Generate Adversarial Network Traffic Based on TMUAP for Reversible Transmission of Data

With the constant change of network environment, traditional methods have been unable to meet the task of high-precision network traffic classification, and deep learning methods rely on efficient feature extraction and other advantages have been widely studied and applied. Network traffic classification based on deep learning can automatically extract traffic characteristics and has achieved excellent results in different classification tasks such as encrypted traffic and malicious traffic, but it also faces the threat of countering network traffic. Adversarial network traffic is generated by adding adversarial perturbation to normal network traffic, which can attack the deep learning classifier and make its classification result predict wrong. Most of the adversarial network traffic generated by existing methods destroys the structural properties of network traffic and cannot achieve reversible data transmission of network traffic. An Adversarial network traffic generation algorithm Twice Modular Operation Universal Adversarial Perturbations (TM-UAP) is proposed in this paper, which realizes the reversible data transmission of adversarial network traffic. Thus, the effective attack on the network traffic classifier and the data availability of both the sender and receiver of network traffic are realized. Finally, this paper selects the commonly used standard network traffic data sets ISCX20 16 and USTC- TFC2016 to conduct simulation experiments on LeNet5, CNN and other classification models. The results show that the proposed method can not only effectively attack network traffic classifiers, but also achieve reversible data transmission of network traffic. It provides a new method for the practical application research of countering network traffic.