Multiclass common spatial patterns and information theoretic feature extraction

We address two shortcomings of the common spatial patterns (CSP) algorithm for spatial filtering in the context of brain--computer interfaces (BCIs) based on electroencephalography/magnetoencephalography (EEG/MEG): First, the question of optimality of CSP in terms of the minimal achievable classification error remains unsolved. Second, CSP has been initially proposed for two-class paradigms. Extensions to multiclass paradigms have been suggested, but are based on heuristics. We address these shortcomings in the framework of information theoretic feature extraction (ITFE). We show that for two-class paradigms, CSP maximizes an approximation of mutual information of extracted EEG/MEG components and class labels. This establishes a link between CSP and the minimal classification error. For multiclass paradigms, we point out that CSP by joint approximate diagonalization (JAD) is equivalent to independent component analysis (ICA), and provide a method to choose those independent components (ICs) that approximately maximize mutual information of ICs and class labels. This eliminates the need for heuristics in multiclass CSP, and allows incorporating prior class probabilities. The proposed method is applied to the dataset IIIa of the third BCI competition, and is shown to increase the mean classification accuracy by 23.4% in comparison to multiclass CSP.     

Multiclass filters by a weighted pairwise criterion for EEG single-trial classification

The filtering technique for dimensionality reduction of multichannel electroencephalogram (EEG) recordings, modeled using common spatial patterns and its variants, is commonly used in two-class brain-computer interfaces (BCI). For a multiclass problem, the optimization of certain separability criteria in the output space is not directly related to the classification error of EEG single-trial segments . In this paper, we derive a new discriminant criterion, termed weighted pairwise criterion (WPC), for optimizing multiclass filters by minimizing the upper bound of the Bayesian error that is intentionally formulated for classifying EEG single-trial segments. The WPC approach pays more attention to close class pairs that are more likely to be misclassified than far away class pairs that are already well separated. Moreover, we extend WPC by integrating temporal information of EEG series. Computationally, we employ the rank-one update and power iteration technique to optimize the proposed discriminant criterion. The experiments of multiclass classification on the datasets of BCI competitions demonstrate the efficacy of the proposed method.     

基于黎曼流形监督降维的矩阵CFAR增强检测

矩阵CFAR检测是从几何流形角度处理雷达目标检测问题的新技术。为进一步提升其在复杂杂波背景下的检测性能,本文提出一种黎曼流形监督降维的矩阵CFAR增强检测方法。首先,将检测问题视为目标与杂波的分类问题,分别构建黎曼流形上目标单元与杂波单元的类内和类间权重矩阵;其次,为增强目标与杂波的可分性,采用保持类内几何距离最小,类间几何距离最大的准则建立降维目标函数,并基于Grassmann流形求解降维优化问题获得映射矩阵;最后,提出一种矩阵CFAR增强检测方法,实现目标增强检测。采用蒙特卡罗方法对仿真数据和实测海杂波数据进行实验分析,结果表明,所提出的方法能够进一步提升检测性能。      

Diffeomorphic sulcal shape analysis on the cortex

We present a diffeomorphic approach for constructing intrinsic shape atlases of sulci on the human cortex. Sulci are represented as square-root velocity functions of continuous open curves in R3, and their shapes are studied as functional representations of an infinite-dimensional sphere. This spherical manifold has some advantageous properties--it is equipped with a Riemannian L2 metric on the tangent space and facilitates computational analyses and correspondences between sulcal shapes. Sulcal shape mapping is achieved by computing geodesics in the quotient space of shapes modulo scales, translations, rigid rotations, and reparameterizations. The resulting sulcal shape atlas preserves important local geometry inherently present in the sample population. The sulcal shape atlas is integrated in a cortical registration framework and exhibits better geometric matching compared to the conventional euclidean method. We demonstrate experimental results for sulcal shape mapping, cortical surface registration, and sulcal classification for two different surface extraction protocols for separate subject populations.     

