Research On Extraction Method Of Road Network From Remote Sensing Image

In today’s transportation system,the importance of road network information in urban planning,map construction and intelligent transportation cannot be ignored.Generally,the mapping of road network is derived from high-resolution images taken by remote sensing satellites.How to extract road network from remote sensing images with high efficiency and high precision has always been an attractive research direction for which many scholars are keen on.As the resolution of remote sensing images continually increase,the feature information in the images becomes more complex,which puts forward the higher requirements for road network extraction.Aiming at the deficiencies in the existing road extraction research work,this thesis focuses on the clustering and segmentation of high-resolution remote sensing images,the road network extraction combined with mathematical morphology,and Hough transform based on the basic characteristics of the road.The main research content and results are as follows.Firstly,the research background and the research significance for the subject extraction are introduced,and some traditional research methods are summarized and analyzed.The difficulties in road network extraction in the algorithm design are described and analyzed.The thesis writing process and the main technical routes are briefly introduced in each chapter.Secondly,the traditional EM algorithm is improved for estimating model parameters,and it can adaptively classify images for clustering and segmentation.The new algorithm uses the Bayesian posterior probability of each pixel as the sample data to fit the Gaussian mixture model.When the EM algorithm estimates the parameters of the model,the spatial position information of the pixel is introduced through the fuzzy theory method,and the Akaike information criterion is adopted.Adaptively,the initial segmentation category is determined,and finally the minimum spanning tree algorithm is applied to achieve the final extraction of the road target in the image.For the application of remote sensing images,a comparison is made between this algorithm,traditional EM algorithm clustering segmentation and K-means algorithm clustering segmentation algorithms.The experimental results show that the improved algorithm effectively improves the accuracy of image segmentation,and reduces the average segmentation time.Finally,a closed operation method combining with multi-directional and multi-scale structure operators and improved Hough transform is proposed to extract road network from above the image segmentation result.This method combines the spectral,geometric,morphological,and contextual features of the road,and uses a multi-directional and multi-scale structure operator to perform feature extraction based on morphological weight processing on the segmented road image according to the special rules,and then according to the results of the feature extraction it performs image fusion,uses the improved Hough transform to extract the straight line of road,and combines the topological connection direction extension algorithm to connect the qualified road strip breaks on the extension line of the straight line,and finally utilizes the skeleton extraction and trimming to extract the road.The experimental results show that the extraction method in this study has the better extraction effect on remote sensing road images,it can completely extract road network,and has a certain adaptability to urban roads with more complex feature information.