一种基于Bagging和混淆矩阵的自适应选择性集成

为了平衡集成学习中差异性和准确性的关系并提高学习系统的泛化性能,提出一种基于Bagging和混淆矩阵的选择性集成方法.基本思想是通过扰动训练集和特征空间生成基分类器,根据每一个基分类器的混淆矩阵构造一个基分类器间相关性的度量矩阵;然后基于相关性度量矩阵对基分类器集合进行子集划分,在每个划分中选择一个基分类器参与集成;最后用多数投票法融合所选基分类器的决策结果,并通过仿真实验验证该方法的有效性.     

基于混淆矩阵的全方位角雷达目标识别

基于高分辨径向距离像HRRP(High Resolution Range Profile)的目标识别一直是雷达目标识别研究的重要方向.HRRP的目标姿态敏感性极大地影响了识别性能,尤其是全方位角目标识别的性能.本文提出一种基于混淆矩阵的分类方法,采用支持向量机(SVM)作为基本的两类分类器(Binary Classifier),使用HAC(Hierarchical Agglomerative Clustering)构造了一个基于"错误纠正"策略的两层层次化分类器(Hierarchical Classifier).实验表明,在复杂度增加不大的情况下,识别性能得到了相当程度的提高.     

基于混淆交叉的多分类支撑向量机树

本文针对复杂模式分类和多分类问题,提出了一种基于混淆交叉的多分类支撑向量机树模型,其整体结构为二叉树,在树的每个中间节点上嵌入了支撑向量机。在训练阶段,引入混淆交叉因子,在同属一个父节点的中间节点样例间进行样例的混淆交叉,将那些对分类曲面有较大影响的样例纳入树型结构更深层次的训练过程,参与更精细的分类超曲面的构建。本文将提出的支撑向量机树与现有的其他方法作了比较,实验结果说明了本文提出的模型在解决复杂模式识别问题及多分类问题上具有高效准确性及优越的泛化性能。     

基于控制流混淆的软件保护方法研究

软件盗版、篡改和逆向工程是软件安全的主要威胁。逆向工程师利用逆向分析技术可以理解软件的行为,并从中提取核心算法和重要数据结构。针对目前大部分的混淆方法难以抵御动态攻击的缺点,文中提出一种基于控制流图多样化的代码混淆方法。实验结果表明,该方法不仅能够有效降低静态反汇编分析准度,同时能够在一个合理的性能开销之内增加动态逆向分析的难度,从而使混淆后的程序具有更高的安全性。     

对Zernike模式法重构19单元哈特曼测量波前的研究

通过对19单元哈特曼测量和模式法重构波前的过程进行数值模拟,研究了Zern ike模式法重构波前时耦合和混淆出现的原因、条件、程度和数学算法的表现。确定了在综合考虑最高可识别模式和计算稳定性同计算工具关系的情况下,选择最佳重构阶数的基本原则。同时对19单元哈特曼复原精度进行了评价,得出模式耦合是影响精度的主要因素,得到了模式法波前测量的适用条件,为利用哈特曼进行光束质量诊断提供了依据。     

一种自整定权值的多分类器融合方法

 本文提出一种自整定权值的融合方法.该方法使用混淆矩阵来衡量分类器性能,并根据分类器输出情况自适应地为各分类器赋予权值,可靠的决策结果获得较大的权值,从而提高决策模板的可信度.对易于被错误分类的样本,在利用其与决策模板的相似性信息的同时,结合它周围的训练样本信息做出判断.通过与DT方法在KDD’99入侵检测数据集和UCI数据库中的8个数据集上的实验对比,表明本文方法具有更好的分类性能.     

经验模态分解中多种边界处理方法的比较研究

经验模态分解(EMD)的一个关键问题是处理边界效应。尽管目前除了Huang申请了NASA专利的边界处理方法,仍没有一个最终的解决方案,但工程上已经提出了多种处理方法。本文实现了工程上常用的5种EMD边界处理方法:线性外延,多项式拟合,镜像法,径向基(RBF)神经网络预测和AR预测方法,设计了一套消除了EMD处理中信号的相互作用及模式混淆影响的测试方法,并利用准周期信号和随机信号对它们的边界效应处理结果进行了定量测试。结果表明镜像法是目前相对最优的EMD边界处理方法。     

混淆恶意JavaScript代码的多特征检测识别与分析

JavaScript目前已经成为交互式网页和动态网页中一项广泛采用的技术,恶意的JavaScript代码也变得活跃起来,已经被当作基于网页的一种攻击手段.通过对大量JavaScript恶意代码的研究,对混淆恶意JavaScript代码进行特征提取与归类,从基于属性特征、基于重定向特征、基于可疑关键词特征、基于混淆特征四大类中总共提取了82个特征,其中47个是四大类中的新特征.从真实环境中收集了总数为5525份JavaScript正常与混淆的恶意代码用于训练与测试,利用多种有监督的机器学习算法通过异常检测模式来评估数据集.实验结果表明,通过引入新的特征,所有分类器的检测率较未引入新特征相比有所提升,并且误检率(False Negative Rate)有所下降.     

