The Algorithm Optimization For De-noised Remote Sensing Fusion Based On Sparse Representation

With the rapid development of the sensor technology,different kinds of imaging sensors have spread to multiple applications.These sensors work in different environments,acquiring images with different information.Different from ordinary optical images,remote sensing images have higher spatial resolution and contain richer detailed information such as textures,which lead to wide use in auxiliary engineering planning,earthquake relief,and meteorological prediction and so on.In recent years,researchers have proposed a lot of remote sensing image processing methods based on sparse representation,which are mainly used in remote sensing image de-noising,and multi-source image fusion.Until now,this kind of methods have made some progress,but in the face of remote sensing images with richer texture information and edge features,there are still many deficiencies,such as no shift invariance,loss of detailed information and so on.In order to solve the aforementioned problems,this dissertation mainly conducts in-depth research and exploration on the basic theory of sparse representation,and carries out systematic research on key technologies of remote sensing image de-noising and fusion,so as to obtain a fused image which is more fully and completely reflected of the scene information.The main works and innovative research results are as follows:1.SAR image de-noising based on sparse representationDue to the special imaging mechanism of SAR images,there is always some speckle noise in SAR images,such as noisy SAR images in SEN1-2 dataset collected by the satellite Sentine-1,which has serious influence on the subsequent image fusion.When SAR images are represented by ordinary sparse representation,the reconstructed image will have edge blur to some extent and spatial resolution degradation due to the compression of the image information,and even lose some key points and edges of the original image.Therefore,for the traditional SAR image de-noising methods based on sparse representation,the improvements are as follows:?1?SAR image de-noising based on residual image fusion and sparse representationCombing the advantages of image de-noising based on sparse representation,and making full use of details such as edges contained in the residual image and the related knowledge about image fusion,one SAR image de-noising method based on residual image fusion and sparse representation is proposed.?2?SAR image de-noising based on shift invariant K-SVD and guided filterIt is well known that images are shift invariant,and only when the overcomplete dictionary which represents an image is also shift invariant,the optimal sparse representing of an image can be obtained.In order to overcome the sensitivity of the traditional K-SVD to the pixel position and phase characteristics in an image,one SAR image de-noising method based on shift invariant K-SVD and guided filter is proposed.2.Remote sensing image fusion based on sparse representationIn order to more comprehensively integrate the large amount of image information collected by multi-sensors and then better provide diversified information reference for the demanders and obtain more concise and intuitive results,two new remote sensing image fusion methods are proposed as follows:?1?Remote sensing image fusion based on sparse representation and guided filterThe traditional image fusion method based on sparse representation not only ignores the inherent spatial continuity of an image itself,but also ignores the relevance of redundant information in source images when fusing sparse coefficients by fusion rule of“₀l-max”,leading to some discontinuous edge features in an image being lost and the incompleteness of the information in the fused image.As a result,one remote sensing image fusion method based on sparse representation and guided filter is proposed.Firstly,the fusion rule based on improved hyperbolic tangent function and“₀l-max”is employed to fuse the sparse coefficients and obtain the fused image based on sparse representation by sparse reconstruction.Image fusion based on spatial domain is performed on source images at the same time.Finally,the guided filtering on the two intial fused images is conducted to obtain the final fused image.?2?Image fusion based on joint sparse representation and optimum theoryTo save the information more completely and obtain a fused image with rich information,one image fusion method based on joint sparse representation and optimum theory is proposed.Joint sparse representation is firstyly employed to separate the complementary and redundant components in source images,and then the proposed fusion rule based on optimum theory is adopted to fuse the complementary sparse coefficients.At last,a fused image can be reconstructed by adding the fused complementary sub-image and redundant sub-image together.3.Noisy remote sensing fusion based on sparse representationWhen processing the remote sensing images,it would bring great convenience to the practical application of an image if image de-noising and fusion can be achieved simultaneously.For this purpose,one noisy remote sensing image fusion method based on joint sparse representation and PCNN is proposed.At first,a redundant dictionary can be obtained by combing the adaptive trained dictionaries based on source images and a fixed dictionary.Since the noise is not sparse and the characteristics in different source images are different,it can be seen as the complement component of source images.The image de-noising can be achieved when fusing the complement sub-images obtained by joint sparse representation.By lots of compared experiments on Oslo city and SEN1-2 dataset,it has been proven that the proposed method can not only achieve the effect of image fusion,but also is robust to image noise,avoiding the noise interference in the fusion process.