Research On Intelligent Extraction Algorithm Of SAR Image Edge Features

Synthetic Aperture Radar(SAR)is a general term for a class of detectors that use the principle of coherent imaging,which can overcome the limitation of natural light intensity,penetrate the surface cover and go deep into the ground.It is widely used in lake and island monitoring,urban surveying,geological monitoring and other fields.SAR satellite uses active imaging method to collect ground object information,and forms a certain angle with the ground object.Therefore,the imaging state of the ground object in SAR image is closely related to its geometric shape and electromagnetic characteristics,and the imaging information is more abundant.However,due to the particularity of the SAR satellite's image generation mechanism,the generated SAR image will have obvious spots compared to the natural light imaging image,and this speckle will interfere with the user's interpretation of the image.As a local feature with continuity cut off,the edge of an image is usually expressed in a certain area where the gray value has a sharp difference.It can outline the physical structure of the imaged object and is a research hotspot in the field of image processing.In this paper,the edge information in the SAR image can not be extracted accurately by the speckle interference,and the in-depth study is carried out.The main work completed can be summarized as follows:(1)An edge detection algorithm of SAR image based on shearlet is proposed.The flow of the algorithm can be described as follows.Firstly,the shearlet transform model is constructed by taking advantage of the shearlet's excellent directional positioning and multi-scale characteristics.Secondly,in order to obtain even symmetric shearlet and odd symmetric shearlet,we use the Hilbert transform to postpone the phase of frequency domain components by 90°(that is,to realize the conversion of odd and even functions).The special transformation transforms even symmetric shearlets constructed using the tensor product of Mexican straw hat wavelets and Gaussian wavelets into odd symmetric shearlets.Thirdly,we perform shearlet transformation on the SAR image to obtain the corresponding odd symmetric and even symmetric shearlet coefficients.Determine the main direction of the edge according to the odd symmetric shearlet coefficients of the SAR image.According to the even symmetric shearlet coefficients and odd symmetric shearlet coefficients,the possibility of the edge on the pixel is calculated.Finally,in view of the problem of false edges and coarse edges in the result,morphological processing including removing false edges,removing edge burrs and edge thinning is used to optimize the results and obtain the final result.(2)A bankline detection algorithm based on shearlet of GF-3 SAR images is proposed.We first use non-local mean filtering to preprocess the GF-3 SAR image,so as to reduce the interference of speckle on bankline detection.Secondly,shearlet is used to detect the bankline of the image.Finally,morphological processing is used to refine the bankline and further eliminate the false bankline caused by the speckle,and accurately display the real bankline.The experimental verification results using two GF-3 SAR images show that the method can deal with the interference of speckles effectively,and can detect the riverside information of GF-3 SAR image completely and smoothly.(3)A road segmentation algorithm for GF-3 SAR image based on capsule network is proposed.We imitate the classic segmentation network U-Net,and use a network structure of encoding and then decoding to achieve road segmentation.Firstly,the PrimaryCapsules layer of the original capsule network is used to construct the segmented capsule model,and the constructed segmented capsule model is applied to the U-Net network.Secondly,we use the dynamic routing mechanism to replace the convolutional layer and the pooling layer in the U-Net network,and use the dynamic routing operation of zero-filling to replace the deconvolution layer in the U-Net network.Finally,we use the interval loss function as the loss function of the network to construct a capsule network-based road segmentation model of GF-3 SAR image similar to the U-Net model.