Light-STF: A Lightweight Multi-Dimensional Feature Fusion DDoS Attack Detection Model Based on SDN
In the current SDN network environment, traffic characteristics exhibit complex characteristics, including high dimensions, strong dynamics, and the coexistence of diversified attacks, which present new challenges to DDoS attack detection. To address the limitations of traditional deep learning detection methods, including large parameters, high computational complexity, and inadequate lightweight design, this paper presents a lightweight multi-dimensional feature fusion detection model for SDN environments-Light-STF. The model employs a three-branch parallel architecture that combines space, time series, and statistics to leverage the complementary information across different dimensions fully. Specifically, spatial branching utilizes deep separable convolution to achieve efficient local feature extraction. The timing branch captures the time-dependent characteristics of the traffic sequence through the Tied Skip-SRU gating unit. The statistical branch uses a lightweight multi-layer perceptron to process the overall distribution characteristics. Considering the difference in the contribution of each dimensional feature in the attack detection task, the model introduces an SE attention mechanism to adaptively weight and fuse the three-branch outputs. Finally, experiments on CICIDS2017, CICDDoS2019, and InSDN datasets show that Light-STF has excellent detection performance, and also has the advantages of few model parameters, small model size, and low computational complexity, so it is suitable for DDoS detection in an SDN environment.
Authors
- Yahong Zhai
- Longyan Xu (ORCID: https://orcid.org/0009-0002-0922-1638)
- Zhiyuan Fan (ORCID: https://orcid.org/0009-0002-4358-6884)
Institutions
- Hubei University of Automotive Technology (CN)
Publication Details
- Journal
- Engineering Research Express
- Published
- 2026-11-20
- DOI
- https://doi.org/10.1088/2631-8695/aec538
- Primary Topic
- Network Security and Intrusion Detection
- Type
- article
- Field-Weighted Citation Impact
- 0.00