Detection of artificial pornographic pictures based on multiple features and tree mode

It is easy for teenagers to view pornographic pictures on social networks. Many researchers have studied the detection of real pornographic pictures, but there are few studies on those that are artificial. In this work, we studied how to detect artificial pornographic pictures, especially when they are on social networks. The whole detection process can be divided into two stages： feature selection and picture detection. In the feature selection stage, seven types of features that favour picture detection were selected. In the picture detection stage, three steps were included. 1） In order to alleviate the imbalance in the number of artificial pornographic pictures and normal ones, the training dataset of artificial pornographic pictures was expanded. Therefore, the features which were extracted from the training dataset can also be expanded too. 2） In order to reduce the time of feature extraction, a fast method which extracted features based on the proportionally scaled picture rather than the original one was proposed. 3） Three tree models were compared and a gradient boost decision tree （GBDT） was selected for the final picture detection. Three sets of experimental results show that the proposed method can achieve better recognition precision and drastically reduce the time cost of the method.

A NOVEL INTRUSION DETECTION MODE BASED ON UNDERSTANDABLE NEURAL NETWORK TREES

Several data mining techniques such as Hidden Markov Model (HMM), artificial neural network, statistical techniques and expert systems are used to model network packets in the field of intrusion detection. In this paper a novel intrusion detection mode based on understandable Neural Network Tree (NNTree) is pre-sented. NNTree is a modular neural network with the overall structure being a Decision Tree (DT), and each non-terminal node being an Expert Neural Network (ENN). One crucial advantage of using NNTrees is that they keep the non-symbolic model ENN’s capability of learning in changing environments. Another potential advantage of using NNTrees is that they are actually “gray boxes” as they can be interpreted easily if the num-ber of inputs for each ENN is limited. We showed through experiments that the trained NNTree achieved a simple ENN at each non-terminal node as well as a satisfying recognition rate of the network packets dataset. We also compared the performance with that of a three-layer backpropagation neural network. Experimental results indicated that the NNTree based intrusion detection model achieved better performance than the neural network based intrusion detection model.

TCP/IP Feature Reduction in Intrusion Detection

Due to the amount of data that an IDS needs to examine is very large, it is necessary to reduce the audit features and neglect the redundant features. Therefore, we investigated the performance to reduce TCP/IP features based on the decision tree rule-based statistical method（DTRS）. Its main idea is to create n decision trees in n data subsets, extract the rules, work out the relatively important features in accordance with the frequency of use of different features and demonstrate the performance of reduced features better than primary features by experimental resuits.

Task-Specific Feature Selection and Detection Algorithms for IoT-Based Networks

As IoT devices become more ubiquitous, the security of IoT-based networks becomes paramount. Machine Learning-based cybersecurity enables autonomous threat detection and prevention. However, one of the challenges of applying Machine Learning-based cybersecurity in IoT devices is feature selection as most IoT devices are resource-constrained. This paper studies two feature selection algorithms: Information Gain and PSO-based, to select a minimum number of attack features, and Decision Tree and SVM are utilized for performance comparison. The consistent use of the same metrics in feature selection and detection algorithms substantially enhances the classification accuracy compared to the non-consistent use in feature selection by Information Gain (entropy) and Tree detection algorithm by classification. Furthermore, the Tree with consistent feature selection is comparable to the ensemble that provides excellent performance at the cost of computation complexity.

Feature Selection Using Tree Model and Classification Through Convolutional Neural Network for Structural Damage Detection

Structural damage detection(SDD)remains highly challenging,due to the difficulty in selecting the optimal damage features from a vast amount of information.In this study,a tree model-based method using decision tree and random forest was employed for feature selection of vibration response signals in SDD.Signal datasets were obtained by numerical experiments and vibration experiments,respectively.Dataset features extracted using this method were input into a convolutional neural network to determine the location of structural damage.Results indicated a 5%to 10%improvement in detection accuracy compared to using original datasets without feature selection,demonstrating the feasibility of this method.The proposed method,based on tree model and classification,addresses the issue of extracting effective information from numerous vibration response signals in structural health monitoring.

Novel Framework for an Intrusion Detection System Using Multiple Feature Selection Methods Based on Deep Learning

Intrusion detection systems (IDSs) are deployed to detect anomalies in real time. They classify a network’s incoming traffic as benign or anomalous (attack). An efficient and robust IDS in software-defined networks is an inevitable component of network security. The main challenges of such an IDS are achieving zero or extremely low false positive rates and high detection rates. Internet of Things (IoT) networks run by using devices with minimal resources. This situation makes deploying traditional IDSs in IoT networks unfeasible. Machine learning (ML) techniques are extensively applied to build robust IDSs. Many researchers have utilized different ML methods and techniques to address the above challenges. The development of an efficient IDS starts with a good feature selection process to avoid overfitting the ML model. This work proposes a multiple feature selection process followed by classification. In this study, the Software-defined networking (SDN) dataset is used to train and test the proposed model. This model applies multiple feature selection techniques to select high-scoring features from a set of features. Highly relevant features for anomaly detection are selected on the basis of their scores to generate the candidate dataset. Multiple classification algorithms are applied to the candidate dataset to build models. The proposed model exhibits considerable improvement in the detection of attacks with high accuracy and low false positive rates, even with a few features selected.

