Study Of The Road Semi-Automatic Extraction Method From Satellite Image

Semi-automatic road extraction is a technology that extracts and identifys the roads in aviation and aerospace by human-computer interaction. In recent years, road extraction in satellite images receives increasing attention with the development of aerospace technology. It has important significance for GIS data update, image matching, target detection and so on. So how to extract the road quickly and accurately becomes the focus of image recognition research.With the background of"UAV Teleoperation and Simulation System", this thesis analyzes the augmented reality technology widely and deeply.Firstly, this thesis summarizes the research significant, status and problems of satellite images. Then it systemically illuminates the methods of road extraction. The feasibility of Semi-automatic road extraction has been obtained by the comparison of Automatic road extraction and Semi-automatic road extraction and the research of this subject's background. Then, according to the needs of UAV location tracking, the thesis analyzes the advantages and weakness of Dynamic Programming template matching method,Snakes or Active Contour template matching method, Active Testing template matching method, template matching method and neural network method. Semi-automatic road extraction of satellite images which is based on the road points prior knowledge in electronic map is proposed.At last, based on the analysis of the existing neural network method, an neural network road extraction with electronic map is proposed. In this method, using of electronic map access the longitude and latitude information of road, and then obtained the satellite image name by calculating the road point where the satellite image. In order to train neural network weights, at first reading the satellite imagery, then using local statistical features and texture features from image as the PCNN neural network's input to. Afterwards using image's demographic characteristics and texture characteristics of each pixel as neural network's input.The SVM to classify the final result of the road extraction through output of the network. The simulation results prove that this method can improve the extraction of road speed and accuracy, receive the expected result.