基于视角-规则的深度TSK模糊分类器及其在多元癫痫脑电信号识别中的应用

在癫痫脑电信号分类检测中,传统机器学习方法分类效果不理想,深度学习模型虽然具有较好的特征学习优势,但其“黑盒”学习方式不具备可解释性,不能很好地应用于临床辅助诊断;并且,现有的多视角深度TSK模糊系统难以有效表征各视角特征之间的相关性.针对以上问题,提出一种基于视角-规则的深度Takagi-SugenoKang (TSK)模糊分类器(view-to-rule Takagi-Sugeno-Kang fuzzy classifier, VR-TSK-FC),并将其应用于多元癫痫脑电信号检测中.该算法在原始数据上构建前件规则以保证模型的可解释性,利用一维卷积神经网络(1-dimensional convolutional neural network, 1D-CNN)从多角度抓取多元脑电信号深度特征.每个模糊规则的后件部分分别采用一个视角的脑电信号深度特征作为其后件变量,视角-规则的学习方式提高了VR-TSK-FC表征能力.在Bonn和CHB-MIT数据集上, VR-TSK-FC算法模糊逻辑推理过程保证可解释的基础上达到了较好分类效果.     

面向不平衡数据的深度TSK模糊分类器

为了进一步提升Takagi-Sugeno-Kang(TSK)模糊分类器在不平衡数据集上的泛化能力和保持其较好的语义可解释性,受集成学习的启发,提出面向不平衡数据的深度TSK模糊分类器(A Deep TSK Fuzzy Classifier for Imbalanced Data, ID-TSK-FC).ID-TSK-FC主要由一个不平衡全局线性回归子分类器(Imbalanced Global Linear Regression Sub-Classifier, IGLRc)和多个不平衡TSK模糊子分类器(Imbalanced TSK Fuzzy Sub-Classifier, I-TSK-FC)组成.根据人类“从全局粗糙到局部精细”的认知行为和栈式叠加泛化原理,ID-TSK-FC首先在所有原始训练样本上训练一个IGLRc,获得全局粗糙的分类结果.然后根据IGLRc的输出,识别原始训练样本中的非线性分布训练样本.在非线性分布训练样本上,以栈式深度结构生成多个局部I-TSK-FC,获得局部精细的结果.最后,对于栈式堆叠IGLRc和所有I-TSK-FC的输出,使用基于最小距离投票原理,得到ID...     

基于脑网络和TSK模糊系统的癫痫脑电识别

脑电检测是癫痫疾病诊断的重要手段,但基于脑电信号特征的人工标记方法,对癫痫发作状态识别的准确度较低。将脑功能网络与TSK模糊系统相结合,提出一种癫痫脑电信号识别的新方法。通过分析多通道脑电信号之间的同步性,构建癫痫患者的脑功能网络,采用复杂网络方法提取特征参数;以脑网络参数为输入特征建立TSK模糊系统模型,通过监督式学习训练分类器,用于识别癫痫发作期的脑电波形。实验结果证明了该方法的有效性,模糊分类器对癫痫发作状态识别的准确度达到98.36%,99.48%敏感度和97.24%特异度。该方法将复杂网络与机器学习算法相融合,为通过脑电检测识别癫痫疾病状态提供了新方法,具有重要的应用价值。     

可解释的深度TSK模糊系统综述

深度神经网络在多个领域取得了突破性的成功,然而这些深度模型大多高度不透明。而在很多高风险领域,如医疗、金融和交通等,对模型的安全性、无偏性和透明度有着非常高的要求。因此,在实际中如何创建可解释的人工智能（Explainable artificial intelligence, XAI）已经成为了当前的研究热点。作为探索XAI的一个有力途径,模糊人工智能因其语义可解释性受到了越来越多的关注。其中将高可解释的Takagi-Sugeno-Kang（TSK）模糊系统和深度模型相结合,不仅可以避免单个TSK模糊系统遭受规则爆炸的影响,也可以在保持可解释性的前提下取得令人满意的测试泛化性能。本文以基于栈式泛化原理的可解释的深度TSK模糊系统为研究对象,分析其代表模型,总结其实际应用场景,最后剖析其所面临的挑战与机遇。     

基于深度卷积神经网络的脑电信号情感识别

为了点对点自动学习脑电信号（Electroencephalogram,EEG）空间与时间维度上的情感相关特征,提高脑电信号情感识别的准确率,基于DEAP数据集中EEG信号的时域、频域特征及其组合特征,提出一种基于卷积神经网络（Convolution Neural Network,CNN）模型的EEG情感特征学习与分类算法。采用包括集成决策树、支持向量机、线性判别分析和贝叶斯线性判别分析算法在内的浅层机器学习模型与CNN深度学习模型对DEAP数据集进行效价和唤醒度两个维度上的情感分类实验。实验结果表明,在效价和唤醒度两个维度上,深度CNN模型在时域和频域组合特征上均取得了目前最好的两类识别性能,在效价维度上比最佳的传统分类器集成决策树模型提高了3.58%,在唤醒度上比集成决策树模型的最好性能提高了3.29%。     