Recognition of motor imagery tasks for BCI using CSP and chaotic PSO twin SVM

Accurate modeling and recognition of the brain activity patterns for reliable communication and interaction are still a challenging task for the motor imagery (MI) brain-computer interface (BCI) system. In this paper, we propose a common spatial pattern (CSP) and chaotic particle swarm optimization (CPSO) twin support vector machine (TWSVM) scheme for classification of MI electroencephalography (EEG). The self-adaptive artifact removal and CSP were used to obtain the most distinguishable features. To improve the recognition results, CPSO was employed to tune the hyper-parameters of the TWSVM classifier. The usefulness of the proposed method was evaluated using the BCI competition IV-IIa dataset. The experimental results showed that the mean recognition accuracy of our proposed method was increased by 5.35%, 4.33%, 0.78%, 1.45%, and 9.26% compared with the CPSO support vector machine (SVM), particle swarm optimization (PSO) TWSVM, linear discriminant analysis LDA), back propagation (BP) and probabilistic neural network (PNN), respectively. Furthermore, it achieved a faster or comparable central processing unit (CPU) running time over the traditional SVM methods.     

Combining spatial filters for the classification of single-trial EEG in a finger movement task

Brain-computer interface (BCI) is to provide a communication channel that translates human intention reflected by a brain signal such as electroencephalogram (EEG) into a control signal for an output device. In recent years, the event-related desynchronization (ERD) and movement-related potentials (MRPs) are utilized as important features in motor related BCI system, and the common spatial patterns (CSP) algorithm has shown to be very useful for ERD-based classification. However, as MRPs are slow nonoscillatory EEG potential shifts, CSP is not an appropriate approach for MRPs-based classification. Here, another spatial filtering algorithm, discriminative spatial patterns (DSP), is newly introduced for better extraction of the difference in the amplitudes of MRPs, and it is integrated with CSP to extract the features from the EEG signals recorded during voluntary left versus right finger movement tasks. A support vector machines (SVM) based framework is designed as the classifier for the features. The results show that, for MRPs and ERD features, the combined spatial filters can realize the single-trial EEG classification better than anyone of DSP and CSP alone does. Thus, we propose an EEG-based BCI system with the two feature sets, one based on CSP (ERD) and the other based on DSP (MRPs), classified by SVM.     

Biologically Inspired Feature Manifold for Scene Classification

Biologically inspired feature (BIF) and its variations have been demonstrated to be effective and efficient for scene classification. It is unreasonable to measure the dissimilarity between two BIFs based on their Euclidean distance. This is because BIFs are extrinsically very high dimensional and intrinsically low dimensional, i.e., BIFs are sampled from a low-dimensional manifold and embedded in a high-dimensional space. Therefore, it is essential to find the intrinsic structure of a set of BIFs, obtain a suitable mapping to implement the dimensionality reduction, and measure the dissimilarity between two BIFs in the low-dimensional space based on their Euclidean distance. In this paper, we study the manifold constructed by a set of BIFs utilized for scene classification, form a new dimensionality reduction algorithm by preserving both the geometry of intra BIFs and the discriminative information inter BIFs termed Discriminative and Geometry Preserving Projections (DGPP), and construct a new framework for scene classification. In this framework, we represent an image based on a new BIF, which combines the intensity channel, the color channel, and the C1 unit of a color image; then we project the high-dimensional BIF to a low-dimensional space based on DGPP; and, finally, we conduct the classification based on the multiclass support vector machine (SVM). Thorough empirical studies based on the USC scene dataset demonstrate that the proposed framework improves the classification rates around 100% relatively and the training speed 60 times for different sites in comparing with previous gist proposed by Siagian and Itti in 2007.     

Transfer Learning of BCI Using CUR Algorithm

The brain computer interface (BCI) are used in many applications including medical, environment, education, economy, and social fields. In order to have a high performing BCI classification, the training set must contain variations of high quality subjects which are discriminative. Variations will also drive transferability of training data for generalization purposes. However, if the test subject is unique from the training set variations, BCI performance may suffer. Previously, this problem was solved by introducing transfer learning in the context of spatial filtering on small training set by creating high quality variations within training subjects. In this study however, it was discovered that transfer learning can also be used to compress the training data into an optimal compact size while improving training data performance. The transfer learning framework proposed was on motor imagery BCI-EEG using CUR matrix decomposition algorithm which decomposes data into two components; C and UR which is each subject’s EEG signal and common matrix derived from historical EEG data, respectively. The method is considered transfer learning process because it utilizes historical data as common matrix for the classification purposes. This framework is implemented in the BCI system along with Common Spatial Pattern (CSP) as features extractor and Extreme Learning Machine (ELM) as classifier and this combination exhibits an increase of accuracy to up to 26% with 83% training database compression.

