| 引用本文： |
-
毕佳琦,徐艳云,冯笑可,陈泓钢,张萌,黄伟庆.面向复杂电磁环境的无人机信号检测与型号识别方法[J].信息安全学报,已采用 [点击复制]
- Bi Jiaqi,Xu Yanyun,Feng Xiaoke,Chen Honggang,Zhang Meng,Huang Weiqing.UAV Signal Detection and Model Identification Method in Complex Electromagnetic Environment[J].Journal of Cyber Security,Accept [点击复制]
|
|
| 摘要: |
| 随着无人机在军事和民用领域的广泛应用，其安全隐患日益凸显，未经授权的无人机入侵可能对国家安全和社会稳定造成威胁。从电磁安全视角出发，建立高效的无人机信号检测与型号识别系统能够及时发现潜在威胁信号，防控非法或异常活动对无线通信环境的影响，保障电磁空间的安全可控。然而，在复杂电磁环境中，共享频段被大量宽带无线通信信号占用，无人机信号在频谱上易与其发生重叠；同时，不同型号无人机之间的时频特征高度相似，尤其是对于采用相同通信协议的设备，这进一步增加了无人机信号准确检测和精细识别的难度。针对上述挑战，本文提出了一种面向复杂电磁环境的无人机信号检测与型号识别方法。该方法首先通过自适应滤波（Adaptive Filtering, AF）对信号进行时频谱滤波与参数自适应优化，以在抑制干扰成分的同时保留无人机信号的关键结构特征；随后，多尺度残差注意力网络（Multi-scale Residual Attention Network, MRAN）结合多尺度残差结构与双模态注意力机制，从通道维度和时频空间维度联合建模，实现无人机信号的鲁棒性检测；最后，双路径交叉注意力网络（Dual-Path Network with Cross-Attention, DPNCA）利用交叉注意力机制融合局部与全局特征，实现无人机型号的精细化识别。在真实无人机信号数据集上的实验结果表明，MRAN 的检测准确率达到 98.2%，DPNCA 的型号识别准确率达到 98.6%，与现有主流方法相比，所提方法在复杂电磁环境下的性能表现具有显著优势，并在不同采集距离及开集场景下依然保持较高的稳定性与泛化能力，充分验证了其在无人机信号检测与型号识别任务中的有效性与鲁棒性。 |
| 关键词: 无人机信号检测 无人机型号识别 电磁安全 自适应滤波 注意力机制 |
| DOI： |
| 投稿时间：2025-09-11修订日期：2026-01-13 |
| 基金项目:国家重点研发计划 |
|
| UAV Signal Detection and Model Identification Method in Complex Electromagnetic Environment |
|
Bi Jiaqi1,2, Xu Yanyun1,2, Feng Xiaoke1, Chen Honggang1, Zhang Meng1,2, Huang Weiqing1,2
|
| (1.Institute of Information Engineering, Chinese Academy of Sciences;2.School of Cyber Security, University of Chinese Academy of Sciences) |
| Abstract: |
| With the widespread deployment of unmanned aerial vehicles (UAVs) in both military and civilian domains, their associated security risks have become increasingly prominent. Unauthorized UAV intrusions can pose significant threats to national security and social stability. From an electromagnetic security perspective, establishing an efficient UAV signal detection and model recognition system is essential for timely identification of potential threats, mitigating the impact of illegal or abnormal activities on wireless communication systems, and ensuring the safety and controllability of the electromagnetic spectrum. However, in complex electromagnetic environments, shared frequency bands are heavily occupied by broadband wireless signals, resulting in UAV signals frequently exhibiting spectrogram overlap. Moreover, the time–frequency characteristics of different UAV models are often highly similar, particularly for devices employing identical communication protocols, which further complicates accurate detection and fine-grained recognition. To address these challenges, this paper proposes a UAV signal detection and model recognition method for complex electro-magnetic environments. The approach first employs Adaptive Filtering (AF) with spectrogram filtering and parameter optimization to suppress interference components while preserving essential UAV signal structures. On this basis, a Multi-scale Residual Attention Network (MRAN) is designed by integrating multi-scale residual blocks with a du-al-modal attention mechanism, enabling robust UAV presence detection by jointly modeling channel-wise and time–frequency saliency. Furthermore, for UAV model recognition, a Dual-Path Network with Cross-Attention (DPNCA) is proposed to effectively fuse local discriminative features and global contextual information through cross-attention interaction, thereby enhancing the model’s ability to distinguish highly similar UAV types. Experimental results on a real-world radio frequency dataset demonstrate the method’s effectiveness. MRAN achieves a detection accuracy of 98.2%, while DPNCA reaches 98.6% for model recognition. Compared with existing state-of-the-art approaches, the proposed method exhibits superior performance in complex electromagnetic environments and maintains high stability and generalization across different acquisition distances and open-set scenarios. These results validate its effectiveness and robustness for UAV signal detection and model identification. |
| Key words: unmanned aerial vehicles signal detection unmanned aerial vehicles model identification electromagnetic safety adaptive filtering attention mechanism |
