Research on Online Anomaly Detection Method for Electromagnetic Signals Based on Hybrid Deep Learning

In the field of dynamic electromagnetic signal anomaly detection, the contradiction between performance and real-time capability is prominent. This paper proposes a hybrid deep learning model integrating 1D CNN, Transformer, and autoencoder, and introduces an online learning mechanism. 1D CNN is used for multi-scale local feature extraction, Transformer is responsible for global dependency modeling, and the autoencoder distinguishes anomalies based on reconstruction error. Experiments show that the model achieves an accuracy of 99.25%, a recall rate of 87.37%, an F1 score of 92.94% on simulation data, with a response time of 0.0166 seconds and a comprehensive performance index of 0.9445. After introducing online learning, the accuracy in new scenarios increases by 9.36%, and the F1 score increases by 4.38%, with the response remaining stable at the millisecond level. Meanwhile, the model demonstrates good generalization under different noise levels, providing an effective solution for real-time anomaly detection in dynamic electromagnetic environments.