基于自适应类别权重的HRRP决策融合识别

针对雷达高分辨率距离像(HRRP)识别中因特殊样本和分类器误判而出现的错误分类问题,提出了一种基于自适应类别权重的多分类器决策融合识别方法。该方法结合K-最近邻思想,利用最近邻和相似度准则挑选与测试样本对应的训练样本集,构造混淆矩阵自适应完成分类器置信度的计算和筛选,最终获得目标各类别权重,输出分类结果。基于实测数据的研究结果表明,相较于以上任意单个分类器和传统决策融合方案,文中提出的融合识别方法识别率有明显提高,并且随着噪声的增大,该方法的优势愈加突显。     

基于多级协同混淆的硬件IP核安全防护设计

传统硬件混淆从物理级、逻辑级、行为级等进行单层次混淆,没有发挥多级协同优势,存在安全隐患。该文通过对物理版图、电路逻辑和状态跳变行为的关系研究,提出多级协同混淆的硬件IP核防护方法。该方案首先在自下而上协同混淆中,采用虚拟孔设计版图级伪装门的方式进行物理-逻辑级混淆,采用过孔型物理不可克隆函数(PUF)控制状态跳变的方式实现物理-行为级混淆;然后,在自上而下协同混淆中,利用密钥控制密钥门进行行为-逻辑级混淆,利用并行-支路混淆线的方法完成行为-物理级混淆;最后提出混淆电路在网表的替换机制,设计物理-逻辑-行为的3级协同混淆,实现多级协同混淆的IP核安全防护。ISCAS-89基准电路测试结果表明,在TSMC 65 nm工艺下,多级协同混淆IP核在较大规模测试电路中的面积开销占比平均为11.7%,功耗开销占比平均为5.1%,正确密钥和错误密钥下的寄存器翻转差异低于10%,所提混淆方案可有效抵御暴力攻击、逆向工程、SAT等攻击。     

基于模糊规则的多分类器融合

用非线性方法解决多分类器融合问题能够取得比较高的识别率, 但是,当前被应用在多分类器融合领域中的非线性方法可理解性较差,给使用者带来一定的困难。而基于模糊规则的模式识别方法是一类可理解性好的非线性方法,但迄今为止还没有被应用于多分类器融合问题中。基于上述考虑,该文将模糊系统应用到多分类器融合中,并且研究了如何设计可理解性好、精度高的模糊系统的问题,提出了一种改进的基于支持向量的模糊系统设计方法。该方法在从ELENA项目数据库和UCI数据库中选出的4个数据集上进行了测试。实验结果表明,该方法能够用可理解性好的模糊系统实现低错误率的多分类器融合。     

An efficient image encryption scheme based on double affine substitution box and chaotic system

Privacy of information is one of the most important and unavoidable issues in our digitally advance era. A huge amount of information transmitted over different servers and networking protocols. The sphere of digitally advanced world is tied with information in different forms of facilitations, which includes online banking systems, ecommerce and many more. Providing the ease of access, anything online makes our confidential information open to different threats. Therefore, to enjoy ease of access and at the same time secure our digital information from theft, we need a robust information security system. In this article, we have designed a novel and an efficient security system, which provides secrecy to our digital information. The designed encryption scheme is fundamentally a combination of chaos and nonlinear confusion components. We have developed a new mechanism of adding confusion, namely S₈ permutation of double affine transformation to construct 40320 new substitution boxes (S-boxes) having nonlinearity 112 from a single S-box. Moreover, nonlinear Lorenz dynamical system is utilized to select any three S-boxes from 40320 newly generated nonlinear components. To add diffusion in our proposed algorithm, we have utilized Chirikov discrete iterative map. The excellence of an offered digital image encryption has been examined and evaluated with standard benchmarks. The simulation results reveal that the quality of the image encryption passes all these tests and is comparable to current benchmarks.     

