Ge Mengmeng, Yu Xiangzhan, Liu Likun
School of Cyberspace Science, Harbin Institute of Technology, Harbin, China.
Front Neurorobot. 2021 Mar 19;15:648374. doi: 10.3389/fnbot.2021.648374. eCollection 2021.
With the rapid popularization of robots, the risks brought by robot communication have also attracted the attention of researchers. Because current traffic classification methods based on plaintext cannot classify encrypted traffic, other methods based on statistical analysis require manual extraction of features. This paper proposes (i) a traffic classification framework based on a capsule neural network. This method has a multilayer neural network that can automatically learn the characteristics of the data stream. It uses capsule vectors instead of a single scalar input to effectively classify encrypted network traffic. (ii) For different network structures, a classification network structure combining convolution neural network and long short-term memory network is proposed. This structure has the characteristics of learning network traffic time and space characteristics. Experimental results show that the network model can classify encrypted traffic and does not require manual feature extraction. And on the basis of the previous tool, the recognition accuracy rate has increased by 8.
随着机器人的迅速普及，机器人通信带来的风险也引起了研究人员的关注。由于当前基于明文的流量分类方法无法对加密流量进行分类，其他基于统计分析的方法需要手动提取特征。本文提出了（i）一种基于胶囊神经网络的流量分类框架。该方法具有多层神经网络，能够自动学习数据流的特征。它使用胶囊向量而非单个标量输入来有效分类加密网络流量。（ii）针对不同的网络结构，提出了一种结合卷积神经网络和长短期记忆网络的分类网络结构。该结构具有学习网络流量时空特征的特点。实验结果表明，该网络模型能够对加密流量进行分类，且无需手动提取特征。并且在前一个工具的基础上，识别准确率提高了8。
