基于排列组合熵的脑电意识任务识别方法的研究

研究基于脑电信号排列组合熵的运动意识任务自动分类方法.求出时变脑电信号所对应的排列组合熵时间序列.它能很好的反映出事件相关去同步(ERD)和事件相关同步(ERS)现象,因此能有效地提取人脑想象左右手运动任务时的特征,最终利用K-近邻法模式分类方法对想象左右手运动任务进行分类决策.对国际脑机接口竞赛相关数据进行测试,最高准确率达到88.57%,最大互信息达到0.42.基于排列组合熵的脑电信号特征,可以作为脑电意识任务的有效分类依据.     

小波包熵和BP神经网络在意识任务识别中的应用

探索了小波包熵和BP神经网络在识别左右手想象运动中的作用.采用脑-机接口2003竞赛中Graz科技大学提供的脑电数据,计算C3、C4电极8～16Hz频带脑电信号的小波包熵,将其作为反应想象左右手运动的特征量,用BP神经网络对大脑想象左右手运动任务进行分类,最大分类正确率可达88.57%,与使用线性判别式算法分类结果相比,效果更好.脑电信号小波包熵随时间的变化与事件相关去同步和事件相关同步现象相一致,可在线识别左右手想象运动,为大脑运动意识任务的特征提取及肢残患者的临床康复提供了新思路.     

基于奇异谱熵的脑电意识任务识别方法的研究

奇异谱分析是脑电信号研究的一种新方法。脑电信号的奇异谱熵可以反映脑电的特征,它有助于研究大脑的动力学行为。时变脑电信号所对应的奇异谱熵时间序列能很好地反映出事件相关去同步(ERD)和事件相关同步(ERS)现象,因此可以提取人脑想象左右手运动任务时的特征,最终利用K-近邻模式分类方法对想象左右手运动任务进行有效的分类决策。最后对国际脑机接口竞赛2003相关数据进行了测试,最高准确率达到85.16%,最大互信息达到0.48。测试结果说明,基于奇异谱熵的脑电信号特征,可以作为脑电意识任务的有效分类依据。     

基于HHT运动想象脑电模式识别研究

脑机接口是一种变革性的人机交互, 其中基于运动想象(Motor imagery, MI)脑电的脑机接口是一类非常重要的脑机交互. 本文旨在探索有效的运动想象脑电特征模式提取方法. 采用在时域、频域同时具有很高分辨率的希尔伯特--黄变换(Hilbert-Huang transform, HHT),进而提取自回归(Auto regressive, AR)模型参数并计算运动想象脑电平均瞬时能量,从而构造特征向量, 最后利用能较好地适应运动想象脑电单次试验分类的支持向量机(Support vector machine, SVM)进行分类. 结果表明在Trial的5.5～7.5s期间, HHT特征提取方法平均分类正确率为81.08%, 具有良好的适应性;最高分类正确率为87.86%, 优于传统的小波变换特征提取方法和未经HHT的特征提取方法;在Trial的8～9s期间, HHT特征提取方法显著优于后两种特征提取方法. 本研究证实了HHT对运动想象脑电这一非平稳非线性信号具有很好的特征提取能力, 也再次验证了运动想象事件相关去同步(Event-related desynchronization, ERD)现象, 同时也表明运动想象脑电的脑--机交互系统性能与被试想象心理活动的质量密切相关. 本文可望为基于运动想象脑电的在线实时脑机交互控制系统的研究打下坚实的基础.     

免疫多域特征融合的多核学习SVM运动想象脑电信号分类

针对多通道四类运动想象(Motor imagery, MI)脑电信号(Electroencephalography, EEG)的分类问题, 提出免疫多域特征融合的多核学习SVM (Support vector machine)运动想象脑电信号分类算法.首先, 通过离散小波变换(Discrete wavelet transform, DWT)提取脑电信号的时频域特征, 并利用一对多公共空间模式(One versus the rest common spatial patterns, OVR-CSP)提取脑电信号的空域特征, 融合时频空域特征形成特征向量.其次, 利用多核学习支持向量机(Multiple kernel learning support vector machine, MKL-SVM)对提取的特征向量进行分类.最后, 利用免疫遗传算法(Immune genetic algorithm, IGA)对模型的相关参数进行优化, 得到识别率更高的脑电信号分类模型.采用BCI2005desc-Ⅲa数据集进行实验验证, 对比结果表明, 本文所提出的分类模型有效地解决了传统单域特征提取算法特征单一、信息描述不足的问题, 更准确地表达了不同受试者个性化的多域特征, 取得了94.21%的识别率, 优于使用相同数据集的其他方法.     

