Research On The Speckle Reduction Algorithm Of Gaofen-3 SAR Image Based On Non-subsampled Shearlet Transform

The GF-3 SAR satellite is the first Synthetic Aperture Radar(SAR)satellite with the 1-meter resolution and the C-band multi-polarized in China.By using the Long March 4C,the GF-3 SAR satellite was successfully launched at the Taiyuan satellite launch center in August 10,2016.The GF-3 SAR satellite is not influenced by the weather conditions such as the rain and the cloud.The GF-3 SAR satellite can be imaged at any time,whether day or night.It is the important foundation for the "Gaofen project".It can observe the object all-day,all-weather and achieve space-time coordination.However,due to the coherent imaging system,the speckle appeared in the GF-3 SAR images formed by the GF-3 SAR satellite,which seriously affect the interpretation and the understanding of the GF-3 SAR images.So speckle smoothing is still a difficult point in GF-3 SAR images processing.Due to the diversified imaging model and high resolution of GF-3 satellite,edges,points,textures and other feature information are so clear in GF-3 SAR images.In another word,GF-3 SAR images own very strong sparsity.Therefore,the promising despeckling method must be constructed based on the sparsity of GF-3 SAR images.In this paper,the depth research has been conducted about how to smooth the speckle effectively in GF-3 SAR images.The main completed work as follows:(1)The GF-3 SAR satellite and the GF-3 SAR images with obvious sparsity are introduced to produce the high quality despeckling method.By visual analysis and statistical distribution analysis of GF-3 SAR images,it can be concluded that GF-3 SAR images have good sparsity,so the ability of the suppressing the speckle will be better if we consider the sparse characteristics of the GF-3 SAR images.(2)The theory and the transformation principle of Shearlet are introduced and the non-subsampled Shearlet transform(NSST)is adopted in the method.Because the down-sampling operation that applied in the traditional Shearlet transform will cause the spectral aliasing,which causes the information in the same direction to be appeared in several different directional subbands and weakens the directional selectivity.So the NSST is used to process the GF-3 SAR images.At the same time,the hard wavelet threshold algorithm is applied to efficiently process high-frequency coefficients.Experiments shows that this method can suppress the speckle effectively and can preserve the edge detail information well compared with the traditional filtering algorithm.(3)The non-local means based on the improved kernel function and the NSST is proposed to process the GF-3 SAR images.And for the problem that the exponential kernel function used in the non-local means algorithm cannot perform the weight assignment well,the improved kernel function is used to improve the traditional non-local means algorithm.By using the NSST operation and then the improved non-local means to deal with the high-frequency coefficients,the method is proposed to process the GF-3 SAR images.The experimental results show that the proposed method can suppress the speckle effectively and the ability of retaining the edge detail information is well.(4)GF-3 SAR images despeckling method based on the improved non-local means using L1/2 norm and the NSST is proposed.The non-local means does not consider the sparsity of the GF-3 SAR images.In order to solve the problems that the speckle in homogeneous regions cannot be suppressed effectively and the capability of edge preserving is not well,an improved non-local means is proposed that uses the L1/2 norm instead of the L2 norm as the measure of the similarity.At the same time,in order to better improve the filtering performance,especially suppressing the speckle effectively in edge regions,the Shearlet as a new multiscale geometric analysis(MGA)tool is used in this method.Experimental results demonstrate that,compared with the traditional and improved non-local means,the proposed algorithm can not only suppress the speckle sufficiently in homogeneous regions,but also effectively preserve edges and fine details,and can better smooth the speckle in edge regions,leading to a satisfying performance for GF-3 SAR images.