Road Extraction Based Decision Support Strategy

Road is a very important image information in the remote sensing. The accurate extraction of the road has a very important practical application in our life. The traditional way to extract road information is spot plotting. With the rapid development of remote sensing technology and the improvement of remote sensing images resolution, more information could be found in remote sensing images. More scholars study automated road extraction by means of various algorithms, greatly improving the efficiency of road extraction.The thesis mainly focuses on extraction of road feature from remotely sensed image. Road feature is obvious in Remote sensing image. A segmentation algorithm is used to extract road feature by means of road line-shape. And context-based road characteristics, we use the classification method to distinguish roads and other surface features. Finally, I introduced skeleton extraction algorithm to extract the road network.This paper studies the segmentation algorithm firstly. The purpose is to segment the lane boundary, thereby performing road feature extraction. This paper adopts an improved algorithm based on GVF-Snake algorithm to segment. Aiming at the weakness and shortage of GVF-Snake in the less tangible of road edge information, a new segmentation algorithm with internal support functions based GVF-Snake is proposed. The example shows that the improved algorithm is better than traditional GVF-Snake agglomerate in results and time performance.Then introduced to a method of skeleton extraction based on maximum disk model. Since this algorithm is not taken into account the connectivity of skeleton. This paper combined Hough transform to judge the connectivity of the skeleton. For the connectivity which satisfies conditions, we using the line extension to connect the connectivity.Based on the contextual features of road, we adopt advanced classification method to extract road information. Once the classification is completed we receive one of two answers: road information and non-road information. So we use the CART classification algorithm. Because the CART is a binary tree, its leaf nodes only two properties, the target and non-target attribute properties.Finally, suburban and urban these two experimental dates were experimented. The experimental results demonstrate the feasibility and universal of the road extraction method mentioned in this paper. And the method can effectively reduce road extraction time, and reached high extraction accuracy.