基于样本增强的加密应用流量识别方法  

Sample-based Augmentation of Encrypted Application Traffic Identification Methods

作　　者：张萌萌 年梅[1] 张俊 ZHANG Mengmeng;NIAN Mei;ZHANG Jun(School of Computer Science and Technology,Xinjiang Normal University,Urumqi 830054;Xinjiang Institute of Physics and Chemistry Technology,Chinese Academy of Sciences,Urumqi 830011)

机构地区：[1]新疆师范大学计算机科学技术学院,乌鲁木齐830054 [2]中国科学院新疆理化技术研究所,乌鲁木齐830011

出　　处：《计算机与数字工程》2023年第11期2638-2642,共5页Computer & Digital Engineering

基　　金：新疆维吾尔自治区高校科研项目(编号:XJEDU2017S032);国家重点研发计划子课题(编号:E1182101)资助。

摘　　要：为解决网络流量数据类别不平衡导致的小类别样本识别率不高问题,提出利用基于样本增强方法平衡流量数据集,提升加密网络流量分类性能。首先,在生成对抗网络中使用深层卷积网络优化生成器,通过生成器与判别器迭代对抗训练,不断优化损失。其次,利用该生成模型扩展公开数据集中的小类别数据,使得数据集类间均衡。最后设计了一种一维卷积神经网络模型,充分挖掘流量中“会话-数据包-字节”层次结构特征,进行加密流量的应用分类。结果表明,平衡后的数据集分类效果相对于原始数据集提高3%左右,准确率达到97.86%。To solve the problem of poor recognition rate of small category samples due to unbalanced network traffic data cate⁃gories,a sample-based enhancement method is proposed to balance the traffic data set and improve the classification performance of encrypted network traffic.Firstly,a deep convolutional network is used in the generative adversarial network to optimise the gener⁃ator,and the loss is continuously optimised by iterative adversarial training of the generator and the discriminator.Secondly,the generative model is used to extend the small class data in the open dataset to make the dataset inter-class balanced.Finally,a one-dimensional convolutional neural network model is designed to fully exploit the"session-packet-byte"hierarchy in the traffic to classify the encrypted traffic applications.The results show that the classification of the balanced dataset is about 3%better than that of the original dataset,with an accuracy of 97.86%.

关 键 词：流量分类 样本增强 生成对抗网络 一维卷积神经网络 

分 类 号：TN711[电子电信—电路与系统]