基于聚合残差网络的加密流量分类方法

Encrypted Traffic Classification Method Based on Aggregation Residual Network

互联网在线应用的迅速发展,使网络中加密流量的数量激增,复杂性增大,这对加密流量的分类问题提出了严峻的挑战。为此,提出一种基于聚合残差网络的加密流量分类方法,该方法使用的模型结合了分组卷积、特征聚合的结构以及残差网络的跳接思想,并充分发挥了一维卷积神经网络在处理一维数据时的优势,实现对加密流量的准确分类。对公开数据集"ICSX VPN-nonVPN"中12种不同类型的流量进行分类的准确率达到了98.1%,在精确率、召回率和F1分数上的均值分别达到了98.2%、97.3%和0.977。

With the rapid development of online applications in the Internet,the quantity and complexity of encrypted traffic in the network increase sharply,which poses a severe challenge to the classification of encrypted traffic.Therefore,An encrypted traffic classification method based on aggregate residual network was proposed,the model used in this method combines the structure of grouping convolution and feature aggregation,and the skip connections thoughts of residual network,at the same time,the advantage of one-dimensional convolution neural network in processing one-dimensional data is brought into play,finally achieve the accurate classification of encrypted traffic.The accuracy of this method for classifying 12 different types of traffic in the public dataset“ICSX VPN-nonVPN”reaches 98.1%,and the mean values of precision rate,recall rate and F1-score reaches 98.2%,97.3%and 0.977.