An Application Traffic Identification Method Based on Deep ResNet
Yingchun Chen, Jingliang Xue, Ou Li, Fang Dong · 2022
Application traffic identification is of great significance to improve network service quality and cyberspace security. Although deep learning has made great progress in the field of traffic identification, many existing methods rely on manually designed features for identification, or rely on inflexible neural networks for limited classification, which makes the implementation of large-scale traffic identification challenging. To solve this problem, this paper proposes a method based on deep ResNet and L2-triplet loss, which learns features from raw traffic data by taking traffic data as images, and outputs traffic features as feature embeddings. Using these feature embeddings, known and unknown application traffic identification can be further realized. This paper also uses feature constraints to improve the adaptability of neural network model in traffic identification task. On the USTC-TFC2016 dataset, the proposed method achieves a good identification performance.