This paper presents a real-time CNN-LSTM-based intrusion detection system (IDS) designed for deployment on heterogeneous embedded router platforms. The proposed system uses the Gotham 2025 dataset to develop a CNN-LSTM model for binary classification of network traffic into Benign and Attack classes. The network traffic data are preprocessed and selected features are used to train and evaluate the model on a computer. The trained model is subsequently converted to ONNX format and deployed using ONNX Runtime on three heterogeneous embedded platforms: Tenda AC9, ONT X740-C, and Raspberry Pi 4. The deployed system performs network traffic capture, feature extraction, preprocessing, and CNN-LSTM inference directly on the embedded platforms. Experimental results demonstrate that the proposed model achieves high classification performance while maintaining a compact model size of approximately 763 KB. The average inference time is 8.350 ms/sample, 2.444 ms/sample, and 0.910 ms/sample on the Tenda AC9, ONT X740-C, and Raspberry Pi 4, respectively. CPU and RAM utilization are also evaluated to characterize the computational requirements of the deployed IDS. The results confirm the feasibility of performing real-time intrusion detection directly on embedded routers and demonstrate the influence of heterogeneous hardware resources on inference performance. This study provides a practical evaluation of deploying deep learning-based IDSs at the network edge under resource-constrained and heterogeneous computing environments.

Intrusion Detection System,CNN-LSTM,Embedded Router,Edge Computing,Real-Time Inference