Research On Azimuth Ambiguity Removal Algorithm For Low Oversampling Rate Staggered SAR

Synthetic aperture radar(SAR)is a common high-resolution microwave imaging sys-tem,which can observe and image the ground all the time.However,traditional SAR can't obtain wide swath in range direction and high resolution in azimuth direction simultane-ously.In order to overcome this shortcoming and the influence of blind area,researchers propose a SAR system based on variable pulse repetition frequency,namely Staggered SAR.In addition,in order to solve the problems of high data load and implementation cost of traditional high-oversampling rate Staggered SAR,researchers have also proposed low-oversampling rate Staggered SAR in recent years.However,the low-oversampling rate mode will cause a large number of high-intensity ambiguities near the real target,which greatly reduces the final imaging quality,making this mode of radar unable to meet the application requirements.To solve this problem,this thesis proposes algorithms,which based on supervised learning and reinforcement learning separately,to achieve azimuth ambiguity suppression in low-oversampling rate Staggered SAR imaging results.The main work of this thesis is as follows:1.The influence of different pulse transmitting modes of traditional SAR and Stag-gered SAR on blind area are introduced.On this basis,the echo modeling of Staggered SAR is carried out under the non-stop assumption,and the modeling process is briefly described.In the imaging results of low-oversampling rate Staggered SAR,the reason of high density and energy ambiguities near azimuth direction of the real target is deduced.Further simulation results verify the theory of this chapter.2.Aiming at the problem of azimuth ambiguity in low-oversampling rate Staggered SAR system,a supervised learning algorithm for azimuth ambiguity suppression is pro-posed.This method Ambiguity Removal U-shaped Network(AR-UNet),which is con-structed by a fully convolution network,obtain the mapping from original image with ambiguity to clear image.In the simulation experiment,the method successfully removes the azimuth ambiguity of the real target,and the effect is better than the traditional algo-rithm,which verifies the effectiveness of this method.3.In order to solve the problem of large amount of computation and low resolution of real target in supervised learning method,the algorithm Ambiguity Removal Policy Gra-dient(AR-PG)based on reinforcement learning is proposed.Different from the AR-UNet method,AR-PG divides the ambiguity suppression into multi-step recursive implementa-tion,and combines the policy gradient algorithm and convolution network to suppress the energy of ambiguity part in the image.Although the effect on ambiguity removal of this method is worse than that of AR-UNet,but it further reduces the required network size and computational complexity,and can better preserve the real target.Simulation results verify the effectiveness of this method.