Research On SAR Image Sidelobe Suppression And Image Quality Improvement

Synthetic aperture radar(SAR)has been widely used in military and civil fields because it can observe the earth all day and all weather.Due to the influence of many non-systematic errors such as sidelobe noise,SAR images tend to be seriously degraded and the image quality is degraded.which is not conducive to subsequent interpretation and application.In order to effectively suppress the side lobe of SAR image and improve the image quality,this thesis studies the algorithms of SAR image side lobe suppression and image quality improvement.In this thesis,aiming at the SAR image side lobe suppression,we analyze the causes of SAR image side lobe and the relevant theoretical basis,and summarize the relevant algorithms in the SAR image side lobe suppression field.SVA(spatial variant apodization)algorithm is a SAR image sidelobe suppression algorithm.By calculating the weight of the value around the pixel point,it can determine whether the corresponding pixel point in the SAR image is in the target main lobe or side lobe region,and then process it separately.This method can suppress the sidelobe energy while keeping the main lobe of SAR image.The traditional SVA algorithm filter has limited side lobe suppression effect on SAR images in complex scenes.Therefore,combined with CS(compressed sensing)theory,this thesis proposes CSSVA(compressed sensing SVA)algorithm,which uses SVA to determine whether the pixel point is located in the main lobe or the side lobe area,and then uses CS to recover the main lobe,while the side lobe position is recovered by the method of background fusion.Through comparison and analysis of point simulation experiments,the peak side lobe ratio and integral side lobe ratio of SAR images in azimuth direction and range direction are reduced by more than 13dB after CSSVA algorithm processing.The lower the peak side lobe ratio and the integral side lobe ratio,the better the side lobe suppression effect.Therefore,it is proved that CSSVA algorithm can effectively suppress the side lobe.After the side lobe suppression,the main lobe will be widened,which will lead to the reduction of SAR image resolution.After the simulation data is processed by CSSVA algorithm,the main lobe will not be widened but will be reduced.This also proves that CSSVA algorithm will not reduce SAR image resolution while suppressing the side lobe.Finally,the performance of CSSVA algorithm,SVA algorithm and XiongMethod on the measured data is compared.The CSSVA algorithm has better side lobe suppression effect in subjective vision.Combined with the objective evaluation indexes such as peak side lobe ratio,integral side lobe ratio and entropy value,it is proved that the CSSVA algorithm proposed in this thesis has good side lobe suppression effect.Aiming at the SAR image quality improvement part,this thesis analyzes the reasons of SAR image degradation,and solves the problem of SAR image quality improvement from the perspective of developing SAR image post-processing module.After SAR image is processed by side lobe suppression algorithm,there will be a large number of zero values in the image.However,the existence of these zero values will lead to the loss of background information and the degradation of SAR image quality.Therefore,this thesis builds a SAR image background restoration model to recover the lost background information of the image and improve the quality of the SAR image.In this thesis,four models are used to recover the background information of SAR images,namely,the BR-CNN(background restoration CNN)model based on CNN(convolutional neural networks),the BR-TCNN(background restoration via transformer and CNN)model based on transformer:BR-RTCNN(background restoration residual connection transformer and CNN),which combines CNN with transform,and BR-PTCNN(background restoration parallel transformer and CNN)proposed in this thesis.BR-RTCNN combines convolution layer and transformer through residual module,and BR-PTCNN combines convolution layer and transformer in parallel.Through the comparison of simulation experiments,it is found that these four models are more effective than the traditional Bicubic algorithm in restoring image background information.Among them,the BR-PTCNN model has the best effect,and its background restoration effect is 3dB higher than the ZPSNR(zero point peak signal to noise ratio)value of Bicubic algorithm,which proves that the BR-PTCNN model can restore SAR image background information well.Finally,combined with the actual data processing results,it is shown that the BR-PTCNN model proposed in this thesis can effectively restore the background information of the image and improve the image quality.