Scholarlink

This research introduces a new method for protecting Internet of Vehicles (IoV) systems using a state-of-the-art Intrusion Detection System (IDS) that functions at both intra-vehicle and external communication levels. Our method converts raw Controller Area Network (CAN) traffic data into image representations using a novel tabular-to-image transformation process that maintains temporal relationships and feature correlations. We compare several deep learning frameworks—namely, a handcrafted CNN model and transfer learning-based Xception, VGG16, ResNet, and EfficientNet—on extensive vehicular security datasets with normal traffic patterns and four types of attacks. We further investigate three ensemble techniques: majority voting, probability averaging, and feature concatenation for improved detection performance. Our testing methodology includes accuracy, precision, recall, F1-score metrics in addition to computational efficiency factors important for highly resource-lean IoV settings. The suggested method illustrates the effectiveness of using cutting-edge computer vision methods in network traffic analysis, yielding a strong defense mechanism against new cyber attacks in vehicular networks while being computationally feasible for real-world implementation.

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 CNN-Transfer Learning Framework for Detecting Intrusions in Vehicles

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shrishail M. Awanti,
  • M. Surya Gagan Reddy,
  • Roopa Ravish,
  • H. B. Prasad

摘要

This research introduces a new method for protecting Internet of Vehicles (IoV) systems using a state-of-the-art Intrusion Detection System (IDS) that functions at both intra-vehicle and external communication levels. Our method converts raw Controller Area Network (CAN) traffic data into image representations using a novel tabular-to-image transformation process that maintains temporal relationships and feature correlations. We compare several deep learning frameworks—namely, a handcrafted CNN model and transfer learning-based Xception, VGG16, ResNet, and EfficientNet—on extensive vehicular security datasets with normal traffic patterns and four types of attacks. We further investigate three ensemble techniques: majority voting, probability averaging, and feature concatenation for improved detection performance. Our testing methodology includes accuracy, precision, recall, F1-score metrics in addition to computational efficiency factors important for highly resource-lean IoV settings. The suggested method illustrates the effectiveness of using cutting-edge computer vision methods in network traffic analysis, yielding a strong defense mechanism against new cyber attacks in vehicular networks while being computationally feasible for real-world implementation.