基于混合二进制灰狼算法的入侵检测特征选择方法

为了减少数据集的冗余特征对入侵检测分类器训练用时和检测精度的影响,对二进制灰狼算法的变异概率问题进行分析,重构逼近向量表达式,改善算法的种群变异机制,加快特征降维,减少分类器训练用时;融入粒子群算法的迭代决策形式,增强算法寻优能力;采用混合二进制灰狼算法进行包裹式特征选择,使得数据集特征结构适合于决策树分类器。经NSL-KDD数据集测试,该方法对DoS、Probe攻击流量的检测精度较好,适合用于数据平衡分布的数据集。

基于组合特征的安卓恶意软件静态检测方法研究

针对目前Android恶意软件的静态检测方法在特征选取上类型单一、同类型数量较多以及得到的检测模型效率不高等问题,提出一种基于组合特征的安卓恶意软件静态检测方法,组合特征集包含权限、组件和预见性3个方面。首先,不同方面的特征通过实验和推理方式选取并保留各自具有代表性的特征为最终特征集;其次,根据特征属性的信息增益优化决策树节点分类规则,构建检测模型;最后,采用十倍交叉验证法进行实验。实验结果表明,改进后的决策树算法模型检测准确率和检测效率均有较大提升,且在相同实验环境下检测结果优于目前流行的随机森林算法、支持向量机算法和朴素贝叶斯算法。

基于奇异谱分析与决策树的GIS振动缺陷检测方法研究

针对气体绝缘开关设备(gas insulated switchgear,GIS)在运行过程中出现的异常振动问题,文中提出了基于奇异谱分析与梯度提升决策树的GIS振动缺陷检测方法。首先通过加速度传感器采集GIS在不同状态下运行的振动信号,采用奇异谱分析的方法处理所获信号,并提取信号的主导分量,再通过时域分析和频域分析方法提取主导分量的特征参量,最后基于梯度提升决策树算法对特征进行递归与分类分析,建立了缺陷诊断模型以得到最终的GIS振动缺陷检测结果。该方法应用在实验室GIS,成功识别出了屏蔽罩松动与地脚螺栓松动两种缺陷。文中重点对所述方法及其应用进行了详细的介绍,为电气设备的故障检测研究提供了参考。

一种用于网络取证分析的模糊决策树推理方法(英文)

网络取证是对现有网络安全体系的必要扩展,已日益成为研究的重点.但目前在进行网络取证时仍存在很多挑战:如网络产生的海量数据;从已收集数据中提取的证据的可理解性;证据分析方法的有效性等.针对上述问题,利用模糊决策树技术强大的学习能力及其分析结果的易理解性,开发了一种基于模糊决策树的网络取证分析系统,以协助网络取证人员在网络环境下对计算机犯罪事件进行取证分析.给出了该方法的实验结果以及与现有方法的对照分析结果.实验结果表明,该系统可以对大多数网络事件进行识别(平均正确分类率为 91.16%),能为网络取证人员提供可理解的信息,协助取证人员进行快速高效的证据分析.

基于主成分分析和决策树的入侵检测方法

特征选择算法能够更好地提高入侵检测系统的检测速度和检测效果,消除冗余数据并减轻噪音特征.结合特征选择算法的优势,提出一种基于主成分分析(PCA)与决策树(C4.5)的入侵检测方法,进而构建出轻量级的入侵检测系统.通过在KDD1999数据集上对该方法进行详细的实验验证,证明该方法一方面确保系统有较高的检测率与较低误报率,另一方面能够比较显著地提高系统的训练时间与测试时间.同时,通过比较实验发现此方法在训练时间、测试时间、检测率、误报率上的效果也优于GA-SVM方法.

基于主成分分析禁忌搜索和决策树分类的异常流量检测方法

真实网络流量包括大量特征属性,现有基于特征分析的异常流量检测方法无法满足高维特征分析要求。提出一种基于主成分分析和禁忌搜索(PCA-TS)的流量特征选择算法结合决策树分类的异常流量检测方法,通过PCA-TS对高维特征进行特征约减和近优特征子集选择,为决策树分类方法提供有效的低维特征属性,结合决策树分类精度和处理效率高的优点,采用半监督学习方式进行异常流量实时检测。实验表明,与传统异常检测方法相比,此方法具有更高的检测精度和更低的误检率,其检测性能受样本规模影响较小,且对未知异常可以进行有效检测。

基于特征融合与决策树级联结构的多姿态人脸检测

为了提高复杂模式下多姿态人脸检测的速度与性能,提出了一种基于特征融合与决策树级联结构相结合的多姿态人脸检测方法.该方法给出了基于形态学梯度的边缘方位场特征,并提出了基于Haar-like特征与边缘方位场特征相融合的AdaboostSVM算法.通过对决策树级联结构进行改进,将特征融合的AdaboostSVM算法与改进的决策树级联结构相结合进行多姿态人脸检测.实验结果表明,该方法能明显改善复杂模式下多姿态人脸检测的速度与性能.

