Wavelet packet feature selection for lung sounds based on optimization

In this paper, a wavelet packet feature selection method for lung sounds based on optimization is proposed to obtain the best feature set which maximizes the differences between normal lung sounds and abnormal lung sounds (sounds with wheezes or rales). The proposed method includes two main steps: Firstly, the wavelet packet transform (WPT) is used to extract the original features of lung sounds; then the genetic algorithm (GA) is used to select the best feature set. The obtained optimal feature set is sent to four different classifiers to evaluate the performance of the proposed method. Experimental results show that the feature set obtained by the proposed method provides a higher classification accuracy of 94.6% in comparison with the best wavelet packet basis approach and multi-scale principal component analysis (PCA) approach. Meanwhile, the proposed method has effective generalization performance and can obtain the best feature set without priori knowledge of lung sounds.     

Feature selection based on mutual information and redundancy-synergy coefficient

INTRODUCTION Feature subset selection (FSS) is a data miningfundamental problem to select out relevant featuresand cast away irrelevant and redundant featuresfrom an original feature set (Liu and Motoda, 1998).If a feature subset satisfies the FSS measure and hasthe minimal size, it is regarded as the optimal fea-ture subset. Complete search strategy is the way toobtain an optimal feature subset. Branch and Bound(Narendra and Fukunaga, 1977), Focus (Almuallimand Dietterich, 199…     

Feature selection for face recognition: a memetic algorithmic approach

The eigenface method that uses principal component analysis （PCA） has been the standard and popular method used in face recognition. This paper presents a PCA - memetic algorithm （PCA-MA） approach for feature selection. PCA has been extended by MAs where the former was used for feature extraction/dimensionality reduction and the latter exploited for feature selection. Simulations were performed over ORL and YaleB face databases using Euclidean norm as the classifier. It was found that as far as the recognition rate is concerned, PCA-MA completely outperforms the eigenface method. We compared the performance of PCA extended with genetic algorithm （PCA-GA） with our proposed PCA-MA method. The results also clearly established the supremacy of the PCA-MA method over the PCA-GA method. We further extended linear discriminant analysis （LDA） and kernel principal component analysis （KPCA） approaches with the MA and observed significant improvement in recognition rate with fewer features. This paper also compares the performance of PCA-MA, LDA-MA and KPCA-MA approaches.     

Intrusion detection using rough set classification

Recently machine learning-based intrusion detection approaches have been subjected to extensive researches because they can detect both misuse and anomaly. In this paper, rough set classification (RSC), a modern learning algorithm, is used to rank the features extracted for detecting intrusions and generate intrusion detection models. Feature ranking is a very critical step when building the model. RSC performs feature ranking before generating rules, and converts the feature ranking to minimal hitting set problem addressed by using genetic algorithm (GA). This is done in classical approaches using Support Vector Machine (SVM) by executing many iterations, each of which removes one useless feature. Compared with those methods, our method can avoid many iterations. In addition, a hybrid genetic algorithm is proposed to increase the convergence speed and decrease the training time of RSC. The models generated by RSC take the form of "IF-THEN" rules, which have the advantage of explication. Tests and compa     

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.     

Improved fast inra prediction algorithm of H.264/AVC

With advanced prediction modes of intra prediction, intra coding of H.264/AVC offers significant coding gains compared with previous video coding standards. It uses an important tool called Lagrangian rate-distortion optimization (RDO) technique to decide the best coding mode for a block, but the computational burden is extremely high. In this paper, we proposed an improved fast intra prediction algorithm including block type selection and mode decision algorithm based on analysis of edge feature of a block. Our algorithm filters out unlikely block type and candidate modes to reduce the RDO calculations. Experimental results showed that the proposed algorithm can reduce the computation complexity of intra prediction from 52.90% to 56.31%, with 0.04 dB PSNR degradation and 2% increase of bit rate.     