基于DTCWT的运动想象脑电特征提取

针对脑电信号采用单一特征识别存在自适应性差和识别率低等问题，提出一种基于双树复小波(DTCWT)的多特征融合的左右手运动想象脑电特征提取方法。对原始脑电信号进行DTCWT变换提取最佳时频段；对所提取的信号频段进行希尔伯特变换与Lempel-Ziv复杂度计算，将得到的时-频域特征与非线性特征组合为特征向量；采用线性判别分析(LDA)完成运动想象任务的分类。实验采用BCI CompetitionⅢ竞赛数据对该方法进行验证，仿真结果表明其识别准确率明显提高，最高可达89.84%。     

基于LMD和CSP的多域融合脑电信号分类方法

运动想象脑电信号非平稳、非线性和微弱性特征明显,采用传统单一维度特征进行分类时存在识别率低、鲁棒性差的问题。提出一种基于局部均值分解(Local Mean Decomposition, LMD)和共空间模式(Common Spatial Pattern, CSP)的多域融合脑电信号分类方法,采用LMD对运动脑电信号进行自适应分解得到多个乘积分量(Product Function, PF),进而从PF中提取反映不同信号差异特性的12维时-频域特征,将PF作为CSP的多通道数据进行分解,并提取18维空域特征。利用相关向量机(Relevance Vector Machine, RVM)分类器对30维时-频-空域特征进行特征选择和分类识别,在自动确定最优分类特征的同时获得理想的分类结果。基于BCI竞赛数据开展实验,结果表明,所提方法可以获得优于95%的正确分类性能,并且在低信噪比条件下具有较强的噪声稳健性。     

一种新的BP学习算法及在运动想象脑电信号分类识别中的应用

对运动想象(MI)脑电信号的正确分类是决定基于运动想象脑电的脑－机接口(BCI)性能的关键因素。为有效地提取MI脑电信号特征、提高分类正确率,提出一种基于单形进化的BP神经网络优化算法(BPSSSE)并运用于MI脑电信号的识别,提取自相关(AR)模型参数和希尔伯特边际谱作为特征输入,通过单形进化算法优化BP神经网络学习性能,实现对MI脑电信号的分类。测试实验中,对BCI竞赛数据进行左右手分类。结果表明在4s~ 8s时间段内平均分类正确率为80.17%,最高分类正确率为87.14%,证明了本文算法在基于MI脑电的脑机交互控制系统中应用研究的有效性和可行性。     

基于多特征和BP神经网络的脑-机接口研究

研究了一种基于运动想象识别的脑-机接口(BCI)系统,通过提取想象过程中的脑电信号(EEG)中Alpha波特征,采用多特征分类的方法,以提高脑-机接口系统运动想象识别的正确率。针对脑电信号单特征分类精确度低、耗时长等缺点,采用自回归模型法、统计特征提取和频域分析的方法对Alpha波提取多个特征值,利用BP神经网络进行分类,对运动想象进行识别。通过实验验证了其识别率较高,取得了预期的效果,证明了多特征融合结合BP神经网络运用于脑机接口系统的可行性。     

左右手运动意识任务的分类

本研究提出了利用事件相关ERD/ERS和相同步提取脑电运动意识任务特征,应用Fisher线性判别式分析法,对想象左右手运动任务进行了分类,并获得了满意的效果。对被试者想象左右手运动过程中,记录的脑电信号采用了能量分析法量化了事件相关去同步ERD和事件相关同步ERS时程,同时提取了相位信息。最后对测试数据进行分类,最大正确分类率达到了83.20%,与用多通道AR模型提取特征方法相比,效果更好,从而为大脑运动意识任务的分类提供了新思路。     

