C3Net: Intrusion Detection for In-Vehicle Network using CNN-ConvLSTM with CBAM

With the popularity of intelligent vehicles, there are numerous concerns regarding the security of the in-vehicle networks. The Controller Area Network (CAN) is a serial communication network with simple structure while connecting Electronic Control Units (ECUs) with high reliability. However, CAN is vulnerable to cyber-attacks due to the lack of security mechanisms, so an efficient intrusion detection system (IDS) is required. In this paper, we proposed a novel intrusion detection model named C3Net. Our model consists of Convolutional Neural Network (CNN), Long Short-Term Memory (ConvLSTM) network and Convolutional Block Attention Module (CBAM). A method converting CAN messages to images is designed to enhance data robustness while encoding potential pattern information into spatial features. CAN image sequences are then fed as input of the model and Fully Connected (FC) subnetwork is used to make classification in the end. Experiments on varying key hyperparameter values were conducted using real-vehicle datasets. Our proposed C3Net achieved excellent results with the highest F1 value of 0.9677 and an average AUC value close to 1.