ResNet-Swin Transformer based intrusion detection system for in-vehicle network

In response to the diverse and complex of attack types in in-vehicle network (IVN), an intrusion detection system based on ResNet-Swin Transformer for IVN is proposed to detect attacks on the CAN bus. ResNet model employs residual connections to extract deep features while retaining shallow information. Meanwhile, Swin Transformer model is capable of capturing both global context and local details in images. By integrating ResNet model with Swin Transformer model, shallow features are preserved along with perception of global information, thereby enhancing the feature representation capability and consequently improving detection accuracy of the model. A Focal smoothing loss function is designed to increase the weight of hard-to-classify samples to improve detection performance and further enhance generalization ability of the model. The simulation results indicate that the proposed model outperforms Swin Transformer model in terms of detection index. Compared to other models, the proposed model achieves the highest detection accuracy for various types of attacks.