SAR Image Segmentation Based On Key Feature Extraction And Multi-level Information Fusion

Synthetic Aperture Radar(SAR)is an active side-view radar system.This system has wide applications in environmental,archaeological,military and other aspects.SAR system can produce high-resolution images and work at any time and in any weather conditions.However,its special imaging mechanism causes SAR images seriously affected by speckle noise.As the basic work of SAR image interpretation,SAR image segmentation refers to dividing a SAR image into several non-overlapping and coherent regions.The pixels in the same region have similarities,while the pixels between different regions have different features.Dividing SAR image into these meaningful areas helps to understand the image from a high level and is convenient for further processing and analysis and promotes the update and development of SAR image processing technology.Information captured by SAR system is the reflection of ground objects on the radar beam.Different ground objects have different reflections on the radar beam.Therefore,SAR images often contain a variety of different information corresponding to different features.In actual applications,only using gray-scale information during segmentation can not achieve ideal result.The key to achieve accurate segmentation of SAR images is to grasp the edge information and reduce the impact of noise.The main research contents of this thesis are as follows:1)A new SAR image segmentation algorithm based on structural information extraction and hierarchical fuzzy clustering is proposed.Firstly,the Bhattacharyya distance of the rotating double window is used to generate the structural information map of the image to be segmented,which contains edge points and noise.Then remaining pixels are texture information points,and the homogeneous points are selected using certain rules.Fuzzy clustering method is used to label them.Finally,a superpixel filter is used the image to determine the class label of each pixel,so that the accurate segmentation result of the entire image can be obtained.2)A SAR image segmentation algorithm based on improved CRF and multi-scale reconstruction is proposed.Specifically,this algorithm uses multi-scale pyramids to reconstruct the input SAR image,and L0 smooth is used to extract the geometric structure information of the image which can make use of more accurate extraction of the geometric structure and improve the robustness of the algorithm.Then several methods is used to optimize the initialization of CRF,which is helpful to improve the accuracy.Finally,based on the extracted geometric structure information,the kernel function is used to improve the CRF binary energy function to achieve the purpose of optimizing the model,which avoids the traditional CRF ignoring the geometric structure.3)A new semantic segmentation method of SAR images based on texture complexity analysis and key superpixels is proposed(TKSFCM).Texture complexity analysis is performed and on this basis,mixed superpixels are selected as key superpixels.Specifically,the algorithm calculates the texture complexity of the input image by a new texture analysis method.Then an improved method called neighborhood information simple linear iterative clustering is used to over-segment the image.For images with high texture complexity,the complex areas are first separated and key superpixels are selected according to certain rules.For images with low texture complexity,key superpixels are directly extracted.Finally,the superpixels are pre-segmented by fuzzy clustering based on the extracted features and the key superpixels are processed at the pixel level to obtain the final result.