Research On Extraction Of Road And Surface Spatial Information From High-resolution SAR Images

The acquisition of road information is of great significance to the environment,economy,military and other fields.In the field of emergency disaster relief,in the face of complex terrain and severe weather,it is particularly important to grasp the real-time information of the road and the surrounding conditions of the road.Synthetic Aperture Radar(SAR),an airborne or spaceborne microwave imaging radar,has been widely used in remote sensing and mapping since its invention due to its all-weather,all-day Earth observation capability.With the rapid development of SAR imaging technology,it is no longer difficult to obtain high-quality and high-resolution SAR images,and the research on road information extraction using high-resolution SAR images continues to attract the attention of scientific community.Based on this background,this dissertation combines remote sensing mechanism analysis and image processing technology to conduct the following research on road and surface spatial information extraction from high-resolution SAR images.The main research topics and results are as follows:(1)Since the road extraction from SAR images is easily confused by water and shadow,this dissertation proposes a semi-automatic method based on the multiplicative Duda operator and the path opening and closing operation using high-resolution SAR images.This method focuses on the key characteristics of the road,which include backscatter coefficient,coefficient of variation,the direction and length of roads.Using the path opening and closing operation,this method transforms the threshold parameter setting of the linear feature response involved in the road segmentation process into a more intuitive length threshold parameter setting.Meanwhile,the proposed method reduces the impact of multiplicative noise,water and shadow targets on the road extraction and improves the road segmentation performance.(2)Aiming at the research on road network extraction from remote sensing images,this dissertation proposes a semi-automatic method for extracting the centerline of road segmentation map based on topological relationship and spline fitting,and establishes a general road linearization model for the road segmentation map.To achieve the road centerlines,the road direction and width are firstly calculated by constructing the ratioof-cross operator.Then,in order to construct the road network,the road is modeled as a set of splines that can be described by curve equations based on the theory of perceptual grouping and topological relationship.By applying the proposed method to the road segmentation maps obtained from different remote sensing images,the experimental results show that it has strong anti-interference ability and wide applicability.(3)In order to improve the completeness and correctness of road extraction and make full use of different types of satellite-based remote sensing images such as microwave and optical images,this dissertation presents a road decision-level information fusion model of SAR and optical remote sensing images that takes into account the contextual information and evidence theory.Based on the road segmentation maps of SAR and optical remote sensing images,the fusion of SAR and optical image road extraction results is achieved through the road fusion method based on overlapping rules,and the road decision-level fusion algorithm which combines the D-S evidence theory and Markov random field model.The proposed fusion model is tested on the Terra SAR-X images from the German Space Agency and the World View-4images from the American company of Digital Globe.It is shown that the fusion model proposed in this dissertation have been significantly improved the completeness and quality of the road extraction results.(4)In order to further obtain the surface space information related to the road,this dissertation finally proposes the extraction of the road slope and land-use information around the road using the high-resolution SAR images.The study is on the basis of the previous research on the extraction of road location information.For the mountain roads with large elevation changes,this dissertation achieves the road and surrounding elevation information based on the Interferometry Synthetic Aperture Radar(In SAR)technology,and then determines the fluctuation of the road segments by the percentage of the elevation difference and the horizontal distance;For the extraction of land cover information around roads,using the unique coherence coefficient characteristics of multi-temporal Interferometry SAR,a land-use classification algorithms for the surrounding objects of the road is proposed based on support vector machine(SVM).The experiment on the multi-temporal Terra SAR-X remote sensing images in Chongzhou,Sichuan province,verifies the applicability and validity of the method in this dissertation.