A Study On Line Target Extraction From SAR Image

Synthetic aperture radar(SAR) has been widely in military and civilian applications.Detection and extraction of Line target in SAR image is of great significance for military reconnaissance and civil surveying and mapping, it is an important clue to target recognition and understanding of the image.Line targets in SAR images look like the banded bright lines or Dark lines which have certain shape information, such as bridges, airport runway, road, river and coastline,etc. This paper systematically studies the extraction technology of SAR line target, and mainly studies the extraction methods of rivers and roads from SAR images.At first,this thesis studied the prior knowledge of rivers in high resolution SAR images,built road model according to river’s characters, analyzed the interference factors in river extraction, then a new method of river extraction from SAR images based on region of interest and prior knowledge was proposed. This method can be divided into prepossessing, region of interest extraction, and river extraction three stages. In prepossessing stage, improved Lee algorithm and image enhancement methods are used to remove the noise from the original image and increase the image contrast; In region of interest extraction stage, histogram threshold segmentation algorithm was used for binary segmentation according to the gray scale difference between river region and background region in SAR images, then we can get the region of interest by removing small blocks; In river extraction stage, a adaptive gray-scale threshold is obtained based on relatively low gray level and relatively high area percentage by which we can remove false river region which has higher gray scale. At last, river area was extracted according to river continuity by removing false river such as pond, lake and so on which is isolated and far from other water area.This thesis also put forward a method of road extraction from SAR images based on Hough transform and mathematical morphology by analyzing road’s prior knowledge and interference factors in road extraction. This method can be divided into prepossessing, region of interest extraction, road direction detection and road extraction four stages. In prepossessing stage, improved Lee algorithm was used to remove the noise from the original image, and enhancement method was used to increase the image contrast; In region of interest extraction stage, OSTU segmentation algorithm was used for SAR image segmentation for binary segmentation according to the gray scale difference between road region and background region in SAR images,and we can get the region of interest by erosion and dilation operations. In road direction detection stage, thinning operation was used to get the skeleton of the road area based on region of interest, then the direction of the main road can be get easily by using Hough transform; In road extraction, by selecting the suitable structuring element, we can get the road section in different directions through morphological open operation based on region of interest and merge the two kinds of roads as the final road extraction result.At last, the two algorithms are verified through multiple sets of measured SAR images,and the results show that the two methods can extract rivers and roads in SAR images efficiently. The two algorithms presented in this paper both are based on threshold segmentation under the premise of accurate. However, threshold segmentation itself has certain limitations, once the segmentation results are not accurate, the subsequent line target extraction results will be incomplete or excessive,thus stable and effective segmentation method need further research.