A Hybrid Few-Shot Learning Based Intrusion Detection Method for Internet of Vehicles

With the rapid development of vehicle networks technologies, cyber security threats in the Internet have gradually penetrated into the Internet of vehicles. In view of the risks and challenges, this paper proposed a hybrid Meta-Learning based intrusion detection method, which core task is to distinguish normal and network flow samples as its learning task. By constructing a feature extraction network based on 3D-CNN, the characteristic values of network flow classification are learned, and then the constructed feature comparison network is used for learning and discrimination. It should be emphasized that the model can obtain enough prior knowledge to realize lightweight intrusion detection by constructing few-shot sample task training. In the experimental section, we first selected Car-Hacking dataset to evaluate the performance of the proposed method and analyze the accuracy, detection rate, precision, false positive rate and F-Score, etc., and extended the testing to the ICSX2012 dataset. The experimental results show that, the method proposed can effectively implement network intrusion detection in few-shot sample scenarios, and has the expansibility in cyber security applications.