A semi-supervised intrusion detection method for resource-constrained vehicular networks

Addressing the issues in traditional in-vehicle network intrusion detection methods, such as the need for a large amount of labeled data and the difficulty in detecting unknown attacks, a lightweight intrusion detection method based on semi-supervised learning and dense pseudo-labeling is proposed. It mainly involves the following three processes: Firstly, using the curriculum-based labeling algorithm to assign pseudo-labels to unlabeled data; Secondly,binary encoding the original labels and the LAN identifier fields of the data with pseudo-labels, converting them into images, and training and testing them using a dense convolutional network; Thirdly,after optimizing the hyperparameters of the dense convolutional network, classifying the mixed dataset. Experimental results show that when the labeled data is 40% of the total data volume of the Automotive Hacker Dataset and a certain intrusion detection dataset, both the accuracy and recall rates are >0.9, and the error rates are <0.1%. This method can effectively shorten detection time and reduce memory usage while detecting known and unknown attacks injected into the controller area network bus.