Urban Road Extraction Based On GF-2 Remote Sensing Image

Road is an organism and one of the important features in urban remote sensing image.It plays an indispensable role in navigation,traffic management,urban planning,geological disaster prevention and control planning and emergency rapid response,so the automatic extraction of urban roads is particularly important.With the rapid development of remote sensing technology,the time,space and spectral resolution of images have been greatly improved,and the research on road extraction by high resolution remote sensing images has gradually emerged.Compared with medium and low resolution remote sensing images,high resolution remote sensing images can show more ground object information,which not only defines the road contour structure,but also contains rich spectral information,texture information and geometric information,making it one of the most important data sources for urban road extraction.At present,the road extraction method emerges in endlessly,and the automatic road extraction method is also common.But urban surface environment is complex,traditional like element method is utilized to extract roads will appear the phenomenon of "foreign body with spectrum",based on the object-oriented method to extract features from the perspective of object,avoids the production of this kind of phenomenon,can be more accurate to extract city road from the complex scene,a universal,become the mainstream of urban road extraction research direction.There are still some problems when using object-oriented method to extract roads.First,the scale of ground object segmentation.Due to the different types of ground objects,the spatial scale suitable for its segmentation is also different.It is difficult to distinguish each terrain type accurately by using a single scale for image segmentation.Secondly,the boundary contour of ground objects is segmented by only spectral and shape information,without considering the edge of road,which is easy to cause the confusion of road and non-road.Third,the choice of the object characteristics.in the process of extraction to various features of remote sensing image classification,classification as the quantity of object features road classification accuracy is higher,but after more than a certain value,features the increased number may cause redundancy,lowered the road extraction accuracy;Fourthly,the problem of image classification.According to different segmentation scales,different ground objects have different appropriate classification levels,and it is impossible to achieve the classification extraction of each ground object only by using a single level.Aiming at the above problems,this paper proposes a multi-level segmentation method of urban road extraction under optimal features based on object-oriented.Firstly,the image is preprocessed and superimposed with the edge information map of the region to generate the initial image,which solves the boundary problem of image object segmentation.Secondly,the initial image image is segmented into image objects at multiple levels,and then the objects are combined and classified to achieve road extraction at multiple levels.To solve the problem of feature redundancy of image objects,a feature optimization method combining Relief F filter and PSO＿OPRF encapsulation is introduced.According to this method,the initial feature set of object is simplified,redundant features are reduced,and the optimal feature subset is selected to improve the extraction accuracy of road.Through the above method,GF-2 remote sensing images were used as data sources to select two study areas with the same pixel size in different cities for urban road extraction.The results show that the accuracy of ground object classification in the two study areas is above90%,and the accuracy of main road extraction is 0.934 and 0.938,respectively,is above 0.9.The extraction accuracy of the secondary trunk was 0.879 and 0.887,respectively.The precision of branch extraction was 0.918 and 0.896,respectively.The overall accuracy is over 0.85,and a good effect is achieved.The automatic extraction of complex urban roads is realized.The results were compared with the road quality extracted by Relief F＿OPRF,Relief F＿PSO＿RF and Relief F＿PSO＿J48 feature optimization methods under multi-level segmentation and the urban road quality extracted by single level segmentation method with optimal feature.It is verified that the road effect extracted by the method in this paper is better,the importance of building multi-level segmentation model under optimal feature for urban road extraction is illustrated.