A Bayesian network-based framework for semantic image understanding

Current research in content-based semantic image understanding is largely confined to exemplar-based approaches built on low-level feature extraction and classification. The ability to extract both low-level and semantic features and perform knowledge integration of different types of features is expected to raise semantic image understanding to a new level. Belief networks, or Bayesian networks (BN), have proven to be an effective knowledge representation and inference engine in artificial intelligence and expert systems research. Their effectiveness is due to the ability to explicitly integrate domain knowledge in the network structure and to reduce a joint probability distribution to conditional independence relationships. In this paper, we present a general-purpose knowledge integration framework that employs BN in integrating both low-level and semantic features. The efficacy of this framework is demonstrated via three applications involving semantic understanding of pictorial images. The first application aims at detecting main photographic subjects in an image, the second aims at selecting the most appealing image in an event, and the third aims at classifying images into indoor or outdoor scenes. With these diverse examples, we demonstrate that effective inference engines can be built within this powerful and flexible framework according to specific domain knowledge and available training data to solve inherently uncertain vision problems.     

基于混合模型的事件触发词抽取

事件结构性语法特征与事件语义特征各有优势,二者融合利于准确表征事件触发词,进而有利于完成事件触发词抽取任务。现有的基于特征、基于结构及基于神经网络模型等的抽取方法仅能捕捉事件的部分特征,不能够准确表征事件触发词。为解决上述问题,提出一种融合了事件结构性语法特征和事件语义特征的混合模型,完成事件触发词抽取任务。首先,在初始化向量模型中融入句子的依存句法信息,使初始向量中包含事件结构性语法特征;然后,将初始向量依次传入神经网络模型中的CNN和BiGRU-E-attention模型中,在捕获多维度事件语义特征的同时,完成事件结构性语法特征与事件语义特征的融合;最后,进行事件触发词的抽取。在CEC中文突发语料库上进行事件触发词位置识别和分类实验,该模型的F值较基准模型的分别提高了0.86%和4.07%;在ACE2005英文语料库上,该模型的F值较基准模型的分别提高了1.4%和1.5%。实验结果表明,混合模型在事件触发词抽取任务中取得了优异的效果。     

中文名实体识别中的特征组合与特征融合的比较

先分析了最大熵模型常用的特征线性组合方法中的权值偏置问题,然后提出了在线性组合之前,对特征进行融合,并根据融合特征和目标类别之间的互信息选择有效复合特征的方法。通过在包含2000个人名的语料库上的测试,表明特征融合能有效地提高名实体识别的精度和召回率。     

Inference and retrieval of soccer event

As to the soccer video, the event is defined as the medium-level spatiotemporal entity interesting to users, having certain context cues corresponding to the specific domain knowledge model. As a medium-level entity, the inference of soccer event is based on the fusion of context cues and domain knowledge model. The shooting event is chosen as research target and the event analysis method is expected to be reusable for other soccer events. According to the analysis of shooting event, the following seven kinds of context cues are extracted, respectively including one kind of caption detection, two kinds of face detection, one kind of audience detection, one kind of goal detection, and two kinds of motion estimation. In the inference of soccer event Bayesian network is used to perform the fusion of context cues. In the experiments the event retrieval is performed based on the video data of World Cup 2002, and the results show that the key to event retrieval is the extraction of context cues related with the user-defined event closely.     

基于实体画像增强网络的事件检测方法

事件检测任务旨在从非结构化的文本中自动识别并分类事件触发词。挖掘和表示实体的属性特征(即实体画像)有助于事件检测,其基本原理在于“实体本身的属性往往暗示了其参与的事件类型”(例如,“警察”往往参与“Arrest-Jail”类的事件)。现有研究已利用编码信息实现实体表示,并借此优化事件检测模型。然而,其表示学习过程仅仅纳入局部的句子级语境信息,使得实体画像的信息覆盖率偏低。为此,该文提出基于全局信息和实体交互信息的画像增强方法,其借助图注意力神经网络,不仅在文档级的语境范围内捕捉实体的高注意力背景信息,也同时纳入了局部相关实体的交互信息。特别地,该文开发了基于共现图的注意力遮蔽模型,用于降低噪声信息对实体表示学习过程的干扰。在此基础上,该文联合上述实体画像增强网络、BERT语义编码网络和GAT聚合网络,形成了总体的事件检测模型。该文在通用数据集ACE 2005上进行实验,结果表明实体画像增强方法能够进一步优化事件检测的性能,在触发词分类任务上的F₁值达到76.2%,较基线模型提升了2.2%。     

