Robust Compressive Sensing SAR Imaging In Low SNR

In traditional Range-Doppler (RD) method, the imaging resolution is related withthe main lobe width and constrained with signal bandwidth. Normally, to get high rangeresolution, signal bandwidth must be broadened; to get high azimuth resolution, thenumber of scene detection must be increased. The increase of bandwidth is means tomass echo data, which will be brought to storage and transmission difficult. Recently,compressive sensing (CS) is given to high attention in high resolution imaging areas.According to the theory of CS, firstly, analyzes the characteristics of radar signal, buildsthe corresponding sparse matrix of the radar signal; secondly, to reduce the echo datarate, it’s need to randomly observe the radar echo data; finally, through the inverseproblem solving, eliminate the influence of point spread function and get highresolution SAR image.According to the prior knowledge that the radar signal is sparse, the performanceof the imaging models based on CS depends on the sparse expression of the radar signal.As is well known,0-norm is the best sparse expression of the radar signal. However,0-norm imaging model is hardly to be resolved.-norm imaging model can besubstituted by1-norm imaging model which is easy to solve. Since1-norm basedCS model cannot fully explore the sparsity property of the signal, the weightedconstraint of the restored coefficients is hardly equal arranged during the reconstruction,and the characteristic of the non-sparsity of the noise will seriously affect the restore ofthe target information in the low signal-to-noise (SNR) condition, which results in muchfalse targets during the imaging, normally the imaging quality is sharply declined.Therefore, Candès puts forward reweighted1-norm model which is much closerto0-norm model. In this paper, after the detailed analysis of the reweighted-normmodel for CS reconstruction, a robust high resolution imaging model with corruptedecho is proposed. The main idea is inspired by the canonical reweighted1-norm basedCS model, the selection of the weights parameters are improved, which makes thevariation and the separation of the large and small weights can be equally penalized, andthe noise components can be effectively suppressed during the imaging, the simulationresults testify the stability and robust of the proposed model in low SNR condition.