Scholarlink

<p>To address the challenges of feature redundancy and insufficient temporal dependency modeling in Internet of Vehicles (IoV) intrusion detection, we propose a lightweight intrusion detection method that integrates SOO-CFS-based feature selection with multi-scale temporal feature extraction. We embed Correlation-based Feature Selection (CFS) into the Stellar Oscillation Optimization (SOO) algorithm and introduce a Top-K mechanism to identify highly discriminative feature subsets, thereby reducing computational overhead and improving real-time detection capability. We further design a cascaded TCN-Swin network, where a Temporal Convolutional Network (TCN) captures local temporal patterns of network traffic, and a one-dimensional Swin Transformer models long-range dependencies in communication behavior. In addition, we incorporate a Squeeze-and-Excitation (SE) channel attention module to enhance the representation of critical traffic features. Experimental results on the Car-Hacking, CAN-Intrusion, and UNSW-NB15 datasets demonstrate that the proposed method achieves competitive overall detection performance while maintaining low inference latency and a compact model size, making it suitable for real-time deployment in resource-constrained IoV environments.</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

A lightweight intrusion detection framework for internet of vehicles with hybrid feature selection and multi-scale temporal modeling

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月20日

查看全文
  • Zhongjun Yang,
  • He Li,
  • Huaici Zhao,
  • Guogang Wang

摘要

To address the challenges of feature redundancy and insufficient temporal dependency modeling in Internet of Vehicles (IoV) intrusion detection, we propose a lightweight intrusion detection method that integrates SOO-CFS-based feature selection with multi-scale temporal feature extraction. We embed Correlation-based Feature Selection (CFS) into the Stellar Oscillation Optimization (SOO) algorithm and introduce a Top-K mechanism to identify highly discriminative feature subsets, thereby reducing computational overhead and improving real-time detection capability. We further design a cascaded TCN-Swin network, where a Temporal Convolutional Network (TCN) captures local temporal patterns of network traffic, and a one-dimensional Swin Transformer models long-range dependencies in communication behavior. In addition, we incorporate a Squeeze-and-Excitation (SE) channel attention module to enhance the representation of critical traffic features. Experimental results on the Car-Hacking, CAN-Intrusion, and UNSW-NB15 datasets demonstrate that the proposed method achieves competitive overall detection performance while maintaining low inference latency and a compact model size, making it suitable for real-time deployment in resource-constrained IoV environments.