Research On Methods Of Typical Road Extraction From High-resolution Remote Sensing Image

Road is typical artificial object, which is a very important part of city structure as the skeleton of the city. With the rapid development of science and technology, the resolution of remote sensing satellite image is in constant increase. Research on extracting target feature information from high-resolution remote sensing image has become a hotshot in recent years. In this paper, the methods of road extraction from high-resolution remote sensing image are studied. Road can be classified into different types according to its curvature and structure feature. Panchromatic images of resolution from zero point four one meters to two meters and multispectral images of four-meter resolution are applied to realize the accurate extraction of road information from different types of road images.At the basis of analysis of road feature, the high-resolution remote sensing image is preprocessed in order to enhance target information and weaken the interference of background noise. Then the image is segmented for the purpose of separation of target and background objects. After the segmentation, road information is accurately extracted by shape index, and then multi-direction morphology structural element is used to operate the image in order to smooth the road information and fill the small holes.In order to solve those two problems existed in the preliminarily extraction results of road information: attached non-road information that can not be deleted and fractures in the main body of road caused by the overlap of some traffic signs. This paper put forward a novel road purification method based on tensor voting, which can delete attached non-road information, connect the gaps so that the pure and complete road information can be extracted. On the basis of complete road extraction result, detecting ball saliency based on tensor voting combined with the characteristics of road intersection can realize the accurately detection and identification of road intersection information.After complete road information has been extracted, the algorithm based on LSD is proposed to optimize the extraction result of road centerline of straight style. The road centerline of curve type is extracted by voting in stick tensor field combined with the characteristics of road intersection. The road networks can be classified according to the structure feature into four types: grid style, ring radial style, free style and hybrid style, then the feature and application of different style road networks are analyzed. The conclusion that the methods proposed in this paper have high accuracies and strong robustness can be arrived by the analysis of experimental data from different satellites with different resolutions and different styles of road information, for that all types road information can be extracted accurately by method of this paper.