Research On Road Extraction From High-resolution Remote Sensing Image Based On Spatial And Spectral Features

With the rapid development of remote sensing technology for earth observation,road extraction based on high-resolution remote sensing images has attracted more and more scholars' attention.However,the current road extraction methods are limited by factors such as shadow of remote sensing images and occlusion of ground objects,which make it difficult to obtain satisfactory results.At the same time,there is a lack of automatic and efficient 3D road extraction method in the creation of high-precision 3D road model.In view of this,the whole process of road extraction based on highresolution image is studied systematically from the aspects of image segmentation,road feature enhancement,seed point connection,point cloud filtering and elevation interpolation.To describe the spatial structure of the road more accurately,a new road shape index is proposed.To solve the heterogeneity of road spectrum in high-resolution remote sensing images,solutions to the problems of road shadow occlusion and road material change are proposed.Finally,a road surface extraction method that takes into account the characteristics of multi-scale space-spectrum is proposed and applied to the 3D road extraction method that integrates multi-source remote sensing data.This study will greatly enrich the selections of road extraction methods from high-resolution remote sensing images.The main work of this dissertation is as follows:(1)High-resolution remote sensing images have the advantage of rich ground information,but they are also more susceptible to shadow and occlusion,resulting in greatly increased uncertainty of road extraction results.In order to solve this problem,a road extraction method based on edge constraint and fast marching is proposed in this dissertation.Firstly,guided image filtering is employed to reduce the negative effects of nonroad pixels while preserving edge smoothness.Then,a multi-feature weighted fusion model based on edge constraints is proposed.The road spectral features,edge curvature features and road centerline features are fused to obtain road probability estimation results.Finally,the road seed points are tracked by fast marching method to obtain road centerline extraction results.Experimental results show that the proposed method improves the accuracy and efficiency of road extraction.(2)In high-resolution remote sensing images,road obstruction or material change will lead to the failure of road extraction.To tackle this issue,this dissertation proposes to use the color space transformation and geodesic method.First,the test image is converted from RGB color space to HSV color space to reduce the material change influence.The geodesic method is subsequently applied to extract initial road segments that link road seed points provided by users.At last,the initial result is adjusted by a kernel density estimation method to produce centered roads.Experiments show that the proposed method yields a substantial improvement over cutting-edge technologies.(3)The spectral characteristics and spatial structure characteristics of roads in high-resolution images are analyzed.Based on the first law of geography and multiscale segmentation,an automatic region expansion algorithm centered on segmented image objects is designed,and the shape index of the expanded region is calculated as the shape index of the central image object.Since the shape of the image object after multi-scale segmentation is irregular,the extended road shape index feature proposed in this dissertation is a shape description of the irregular image object considering the characteristics of the spatial extension of the road in a specific direction.Experimental results show that combining this feature with the spectral features of the road improves the accuracy of road extraction from high-resolution remote sensing images.(4)Based on the multi-scale characteristics of road features in high-resolution remote sensing images,the road probability estimation method considering multi-scale collaborative representation of spatial and spectral features is studied,and road surface extraction results are obtained by graph cut method.In order to obtain high-precision 3D road data,the airborne Li DAR data and road surface data extracted from highresolution remote sensing images are combined for 3D road extraction.New data processing algorithms are designed in three stages,which are automatic road layering,point cloud filtering and road elevation interpolation to complete 3D road extraction.At the same time,new quality evaluation factors are designed to evaluate the quality of the extracted 3D road data.The experimental results show that the accuracy of the threedimensional road data created by the proposed method meets the requirements of road models of high-precision maps.