A knowledge model for gray scale image interpretation with emphasis on welding defect classification—An ontology based approach

Image interpretation is the process of mapping the content of the image to a real world object that is easily understandable by any user. To perform any image interpretation, the image information is extracted through feature extraction and is then mapped to the known objects of any domain. In order to retain the extracted feature information of the domain for reusability, a proper modeling of the image content is required. This helps in maximizing the leverage of knowledge in image interpretation of specific domain through a computer interpretable model which results as a knowledgebase. This paper focuses on such a modeling for gray scale image interpretation emphasizing on welding defect classification which resulted in domain ontology of welding defects. Domain ontology is created by formalizing the information related to the gray scale image and its significance in welding defects. The developed system is evaluated using industrial radiographs to detect and classify welding defects.     

Context-Free Event Domains Are Recognizable

The possibly non-distributive event domains which arise from Winskel's event structures with binary conflict are known to coincide with the domains of configurations of Stark's trace automata. We prove that whenever the transitive reduction of the order on finite elements in an event domain is a context-free graph in the sense of Müller and Schupp, the event domain may also be generated from a finite trace automaton, where both the set of states and the concurrent alphabet are finite. We show that the set of graph grammars which generate event domains is a recursive set. We obtain altogether an effective procedure which decides from an unlabeled graph grammar whether it generates an event domain and which constructs in that case a finite trace automaton recognizing that event domain.     

A model-driven approach for facilitating user-friendly design of complex event patterns

Complex Event Processing (CEP) is an emerging technology which allows us to efficiently process and correlate huge amounts of data in order to discover relevant or critical situations of interest (complex events) for a specific domain. This technology requires domain experts to define complex event patterns, where the conditions to be detected are specified by means of event processing languages. However, these experts face the handicap of defining such patterns with editors which are not user-friendly enough. To solve this problem, a model-driven approach for facilitating user-friendly design of complex event patterns is proposed and developed in this paper. Besides, the proposal has been applied to different domains and several event processing languages have been compared. As a result, we can affirm that the presented approach is independent both of the domain where CEP technology has to be applied to and of the concrete event processing language required for defining event patterns.     

基于事件本体的突发事件领域知识建模

基于本体的知识建模方法有很多,在某些特定领域采用传统的本体建模方法存在着一些不足。以突发事件领域为例,提出了基于事件本体的知识建模方法。该模型分为上层事件类、下层事件类和事件实例,上层事件类描述的抽象的事件的分类体系,下层事件类是通过事件类关系组成的事件格结构。该模型不仅可以描述事件的时间、地点、对象等要素,还能描述事件类之间的关系。采用本体建模工具Protégé来构建突发事件领域本体,并以“恐怖袭击”作为实例验证了该模型的可用性。研究结果表明,该模型可以清晰地描述事件类的完整性,语义清晰,扩展性强。     

SVAS: Surveillance Video Analysis System

This paper introduces a Surveillance Video Analysis System, called SVAS, for surveillance domain, in which the semantic rules and the definition of event models can be learned or defined by the user for automatic detection and inference of complex video events. In the scope of SVAS, an event model method named Interval-Based Spatio-Temporal Model (IBSTM) is proposed. SVAS can learn action models and event models without any predefined threshold values and generates understandable and manageable IBSTM event models. Hybrid machine learning methods are proposed and used. A set of feature models named Threshold Model, which reflects the spatio-temporal motion analysis of an event, is kept as the first model. As the second model, Bag of Actions (BoA) model is used in order to reduce the search space in the detection phase. Markov Logic Network (MLN) model, which provides understandable and manageable logic predicates for users, is kept as the third model. SVAS has high performance event detection capability due to its interval-based hierarchical manner. It determines related candidate intervals for each main model of IBSTM and uses the related main model when needed rather than using all models as a whole. The main contribution of this study is to fill the semantic gap between humans and video computer systems such that, on the one hand it decreases human intervention through its learning capabilities, but on the other hand it also enables human intervention when necessary through its manageable event model method. The study achieves all of them in the most efficient way through its machine learning methods. The proposed system is applied to different event datasets from CAVIAR, BEHAVE and our synthetic datasets. The experimental results show that our approach improves the event recognition performance and precision as compared to the current state-of-the-art approaches.