| 引用本文: | 普运伟，李文康，何娅琳，田春瑾. 基于模糊函数多域特征自适应融合的雷达辐射源信号识别[J]. 雷达科学与技术, 2026, 24(3): 332-342.[点击复制] |
| PU Yunwei, LI Wenkang, HE Yalin, TIAN Chunjin. Radar Emitter Signal Recognition Based on Multi-Domain Feature Adaptive Fusion with Ambiguity Function[J]. Radar Science and Technology, 2026, 24(3): 332-342.[点击复制] |
|
| 摘要: |
| 针对当前复杂体制雷达辐射源信号识别方法特征利用不充分、抗噪性能差等问题，提出了一种基于模糊函数多域特征自适应融合的识别方法。首先采用积分加速的基于结构张量的自适应非局部均值（ST-NLM）算法对信号模糊函数进行去噪处理，通过结构张量自适应调节滤波系数；然后从时延域、多普勒频域和时延-多普勒联合域提取多域投影特征；最后构建残差神经网络+自适应注意力特征融合（ResNet+AAFF）模型，利用多尺度空洞卷积、深度可分离卷积和高效通道注意力（ECA）注意力机制实现特征自适应融合。实验结果表明，该方法在信噪比为0 dB以上均能保持100%的准确率，即使在信噪比为-4 dB时，识别率仍可达98.12%。验证了所提出方法在低信噪比下具有一定的有效性和可行性。 |
| 关键词: 雷达辐射源信号 模糊函数 多域特征融合 深度学习 注意力机制 |
| DOI：DOI:10.3969/j.issn.1672-2337.2026.03.010 |
| 分类号:TN974 |
| 基金项目:国家自然科学基金（61561028）； 昆明理工大学人培基金（KKZ3202403190） |
|
| Radar Emitter Signal Recognition Based on Multi-Domain Feature Adaptive Fusion with Ambiguity Function |
|
PU Yunwei, LI Wenkang, HE Yalin, TIAN Chunjin
|
|
1. School of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, China;2. Library of Kunming University of Science and Technology, Kunming 650500, China
|
| Abstract: |
| To address the problems of insufficient feature utilization and poor noise resistance in current radar emitter signal recognition methods for complex systems, a recognition method based on multi-domain feature adaptive fusion with ambiguity function is proposed. First, an integral-accelerated ST-NLM algorithm is employed to denoise the signal ambiguity function, and the filtering coefficient is adaptively adjusted through the structure tensor. Then, multi-domain projection features are extracted from the time-delay domain, Doppler frequency domain, and joint time delay-Doppler domain. Finally, a ResNet+adaptive attention feature fusion （AAFF） model is constructed, utilizing multi-scale dilated convolution, depthwise separable convolution, and efficient channel attention（ECA） mechanism to achieve adaptive feature fusion. Experimental results show that the method can maintain 100% accuracy when the signal-to-noise ratio（SNR） is above 0 dB, and even at -4 dB SNR, the recognition rate can still reach 98.12%. It validates that the proposed method has certain effectiveness and feasibility under low SNR conditions. |
| Key words: radar emitter signal ambiguity function multi-domain feature fusion deep learning attention mechanism |