并行集成具有高可解释的TSK模糊分类器

针对分层Takagi-Sugeno-Kang (TSK)模糊分类器可解释性差,以及当增加或删除一个TSK模糊子分类器时Boosting模糊分类器需要重新训练所有TSK模糊子分类器等问题,提出一种并行集成具有高可解释的TSK模糊分类器EP-Q-TSK.该集成模糊分类器每个TSK模糊子分类器可以使用最小学习机(LLM)被并行地快速构建.作为一种新的集成学习方式,该分类器利用每个TSK模糊子分类器的增量输出来扩展原始验证数据空间,然后采用经典的模糊聚类算法FCM获取一系列代表性中心点,最后利用KNN对测试数据进行分类.在标准UCI数据集上,分别从分类性能和可解释性两方面验证了EP-Q-TSK的有效性.     

基于中心型TSK模糊模型的分层模糊系统

模糊系统随着输入维数的增加,其中模糊规则和辨识参数的数量将按指数级增长,针对这一问题,采用分层模糊系统是一种很好的解决方法,但分层模糊系统中各层的辨识变量没有明确的物理含义,无法进行合理的模糊化设计和解释。基于一种分层模糊系统,引用中心性TSK模糊系统思想,从而构造了一种新型的模糊系统。这种新型模糊系统保留了分层模糊系统的结构优势,极大地减少了模糊系统的模糊规则数量和辨识参数数量,又能对用到的内部参数进行很好的解释。并通过实例仿真表明基于中心型TSK模糊模型的分层模糊系统具有较好的逼近性能和更简单的结构。     

适合大规模数据集且基于LLM的0阶TSK模糊分类器

针对传统分类器的泛化性能差、可解释性及学习效率低等问题, 提出0阶TSK-FC模糊分类器.为了将该分类器 应用到大规模数据的分类中, 提出增量式0阶TSK-IFC模糊分类器, 采用增量式模糊聚类算 法(IFCM($c+p$))训练模糊规则参数并通过适当的矩阵变换提升参数学习效率.仿真实验表明, 与FCPM-IRLS模糊分类器、径向基函数神经网 络相比, 所提出的模糊分类器在不同规模数据集中均能保持很好的性能, 且TSK-IFC模糊分类器在大规模数据分类中尤为突出.     

基于数据增强的运动想象脑电分类

针对运动想象脑电（MI-EEG）多分类问题,在已有研究的基础上进行改进,构建了基于深度可分离卷积的轻量级卷积神经网络（L-Net）和轻量级混合网络（LH-Net）,并在BCI竞赛Ⅳ-2a四分类数据集上进行了实验和分析,结果表明：L-Net比LH-Net可以更快地拟合数据,训练时间更短;但LH-Net的稳定性比L-Net更好,在测试集上的分类性能具有更好的稳健性,平均准确率和平均Kappa系数比L-Net分别提高了3.6个百分点和4.8个百分点。为了进一步提升模型分类性能,采用了基于时频域的高斯噪声添加新方法对训练样本进行数据增强（DA）,并针对噪声的强度进行了仿真验证,推测出了两种模型的最优噪声强度的取值范围。仿真结果表明使用了该数据增强方法后,两种模型的平均准确率最少提高了4个百分点,四分类效果均得到了明显提升。     

监测麻醉深度的脑电信号的近似熵特征研究

近似熵是一种用来量化时间序列复杂度的新方法。实验表明脑电信号的近似熵在监测麻醉深度过程中能很好地反映全麻时的深浅程度，且具有算法简单等特点，特别适用于分析脑电等生物信号。文章主要用近似熵监测了SD大鼠的麻醉过程，结论和事实非常吻合。     

基于卷积特征和贝叶斯分类器的人脸识别

为解决传统人脸识别算法特征提取困难的问题,提出了基于卷积特征和贝叶斯分类器的人脸识别方法,利用卷积神经网络提取人脸特征,通过主成分分析法对特征降维,最后利用贝叶斯分类器进行判别分类,在ORL（olivetti research laboratory）人脸库上进行实验,获得了99.00%的识别准确率。实验结果表明,卷积神经网络提取的人脸图像特征具有很强的辨识度,与PCA（principal component analysis）和贝叶斯分类器结合之后可有效提高人脸识别的准确率。     

