Research On SAR Image Segmentation Based On Watershed

Synthetic Aperture Radar(SAR)is an active imaging radar,as an important means of remote sensing observation,it has the advantages of all-weather,all-day,multi-band,high resolution,side-view imaging,and large-area data acquisition capability and so on,more widely used in the military and civilian fields.SAR image segmentation is a critical task in the interpretation of SAR images and the most basic and key technology for further processing such as target recognition,data compression,and delivery.However,due to the existence of a large number of multiplicative speckle noise inherent in SAR images,SAR image segmentation work becomes very difficult,and many conventional segmentation algorithms cannot obtain ideal segmentation results.In this thesis,two novel SAR image segmentation methods based on watershed algorithm were proposed;they are watershed based perceptual hash algorithm(WSp H)and watershed superpixel based affinity propagation clustering algorithm(WSAP).The details of the two methods are described as follows:(1)WSp H method is a segmentation method based on region merging.Due to the influence of speckle noise on SAR image segmentation,before the segmentation process,the effects of some common local adaptive speckle filtering algorithms on the denoising of the SAR image are analyzed,and it is determined that the Lee filter algorithm is used to denoise the image.Then the initial segmentation is performed using a gradient-plus-mask three-step watershed method,which is effective in reducing small,isolated closed regions,compared with traditional watershed segmentation algorithms.Afterwards,DCT-based perceptual hash algorithm is used to perform region merging on the initial segmentation regions of the previous stage,reducing the effect of over-segmentation,improving the accuracy of the result,and obtaining an ideal segmentation result.Finally,the validity of the algorithm is verified by experimental comparison.(2)WSAP method uses the affinity propagation clustering algorithm for merging,after obtaining the segmentation result by the watershed superpixel segmentation algorithm.The difference from the traditional affinity propagation clustering algorithm is that the WSAP method fully considers the grayscale feature of the SAR image.In this method,the feature distance of superpixels obtained by combining gray feature distances,covariance matrix distances and spatial distances,is the measure of similarity in the process of clustering,which effectively improves the accuracy of the segmentation result.In the process of merging using clustering algorithm,the data points are the superpixels obtained by the watershed algorithm,which greatly reduces the computational complexity of affinity propagation clustering algorithm.In addition,the initial segmentation method for generating superpixels is a classical marker-based watershed algorithm using the gradient information and spatial constraint to process pixel priorities.The superpixel obtained by this method is relatively compact and has a relatively uniform shape,further improving the accuracy of the segmentation and finally obtaining the ideal segmentation result.Finally,the effectiveness of the proposed algorithm is verified by experiments on real SAR images.