基于决策树的面向对象变化信息自动提取研究

为了从不同时相的遥感影像数据中自动提取变化信息且保证其效率,本文结合面向对象分析技术,提出了一种基于决策树变化信息自动提取的新方法。该方法利用影像的特征指数及形状特征、光谱特征、纹理特征等作为特征集,将其作为知识库应用到决策树控制模型中,进而利用该模型实现自动分类。对所得到的分类后影像对象,组织分析其综合属性并作为决策规则再次分类,通过"双重分类"的方式实现面向对象的遥感影像变化信息自动提取。该方法为遥感影像变化信息自动提取提供了新的思路。

基于决策树的网络流量异常分析与检测

针对现有网络流量异常检测方法准确率较低的问题,提出基于决策树的网络流量异常分析与检测方法。研究网络流量结构特征及流量异常的交叉熵表示方法。采用C4.5算法建立决策树模型,将具有连续性的属性值离散化,根据最大信息增益比逐层选取分类属性,依此规则对流量数据进行分类。实验结果表明,当该方法的检测准确率达90%以上时,误报率可控制在5%以内,与同类方法相比能更准确地发现网络流量异常并进行分类。

用于网络入侵检测的VFSA-C4．5特征选择算法

提出了一种新的用于网络入侵检测的特征选择算法——VFSA．C4．5算法。该算法采用快速模拟退火（VFSA）搜索策略对特征子集空间进行随机搜索，然后利用提供的数据在C4．5决策树上的分类错误率作为特征子集的评估标准来为入侵检测获取最优特征子集。在著名的KDD1999入侵检测数据集上进行了大量的实验，结果表明该算法相对于其它一些应用于入侵检测的特征选择算法，在保证较高检测率的前提下，可有效地降低误报率、入侵检测的计算复杂度和提高检测速度，能更适用于现实高速网络应用环境。

基于特征点识别的头部姿态计算

为了提升使用随机回归森林进行头部姿态分析的精度,提出了一种基于特征点识别分析头部姿态的计算框架.考虑到高误差投票的干扰,该计算框架以随机森林的特征点识别为基础从而避免异常投票干扰,将头部姿态计算问题转换为空间鼻尖特征点和朝向特征点的识别问题.在随机森林的训练中,决策函数使用了高斯曲率和平均曲率作为图形特征,根据微分熵的信息增益在随机生成的决策函数库中搜索最优化决策函数.在训练完成的随机回归森林的叶子节点中,通过分析保存的样本数据,可以得到目标特征点的高斯分布估计.根据实验测试结果,在适当的阈值设定的情况下,该方法可以实现较高的识别成功率,使用曲率后明显提高了识别精度,能够在一定程度上处理有遮挡的数据,并且该方法已经成功应用于虚拟座舱的实时头部姿态分析计算系统.

TCP/IP审计数据缩减技术在入侵检测中的可行性研究

目前的一些入侵检测系统是利用网络层的TCP/IP数据包里的特征进行分析建模,但TCP/IP的特征属性对检测过程的贡献不同,因而如果能够在不影响检测准确性的前提下,适当缩减特征属性的数量,那么对于提高IDS的检测率和实时性势必产生有益的影响,鉴于此该文提出基于决策树的规则统计方法(DTRS)来缩减TCP/IP的特征属性。它的基本思想是通过在n个子数据集上建立n棵决策树,提取其中的规则,根据特征属性使用频度的不同,计算出相对重要的属性,并通过实验验证了其可行性和有效性。

一种资源三号卫星影像的云量判读方法与应用

鉴于遥感影像上云的存在是影响影像质量判定的一个重要因素,对下垫面地物的识别与分类造成困难,降低了影像的利用率,分析了不同类型的云斑在资源三号卫星影像上的光谱亮度特征与直方图曲线分布特征,使用灰度共生矩阵提取云的纹理特征,依次构建云的光谱特征和纹理特征的判别准则,采用决策树分类法识别检测含云量,提出了一种针对资源三号卫星影像云量判读的算法。实验结果表明,该算法能够定性定量地判读出资源三号卫星影像上不同类型的云含量,正确率高达96%以上,具有效果好且速度快的特点。

基于Adaboost算法的汉语儿向语音检测

儿向语音对早期儿童成长有较大影响,正确检测并充分利用儿向语音具有现实意义。为此,构建一种基于Adaboost算法的汉语儿向语音检测模型,以提高检测准确率。使用决策树作为弱分类器对提取的汉语儿向语音特征进行学习,并组成弱分类器元组,同时对该弱分类器组的分类结果进行加权,区分待测语音的类别。实验结果表明,汉语儿向语音的元音持续时长超过非儿向语音的元音持续时长;提升弱分类器的数量可提高汉语儿向语音检测正确率;分段语音时间越长,汉语儿向语音检测正确率越高;采用改进的Adaboost算法比采用v-SVM算法具有更高的准确率和精度,同时可增强系统的鲁棒性。