Radar Signal Intra-Pulse Modulation Recognition Based on Contour Extraction

Zhengyang Yu, Jianlong Tang

Abstract: Recent evidence suggests that several methods are focused on changing the convolution neural network (CNN) structure to improve the recognition rate of radar signal intra-pulse modulation. As a result, the CNN structure is becoming significantly complex with poor interpretability. Aiming at this problem, we proposed a recognition method based on contour extraction. The signal binary map is replaced by the contour of signal sending to the CNN for recognition. Because the input is limited to the signal contour, CNN only considers the contour feature extraction, without complex structures. Simulations show that the method can effectively identify five kinds of radar signals when the SNR > 2dB. When the SNR > 4dB, the method has a recognition rate of 2.6% higher than the method without contour extraction.