Research On Registration Technology Of Sar Optical Remote Sensing Image

In the past decades,the feature-based registration method has achieved good registration results in optical image registration,but it is not good to apply the feature-based method to optical and synthetic aperture radar（SAR）registration.Optical and SAR image registration is a challenging problem.There are three difficulties:different geometric characteristics,different intensity characteristics and speckle noise.Aiming at the above problems,this paper studies the registration of SAR and visible images.The specific work of this paper is as follows:Firstly,the common registration methods and applications of SAR and visible light images are studied,the characteristics of SAR and visible light images are analyzed,and the preprocessing methods of SAR images are introduced.The SAR images used in this paper are preprocessed SAR images,and the principle of image registration and the common evaluation criteria of image registration are described.Secondly,aiming at the problem of coarse correction optical and SAR image registration algorithm,sift is optimized in five aspects: consistent gradient calculation,key point detection,dominant direction allocation,support region selection and matching strategy.The consistent edge information is obtained to improve the repeatability of key point detection in optical and SAR satellite image matching.Then,only the first octave image is obtained,and the redundant information is eliminated.And the diagonal points are detected to improve the stability.Skip the dominant direction assignment,and use multiple support regions to construct descriptors,which improves the computational efficiency and increases the registration accuracy.In terms of registration strategy,the two-way k-nearest neighbor ratio method combined with confidence is used.Used to estimate transform parameters for image registration.Experiments show that the improved algorithm is competent for optical and SAR image registration after rough correction.Finally,aiming at the registration of high complexity SAR and optical images with more detailed information,a registration algorithm suitable for high complexity images is proposed.The algorithm mainly includes three parts: detecting key points in Harris scale space,allocating dominant directions based on phase consistency expansion,and constructing descriptors to match key points.According to the different characteristics of SAR and visible images,the consistent edge information is calculated by Sobel operator and exponentially weighted mean ratio operator（roewa）respectively,and two Harris scale spaces are constructed on this basis.By finding the local maximum in the scale space,and then refining the location according to the spatial relationship of the key points,the key points are screened through the Gaussian dynamic threshold method,which improves the repeatability of the key points.In addition,multiple image blocks are used to extract the gradient position direction descriptor to increase the discrimination,and the descriptor is modified to increase the complexity of the descriptor.In the aspect of registration strategy,two-way k-nearest neighbor ratio method combined with confidence is used.Experimental results on several high complexity optical images and SAR images show that the proposed algorithm is robust to high complexity optical to SAR images,and is superior to other advanced algorithms in registration accuracy.