Urban Road Network Information Extraction Method Using Very High Resolution Remote Sensing Image

Road is a component of fundamental geographical infrastructure.The timely acquisition and updating of road information is of great importance to urban planning and construction,commercial development and applications,as well as the deployment of military resources.Road network is generally composed of a variety of roads.Urban road network is composed of all kinds of roads,such as urban expressway,trunk road,secondary trunk road,branch road,which form the framework of synthetic urban transportation system.Very high resolution(VHR)remote sensing images provide detailed information for extracting the urban road network.However,VHR images have a huge amount of data and grow rapidly.Manual interpretation alone is far from meeting the huge demand for timely access to road development and change information.The traditional pixel-based image analysis method has been proved difficult and ineffective in practice.By contrast,object based image analysis has been proved as an effective method to automatically extract geographic information from VHR image.However,nowadays methods perform well only for simple roads,they do not perform well on the extraction effect of roads with small linear features of branching and equal road width that are obscured by other features in VHR images,.How to improve the extraction accuracy of urban road network with multiple types of roads is urgently needed to be solved.To solve the above problem,this paper uses object-based image classification method,using WorldView 2/3 and domestic high score series(GF 1 / 2)satellite image as data source,synthetically using spectrum,texture,geometry and context,etc.Based on the optimal segmentation pattern,the multi-scale optimal segmentation hierarchical road extraction model is established,and the road network is extracted based on the independent data mining algorithm.The extraction results are classified and optimized based on the mathematical morphology method.Reconstruction of road networks.Through the above research,the following conclusions are obtained:(1)A multi-scale optimal segmentation hierarchical road extraction model based on fractal network evolutionary segmentation algorithm is proposed.The method can solve the problem that it is difficult to guarantee the precision of road extraction in the urban road network where many types coexist.According to the characteristics of different road width,the model divides the multi-scale urban road network into several parallel road levels.The multi-scale optimal hierarchical segmentation model is constructed to realize the traversal and cyclic extraction of the roads with different amplitudes in the urban road network,which improves the accuracy of road network classification and extraction.(2)Random forest with independent data mining platform,J48 decision tree algorithm based on classification and extraction method of object image analysis method,obviously on the city high resolution remote sensing images in complex scenes of the road network extraction effect.The method can automatically select the classification and threshold,thus avoiding the classification algorithm parameter cannot fine tune affect the classification accuracy.(3)After the classification and optimization of road extraction results based on mathematical morphology,it is better to extract the skeleton of urban multi-type roads and reconstruct road network.Based on maintaining the original information of roads,mathematical morphology can effectively connect broken roads.Fill the road hole,smooth the sawtooth edge,remove the road information,refine the road extraction result,remove the burr and optimize the extraction result,and reconstruct the whole urban road network.(4)The evaluation results of the quantitative accuracy of road network extraction by using the indexes of integrity,accuracy and quality are obviously superior to the experimental results in some literatures.In this paper,the complete rate and correct rate of road network extraction are obtained from four study areas.In this paper,the complete rate,correct rate and extraction quality results of the four research areas were 94.42%,89.23% and 87.61% respectively.95.08%,90.71%,88.34%;89.51%,84.70%,82.25%;92.79%,93.69% and 90.17%,the experimental results were satisfactory.