High Resolution Synthetic Aperture Radar Imaging Based On Compressed Sensing

The characteristic advantages of synthetic aperture radar imaging make it develop rapidly in military and civilian applications.Under the framework of Nyquist sampling theorem,high resolution radar imaging is limited by the high sampling rate and large amount of post-processing data.Therefore,a new theory of sampling and signal reconstruction is needed.In recent years,the theory of compressed sensing has outstanding performance in dealing with sparse or compressible signals.It is recognized and widely used by researchers in many fields.The theory of radar scattering center agree with the requirements of signal sparsity,which makes it possible to apply compressed sensing to synthetic aperture radar imaging.Focusing on the combination of compressed sensing theory and synthetic aperture radar imaging,this paper improves the sparse Bayesian learning signal reconstruction algorithm based on compressed sensing theory,aiming at reducing the computational complexity and improving imaging accuracy.The main work is divided into the following two points:1)On the basis of two-dimensional imaging model,namely multi-measure vector joint sparse optimization,the Minimum L0 norm processing method is introduced in the sparse Bayesian algorithm to correct the mean matrix of the signal,speed up the convergence and improve the imaging quality.The improved sparse Bayesian learning algorithm with Minimum L0 norm processing method discretizes the radar echo and imaging scene,and performs Fourier transform on the distance dimension of the echo model to obtain a two-dimensional model.Then add a Minimum L0 norm processing correction step after the mean matrix update step of the classical sparse Bayesian learning algorithm.This correction step can widen the difference between the scattering center and the interference during each iteration,so that it can quickly converge to the specified condition.The correction step can also eliminate the distance dimensional fringe interference introduced by the 2D model.2)Considering the clustering effect of the target scattering center,a block structure hypothesis that can dynamically approximate the real state of the target is introduced in the sparse Bayesian algorithm.The method of two reconstructions is used to eliminate the fringe interference introduced by the two-dimensional model.This can improve the image quality while ensuring the speed of operation.The improved block sparse Bayesian learning algorithm first performs Fourier transform on the distance dimension,and divides the columns of the two-dimensional model into blocks of the same length.In each column,the non-zero blocks can be placed at any position.Therefore,it can be overlapped into larger non-zero blocks.This makes the assumption of the same length unrestricted,and can dynamically approximate the real structure of the target during the iterative process.So,this block model is very flexible.Keep the image after iteration.Then perform a Fourier transform on the azimuth dimension,repeat the calculation,and retain the imaging result.The intersection of the two results takes off the stripe interference of the respective dimensions.The two algorithms proposed in this paper greatly reduce the computational complexity of the sparse Bayesian learning algorithm.At the same time,it retains its advantages of high precision,anti-interference,no need to manually adjust parameters,etc.