Scholarlink

<p>Vehicular Ad Hoc Networks (VANET) are highly vulnerable to semantic attacks, where adversaries manipulate the meaning of safety messages while preserving protocol compliance, making detection particularly challenging. Existing VANET intrusion detection systems predominantly rely on rule-based validation, convention deep learning (DL) models with limited temporal awareness. Existing federated learning (FL) frameworks that focus on privacy preservation but fail to capture semantic-level inconsistencies in V2V communications. To address these issues, this paper proposes a novel FL assisted hybrid DL model for semantic attack detection in VANETs. Unlike prior FL-based approaches that employ standalone CNN, RNN or shallow models, the proposed approach introduces a Hybrid One-Dimensional Vision Transformer-based Residual Dilated Convolutional Network (Hy-OneVReTr) for multi-scale local feature extraction and ViT to model long-range temporal and contextual dependencies in V2V message sequences. Furthermore, a Random replace Chaotic Greylag Goose Optimization (RanCh-GLOp) is used to fine-tune the Hy-OneVReTr parameters, which effectively mitigates convergence stagnation and non-IID data challenges in vehicular environments. The framework preserves data privacy by avoiding raw data exchange while enabling collaborative learning across distributed vehicles. Extensive experiments conducted on the ETSI-V2V dataset demonstrate that the proposed framework consistently outperforms state-of-the-art models such as CNN, random forest, XG Boost and recent FL based intrusion detection models. The proposed model achieves an accuracy of 98.6% along with superior precision and recall, while maintaining low positive and false negative rates. These results confirm the effectiveness and robustness of the proposed framework for privacy-preserving and real time semantic attack detection in VANET 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

Federated learning-assisted hybrid 1D vision transformer with chaotic random replacement for semantic attack detection in VANETs

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • Divya Babu,
  • Terli Sankara Rao

摘要

Vehicular Ad Hoc Networks (VANET) are highly vulnerable to semantic attacks, where adversaries manipulate the meaning of safety messages while preserving protocol compliance, making detection particularly challenging. Existing VANET intrusion detection systems predominantly rely on rule-based validation, convention deep learning (DL) models with limited temporal awareness. Existing federated learning (FL) frameworks that focus on privacy preservation but fail to capture semantic-level inconsistencies in V2V communications. To address these issues, this paper proposes a novel FL assisted hybrid DL model for semantic attack detection in VANETs. Unlike prior FL-based approaches that employ standalone CNN, RNN or shallow models, the proposed approach introduces a Hybrid One-Dimensional Vision Transformer-based Residual Dilated Convolutional Network (Hy-OneVReTr) for multi-scale local feature extraction and ViT to model long-range temporal and contextual dependencies in V2V message sequences. Furthermore, a Random replace Chaotic Greylag Goose Optimization (RanCh-GLOp) is used to fine-tune the Hy-OneVReTr parameters, which effectively mitigates convergence stagnation and non-IID data challenges in vehicular environments. The framework preserves data privacy by avoiding raw data exchange while enabling collaborative learning across distributed vehicles. Extensive experiments conducted on the ETSI-V2V dataset demonstrate that the proposed framework consistently outperforms state-of-the-art models such as CNN, random forest, XG Boost and recent FL based intrusion detection models. The proposed model achieves an accuracy of 98.6% along with superior precision and recall, while maintaining low positive and false negative rates. These results confirm the effectiveness and robustness of the proposed framework for privacy-preserving and real time semantic attack detection in VANET environments.