A Research On The Method Of Road Extraction From Remote Sensing Image

As the basic geographic information data,road is also one of the geographical information frequently updated in the urban database.The timeliness of road information directly affects the application of map drawing,intelligent traffic management,unmanned navigation and other aspects.It is a challenge to obtain meaningful road information quickly and accurately from massive remote sensing image data.At present,the method of road extraction from remote sensing images is mainly accomplished by manual vision.Due to the limitation of computer,theoretical algorithm and other technical levels,there is no perfect system for intelligent and universal road extraction.In addition,with the improvement of the resolution of remote sensing images,the information characteristics of ground objects in the images have become increasingly complex.To some extent,these information features cause interference to road identification,which further increases the difficulty of road extraction.Therefore,the research of remote sensing image road extraction is of great significance.Based on the basic features of roads,the paper explored the automatic extraction technology of roads in remote sensing images and proposed two feasible methods for road extraction,which are mainly as follows:Firstly,according to the radiation characteristics of the road,using the difference of gray value of the image,an improved threshold segmentation method based on genetic algorithm for OTSU was proposed to segment the remote sensing image.The genetic algorithm can make OTSU quickly and accurately find the optimal solution to achieve the best segmentation.After the median filtering process,the basic operations of the mathematical morphology(corrosion,expansion,open operation,closed operation)are used to remove the non-road area in the binary image,and then the edge detection "Sobel" operator is used to extract the extracted road.The net is placed on the grayscale image of the original image for comparison,which proves the feasibility of the method.Finally,the extraction results obtained by the improved segmentation algorithm are compared with the results of the original segmentation method,and qualitative and quantitative analysis are carried out,and the improved algorithm is better.Secondly,a road extraction method based on SVM combined with FCM clustering is proposed.Due to the shortcomings of the SVM algorithm for the easy classification of samples in a very small area,a FCM clustering method is proposed to improve the SVM deficiency.In the first,the unsupervised clustering of remote sensing images is performed by FCM clustering.It can expand the distance between different groups according to membership degree,and separate the road clustering and other clusters in the image.Then,the separated road cluster images are further classified by SVM,and then the blocky speckle noise still exists by mathematical morphology,and finally the road is extracted.Quantitative analysis of the accuracy of the results proves the feasibility of the method.