Spectrum anomaly detection and defense algorithm based on attention mechanism residual neural network

Spectrum anomaly detection involves identifying abnormal signals within the electromagnetic spectrum. Most existing methods focus on traditional abnormal signals and can achieve high detection rates through training and testing of neural network models. However, the presence of adversarial samples greatly reduces the performance of existing network models. To address this issue, a spectrum anomaly detection algorithm based on attention mechanism residual neural network is proposed, which can significantly improve the robustness of network models without adversarial training. This method integrates the attention mechanism module and the Residual Network (ResNet), using ResNet18 to process spectrogram data, avoiding gradient vanishing and overfitting during iteration, and using attention mechanism to enhance the model's attention to local information in spectrogram data. The experimental results show that the predictive performance of the proposed algorithm is superior to the compared algorithms. Compared with ResNet18 and feature transformation based image retrieval adversarial defense algorithms, the new algorithm has increased the AUC value for abnormal signal detection by approximately 47% and 24%, respectively, effectively improving the robustness of the network model and reducing the negative impact of adversarial samples on the neural network model. The experimental results verify the effectiveness and superiority of the residual neural network model based on attention mechanism.