EEG情感识别中基于集成深度学习模型的多分析域特征融合

提出一种基于集成深度学习模型的情感状态检测方法.首先从脑电信号的时域、频域和时频域中提取4种表征情绪状态显著信息的初始特征;然后使用胶质细胞链改进的深度信念网络分别提取这些特征的高层抽象表示;最后利用判别式受限玻尔兹曼机对高层抽象特征进行融合,进行情感状态预测.在DEAP数据集上进行的实验显示,胶质链能够挖掘和利用EEG不同通道之间的相关性信息,而集成深度学习模型能够有效集成EEG信号在时域、频域和时频域蕴含的情感状态相关的显著性信息.     

An efficient fuzzy classifier with feature selection based on fuzzyentropy

This paper presents an efficient fuzzy classifier with the ability of feature selection based on a fuzzy entropy measure. Fuzzy entropy is employed to evaluate the information of pattern distribution in the pattern space. With this information, we can partition the pattern space into nonoverlapping decision regions for pattern classification. Since the decision regions do not overlap, both the complexity and computational load of the classifier are reduced and thus the training time and classification time are extremely short. Although the decision regions are partitioned into nonoverlapping subspaces, we can achieve good classification performance since the decision regions can be correctly determined via our proposed fuzzy entropy measure. In addition, we also investigate the use of fuzzy entropy to select relevant features. The feature selection procedure not only reduces the dimensionality of a problem but also discards noise-corrupted, redundant and unimportant features. Finally, we apply the proposed classifier to the Iris database and Wisconsin breast cancer database to evaluate the classification performance. Both of the results show that the proposed classifier can work well for the pattern classification application.     

基于网格特征和模糊匹配的徽标识别

针对图像智能处理中对徽标识别的应用需求,提出了一种基于网格特征和模糊匹配的徽标识别算法.该算法通过提取徽标的网格特征并映射为模糊集隶属度特征,根据模糊匹配技术进行识别,显著增强了对质量不佳图像的适应性和抗干扰性.实验表明,该算法的识别准确率能达到95.5％,并能有效提高徽标识别系统的鲁棒性.     

Hybrid one-class classifier ensemble based on fuzzy integral for open-lexicon handwritten Arabic word recognition

Pattern Analysis and Applications - One-class classifier (OCC) is involved for solving different kinds of problems due to its ability to represent a class distribution regardless the remaining...     

A faster convergence and concise interpretability TSK fuzzy classifier deep-wide-based integrated learning

Hierarchical TSK fuzzy system was proposed to approach the exponential growth of IF-THEN rules which named “fuzzy rule explosion”. However, it could not get better performance in few layers for instability of TSK fuzzy system, such that hierarchical TSK fuzzy system suffers from bad interpretability and slow convergence along with too much layers. To get a better solution, this study employs a faster convergence and concise interpretability TSK fuzzy classifier deep-wide-based integrated learning (FCCI-TSK) which has a wide structure to adopt several ensemble units learning in a meantime, and the best performer will be picked up to transfer its learning knowledge to next layer with the help of stacked generalization principle. The ensemble units are integrated by negative correlation learning (NCL). FCCI-TSK adjusts the input of the next layer with a better guidance such that it can quicken the speed of convergence and reduce the number of layers. Besides, leading with guidance, it can achieve higher accuracy and better interpretability with more simple structure. The contributions of this study include: (1) To enhance the performance of fuzzy classifier, we mix NCL and stacked generalization principle together in FCCI-TSK; (2) To overcome the phenomenon of “fuzzy rule explosion”, we adopt deep-wide integrated learning and information discarding to accelerate convergence and obtain concise interpretability in the meantime. Comparing with other 11 algorithms, the results on twelve UCI datasets show that FCCI-TSK has the best performance overall and the convergence of FCCI-TSK is also examined.     

EEG signal preprocessing for biometric recognition

Electroencephalography (EEG) has been recently investigated as a biometric modality for automatic people recognition purposes. Several studies have shown that brain signals possess subject-specific traits that allow distinguishing people. Nonetheless, extracting discriminative characteristics from EEG recordings may be a challenging task, due to the significant presence of artifacts in the acquired data. In order to cope with such issue, in this paper we evaluate the effectiveness of some preprocessing techniques in automatically removing undesired EEG contributions, with the aim of improving the achievable recognition rates. Specifically, methods based on blind source separation and sample entropy estimation are here investigated. An extensive set of experimental tests, performed over a large database comprising recordings taken from 50 healthy subjects during three distinct sessions spanning a period of about one month, in both eyes-closed and eyes-open conditions, is carried out to analyze the performance of the proposed approaches.