Research On Road Extraction Of High Resolution SAR Images

As an important ground feature target,road extraction from remote sensing images is a very popular research topic and it is widely used.SAR images make up for the lack of light effects due to their all-day and all-weather working characteristics and play an irreplaceable role in the field of remote sensing.With the continuous development of SAR imaging technology,its spatial resolution has become higher and higher,and roads no longer show linear features at low resolution in the image,but have narrow and long planar features.This results in that the previous low resolution road extraction methods can not be effectively used in high resolution,and a new high-resolution road extraction method is required.Based on the characteristics of roads in high-resolution SAR images,this paper studies the method of road extraction based on region segmentation.The main works are as follows:1.Aiming at the shortcomings of large data and the time-consuming process of coherent speckle noise suppression in large scenes of high-resolution SAR images,a GPU-accelerated coherent speckle noise suppression method is proposed.In this paper,Lee filtering is used as an example to compare the coherent speckle noise suppression process in different program environments and different filtering windows.The CUDA-based GPU acceleration method was used to improve Lee filtering,which significantly improved the filtering speed and greatly improved practicability.2.Aiming at the disadvantages of the coherent speckle noise suppression process in the preprocessing stage,which will affect the subsequent road extraction speed,delete useful image information,and the slowness of road element extraction in large scenes,a GPU-accelerated gray entropy threshold segmentation Method,this method can effectively avoid the step of coherent speckle noise suppression during preprocessing,greatly simplify the road extraction process,and improve the processing speed.3.Aiming at the shortcomings of different road scene segmentation methods and their shortcomings in road scene extraction,a GPU-accelerated level set segmentation method is proposed.After experimental comparison,the segmentation speed has been significantly improved.4.Aiming at the shortcomings of road elements in the road extraction preprocessing stage and the slow connection of "broken" road elements in large scenes,a GPU-accelerated dynamic programming method was proposed to connect the divided road elements to further improve the road extraction effectiveness.