BNS-CNN: A Blind Network Steganalysis Model Based on Convolutional Neural Network in IPv6 Network

Abstract

There still exists the difficulties in the feature extraction and few approaches can detect multiple network steganographic algorithms currently in the IPv6 network. A unified network steganalysis model based on convolutional neural network, abbreviated as BNS-CNN, is proposed to detect multiple network storage steganographic algorithms. After preprocessing the network traffic, the model divides them by field to preserve the integrality of traffic feature to the maximum extent to build a matrix. Multiple convolution kernels and the K-max pooling are effectively combined to perform the feature extraction to speed up the model convergence; the full connection layer is designed to improve the ability of feature integration and boosts up the robustness of the model. Compared with the traditional network steganalysis method, the model can automatically extract data features and identify multiple storage covert channels at the same time. The experimental results show that the detection accuracy of BNS-CNN model is as high as 99.98% with low time complexity and favorable generalization performance.

This work was supported in part by the NSFC-General Technical Research Foundation Joint Fund of China under Grant U1536113, and in part by the CERNET Innovation Project under Grant NGII20180405.