Research On Road Extraction From High Spatial Resolution Satellite Image

As foundational geographical Information, the identification and accurate positioning ofroad network is significant for image understanding, GIS data obtaining, cartography and actingas the reference of other objects. For the demand of remote sensing and mapping production, thispaper mainly studies the road characteristics of multi-source high spatial resolution satelliteremote sensing images, and researches the Angle Texture Signature for the intelligent roadextraction in the remote sensing images. The method is intelligent extraction road algorithmbased on the angular texture features templates and the grayscale template matching, and focuseson the road width acquisition, road centerline extraction and the road trajectory prediction model.The main work and innovations are summarized as follows:1. Based on a deep analysis of the status of the road network extraction in remote sensingimages, the problems of existing models, methods and the development trends in the future aresummarized.The paper rofoundly analyzes the road characteristics and the challenge ofautomation road network extraction in high-resolution remote sensing images, and describes thedetail of the advantages of texture feature extraction method, establishes the the foundation ofthe model.2. A road extraction model is constructed in this paper,starting on the angle texture signatureof the concept and the extraction methods.The road extraction accuracy greatly impacts on theroad seed points selected, The intelligent selection method to calculate the width and direction ofthe road are studied, which can achieve good results through one or three seed points.Theexperiment proves that the method can improve the robustness of the algorithm and theextraction accuracy of road extraction in high spatial resolution satellite remote sensing images.3. The vehicle occlusion handling policy is introduced into the angle texture signatureextraction model, the accuracy and reliability of the algorithm are greatly improved.4.According to the road characteristic gradation similarity, the grayscale template matchingmethods is constructed to road intelligent extraction, and the template matching measure isimproved,which is based on the variance, the mean value, the energy of the template secondmoment normalized, then weighted similarity matching is measured.5. The principle of the road trajectory prediction model is introduced the road intelligentextraction model, and through the parabola fitting and linear regression analysis, and thecurvature constraints to optimize road extraction results, the precision of road network extractionis greatly improved. 6Based on the angle texture signature model and the gray template matching model, theprototype system of road centerline extraction is achieved, and the results of road extraction isevaluated and analysised.