In recent years, with the widespread application of network encryption technologies, the proportion of encrypted traffic has continued to grow. While encryption technologies protect user privacy, they also pose severe challenges to malicious traffic detection. Current encrypted traffic detection faces the following difficulties: severe class imbalance caused by the scarcity of malicious samples, a large number of noisy labels introduced in the automated annotation process, and adversarial evasion by attackers through modifying traffic features, all of which lead to difficulties in malicious encrypted traffic detection.

To address the above problems, the Cybersecurity Team of our center has proposed the METRA framework. This framework deeply integrates network protocol domain knowledge with deep learning techniques, extracts deep patterns from limited data through protocol-aware representation learning, tackles annotation errors via a probability-driven soft label denoising mechanism, and enhances model detection robustness based on dual-dimensional adversarial training. Experiments on multiple public benchmark datasets and real-world TLS traffic datasets show that METRA significantly outperforms existing methods in detection accuracy. It demonstrates outstanding robustness in adversarial attack scenarios, and its performance advantages are more pronounced in data-scarce and high-noise environments, fully verifying the practical value of this method in real complex environments.

This research achievement has been accepted by ICASSP 2026, a recommended Class B conference by the China Computer Federation (CCF) and a top academic conference in the field of signal processing and audio technology. The first author of the paper is Wang Yaohui, a doctoral candidate at our center, and the corresponding author is Senior Engineer Long Chun. This work was jointly supported by the Young Scientist Project of the National Key R&D Program of China (2023YFB3106700), the National Key R&D Program of China (2023YFC3304704), and the Youth Innovation Promotion Association of the Chinese Academy of Sciences (2023181).

System Architecture Diagram of the METRA Framework

Related Publication：Y. Wang, D. Sun, W. Wan, J. Zhao, G. Du and C. Long, "METRA: Robust Encrypted Traffic Detection Against Adversarial Attacks via Multi-Task Learning and Label Denoising," in 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Barcelona, Spain, 2026.