Research On Key Algorithms Of SAR Image Super-resolution Enhancement

Synthetic Aperture Radar(SAR),as an advanced microwave remote sensing radar,has unique ability of interferometry and multi-polarization measurement,At the same time,it also can work day-and-night under all-weather conditions.It has incomparable advantages in military and meteorological applications compared with optical remote sensing.The quality of a SAR image often determines the application value of its SAR products,which is mainly limited by the resolution index.The higher the resolution of SAR image,the stronger the target features,and the richer the geographic information can be obtained from it.But in many cases,the resolution of SAR image we get is not very high,which can not meet the requirements of practical application.Therefore,we need image processing methods to improve the resolution of SAR image.Based on the booming technology of image super-resolution(SR)enhancement in recent decades,this paper analyses the characteristics of SAR image(multi-polarization,sparsity,speckle-rich noise)which is different from ordinary optical image,and introduces several image preprocessing methods aiming at SAR image characteristics in detail.On this basis,two classical regularization methods in the field of image reconstruction—Tikhonov regularization method and bilateral total variation regularization method,are introduced.Realizing these two algorithms by programming and using real SAR images in Qionglai and Maoxian areas to verify the feasibility of these two algorithms to enhance SAR images.Aiming at the problem of obvious edge oscillation effect in SAR images reconstructed by bilateral total variation algorithm,an improved method based on modified point spread function is proposed,which makes the improved algorithm can effectively suppress edge oscillation.The quality of construction results has been significantly improved.Then,aiming at the low efficiency of regularization method in fusing image priori information,the Projection Onto Convex Set(POCS)method is introduced,which is an image reconstruction method based on set theory.Although this method can effectively incorporate image priori knowledge into the reconstruction iteration process,but in many cases,the quality of reconstruction results can be guaranteed only when the low-resolution images are enough.To solve this problem,this paper proposes a method of adding all polarization mode images of multi-polarization SAR images into the low-resolution sample sequence,which greatly enhances the low-resolution input data set of SAR image reconstruction experiment.The quality of the reconstructed image is better than that of the reconstructed SAR image using only single polarization mode.In most cases,the quality of a reconstructed SAR image can not be evaluated with reference image quality assessment(such as PSNR,SSIM)because of the lack of high-resolution SAR image of the original scene as a reference.On the other hand,this paper selects several non-reference evaluation indexes which have good effect in the field of optical image evaluation,and by means of subjective evaluation and ranking of the reconstruction results of the above algorithms,the non-reference evaluation indexes which are suitable for evaluating the sharpness of SAR image have been further screened,the final two selected indexes have good evaluation effect and is consistent with the subjective evaluation of human eyes.