蝙蝠算法联合选择特征和分类器参数的入侵检测

针对入侵检测的特征和分类器参数选择问题,采用极限学习机ELM(extreme learning machine)进行构建分类器,提出一种蝙蝠算法(BA)联合选择特征和分类器参数的网络入侵检测模型(BA-ELM)。首先将特征子集和极限学习机参数编码成蝙蝠个体,以入侵检测准确率和特征数加权组成个体适应度函数;然后通过个体和群体更新的规则引导蝙蝠向最优解飞行,从而找到最优的子特征集和极限学习机参数;最后建立最优的入侵检测模型,并通KDD CUP 99数据集进行仿真性能分析。结果表明,BA-ELM较好地解决了入侵检测特征选择与分类器参数不匹配难题,提高了网络入侵检测率和检测效率,更加适合于网络入侵检测的实时要求。

特征选择和分类器参数优化联合进行的人体行为识别

特征选择和分类器参数优化是提高人体行为识别率的关键技术,针对当前模型没有考虑两者之间的联系不足,为了提高人体行为的识别率,提出了一种特征选择和分类器参数优化联合进行的人体行为识别模型。首先,分析当前人体行为识别研究的现状,并建立人体行为识别特征和分类器参数优化的数学模型;然后,利用改进粒子群算法对数学模型进行求解,建立最优的人体行为识别模型;最后,通过仿真实验测试其性能。结果表明,其模型克服了人体行为识别模型的缺陷,提高了人体行为识别率,识别速度也要快于对比模型。

一种基于PSO同步进行特征选择及参数优化的核K近邻分类算法

特征选择及分类器参数优化是提高分类器性能的两个重要方面,传统上这两个问题是分开解决的.近年来,随着进化优化计算技术在模式识别领域的广泛应用,编码上的灵活性使得特征选择及参数的同步优化成为一种可能和趋势.为了解决此问题,本文研究采用二进制PSO算法进行特征选择及核K近邻分类器参数的同步优化.实验表明,该方法可有效地找出合适的特征子集及核函数参数,并取得较好的分类效果.

基于多源遥感数据的输电线走廊树种分类

针对目前树冠提取中受背景影响和易出现过度分割的问题,首先,采用可见光差异植被指数和双边滤波对传统的单木树冠分割方法进行了改进;然后,以单木树冠为对象提取多维特征,并利用XGBoost算法进行特征重要性排序和特征选择;最后,使用随机森林、支持向量机、人工神经网络3种非参数分类器,设计了12种分类方案,进行了单木树种分类和精度评价。结果表明,改进的单木分割方法可以有效提高树冠提取精度,得到的树冠分割精度在80%以上;将Li DAR数据和航空正射影像相结合,采用XGBoost算法进行特征选择后,使用ANN分类器的分类方案精度最高,总体精度为86.19%,说明多源数据协同作用和特征选择可以提高树种分类精度,在单木尺度上ANN分类器对现有树种类型的分类能力最强。

半监督技术和主动学习相结合的网络入侵检测方法

针对当前网络入侵具有多样性和易变性,单一方法很难获得理想网络入侵检测结果的问题,为提高网络入侵检测正确率,有效拦截各种网络入侵,提出一种将半监督技术与主动学习相结合的网络入侵检测方法.首先,采集网络入侵数据,提取网络入侵特征,并采用半监督技术根据特征对网络入侵数据进行聚类处理;其次,采用主动学习算法对聚类后的数据进行训练,构建网络入侵检测的分类器,并引入蚁群算法对构建网络入侵检测的分类器进行优化;最后,采用标准数据集对网络入侵检测方法进行仿真测试.测试结果表明,该方法解决了当前入侵检测方法存在的缺陷,提升了网络入侵检测正确率,漏检率和误检率明显少于经典网络入侵检测方法,同时缩短了网络入侵检测时间,改善了网络入侵检测效率,能更好地保证网络通信和数据传输安全.

