Scholarlink

<p>With the speedy advancement of encryption technology and the exponential increase in applications, network traffic classification has become an increasingly important research topic. Existing methods for classifying encrypted traffic have certain limitations. For example, they only extract session-level features and cannot mine the potential correlation between bytes, and traditional traffic classifiers lack attention to critical information, which makes it difficult to capture effective features between bytes and requires a large amount of resource consumption. Based on the above limitations, we propose a novel and effective classification model: fine-grained feature residual network (FGFR-Net). We constructed byte-level granularity graphs for the traffic data and used graph isomorphism network with stronger structural discrimination for graph encoding. Additionally, we design a classifier based on a residual network, incorporating deformable and depthwise separable convolutional layers. A channel attention mechanism is added between stages to capture key implicit features, improving classification performance with reduced complexity. We evaluated FGFR-Net on two public datasets, ISCX-VPNonVPN and USTC-TFC, achieving an F1-score of 99.08% and 99.21%. The results demonstrate that our method enhances encrypted traffic classification performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

FGFR-Net: An Improved Residual Network Encrypted Traffic Classification Model Based on Byte-Level Traffic Graphs

  • 期刊论文

  • 发表日期：

    2026年1月23日

查看全文
  • Yi Zhang,
  • Shanshan Wang,
  • Zhenxiang Chen,
  • Bo Yang

摘要

With the speedy advancement of encryption technology and the exponential increase in applications, network traffic classification has become an increasingly important research topic. Existing methods for classifying encrypted traffic have certain limitations. For example, they only extract session-level features and cannot mine the potential correlation between bytes, and traditional traffic classifiers lack attention to critical information, which makes it difficult to capture effective features between bytes and requires a large amount of resource consumption. Based on the above limitations, we propose a novel and effective classification model: fine-grained feature residual network (FGFR-Net). We constructed byte-level granularity graphs for the traffic data and used graph isomorphism network with stronger structural discrimination for graph encoding. Additionally, we design a classifier based on a residual network, incorporating deformable and depthwise separable convolutional layers. A channel attention mechanism is added between stages to capture key implicit features, improving classification performance with reduced complexity. We evaluated FGFR-Net on two public datasets, ISCX-VPNonVPN and USTC-TFC, achieving an F1-score of 99.08% and 99.21%. The results demonstrate that our method enhances encrypted traffic classification performance.