A Model-Driven To Sample-Driven Method For Rural Road Extraction

The road network extraction by remote sensing image has gradually become the main way to update road information.With the revitalization of rural areas,transportation takes the lead,rural roads play an increasingly significant role in the development of rural regional economy.There are irregular curvature changes,narrow road width and diversified pavement materials in rural roads.When the existing methods extract rural roads,the problems of low automation exist when the accuracy of road extraction results is high and the recall rate is high.Therefore,this paper proposes a method of rural road extraction from model driven to sample driven,and the main work is as follows:(1)Remote sensing image preprocessing.Firstly,from the perspective of highlighting road structure information,an algorithm for segment sequence extraction is proposed.In this method,the edge of the object is expressed in the way of line arrangement by phase constraint and distance constraint,which improves the ability of road edge analysis;Secondly,through the quantitative analysis of a variety of filtering methods,L0 smooth filtering method is selected to increase the homogeneity inside the road and the difference inside and outside the road;finally,in order to further improve the difference between the road and non-road,this paper fuses the high-resolution panchromatic remote sensing image and the corresponding multispectral image to generate the multispectral remote sensing image with rich color.At the same time,in order to quantitatively simulate the color difference of human perception,this paper transforms RGB color space into HSV nonlinear space,so as to accurately analyze the difference between road and non-road.(2)A model driven rural road extraction method is proposed.In view of the narrow and long geometric characteristics of the road,this paper adopts the line segment sequence to express the edge of the ground object,which overcomes the problem of the lack of topological relationship only relying on the line segment,greatly improves the expression ability of the curved road edge,and provides the prior length information for the determination of the road edge line.On this basis,the edge of the road is determined by the length constraint and gray contrast constraint of the line segment sequence.The center of the road is extracted by the overall homogeneity constraint of the road,and part of the road is extracted by topological connection analysis;(3)Based on the results of model driven road extraction,a-driven rural road extraction method is proposed.Firstly,the two end points of the road point set obtained by the model driven method and the manually input road center point are used as the initial sample points;Secondly,a local road direction prediction model is established.Considering the curvature mutation of rural roads,the length of a single line segment is difficult to ensure the accuracy of direction prediction.Therefore,the MLSOH model is improved by using line segment sequence to reduce the interference of other direction structure information on road structure information and expand the perception ability of the machine on road structure.Then,a multi circle template is designed to analyze the difference Finally,an interactive matching model between panchromatic image and HSV space is proposed.Aiming at the problem of insufficient information of panchromatic image,this model uses dynamic weight allocation to strengthen the difference between road and non-road in HSV color space and further improve the accuracy of matching.Combining the above three steps,this paper proposes a rural road extraction method from model driven to data driven.Three high-resolution remote sensing images of different types,different regions and different scenes are selected in the experiment,and the proposed method is compared with the three road extraction methods.The experimental results show that the algorithm proposed in this paper can greatly reduce the degree of manual participation on the basis of ensuring high accuracy of road extraction results,high recall rate and vector storage.