Radar Intra-Pulse Modulation Signal Recognition Using Multi-Branch Denoising Convolutional Neural Network and Inception- ResN et-v2

The Radar intra-pulse modulation signal recognition is one of the important technologies in electronic warfare. In this article, a novel multi-branch denoising convolutional neural networks (MBDnCNN) is proposed for denoising time-frequency images under low signal-to-noise ratio (SNR). Additionally, Convolutional Block Attention Module (CBAM) is utilized to compress the training time of Inception-ResN et-v2. Firstly, the grayscale Cohen-class Time Frequency Distribution (CTFD) image of the radar signal is obtained. Then, the time-frequency image is denoised with MBDnCNN. Finally, the modulation label of radar signal is identified using the improved Inception-ResNet-v2. To evaluate the method, comparative experiment and ablation experiment are conducted. Experimental results show that the proposed algorithm can recognize 12 kinds of radar signals with an overall precision of 97% when SNR is -8 dB.