Securing UAVs Under Electromagnetic Interference: A CELR-Net Framework for Effect Mitigation and Experimental Verification

Abstract

In unmanned aerial vehicle (UAV) systems, external electromagnetic interference (EMI) severely threatens flight safety by disrupting the normal operations of onboard devices. Through extensive UAV flight experiments under EMI, hardware-level disruptions in onboard devices are effectively reflected via distinct flight data anomalies, and two categories consisting of five types of EMI-induced anomalies are summarized to quantitatively characterize EMI effects at the data level. Experimental results indicate that EMI-induced anomalies exhibit coupled spatial–temporal characteristics across multiple flight parameters. To extract spatial correlations caused by the varying electromagnetic sensitivity of different onboard devices, the convolutional neural network and efficient channel attention are employed. To capture the complex temporal dependencies in flight data, long short-term memory is further incorporated. Meanwhile, a residual filtering technique is employed to suppress random noise in practical measurements. Combining spatial correlation extraction, temporal dependence capture, and noise suppression, a unified framework, termed the Convolutional-Efficient Channel Attention-Long Short-Term Memory-Residual Filtering Network (CELR-Net), is proposed for data-level EMI effect mitigation. To verify the performance of the proposed CELR-Net method, experiments are conducted on both simulated and real-world platforms, and the results demonstrate superior performance in accuracy compared to some commonly used methods. Ablation studies further validate the effectiveness and contribution of each component in the proposed network architecture.