| Synthetic Aperture Radar(SAR)as an active microwave imaging radar can obtain high-resolution radar images,and SAR can still perform well even in low visibility conditions.The image features obtained by SAR are rich in information,which compensates for the shortage of other imaging methods,such as visible light and infrared light imaging.Therefore,SAR imaging is one of the most valuable data sources for analysis.However,SAR is inherently affected by speckle noise,which strongly degrades the appearance of images visually and diminishes the implementation of subsequent processing such as target detection,image segmentation,and recognition.Therefore,it is very important to remove speckle noise as a preprocessing step in SAR image application.How to effectively suppress speckle noise and preserve texture details is the key and difficulty of research.Although numerous despeckling methods have been proposed over the past three decades,SAR despeckling remains a challenging research subject due to its uniqueness and complexity.In this thesis,the related methods of SAR image despeckling and the related theory of Non-Local Means(NLM)are studied.To overcome the shortcomings of Weighted Nuclear Norm Minimization(WNNM)and Guided Non-Local Means(GNLM),two SAR image despeckling algorithms are proposed.The main research work and innovation points of this thesis are as follows:(1)A blind SAR image despeckling method based on fast weighted nuclear norm minimizationTo address the difficulty of NLM-based techniques in SAR image despeckling can not select similar patches in highly noisy and boundary areas,this thesis proposes a novel SAR image despeckling method based on the framework of WNNM.First,a Gaussian function is used to approximate the distribution of Good Similar Patches(GSP).Clustering is then used to select the optimal patches for the GSP matrix.WNNM can then estimate the underlying noiseless component of the GSP matrix.To increase the speed of the WNNM,this method uses truncated Random Singular Value Decomposition(RSVD)to replace the original singular value decomposition.In addition,a noise variance estimation method is introduced that enables blind despeckling with the proposed method.The experimental results demonstrate that the proposed algorithm yields superior subjective visual inspection and objective indices such as Peak Signal-To-Noise Ratio(PSNR)compared with other algorithms.(2)A collaborative despeckling method for SAR images based on optical guide and texture classificationSAR images usually contain many different types of regions,including homogeneous and heterogeneous regions.Some filters could despeckle effectively in homogeneous regions but could not preserve structures in heterogeneous regions.Some filters preserve structures well but do not suppress speckle effectively.Following this theory,this thesis design a combination of two state-of-the-art despeckling tools that can overcome their respective shortcomings.Superpixel-Based Fast Fuzzy C-Means(SFFCM)clustering and Gray-Level Co-Occurrence Matrices(GLCM)are used for image classification and weighting,respectively.Since the structural information between SAR images and their co-registered optical images is consistent,and the optical images are cleaner than SAR images,both SFFCM and GLCM process the co-registered optical images of SAR images.The experimental results on synthetic and real-world SAR images show that our proposed algorithm can preserve structural details and suppress speckle noise under a strong noise level compared with other algorithms. |
