Chinese Journal of Computers   Full Text
Title	Automatic Extraction of Road Network from SAR Imagery Based on Genetic Algorithm
Authors	JIA Cheng-Li1) ZHAO Ling-Jun1) WU Qi-Chang2) KUANG Gang-Yao1)
Address	1)(School of Electronic Science and Engineering, National University of Defense Technology, Changsha 410073) 2)(Remote Sensing Information Research Institution, Beijing 100085)
Year	2007
Issue	No.7(1186—1194)
Abstract &   Background	Abstract This paper proposes a new approach for automatic detection of road network from SAR imagery based on genetic algorithm. Since the roads on low resolution SAR imagery can be modeled as dark linear structure, it first detects line feature pixels in order to get potential road pixels. Followed that, extracting primitive line segments are accomplished by Radon Transform on every set of connected pixels. Then it takes the longest primitive line segment as initial seed, searches line segments in a search region around it, and uses Genetic algorithm to select the optimum line segments to be grouped to the seed line. This process iterates until all the lines are grouped, and the candidate road segments are got. The snake model is used to adjust the position of the candidate road segments in order to make road segments discrimination using the road’s linear feature more effective. Finally, road network is constructed after extracting crossings. The feasibility of the approach is slowed not only by presenting reasonable results but also by evaluating them quantitatively based on ground truth. keywords SAR imagery; road detection; genetic algorithm; primitive line segment extraction; linear feature detecting background The research of this paper belongs to the research field of automatic objects extraction from SAR images. According to the state of the art on automatic road extraction techniques, most of the algorithms are on rural areas. In fact, even automatic extraction of roads from rural SAR images is a complicated problem, however, in urban areas, the roads are denser, the finer details are visible, thus urban areas are full with disturbing context, the complexity of the scene grows greatly. To reduce the difficulty but also make advances in road extraction of urban areas, the authors decide to do the work in the semi-urban areas, which are the areas between urban areas and rural areas. The main contribution of this paper is proposing one solution of primitive line segment grouping of complex SAR scene based on Genetic algorithm. The research team have achieved in relative research areas such as statistic modeling of SAR image and speckle filtering, target detecting & discrimination & recognition, edge detecting, road and building extraction, and have published about 50 pieces of paper in domestic and foreign core journal and academic meeting.