Consensual and Hierarchical Classification of Remotely Sensed Multispectral Images

Consensual and hierarchical approaches are developed for the classification of remotely sensed multispectral images. The proposed method consists of preprocessing of input patterns, generating multiple classification results by hierarchical neural networks, and a combining scheme to generate a consensus of multiple classification results. Transformations of input patterns by random matrices and nonlinear filtering are used for preprocessing. By varying the input patterns, the multiple classification results are generated with sufficiently independent errors by using a single type of classifier. This helps to improve classification performance when the multiple classification results are combined. Hierarchical neural networks involve the use of successive classifiers which are tuned to reduce the remaining errors to increase the classification performance. This structure includes detection schemes to decide whether successive classifiers are utilized for each input. Consensual and hierarchical approaches generate more reliable and accurate results based on group decision.     

基于K-L散度的最大后验弧主导的混淆网络生成算法

为快速生成高质量混淆网络,该文提出一种最大后验弧主导的快速生成算法。它只需遍历一遍Lattice,具有线性时间复杂度。采用K-L散度(Kullback-Leibler Divergence,KLD)来度量弧标号之间的发音相似性,改善了混淆网络生成中弧对齐的准确性。实验结果显示,所提算法在生成速度上和Xue的快速算法是可比的,而生成质量更好。通过采用KLD作为弧标号相似性测度,生成混淆网络的质量得到了进一步提高。     

Anomaly detection model based on multi-grained cascade isolation forest algorithm

The isolation-based anomaly detector,isolation forest has two weaknesses,its inability to detect anomalies that were masked by axis-parallel clusters,and anomalies in high-dimensional data.An isolation mechanism based on random hyperplane and a multi-grained scanning was proposed to overcome these weaknesses.The random hyperplane generated by a linear combination of multiple dimensions was used to simplify the isolation boundary of the data model which was a random linear classifier that can detect more complex data patterns,so that the isolation mechanism was more consistent with data distribution characteristics.The multi-grained scanning was used to perform dimensional sub-sampling which trained multiple forests to generate a hierarchical ensemble anomaly detection model.Experiments show that the improved isolation forest has better robustness to different data patterns and improves the efficiency of anomaly points in high-dimensional data.     

Intelligent Decision Making System for Digital Modulation Scheme Classification in Software Radio Using Wavelet Transform and Higher Order Statistical Moments

This paper proposes a neural network (NN) based intelligent decision making system for digital modulation classification using wavelet transform, histogram peak and higher order statistical moments. The decision making system is developed to classify the modulation schemes buried in additive white Gaussian noise and channel interference utilizing NN classifier. The performance is verified and validated for M-ary PSK, M-ary FSK, M-ary QAM and GMSK modulation schemes by examining the receiver operating characteristics, confusion matrix and probability of correct identification for various signal-to-noise ratios (SNR) and also for various decision parameters. The performance of the proposed system also has been compared with existing methods and found that this method can be considered as reliable classification method for Digital Modulation Scheme with lower SNR upto  − 5 dB.

Voronoi域模式分类器

提出基于模式训练的Voronoi划分向量分类器结构,指出该结构是以欧几里得距离为测量的最优模式分类器,给出M类k维(k2)Voronoi域分类器的两个判决准则和三种结构。     

基于二维加权主元分析的人脸识别研究

提出了一种基于二维加权主元分析的方法进行人脸识别。该方法考虑了人脸的不同部位所包含的识别信息量不同，对人脸的不同部位赋予不同的权重，并结合二维主元分析方法求解加权子空间，然后将人脸样本向该子空间进行投影来提取人脸特征，最后采用最近邻距离分类器进行分类。该方法在NUST603人脸图像库中进行了实验，实验结果表明了该方法的有效性。     

Exploiting spatial correlation features for SAR image analysis

Spatial information is of great importance in Synthetic Aperture Radar (SAR) image analysis and recently, many methods have been developed that take this feature into account. This paper deals with a supervised approach to SAR image classification that exploits spatial features within a hierarchical classification framework. In contrast to the classical approach, which makes the hypothesis about sample data independence, in the proposed method, the spatial dependence of neighboring pixels is taken into account to estimate relatively simple statistical features such as sample spatial mean and sample spatial variance, thus allowing contextual information to be easily handled. The Bhattacharyya distribution distance is used during the training phase, and the generated tree is applied during the test phase. After this, both phases are based on the proposed features. As a result, second-order statistics play a major role in the present classification problem. Experimental results on different SAR data sets are reported. It is shown that the accuracy of the proposed method is better than that of the hit classifier and that the new method is also computationally more convenient