     

一种基于FPGA的脑电分类算法实现

通过对脑电信号特征的分析,利用小波变换的多尺度分析技术对脑电信号进行特征提取,进而使用主成分分析算法对特征进行降维,并对降维后的信号使用Fisher线性判别方法进行分类。最后,利用VerilogHDL硬件编程语言设计实现了Mallat分解算法、PCA算法和LDA算法模块,并在FPGA应用板上实现了脑电分类功能。系统对2008年BCI大赛的数据进行了测试,分类准确率达到92.31%,表明该方法对开发便携式脑机接口系统具有良好的应用价值。     

基于相关性和稀疏表示的运动想象脑电通道选择方法

在基于运动想象(MI)的脑机接口(BCI)中,通常采用较多通道的脑电信号(EEG)来提高分类精度,但其中会有包含与MI任务无关或冗余信息的通道,从而影响BCI的性能提升。该文针对运动想象脑电分类中的通道选择问题,提出一种采用相关性和稀疏表示对通道进行选择的方法(CSR-CS)。首先计算训练样本每个通道的皮尔逊相关系数来选择显著通道,然后提取显著通道所在区域的滤波器组共空间模式特征拼接成字典,利用由字典所得到的非零稀疏系数的个数表征每个区域的分类能力,选出显著区域所包含的显著通道作为最优通道,最后采用共空间模式和支持向量机分别进行特征提取与分类。在对BCI第3次竞赛数据集IVa和BCI第4次竞赛数据集I两个二分类MI任务的分类实验中,平均分类精度达到了88.61%和83.9%,表明所提通道选择方法的有效性和鲁棒性。     

一种基于AR模型的矩阵CFAR检测器

矩阵恒虚警（Constant False Alarm Rate,CFAR）检测器是根据信息几何理论利用检测单元相关矩阵与参考单元矩阵黎曼均值间的测地线距离作为检测统计量进行检测.本文根据尤尔-沃克方程中相关矩阵与自回归（Auto-Regressive,AR）谱之间的一一对应关系,将矩阵CFAR检测器中两矩阵间的距离转化为两AR谱之间的距离,提出了一种基于AR模型的矩阵CFAR检测器,该检测器能利用分辨率更高的AR模型进行谱估计,并将功率谱整体进行检测,能充分利用回波的多普勒信息,非常适用于频域扩展目标的检测,通过利用具有频域扩展特性的尾流实测数据对该方法和传统方法的检测性能进行比较分析,验证了本文所提检测器的有效性.     

A brain-controlled switch for asynchronous control applications

Asynchronous control applications are an important class of application that has not received much attention from the brain-computer interface (BCI) community. This work provides a design for an asynchronous BCI switch and performs the first extensive evaluation of an asynchronous device in attentive, spontaneous electroencephalographic (EEG). The switch design [named the low-frequency asynchronous switch design (LF-ASD)] is based on a new feature set related to imaginary movements in the 1-4 Hz frequency range. This new feature set was identified from a unique analysis of EEG using a bi-scale wavelet. Offline evaluations of a prototype switch demonstrated hit (true positive) rates in the range of 38%-81% with corresponding false positive rates in the range of 0.3%-11.6%. The performance of the LF-ASD was contrasted with two other ASDs: one based on mu-power features and another based on the outlier processing method (OPM) algorithm. The minimum mean error rates for the LF-ASD were shown to be significantly lower than either of these other two switch designs.     

Active contour evolved by joint probability classification on Riemannian manifold

In this paper, we present an active contour model for image segmentation based on a nonparametric distribution metric without any intensity a priori of the image. A novel nonparametric distance metric, which is called joint probability classification, is established to drive the active contour avoiding the instability induced by multimodal intensity distribution. Considering an image as a Riemannian manifold with spatial and intensity information, the contour evolution is performed on the image manifold by embedding geometric image feature into the active contour model. The experimental results on medical and texture images demonstrate the advantages of the proposed method.     

Common Spatial Pattern Ensemble Classifier and Its Application in Brain-Computer Interface

Common spatial pattern (CSP) algorithm is a successful tool in feature estimate of brain-computer interface (BCI). However, CSP is sensitive to outlier and may result in poor outcomes since it is based on pooling the covariance matrices of trials. In this paper, we propose a simple yet effective approach, named common spatial pattern ensemble (CSPE) classifier, to improve CSP performance. Through division of recording channels, multiple CSP filters are constructed. By projection, log-operation, and subtraction on the original signal, an ensemble classifier, majority voting, is achieved and outlier contaminations are alleviated. Experiment results demonstrate that the proposed CSPE classifier is robust to various artifacts and can achieve an average accuracy of 83.02%.     

