Spatio-temporal feature-based encrypted traffic classification model ASTNet

Encrypted traffic classification can effectively supervise and manage the data traffic transmitted in the network. Most deep learning-based encrypted traffic classification models focus on modeling only one characteristic of the network, but the network traffic has both spatial and temporal characteristics, and part of the research adopts the recurrent neural network training method to grasp the temporal characteristics of the traffic, but there is a low efficiency problem when training the model on sequential data. The model is not efficient. In this paper, we propose ASTNet, an encrypted traffic classification model based on spatio-temporal features. Firstly, the session stream is cut according to the length of 784 bytes and then converted into a grayscale map as the input to the spatial feature extraction sub-module of the ASTNet model. Take out the first 8 data packets in the session stream, then intercept 256 bytes in each data packet, sort them according to the timestamp of the data packets, and use the processed time series as the input of the time feature extraction sub-module of the ASTNet model. Then the output features of the two feature modules are feature fused to obtain the spatiotemporal features of the encrypted traffic, and finally the final classification result is output by the classifier. We conducted on the public dataset ISCXVPN2016 (VPN-nonVPN dataset) experiments were conducted to compare the experimental results with the baseline method, and Experimental results show that our model achieves better results in encrypted traffic classification.