The accelerated advancement of edge computing technology has resulted in a heightened risk of network security breaches for devices situated at the network edge. These devices, which are capable of providing immediate services and data processing, are particularly vulnerable to cyber threats. The traditional network intrusion detection system (IDS) is confronted with two significant challenges: the limitations of resources and the necessity for high real-time performance in edge computing environments. In order to effectively address these challenges, we propose an edge computing network intrusion detection method based on eBPF (extended Berkeley Packet Filter) and lightweight networks. The method employs eBPF technology to efficiently capture network traffic data at the Linux kernel level. Subsequently, the captured traffic data are converted and input into an improved lightweight MobileViT-based network model. The MobileViT model effectively addresses the neuron death problem by introducing the GELU (Gaussian Error Linear Unit) activation function in place of the traditional ReLU6, thereby improving the convergence speed and detection accuracy of the model. The experimental results demonstrate that this method markedly reduces the number of model parameters and detection delay while maintaining high detection accuracy, thereby satisfying the rigorous resource and response time requirements of edge computing environments.