Synthetic Aperture Radar Imaging And Moving Target Clutter Suppression Based On Sparse Constraint

Synthetic Aperture Radar(SAR)is a remote sensing imaging system with all-weather and allweather operation capabilities,and has been widely used in both military and civilian fields.The rapid development of Compressed Sensing(CS)technology has broken through the bottleneck of traditional SAR technology to a certain extent and brought new development ideas to SAR imaging.CS-based SAR imaging can reconstruct high-quality SAR images with very little echo data,and suppress sidelobes and noise of the response function.However,compared with the traditional SAR imaging algorithm,the computational complexity and storage cost of this method are much higher,and the computational efficiency is low.In addition,when there are multiple moving targets(MT)in the imaging scene,it will often lead to tailing and geometric position deviation in the constructed image,which will suppress the clutter of the stationary target in the echo and the repetition of the moving target image.Focus processing becomes crucial.Finally,with the vigorous development of deep learning technology,it has made major breakthroughs in common areas such as speech recognition,natural language processing,and image processing.In recent years,this technology has also been successfully introduced into SAR imaging,automatic target recognition and other fields.The research content of this article is as follows:(1)A spotlight SAR compressed sensing imaging algorithm based on approximate observation is proposed.First,the algorithm replaces the precise observations in the traditional CS-SAR framework with the approximate observations of the inverse imaging process of the Polar Format Algorithm(PFA).Secondly,the CS and PFA imaging processes are incorporated into a sparse regularization framework and solved by iterative soft threshold shrinking algorithm(Iterative Soft Thresholding Algorithm,ISTA).The simulation results prove that the proposed CS-PFA imaging method can not only obtain high-quality and high-resolution images with significantly reduced measurement values,but also significantly reduce the computational complexity and storage cost of the algorithm.(2)A spotlight SAR clutter suppression and moving target refocusing algorithm based on approximate observation under single channel conditions is proposed.First,the azimuth filter function is constructed based on the motion error spectrum of unistatic PFA moving target imaging.Secondly,the moving target parameter estimation based on Fractional Fourier Transform(FRFT)is combined with the filtering operation,and the dictionary of moving target radar projection matrix is constructed by the method of grid division.After that,a joint projection operator for moving and static targets is constructed,and the approximate observation process is incorporated into the ISTA framework.This algorithm can not only achieve clutter suppression and refocusing of moving targets with different speeds at the same time,but also greatly reduce the computational complexity and storage memory.Through the simulation of the point target and the measured data,the effectiveness of the algorithm is verified.That is,in the case of low signal-to-clutter ratio(SCR),it is possible to obtain separated static target and refocused moving target image and Reduce the sidelobe value.(3)The feasibility of deep learning in SAR imaging processing is verified.First,a model-driven deep learning method is used to map each iteration step of the Fast Iterative Soft Thresholding Algorithm(FISTA)to a layer in the deep network,and then construct a recurrent neural network(Recurrent Neural Network).Networks,RNN)to get the variables of interest.Secondly,by inserting an additional linear stage,the estimated measured value is obtained.Then the measured value and the input real measured value are put into the mismatch function for comparison,so the network can be regarded as a cyclic autoencoder,which realizes unsupervised learning.Finally,the simulation of point targets verifies that the RNN framework can better adapt to the problem of image reconstruction and achieve better imaging results.