High Resolution Remote Sensing Image Based On Object-oriented Road Extraction Research

The road is the important features in high-resolution remote sensing image, constitutes the skeleton of the city, is one of the fastest update fundamental geographic information of the high-speed development of the city. There are many algorithms for the road extraction, however, for high-resolution remote sensing image of the road extraction, although scholars have studied a lot, but in the present research, the studies are still in the experimental and theoretical stage, not fully applied to production practice. Therefore, the study on the road extraction from high-resolution remote sensing images has academic significance and practical value.This paper uses object-oriented image analysis method to research the method of the road extraction on the different scale of high-resolution remote sensing image. Firstly, on the basis of theoretical research, the paper studied the object-oriented image analysis of scale transforming, scale effect, segmentation parameter selection. According to the image information loss degree in the process of scale transforming of experimental evaluation, the paper test and verify the scale effect which is led by the different segmentation scale, determine the optimal parameters on multi-scale segmentation, proposed a kind of method of optimal segmentation scale based on the training sample, and identify the optimal segmentation scale on different road types; Secondly, the paper analysis the different road features of the research area, introduce a kind of visualization of feature selection method to achieve effective combination features, and create the image object hierarchy through the road scale conversion and selection. Combining the two classification method of nearest neighbor and fuzzy logic, the paper optimize the feature space, construct the knowledge base of the roads, and identify the road network; Thirdly, using the mathematical morphology, the paper get the whole road network in the post processing; Finally, the paper evaluate the multi scale road extraction accuracy by using the classification of stability and the confusion matrix, compare the method with the other, and analyze the contrastive result.The experimental results show that the method of object-oriented multi-scale road extraction can effectively solve the difficulties in the high-resolution remote sensing image road extraction, fully embodies the feasibility and efficiency of this road extraction method.