Research On Road Network Extraction Algorithm In High Resolution Remote Sensing Images

Road extraction has always been a hot research topic in the field of image processing.With the emergence and rapid development of remote sensing technology,the imaging and road extraction methods are being pragressed.At the same time,the update of imaging technology has made the significant improvement of image resolution.The maximum resolution of remote sensing images can be up to 1 m/p,and the image acquisition method is becoming easier and easier.This greatly facilitates road extraction from high-resolution remote sensing images.However,due to the large and complex data information in high-resolution images,it is difficult to extract road network.Therefore,the traditional algorithms cannot effectively extract road information.And because the images have different characteristics in diverse types of scene,the algorithms for different scenarios are needed.In this thesis,the image enhancement algorithm and two kinds of road extraction algorithms are proposed for images of different scenes.The contents are as follows.(1)Image enhancement is an important step in image preprocessing.For reducing interference of non-road objects on the road,an enhancement algorithm to highlight the road information is proposed.In this algorithm,random projection depth function is used to obtain image depth.And the depth values are sorted according to the distance from central data point to obtain the depth field image.In the sort,the set of road data points is always far away from the central point,so the road information can be effectively enhanced according to this kind of feature.(2)In rural road image,because the image is simple(few ground objects),the image segmentation algorithm can be applied to quickly extract road.But in the segmentation processing,the modified gradient image is proposed to overcome over-segmentation problem of traditional watershed algorithm.Further,the random projection depth function is applied to process the image,and then highlighting road information.Next,the gradient image is obtained from the previous results,and it applied the adaptive threshold algorithm to get threshold value to generate its binary image and carry out object labeling.Finally,the watershed algorithm is used for segmentation to obtain the desired rural road information.The comparison experiments show that this algorithm can suppress the over-segmentation phenomenon and the image segmentation effect is satisfactory.(3)The road scene in the town is characterized by many and complex features.The image segmentation algorithm mentioned might not quickly and effectively extract town roads,so a combination of MRF and SVM algorithm is studied for road extraction.In order to overcome the defects that SVM algorithm is easy to be misclassified for small sample data in dataset,MRF is utitised for initial classification after enhancing road information in the depth field image.In addition,improving kernel function,namely synthetic polynomial kernel function is used to weigh the influence of different features in the road feature space on the extracted results.Next,the mathematical morphology algorithm is used to optimize the extracted road,to eliminate non-road areas and repairs and fills the road areas for more complete and smooth road areas.Finally,the skeleton extraction algorithm is applied to obtain the complete road network.The feasibility of the algorithm is proved by the quantitative and qualitative analysis of the experimental results.