Attack Detection for Intelligent Vehicles via CAN- Bus: A Lightweight Image Network Approach
Sheng Gao, Linchuan Zhang, Lei He, Xiaoyang Deng, Huilin Yin, Hao Zhang
Abstract
This article investigates the security detection mechanism of intelligent vehicles under DoS attack. A lightweight CanNet-based attack detection mechanism is developed, which is suitable for both periodic and aperiodic DoS attacks. Different from the existing attack detection mechanisms, the proposed one utilizes a designed CAN image generation scheme to convert CAN traffic data into CAN images, which makes the attack visible and traceable. And the lightweight CanNet image classification network is constructed to detect the abnormalities in the generated CAN images. In order to exploit the maximum correlation of data attributes among the acquired CAN traffic data, copula entropy, the equivalent form of mutual information, is introduced into the CAN image generation scheme. In addition, an extended CAN image color coding scheme is also designed to cope with the secure detection of traffic data beyond the 12-bit CAN ID. Finally, the effectiveness of the designed detection mechanism is validated by the comparative experiments employing the CAN traffic data from YF Sonata (the type of car).