A Multi-Objective Genetic Algorithm for Optimal Portfolio Problems

This paper concerns with modeling and design of an algorithm for the portfolio selection problems with fixed transaction costs and minimum transaction lots. A mean-variance model for the portfolio selection problem is proposed, and the model is formulated as a non-smooth and nonlinear integer programming problem with multiple objective functions. As it has been proven that finding a feasible solution to the problem only is already NP-hard, based on NSGA-II and genetic algorithm for numerical optimization of constrained problems (Genocop), a multi-objective genetic algorithm (MOGA) is designed to solve the model. Its features comprise integer encoding and corresponding operators, and special treatment of constraints conditions. It is illustrated via a numerical example that the genetic algorithm can efficiently solve portfolio selection models proposed in this paper.This approach offers promise for the portfolio problems in practice.     

Joint application of feature extraction based on EMD-AR strategy and multi-class classifier based on LS-SVM in EMG motion classification

This paper presents an effective and efficient combination of feature extraction and multi-class classifier for motion classification by analyzing the surface electromyografic(sEMG) signals. In contrast to the existing methods,considering the non-stationary and nonlinear characteristics of EMG signals,to get the more separable feature set,we introduce the empirical mode decomposition(EMD) to decompose the original EMG signals into several intrinsic mode functions(IMFs) and then compute the coefficients of autoregressive models of each IMF to form the feature set. Based on the least squares support vector machines(LS-SVMs) ,the multi-class classifier is designed and constructed to classify various motions. The results of contrastive experiments showed that the accuracy of motion recognition is improved with the described classification scheme. Furthermore,compared with other classifiers using different features,the excellent performance indicated the potential of the SVM techniques embedding the EMD-AR kernel in motion classification.     

Patch-guided facial image inpainting by shape propagation

Images with human faces comprise an essential part in the imaging realm. Occlusion or damage in facial portions will bring a remarkable discomfort and information loss. We propose an algorithm that can repair occluded or damaged facial images automatically, named ‘facial image inpainting＇. Inpainting is a set of image processing methods to recover missing image portions. We extend the image inpainting methods by introducing facial domain knowledge. With the support of a face database, our approach propagates structural information, i.e., feature points and edge maps, from similar faces to the missing facial regions. Using the interred structural information as guidance, an exemplar-based image inpainting algorithm is employed to copy patches in the same face from the source portion to the missing portion. This newly proposed concept of facial image inpainting outperforms the traditional inpainting methods by propagating the facial shapes from a face database, and avoids the problem of variations in imaging conditions from different images by inferring colors and textures from the same face image. Our system produces seamless faces that are hardly seen drawbacks.     

主流形和非线性维数化简的切向空间校正

We present a new algorithm for manifold learning and nonlinear dimensionality reduction. Based on a set of unorganized da-ta points sampled with noise from a parameterized manifold, the local geometry of the manifold is learned by constructing an approxi-mation for the tangent space at each point, and those tangent spaces are then aligned to give the global coordinates of the data pointswith respect to the underlying manifold. We also present an error analysis of our algorithm showing that reconstruction errors can bequite small in some cases. We illustrate our algorithm using curves and surfaces both in 2D/3D Euclidean spaces and higher dimension-al Euclidean spaces. We also address several theoretical and algorithmic issues for further research and improvements.     

基于LDA变换的验证码识别技术初探

本文采用基于递归算法的去除离散点法消除孤立噪声,选用扫描边界的方法分割字符,来研究验证码自动识别技术,选择和提取稳定而又便于表示的特征向量是本系统的核心之一。本文提出了简单的字符特征提取方法：采用网格灰度特征并对该特征进行线性鉴别分析（LDA,Linear discrimlnant analysis）变换,结合最小距离分类器完成字符识别过程,通过提高训练样本数,有效解决了形近字符识别率低的问题,取得了很好的识别效果。     

一种垂直分布环境下的特征选择及规则提取算法

特征选择及规则提取是数据挖掘过程中的重要环节.Rough集理论提供了一种新的属性约简即特征选择及规则提取工具,但目前Rough集理论研究主要针对单个决策表(或信息系统),分布式环境下的粗糙集理论研究还不多见.文章提出一种垂直分布环境下的特征选择及规则提取算法,算法分析结果表明,该种算法是有效可行的.     

