症病人的左右颞叶脑区与正常人存在明显差异。

Individualization of Data-Segment-Related Parameters for Improvement of EEG Signal Classification in Brain-Computer Interface

In electroencephalogram (EEG) modeling techniques, data segment selection is the first and still an important step. The influence of a set of data-segment-related parameters on feature extraction and classification in an EEG-based brain-computer interface (BCI) was studied. An auto search algorithm was developed to study four datasegment-related parameters in each trial of 12 subjects’ EEG. The length of data segment (LDS), the start position of data (SPD) segment, AR order, and number of trials (NT) were used to build the model. The study showed that, compared with the classification ratio (CR) without parameter selection, the CR was increased by 20% to 30% with proper selection of these data-segment-related parameters, and the optimum parameter values were subject-dependent. This suggests that the data-segment-related parameters should be individualized when building models for BCI.

Noise-assisted MEMD based relevant IMFs identification and EEG classification

Noise-assisted multivariate empirical mode decomposition(NA-MEMD) is suitable to analyze multichannel electroencephalography(EEG) signals of non-stationarity and non-linearity natures due to the fact that it can provide a highly localized time-frequency representation.For a finite set of multivariate intrinsic mode functions(IMFs) decomposed by NA-MEMD,it still raises the question on how to identify IMFs that contain the information of inertest in an efficient way,and conventional approaches address it by use of prior knowledge.In this work,a novel identification method of relevant IMFs without prior information was proposed based on NA-MEMD and Jensen-Shannon distance(JSD) measure.A criterion of effective factor based on JSD was applied to select significant IMF scales.At each decomposition scale,three kinds of JSDs associated with the effective factor were evaluated:between IMF components from data and themselves,between IMF components from noise and themselves,and between IMF components from data and noise.The efficacy of the proposed method has been demonstrated by both computer simulations and motor imagery EEG data from BCI competition IV datasets.

An efficient approach of EEG feature extraction and classification for brain computer interface

In the study of brain-computer interfaces,a method of feature extraction and classification used fortwo kinds of imaginations is proposed.It considers Euclidean distance between mean traces recorded fromthe channels with two kinds of imaginations as a feature,and determines imagination classes using thresh-old value.It analyzed the background of experiment and theoretical foundation referring to the data sets ofBCI 2003,and compared the classification precision with the best result of the competition.The resultshows that the method has a high precision and is advantageous for being applied to practical systems.

Alertness Staging Based on Improved Self-Organizing Map

In order to classify the alertness status, 19 channels of electroencephalogram(EEG) signals from 5 subjects were acquired during daytime nap. Ten different types of features(including time domain features, frequency domain features and nonlinear features) were extracted from EEG signals, and an improved self-organizing map(ISOM) neuron network was proposed, which successfully identify three different brain status of the subjects: awareness, drowsiness and sleep. Compared with traditional SOM, the experiment results show that the ISOM generates much better classification accuracy, reaching as high as 89.59%.

P300 EEG Recognition Based on SVM Approach

In this paper, we used SVM method to detect P300 signal. Before training a classification parameter for the SVM, several preprocessing operations were applied to the data including filtering, downsampling, single trial extraction, windsorizing, electrode selection et al. With the SVM algorithm, the classification accuracy could be up to above 80%. In some cases, the accuracy could reach 100%. It is suitable to use SVM for P300 EEG recognition in the P300-based brain-computer interface (BCI) system. Our further work will include the improvement to yield higher classification accuracy using fewer trials.

EEG classification based on probabilistic neural network with supervised learning in brain computer interface

Aiming at the topic of electroencephalogram （EEG） pattern recognition in brain computer interface （BCI）, a classification method based on probabilistic neural network （PNN） with supervised learning is presented in this paper. It applies the recognition rate of training samples to the learning progress of network parameters. The learning vector quantization is employed to group training samples and the Genetic algorithm （GA） is used for training the network＇ s smoothing parameters and hidden central vector for detemlining hidden neurons. Utilizing the standard dataset I （a） of BCI Competition 2003 and comparing with other classification methods, the experiment results show that the best performance of pattern recognition Js got in this way, and the classification accuracy can reach to 93.8%, which improves over 5% compared with the best result （88.7 % ） of the competition. This technology provides an effective way to EEG classification in practical system of BCI.

A classification-based method to estimate event-related potentials from single trial EEG

A novel method based on machine learning is developed to estimate event-related potentials from single trial electroencephalography. This paper builds a basic framework using classification and an optimization model based on this framework for estimating event-related potentials. Then the SingleTrialEM algorithm is derived by introducing a logistic regression model, which could be obtained by training before SingleTrialEM is used, to instantiate the optimization model. The simulation tests demonstrate that the proposed method is correct and solid. The advantage of this method is verified by the comparison between this method and the Woody filter in simulation tests. Also, the cognitive test results are consistent with the conclusions of cognitive science.