A Method For Terrain Feature Points Extraction Of Grid DEM Based On Sample Enhancement Using Random Forest

Topographic data is the basis of geological analysis and application,and is widely used in the fields of hydrology,agriculture,geology and geomorphology,land utilization,natural disasters and so on.However,the DEM data of grid used to represent terrain is huge and includes redundant information of terrain.In the analysis and application of Geosciences,it is usually necessary to capture the main terrain features with rich terrain information from DEM data,and expect to express the terrain surface as accurately as possible with as few terrain features as possible,so as to improve the storage and processing efficiency of terrain data.Therefore,it is of great significance to extract terrain features accurately in reducing the redundancy of terrain data,meeting the multi-scale requirements of spatial data and building efficient spatial database.In this paper,the paper proposes a new method of extracting terrain feature points based on the random forest samples,aiming at the limitation of the insufficient consideration of the terrain fluctuation in the grid when extracting the terrain feature points by the maximum Z tolerance method and the important point method.The paper gives full play to the advantages of machine learning in processing high-dimensional data and nonlinear fitting,and proposes a method of extracting terrain feature points more accurately,Keep more terrain details and richer terrain features.Firstly,the basic idea of extracting terrain feature points based on sample enhancement is given: Taking the spatial coordinates of the original DEM grid center point as the feature variable and the elevation as the dependent variable,the regression model of random forest is constructed.The elevation value of DEM data is enhanced by space interpolation through the spatial coordinates of the random points inserted in the grid;The difference between grid center point and the mean elevation of random point in grid is taken as the measurement information of grid importance.The grid node with high importance is selected as the terrain feature point,so as to realize the extraction of terrain feature points.Then,based on the research idea,the paper carries out the experiment of extracting terrain feature points based on sample enhancement,draws the spatial distribution map of terrain feature points,and reconstruct the DEM of original scale according to the terrain feature points.Taking the grid points of the original DEM as the inspection point,the error statistical index of reconstruction DEM is calculated.The paper compares the maximum Z tolerance method and the important point method,and evaluates the extraction effect of the terrain feature points by combining qualitative and quantitative methods from three aspects: the spatial distribution of terrain feature points,the error statistical index of reconstruction DEM and the matching of feature lines.Finally,the average slope of terrain is selected as the measurement index of terrain complexity,and four test areas with different terrain complexity are selected for the extraction of terrain feature points.The error statistical index of DEM reconstruction under different simplification rates in different test areas is obtained,and the relationship between error statistics index and average slope is drawn,The error statistical index of sample enhancement method and other two methods is analyzed.The experimental results show that the method of extracting terrain feature points based on sample enhancement can extract the terrain feature points effectively,and can take into account more terrain details while retaining the overall structure of the terrain;Compared with the maximum Z tolerance method and the important point method,the DEM reconstruction method has higher accuracy in the terrain area with high terrain complexity.The method of extracting DEM terrain feature points based on machine learning sample enhancement and the related research results have certain reference function and applicability in the digital terrain analysis and application,which can provide a new way for high-precision terrain simplification.