Common Spatial Pattern Ensemble Classifier and Its Application in Brain-Computer Interface

Abstract-Common spatial pattern （CSP） algorithm is a successful tool in feature estimate of brain-computer interface （BCI）. However, CSP is sensitive to outlier and may result in poor outcomes since it is based on pooling the covariance matrices of trials. In this paper, we propose a simple yet effective approach, named common spatial pattern ensemble （CSPE） classifier, to improve CSP performance. Through division of recording channels, multiple CSP filters are constructed. By projection, log-operation, and subtraction on the original signal, an ensemble classifier, majority voting, is achieved and outlier contaminations are alleviated. Experiment results demonstrate that the proposed CSPE classifier is robust to various artifacts and can achieve an average accuracy of 83.02%.     

Single-Trial EEG Source Reconstruction for Brain–Computer Interface

A new way to improve the classification rate of an EEG-based brain-computer interface (BCI) could be to reconstruct the brain sources of EEG and to apply BCI methods to these derived sources instead of raw measured electrode potentials. EEG source reconstruction methods are based on electrophysiological information that could improve the discrimination between BCI tasks. In this paper, we present an EEG source reconstruction method for BCI. The results are compared with results from raw electrode potentials to enable direct evaluation of the method. Features are based on frequency power change and Bereitschaft potential. The features are ranked with mutual information before being fed to a proximal support vector machine. The dataset IV of the BCI competition II and data from four subjects serve as test data. Results show that the EEG inverse solution improves the classification rate and can lead to results comparable to the best currently known methods.     

Single-trial EEG source reconstruction for brain-computer interface.

A new way to improve the classification rate of an EEG-based brain-computer interface (BCI) could be to reconstruct the brain sources of EEG and to apply BCI methods to these derived sources instead of raw measured electrode potentials. EEG source reconstruction methods are based on electrophysiological information that could improve the discrimination between BCI tasks. In this paper, we present an EEG source reconstruction method for BCI. The results are compared with results from raw electrode potentials to enable direct evaluation of the method. Features are based on frequency power change and Bereitschaft potential. The features are ranked with mutual information before being fed to a proximal support vector machine. The dataset IV of the BCI competition II and data from four subjects serve as test data. Results show that the EEG inverse solution improves the classification rate and can lead to results comparable to the best currently known methods.     

Distance based kernels for video tensors on product of Riemannian matrix manifolds

In this paper, we explore the inherent geometry of video tensors by modeling them as points in product of Riemannian matrix manifolds. A video tensor is decomposed into three modes (factors) using matrix unfolding operation and each mode is represented as a point in a product space of Grassmannian and symmetric positive definite (SPD) matrix manifold. Hence a video is represented as a point in the Cartesian product of three such product spaces. Being a manifold valued (non-Euclidean) representation, application of several state-of-the-art Euclidean machine learning algorithms lead to inferior results. To overcome this, we propose positive definite kernels which map the points from product manifold space to Hilbert space. The proposed kernel functions implicitly make use of geodesic distance on product manifold to obtain a similarity measure and generate a kernel-gram matrix. In addition, we generate a discriminative feature representation for each manifold valued point using kernel-gram matrix diagonalization. Classification is performed in a sparse framework. The proposed methodology is tested over three publicly available datasets for hand gesture, traffic signal and sign language recognition. Experimentation performed over these datasets show that the proposed methodology is powerful in terms of classification accuracy in comparison with the state-of-the-art methods.     

A comparison of common spatial patterns with complex band power features in a four-class BCI experiment

We report on the offline analysis of four-class brain-computer interface (BCI) data recordings. Although the analysis is done within defined time windows (cue-based BCI), our goal is to work toward an approach which classifies on-going electroencephalogram (EEG) signals without the use of such windows (un-cued BCI). To that end, we provide some elements of that analysis related to timing issues that will become important as we pursue this goal in the future. A new set of features called complex band power (CBP) features which make explicit use of phase are introduced and are shown to produce good results. As reference methods we used traditional band power features and the method of common spatial patterns. We consider also for the first time in the context of a four-class problem the issue of variability of the features over time and how much data is required to give good classification results. We do this in a practical way where training data precedes testing data in time.