SAR Image Segmentation Based On Smooth Noise Reduction And Label Correction

Synthetic Aperture Radar(SAR)is widely used in various military and civilian fields due to its excellent data acquisition and imaging capabilities.However,the imaging mechanism of synthetic aperture radar results in the existence of multiplicative speckle noise that interferes with normal data in the SAR images.It is difficult to obtain ideal results for general optical image processing methods on SAR images.Therefore,a special processing method for SAR images needs to be designed according to its characteristics.As a common SAR image preprocessing method,SAR image segmentation can simplify the display mode of SAR images,making the information of SAR images easier to analyze and understand.Among them,the SAR image unsupervised segmentation algorithm is widely used in SAR image segmentation processing because it can obtain segmentation results simply and quickly.This thesis has carried out a series of research and experiments on the existing unsupervised segmentation methods of SAR images.Based on the problems of error-prone segmentation edges,slow segmentation speeds,and poor segmentation capabilities of small targets that often occur in SAR image segmentation,several improvements are proposed.The main contents are as follows:(1)A SAR image segmentation algorithm based on edge description and superpixel smoothing is proposed.First,Gabor function templates and three edge detection methods are used to detect and extract the strong edges of the image.Next,we use simple linear iterative clustering algorithm(SLIC)to generate image superpixels and use them for the representation of weak edges.Then,the superpixel boundaries are used to carry the strong edge information of the image.And the edge-constrained smoothing algorithm is performed to achieve superpixel smoothing.Finally,we use K-Means algorithm to cluster the superpixels to obtain the final segmentation results.This method uses superpixel boundaries to carry the strong and weak edges of the image so that they can participate in the final segmentation,making the segmented edges of the image more accurate.(2)A SAR image segmentation algorithm based on directional template smoothing and label correction is proposed.First,the direction detection templates are used to detect the direction of each coordinate in the image,and the corresponding smoothing templates are used to smooth the edge region of the image.Then,Gaussian smoothing is applied to the homogeneous region according to the direction-difference map generated by direction detection.The two smoothed results are combined and the K-Means clustering is used to obtain the preliminary segmentation results.Finally,the region growing majority voting method is used to achieve the label correction and obtain the final segmentation results.This method significantly improves the segmentation efficiency of SAR images by using directional template smoothing and simple and effective label correction operation.(3)A SAR image segmentation algorithm based on constrained smoothing and iterative label repair is proposed.First,edges are extracted from the SAR image after edge region smoothing,and edge-constrained Gaussian smoothing is performed on homogeneous regions of the image.Next,K-Means clustering is performed for preliminary segmentation,and Markov random field iterative conditional mode(ICM)algorithm are used to repair the the label of corner pixels.Finally,the pixel-group-counting comparison and region-growth label correction algorithm are used to obtain the final segmentation results.This method uses the Markov random field ICM algorithm to make small targets correctly labeled,which effectively improves the algorithm's ability to segment small targets in SAR images.