Research On Road Network Extraction From High-resolution Remote Sensing Images

Road information is one of the foundational parts of Geographic InformationSystem (GIS). It is useful in many fields, such as acquiring and updating GIS data,mapping and image interpretation. With the rapid development of remote sensingtechnology, image resolution is getting higher and higher and high-resolution remotesensing image is becoming the main source of road network extraction. Road networkextraction from high-resolution remote sensing image is still a challenging topicbecause of the high complexity of ground-object environment and noise effect inhigh-resolution images. In this dissertation, a research on road network extraction fromhigh-resolution remote sensing image is proposed. The main works are listed as follows:(1) Representative methods of road network extraction are analyzed and classifiedfrom the view of element’s level. The representative methods of road extraction areclassified as: a) road extraction methods based on edge detection b) road extractionbased on road model. In this way, the intrinsic character of road network extractionmethod is exposed which can help inspire new ideas on road network extraction.(2) The main features of the road in high-resolution remote sensing image isconcluded. This dissertation concludes four main features of road in high-resolutionremote sensing image: radiation feature, geometric feature, topological feature andcontext feature. The study of road’s main features in high-resolution remote sensingimage is the key to road network extraction.(3) A method based on relative entropy segmentation and mathematicalmorphology for road network extraction from high-resolution remote sensing image isproposed and implemented. Firstly, relative entropy segmentation is used for detectingroad area from the classified high resolution image, and then mathematical morphologyoperation is used for eliminating the non-road segment, and thinning the road lines. Byusing this method, other linear non-road residual objects in the image can avoid to beincorrectly classified as roads.(4) A method based on wavelet transform and Hough transform for road networkextraction from high-resolution remote sensing image is proposed and implemented.Firstly, wavelet transform is used for image denoising and detecting edges of roads.Then the image is segmented into road area and non-road area. Hough transform is usedto extract roads after segmentation. The effectiveness of the method is testified by theresults of the experiment.