Scholarlink

<p>With the exponential proliferation of cloud computing and internet services, distributed denial of service (DDoS) attacks are becoming increasingly frequent and sophisticated. Traditional machine learning approaches for DDoS detection often suffer from suboptimal detection rates, high false positive rates, and excessive latency. To address these limitations, we propose a novel deep transfer learning framework. Specifically, we utilize the fast Fourier transform (FFT) to map high-dimensional traffic vectors from the CICDDoS2019 dataset into image representations, enabling a fine-tuned VGG19 model to extract complex attack features effectively. To mitigate the substantial computational overhead inherent in FFT transformations and deep convolutional neural network (CNN) matrix operations, we implement GPU-accelerated parallel computing techniques. This architecture significantly reduces processing latency, making real-time threat detection feasible—a capability often unattainable with conventional serial processing. Experimental results demonstrate that our method achieves superior reliability, yielding an accuracy of 99.11% and a False Positive Rate (FPR) of only 0.42%. Furthermore, the system exhibits robust real-time performance, attaining an average inference time of 0.0183&#xa0;s per sample (54.6 FPS), thereby satisfying the stringent requirements for online high-speed detection.</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

DDoS attack detection using deep transfer learning and FFT-based data transformation

  • 期刊论文

  • 发表日期：

    2026年2月10日

查看全文
  • Guohao Shen,
  • Yuejin Zhang,
  • Guoqing Zhong,
  • Zhongyuan Gong

摘要

With the exponential proliferation of cloud computing and internet services, distributed denial of service (DDoS) attacks are becoming increasingly frequent and sophisticated. Traditional machine learning approaches for DDoS detection often suffer from suboptimal detection rates, high false positive rates, and excessive latency. To address these limitations, we propose a novel deep transfer learning framework. Specifically, we utilize the fast Fourier transform (FFT) to map high-dimensional traffic vectors from the CICDDoS2019 dataset into image representations, enabling a fine-tuned VGG19 model to extract complex attack features effectively. To mitigate the substantial computational overhead inherent in FFT transformations and deep convolutional neural network (CNN) matrix operations, we implement GPU-accelerated parallel computing techniques. This architecture significantly reduces processing latency, making real-time threat detection feasible—a capability often unattainable with conventional serial processing. Experimental results demonstrate that our method achieves superior reliability, yielding an accuracy of 99.11% and a False Positive Rate (FPR) of only 0.42%. Furthermore, the system exhibits robust real-time performance, attaining an average inference time of 0.0183 s per sample (54.6 FPS), thereby satisfying the stringent requirements for online high-speed detection.