基于卷积神经网络的网络流量识别技术研究

The Research of network traffic identification based on Convolutional neural network

摘要:

近年来，深度包检测技术和基于统计特征的网络流量识别技术迅速发展，但它们分别存在不能识别加密流量和依赖人对特征主观选择的缺陷。文章提出了基于卷积神经网络的流量识别方法，将网络数据按照一定的规则转换为灰度图像进行识别，并根据TCP数据包的有序性和UDP数据包的无序性，对原始的网络数据进行了扩展，以进一步提高识别率。实验数据表明，该方法对应用程序和应用层协议两个层次的网络流量具有较高的检测率。

Abstract:

In recent years, the deep packet inspection technology and traffic identification technology based on the statistical characteristics of data packet have developed rapidly. But they have some disadvantages. The deep packet inspection technology can’t identify the encrypted network traffic, and the other technology heavily relies on subjectively chosen statistical features. A network traffic identification method based on convolutional neural network algorithm is proposed in this paper. According to certain rules, the network data is converted to gray images. In order to improve the recognition rate, the original network data is extended according to the order of the TCP packets and the disorder of the UDP packets. Experimental data shows that this method has a high detection rate both in the application and application layer protocol.