结合Delaunay三角网的自适应多尺度图像重叠域配准方法

为改进随机采样一致性算法模型参数估计可能不是最优导致图像特征点配准率不高的问题,缩短特征点提取时间,提出一种结合Delaunay三角网格约束的自适应多尺度图像重叠域配准方法。采用自适应通用加速分割检测算法,快速检测出均匀稳定的特征点,并且用二进制特征描述子解决尺度不变性和旋转不变性问题。因传统随机采样一致性算法阈值选取和迭代次数的局限性,会掺杂部分难筛的误配点,在此前提下借助Delaunay算法剖分粗匹配点集,遍历计算网格间对应三角形相似度并储存在相似性度量矩阵中。依据Delaunay三角网特性,剔除相似度差异大的三角形,重构网格保存余下的匹配点集。实验结果表明,该方法特征点提取速率比FAST快15%~20%,特征点正确配准率比随机采样一致性算法提高约4.9%,不仅可自适应多尺度快速提取特征点,而且在保证特征点正确配准率基础上尽量多地保留有效特征点数量。     

改进遗传算法在文本聚类中的应用研究

分析了K均值聚类算法（K-means）存在的不足和改进遗传算法的全局优化能力,提出一种基于改进遗传算法的文本聚类方法,该方法将原始文档转化成用向量空间模型来描述的文本向量,首先随机产生若干个文档向量作为初始聚类中心形成遗传算法的染色体种群,经过改进遗传算法的选择、交叉、变异进化运算,得到较为优化的K均值聚类算法的初始聚类中心。实验表明该算法文本聚类提高了查准率和查全率,算法的高效性也得到了验证。     

基于PCA的决策树优化算法

为了改善传统ID3算法在分类属性选择上存在多值偏向性的不足,提出基于PCA的决策树优化算法。在普通基于PCA 的决策树改进算法中,存在数据经降维处理后代表性不强的问题,导致算法需经过多次数据运行后,准确率才能小幅提升。在ID3算法基础上,在分类前两次提取属性特征值,并计算了需要分类的数据量,也即对原始数据进行最重要的属性选择。在子树建立之后,再进行数据的降维合并选择。采用UCI数据库中的3个数据集对改进算法进行验证,结果表明改进算法的平均准确率达到94.6%,相比传统ID3算法与普通PCA决策树优化算法分别提升了1.6%和0.6%。因此,基于PCA的决策树算法能在一定程度上提升结果准确率,具备一定的应用价值。     

Android平台下恶意软件分析与检测

针对Android平台下恶意软件侵扰问题,提出一种基于权限—敏感API特征的加权朴素贝叶斯分类算法的检测方案。首先对Android应用程序中的配置文件进行解析,然后利用Apktool工具对APK文件进行反编译,提取出权限—敏感API特征集,并通过信息增益算法和卡方检验算法过滤冗余数据,最后利用加权朴素贝叶斯分类算法的恶意软件检测模型进行分类判断。实验结果证明,该系统能有效提高分类器的效率和恶意软件的检测率。     

一种启发式聚类特征选择方法

为了使聚类可以在低维数据空间中进行,提出了一种新的聚类特征选择方法.该方法分3个步骤,每个步骤都在一个wrapper框架中执行.首先,将所有原始特征都按照重要性进行排序,引入一个特征重要性评价函数E(f);然后,顺序地选择特征组成重要特征子集;最后,去除重要特征子集中可能存在的冗余特征.由于是顺序选择特征而不是在巨大的特征子集空间中进行搜索,因此算法效率很高.实验结果表明该方法可以找出有助于聚类的重要特征子集,并且可以去掉那些不利于聚类的特征.     

基于类别的CHI特征选择方法

文本分类问题中,卡方特征选择是一种效果较好的特征选择方法。计算单词的卡方值时,先计算单词针对每个类别的卡方值,再通过类别概率将卡方值调和平均,作为单词相对于整个训练集合的卡方值,这种全局方法忽视了单词和类别间的相关性。针对这一问题,提出基于类别的卡方特征选择方法。基于类别的方法针对每个类别遴选特征词,特征词数量根据事先设定的阈值、类别的文档数和整个训练集合文档数计算得到,不同类别的特征空间可能包含相同的特征词。采用KNN分类方法,将基于类别的方法与全局方法进行比较,实验结果表明,基于类别的方法能够提高分类器的总体性能。