Recognition of LPI Radar Signal Intrapulse Modulation Based on CNN and Time-Frequency Denoising

Aiming at the problem of low probability of intercept (LPI) radar signal recognition accuracy under low signal-to-noise ratio (SNR), a method for LPI radar signal recognition based on convolutional neural network (CNN) and time-frequency denoising is proposed. Firstly, the Smoothed Pseudo Wigner-Ville Distribution (SPWVD), which performs well under low SNR, is applied for time-frequency analysis of radar signals. Then, a frequency domain filter is designed using the K-means clustering method to reduce noise in the signal. Finally, the basic structure of the CNN network is studied, and a CNN network structure is designed and developed for the proposed LPI radar signal recognition system. Suitable hyperparameters are determined for it through parameter tuning. Time-frequency images are input into the CNN network to extract and learn deep features for radar signal recognition. Experimental results show that when the SNR is -8 dB, the overall recognition accuracy of 12 kinds of LPI radar signals reaches 91.67% using this method.