Research On SAR Image Segmentation Method Based On Fuzzy Clustering

Synthetic Aperture Radar(SAR)is a coherent microwave radar,which has the advantage of all-day data acquisition capability and is not affected by climate and illumination changes.It is an important means of remote sensing observation at present.In the field of SAR image interpretation,SAR image segmentation is the basis of SAR image understanding,and it is an indispensable key technology in subsequent steps such as feature extraction and target recognition.However,due to the principle of coherent imaging and the rich coherent speckle noise in SAR images,SAR image segmentation becomes difficult,and many traditional image segmentation algorithms cannot achieve ideal segmentation results.This dissertation mainly takes fuzzy clustering theory as the cornerstone to study the SAR image segmentation methods based on fuzzy clustering.The main research work is as follows:(1)The current status of domestic and international research in the field of SAR image segmentation and the problems that still exist are analyzed and summarized.The concepts and basic principles of fuzzy clustering theory are elaborated,and the related traditional fuzzy clustering algorithms are sorted out and the formulas are derived in detail.(2)The traditional fuzzy clustering algorithm has uncertainty and randomness in the determination of the initial clustering center.If the initial clustering center is not selected properly,the number of iterations of the algorithm will be increased,and the algorithm may even fall into a local extreme point,which will affect the clustering performance.An adaptive genetic algorithm is introduced to determine the initial clustering center,and the influence of individual fitness,fitness change speed and genetic generation is integrated,and the calculation method of crossover and mutation probability is improved.Meanwhile,Lagrange number multiplication method is used to perform a strict mathematical derivation of the iterative formulas of affiliation and clustering center.The experiments show that the improved algorithm can solve the affiliation and clustering center more accurately,and the segmentation performance and the operation efficiency of the algorithm are improved.(3)The traditional Euclidean distance and image block distance both do not consider the edge information of the image and cannot accurately reflect the degree of association between pixels.In order to make full use of the grayscale and spatial information between pixels,firstly the weighted kernel function for the assignment of non-local pixel similarity weights in the NLFCM algorithm is improved,and then a new method for describing the similarity of local information is proposed,followed by combining both local and non-local spatial information of the image to improve the distance measure from pixel points to the clustering center,which can adaptively calculate the weights of local and non-local terms based on the pixel information.The experiments show that the improved algorithm can better protect the SAR image target edges and details while suppressing the coherent speckle noise of SAR images,and has better SAR image segmentation effects.(4)A visualization platform of SAR image segmentation system based on fuzzy clustering is designed and implemented,containing two functional modules of image pre-processing and image segmentation,which can significantly improve the efficiency of SAR image processors.The dissertation concludes with a summary and outlook of the research work.