基于注意力机制的多模态加密流量分类

Multimodal Encrypted Traffic Classification Based on Attention Mechanism

加密流量分类既可以保证网络的服务质量,也可以通过拦截恶意流量保证网络安全。针对多数深度学习加密流量分类方法没有充分利用加密流量数据的异构特性的问题,提出了一种基于注意力机制的多模态加密流量分类方法。该方法构造有效载荷模态数据和统计信息模态数据作为加密流量的多模态表示,引入了长度标识和方向标识来优化有效载荷模态数据的构造,并构建特征融合模块优化通道注意力和空间注意力的特征提取效果,提取多模态数据重要特征,从而实现准确的加密流量分类预测。实验结果表明,该加密流量分类方法的分类精确度在ISCX VPN-nonVPN公开数据集和人工采集数据集上分别达到97%和92%以上。

Encrypted traffic classification can not only ensure the quality of service of the network,but also ensure network security by intercepting malicious traffic.Aiming at the problem that most deep learning encrypted traffic classification methods do not fully utilize the heterogeneous characteristics of encrypted traffic data,a multimodal encrypted traffic classification method based on attention mechanism is proposed in this paper.This method constructs payload modal data and statistical information modal data as a multi-modal representation of encrypted traffic,introduces length flags and direction flags to optimize the construction of payload modal data,and builds a feature fusion module to optimize channel attention.The feature extraction effect of spatial attention can extract important features of multimodal data,so as to achieve accurate encrypted traffic classification prediction.The experimental results show that the classification accuracy of this encrypted traffic classification method is over 97%and 92%on the ISCX VPN-nonVPN public data set and the manually collected data set.