Gabor滤波器和支持向量机相融合的人脸识别

人脸识别是一种具有实际应用前景的技术,针对人脸识别过程中特征提取和分类器构建等问题,提出一种基于Gabor滤波器和支持向量机相融合的人脸识别方法。首先收集人脸样本组成图像训练库,并提取人脸图像的Gabor特征,生成人脸识别数据,然后通过支持向量机对人脸图像库进行训练,建立人脸图像的分类器,最后采用ORL人脸库进行测试实验。实验结果表明,与其他人脸识别方法相比,本文方法可以实现更加精准的人脸分类与识别,对人脸识别更具有适用性。

A NON-PARAMETER BAYESIAN CLASSIFIER FOR FACE RECOGNITION

A non-parameter Bayesian classifier based on Kernel Density Estimation (KDE)is presented for face recognition, which can be regarded as a weighted Nearest Neighbor (NN)classifier in formation. The class conditional density is estimated by KDE and the bandwidthof the kernel function is estimated by Expectation Maximum (EM) algorithm. Two subspaceanalysis methods-linear Principal Component Analysis (PCA) and Kernel-based PCA (KPCA)are respectively used to extract features, and the proposed method is compared with ProbabilisticReasoning Models (PRM), Nearest Center (NC) and NN classifiers which are widely used in facerecognition systems. The experiments are performed on two benchmarks and the experimentalresults show that the KDE outperforms PRM, NC and NN classifiers.

Modeling the effects of mechanical parameters on the hydrodynamic behavior of vertical current classifiers

This study modeled the effects of structural and dimensional manipulations on hydrodynamic behavior of a bench vertical current classifier. Computational fluid dynamics （CFD） approach was used as modeling method, and turbulent intensity and fluid velocity were applied as system responses to predict the over- flow cut size variations. These investigations showed that cut size would decrease by increasing diameter and height of the separation column and cone section depth, due to the decrease of turbulent intensity and fluid velocity. As the size of discharge gate increases, the overflow cut-size would decrease due to freely fluid stream out of the column. The overflow cut-size was significantly increased in downward fed classifier compared to that fed by upward fluid stream. In addition, reforming the shape of angular overflow outlet＇s weir into the curved form prevented stream inside returning and consequently unselec- tire cut-size decreasing.

The modulation recognition based on decision-making mechanism and neural network integrated classifier

A neural network integrated classifier(NNIC) designed with a new modulation recognition algorithm based on the decision-making tree is proposed in this paper.Firstly,instantaneous parameters are extracted in the time domain by the coordinated rotation digital computer(CORDIC) algorithm based on the extended convergence domain and feature parameters of frequency spectrum and power spectrum are extracted by the time-frequency analysis method.All pattern identification parameters are calculated under the I/Q orthogonal two-channel structure,and constructed into the feature vector set.Next,the classifier is designed according to the modulation pattern and recognition performance of the feature parameter set,the optimum threshold is selected for each feature parameter based on the decision-making mechanism in a single classifier,multi-source information fusion and modulation recognition are realized based on feature parameter judge process in the NNIC.Simulation results show NNIC is competent for all modulation recognitions,8 kinds of digital modulated signals are effectively identified,which shows the recognition rate and anti-interference capability at low SNR are improved greatly,the overall recognition rate can reach 100%when SNR is12dB.

基于MLER的语音/音乐分类方法

音频分类是音频信号处理中一项重要的预处理工作。该文描述了一种基于能量的分类方法,将音频信号分为语音和音乐2种类型。分类的过程分为3个阶段,首先计算优化低能量率MLER(modified low energy ratio)作为特征,然后利用初级分类器得到初步分类的结果,最后利用音频类别的前后相关性,使用上下文分类器修正初始分类得到最终分类的结果。该文重点对MLER中参数的合理选取范围进行了讨论,并对传统的初始分类器作了改进,用非参数分类器和参数分类器代替原有的Bayes硬判决的方法,避免了由于门限选择不当所带来的分类错误。实验表明,使用参数分类器时,对纯语音和纯音乐分类效果很好,正确率达99%以上。

基于鸡群算法和核极限学习机的入侵检测模型

为了提高入侵检测系统的检测精度和效率,提出一种基于改进鸡群算法(ICSO)和核极限学习机(KELM)的入侵检测模型(ICSO-KELM)。考虑到模型中特征选择与分类器参数优化的相互影响,利用具有较好全局优化能力的改进鸡群优化算法优化核极限学习机正规化系数和核函数参数的同时选择最优特征子集。仿真实验结果表明,该方法有效地适配了入侵检测中的特征选择和分类器参数,与SVM、KNN等方法相比,其检测准确率和效率有明显提升,误报率也有所降低。