Content Recognition of Network Traffic Using Wavelet Transform and CNN

Abstract

With the development of the Internet, the content types of the network traffic become more and more diverse, including video, news, music, image and so on. Traffic identification plays an important role in network management, security defense and performance optimization. Traditionally, the network traffic analysis focuses on the protocol identification and application classification, which has been well studied in the past two decades. However, as a large number of existing general protocols and legal applications can be abused to hide and transmit the data of different content types, illegal content may penetrate the traffic analysis system, and lead to inefficient network management and cause potential risks for internal networks. Different from the traditional work on the identification of the protocols or applications, in this paper, we propose a new method for recognizing the content types for the network traffic. The proposed method is based on two technologies including the wavelet transform and CNN. The wavelet transform is exploited to process the time-frequency signals of the observed network traffic that is further classified by the CNN. Experiment results are presented to validate the performance of the proposed scheme.