The battlefield environment is characterized by rapid dynamics and highly complex electromagnetic conditions, posing significant challenges for radar systems. In particular, accurately and efficiently identifying active jamming signals from limited sample data under strong electromagnetic jamming remains a critical issue. To address this challenge, we propose a novel P-MSGC-AM-CNN network algorithm. Firstly, a bottleneck-structured multi-scale ghost convolution is integrated with channel attention and spatial attention modules to construct the MSGC-CA and MSGC-SA modules, thereby enhancing multi-scale feature extraction while simultaneously strengthening the representation of channel-wise and spatial features. Secondly, these modules are further optimized through the adoption of asymmetric convolutions and the replacement of fully connected layers, effectively reducing model parameters and mitigating the risk of overfitting. Finally, a systematic analysis was conducted on the impact of network parameters on both the recognition accuracy of active jamming signals and training time. Furthermore, the proposed P-MSGC-AM-CNN network algorithm was comprehensively evaluated through comparative experiments against the classic ResNet18 network and three network architectures reported in the existing literature. The simulation results demonstrate that in a strong noise environment with a jamming-to-noise ratio of -10 dB, the P-MSGC-AM-CNN algorithm achieves an average recognition accuracy exceeding 89.8% across 28 types of active jamming signals, outperforming other comparative networks in recognition accuracy. The method is particularly well-suited for resource-constrained real-time applications and exhibits significant potential for practical engineering deployment.
Zhao et al. (Tue,) studied this question.