Research On CycleGAN-based Image Domain Clutter Suppression In High Frequency Passive Bistatic Radar

High Frequency Passive Bistatic Radar(HFPBR)is a bistatic or multistatic radar that passively receives non-cooperative short-wave signals to detect corresponding radar targets.Due to the use of non-cooperative illuminators,the extraction of weak target information from HFPBR faces the problem that the radar target echoes would be submerged by direct-path wave and strong multipath echoes(collectively called clutter).To address this issue,the existing traditional rejection algorithms of HFPBR are mainly designed by professionals based on the different characteristics between the radar target echoes and the clutter in the time,space,or frequency domains.In recent years,with the development of artificial intelligence technology,data-driven deep learning has been widely used in the image field.With the help of deep learning in automatically extracting data features,this paper explores a new clutter rejection method of HFPBR in image domain based on the cycle-consistent generative adversarial network(CycleGAN)and its improved network.The specific research contents are as follows:(1)Starting with the principle of HFPBR,this paper introduces the common signal receiving device,the principle of signal processing,matched filtering technology of HFPBR,and several common traditional clutter rejection algorithms.The acquisition principle of the range-doppler(RD)map of HFPBR and the relevant basic knowledge of image migration based on generative adversarial networks are also introduced,which paves the way for the application of deep learning algorithms in the field of clutter rejection.(2)Aiming at solving the problem that traditional clutter rejection algorithms consume more labor costs,combined with the great success of CycleGAN in the field of computer vision,it is proposed to use the CycleGAN to automatically learn the feature mapping relationship between RD maps before and after clutter rejection in HFPBR.In this method,the clutter rejection problem is turned into an image-to-image translation problem between RD maps before and after clutter rejection innovatively.The effectiveness of the method is proved by qualitatively and quantitatively comparing with the traditional algorithms on the measured data set.(3)Aiming at solving the problem that the shortcomings of the original CycleGAN such as too many network calculation parameters and poor clutter rejection performance,this paper proposes to replace the residual network(Res Net)with the densely connected convolutional networks(Dense Net)in the original CycleGAN network.Through the establishment of generators based on dense convolution connections,the specific working principle of the Dense Net in generators is analyzed,and the clutter rejection performance of the improved algorithm is compared and analyzed from the quality of the generated RD maps and the degree of clutter rejection.The impact of the number of dense blocks on network performance is also analyzed in this paper.