A Lightweight Edge Network Intrusion Detection System Based on MobileVit

Aiming at the current problems of limited computational and storage resources in edge networks, poor robustness of traditional machine learning algorithms, and the large resource overhead required to deploy deep learning algorithms to edge networks, we proposes a lightweight intrusion detection system based on MobileViT, which can detect cyber-attack traffic in edge networks with little resource consumption. The system uses smote to balance the dataset to make the detection model more sensitive to a small number of samples, and then transforms the network traffic into a custom grey scale graph, which reduces the time of transformation and improves the accuracy of the model, which is then fed into a modified minor-MobileViT network for training, and then tested in a test set. Experimental results with relatively few model parameters show that our proposed method has good detection performance, while the model requires less storage space and has low computational overhead, making it suitable for network traffic detection and classification under edge networks.