The Research On Extraction Of Roads And Airport From SAR Imagery

This thesis mainly deals with the automatic extraction of two kinds of linear man-made objects, namely road and airport, from the SAR imagery. Since the road and the airport both have distinct line feature, a uniform flow can be used to extract them: edge detection (low level processing), primitive line segments extraction (middle level processing), primitive line segments linking and object extraction (high level processing). According to this scheme, this thesis studies edge detection from SAR imagery, primitive line segments extraction, automatic extraction of road network from semi-urban SAR imagery, automatic extraction of curve roads from SAR imagery, and extraction runway and region of airport from complex scene of SAR imagery.The subject of chapter 2 is edge detection in SAR images. According to the different edge types, the principles, the implement and the detection performance in theory and in practice of classic step and line edge detectors of SAR imagery are studied respectively. A modified ROEWA edge detector is presented.In chapter 3 approaches of primitive line segments extraction are studied. We analyze the principles and implement of three typical primitive line segments extraction methods, present a series of quantitative performance evaluation measures and evaluate these three methods. The proper line segments extraction approache for linear object extraction can be selected based on the evaluation results. The influence of speckle reduction on primitive line segments extraction is also studied and instructive conclusion is presented.In chapter 4 we focus on the extraction of roads from SAR imagery. Considering the deficiencies in current research of this area, we mainly solve the problems of automatic extraction of road network from semi-urban SAR imagery and automatic extraction of curve roads. For the former problem, we present a line segments linking method for complex scenes--a GA based primitive line segments linking method. Based on this linking method as well as edge detection and primitive line segments extraction studied in chapter 2 and chapter 3 and according to the road model constructed, a completed approach of automatic extraction of roads and corners from semi-urban SAR imagery is proposed. For the latter problem, the heuristic curve linking method is studied and a tensor voting based curve linking method is proposed. And based on these two curve linking methods, the straight road extraction approach is extended to the extraction of curve roads. We also integrate a series of quantitative evaluation measures to evaluate the performance of road extraction algorithms presented in this thesis.In chapter 5 extraction of the airport from SAR imagery, mainly the automatic extraction of runway and region of the airport from complex scene of SAR imagery, is studied. According to the knowledge of the runway, a line segments linking method based on clustering is proposed, among which, a new distance metric is constructed. Based on this linking method as well as edge detection and primitive line segments extraction studied in chapter 2 and chapter 3, an algorithm is proposed for extraction the airport runway in complex scene. Based on the runway extraction result, the location and the aspect of the airport region are detected using region growing and KL transform.