| 摘要: |
| 针对小样本条件下且低信噪比时低截获概率(Low Probability of Intercept,LPI)雷达信号识别精度低的问题，本文提出了一种基于局部最大化同步压缩变换 (Local maximum synchrosqueezing transform,LMSST)与平滑伪维格纳维尔变换(Smoothed Pseudo Wigner-Ville Distribution,SPWVD)的双通道特征融合网络模型。利用LMSST和SPWVD对仅有的小样本LPI雷达信号分别进行时频分析，获取二维时频图像；使用循环对抗生成网络对其进行扩充并送入双通道网络对其进行特征提取和特征早融合；采用Softmax分类器对融合后的特征进行分选识别。研究结果表明，在信噪比为-8dB时，所设计的模型的整体识别率达到93.1%；相较于单通道识别模型，在小样本条件下的识别精度有效提高约6%~7%。此研究为小样本时LPI雷达信号的识别提供了一种理论依据。 |
| 关键词: 小样本信号识别 时频分析 循环对抗生成网络 早特征融合 双通道网络 |
| DOI： |
| 分类号:TN 974 |
| 基金项目:国家自然科学基金项目（面上项目，重点项目，重大项目） |
|
| Dual-channel recognition of LPI radar signals under small sample conditions |
|
|
|
|
| Abstract: |
| Aiming at the problem of low recognition accuracy of radar signals with low interception probability under small sample conditions and at low signal-to-noise ratios (SNRs), this paper proposed a dual-channel feature fusion network model based on the Smoothed Pseudo-Wignerville Distribution (SPWVD) and Local Maximum Synchrosqueezing Transform (LMSST). Only a small sample of LPI radar signals are analyzed separately using LMSST and SPWVD to obtain 2D time-frequency images; This is expanded using Cycle Generative Adversarial Networks (CycleGAN) and fed into a dual-channel network for feature extraction and early fusion of features; The fused features are sorted and recognized using Softmax classifier. The results show that the overall recognition rate of the designed model reaches 93.1% at a signal-to-noise ratio of -8dB; compared with the single-channel recognition model, the recognition accuracy under small sample conditions is effectively improved by 6% to 7%. This research provides a theoretical basis for the recognition of LPI radar signals at small samples. |
| Key words: small-sample signal recognition time-frequency analysis cyclegan early feature fusion dual-channel network |