Research On Road Information Extraction Method Based On Multi-scale Remote Sensing Image

In recent decades, remote sensing technology has developed rapidly, and its Big Data era has arrived. How can we obtain valuable information from massive data quickly and accurately has become a hot issue in research today. Road is a basic geographic information data. And the extraction efficiency, the speed, the accuracy of positioning of the ways to extract the roads have a profound impact on geographic information data updating, cartography, remote sensing image interpretation, traffic control and urban planning, etc.. From the practical point of view, taking multi-scale remote sensing image as the study object of road extraction, this article used the least squares template matching method, adaptive region growing method and support vector machine to extract the roads from images, and did feasibility experiment on the extraction method. And then we did quantitative comparative analysis to the experiment results, and achieved certain achievements.In this paper, aiming to research on the road extraction algorithm based on remote sensing images, we did main works as follows:(1) We summarized the status of road extraction algorithm based on remote sensing images, and did comparative analysis of the pros and cons of different methods at different times, learned and referenced the results of previous studies, in order to make foundation of putting forward our own extracting method.(2) We summarized the classification of our country’s roads, and analyzed the different characteristics of roads in different resolution remote sensing images.(3) In order to make the remote sensing images more suitable for road extraction, and weaken the noise interference, before the road extraction from remote sensing images, we needed to do some pre-processing to the remote sensing images. Remote sensing image preprocessing can largely affect the extraction results. So, combined with previous experience, we used Dudo’s road operator to do the remote sensing image sharpening, and did image smoothing with the method of median filtering or bilateral filtering.(4) For high-resolution remote sensing images, due to the complexity of the road information and too much noise, we proposed and implemented an improved road tracking method based on least squares rectangle template matching. Via the comparative experiment of the high-resolution remote sensing images’road tracking extraction methods of pattern matching, profile matching, angular texture matching which were improved in this article, we found that the integrity of pattern matching was 90.7%, which was higher than profile matching (81.5%) and angular texture matching (85.3%). The correct rate of pattern matching was over 90% and its quality was over 81.6%. The degree of automation of pattern matching was higher than profile matching by 46.9% and angular texture matching by 37.6%. The extracting speed of pattern matching was 45.8m/s, which was higher than profile matching (22.9m/s) and angular texture matching (30.8m/s). So we can prove that the method of pattern matching can accurately extract roads in high-resolution remote sensing images, and there was a certain practicality.(5) For medium-resolution remote sensing images, road information is relatively simple, gray scale is uniform, noise is little. This paper proposed an extracting method of the coupling of support vector machine and region growing. And, based on qualitative and quantitative experiments, we did comparative analysis on the three methods of medium-resolution remote sensing images’road information extracting based on support vector machine, region growing and their coupling. For the method of the coupling, the missing division error was 14.5%, the wrong division error was 8.8%, the overall classification accuracy was 90.7%, and is better than the method of support vector machine and region growing. The Kappa coefficient of the method of the coupling was 0.91, which was higher than the method of support vector machine (0.81) and region growing (0.72). So we found that the method of the coupling of support vector machine and region growing can realize the quick and effective road extraction, and can significantly improve the extracting accuracy.(6) In order to evaluate the effectiveness of the extraction of road information more effectively, according to the existing research results and combining with the actual situation of this study, we proposed to use integrity, accuracy, extracting quality, degree of automation, drawing time and other indicators to evaluate the effectiveness of the extraction method of road information based on remote sensing images.