基于降噪模糊函数和EfficientNet的雷达信号识别

A Radar Signal Recognition Method Based on Denoised Ambiguity Function and EfficientNet

雷达信号识别是电子对抗的关键技术,鉴于模糊函数等高线较强的雷达信号内在结构表征能力,结合深度神经网络强大的识别性能,论文提出一种基于EEMD降噪模糊函数等高线和EfficientNet的识别方法。首先选取合适的EEMD参数,对时域信号进行降噪并构建模糊函数等高线数据集,然后建立基于EfficientNet-B0模型的深度网络,结合迁移学习完成标签数据的训练,最后基于该网络实现雷达信号识别。仿真实验表明,信噪比大于-10dB时,所提方法对于BPSK、BFSK、FMCW、QPSK、LFM-BC和MSEQ六类调制信号,能达到99.67%以上的平均识别准确率,并具有较好的泛化能力和较强的鲁棒性。

Radar signal recognition is the key technology of electronic countermeasures.In view of the ambiguity function's unique effect on characterizing signal inherent structure,and strong recognition performance combined with deep neural network,this paper proposes a recognition method based on contour lines of ambiguity function with Ensemble Empirical Mode Decomposi⁃tion(EEMD)noise reduction and EfficientNet.First,the appropriate EEMD parameters to denoise the time domain signals and con⁃tour lines'datasets are created.Then,a Deep Network based on EfficientNet-B0 model is established and transfer learning is used to complete the training of labeled data.Finally,the network is used to achieve radar emitter signals recognition.The simulated ex⁃periments show that the average recognition accuracy rate of six kinds of complex modulated signals,i.e.,BPSK,BFSK,FMCW,QPSK,LFM-BC and MSEQ,by proposed method keeps above 99.67%in fixed SNR environment above-10dB with good general⁃ization ability and strong robustness.