中国岩溶2025，Vol.44Issue(1)：89-99,11.DOI:10.11932/karst20250106
茂密植被山区岩溶漏斗遥感识别方法
Research on remote sensing identification of dolines with dense vegetation cover based on point cloud principal component analysis point cloud principal component analysis
摘要
Abstract
China's karst areas are widely distributed,with a total area of over 3.4 million square kilometers,accounting for one-third of the country's total land area.Karst engineering geological problems are widespread and difficult issues in the development of water conservancy and hydropower in Southwest China.In order to achieve the goals of"carbon peak"and"carbon neutrality",China is currently accelerating the construction of pumped storage power stations.However,karst leakage is one of the most important engineering geological problems faced by the construction of pumped storage power stations in areas with carbonate rock development.Therefore,in order to avoid reservoir leakage,it is of great significance to comprehensively,quickly,and accurately identify locations of dolines in reservoir and dam areas for water conservancy and hydropower engineering construction. The traditional investigation of dolines in reservoirs mainly uses the method of manual ground investigation and drilling.However,in the mountainous areas with dense vegetation cover in Southwest China,manual investigation is often very inefficient and limited by the strong concealment of dolines,making it difficult to achieve accurate and efficient investigation of large-scale dolines.To solve the problem of low efficiency in the investigation of dolines in dense vegetation areas,this article takes a reservoir area of pumped storage power station as a study area and proposes an automatic identification method for dolines based on point cloud principal component analysis to quickly identify and extract dolines in the reservoir area. The study area is located in the northeast of Sichuan Province,with an area of 10 km2.It is situated at the junction of the Qianlongmen mountains and the northwest of the Sichuan basin,in the tilted core of the Yangtianwo area.The terrain slope is generally above 30°,and the maximum relative height difference in the area is 1,070 m.The main vegetation in the area is forest,consisting of evergreen broad-leaved secondary forests and shrubs.In the study area,carbonate rock formations,such as the Upper Devonian Maoba Formation(D3m),Shawozi Formation(D3s),and Middle Devonian Guanwushan Formation(D2g),are mainly exposed with strong karst development. Firstly,airborne LiDAR technology was used to obtain ground 3D point cloud data in the study area after vegetation was filtered out.Then,in response to the strong directional characteristics of the concave shape of dolines,a preliminary extraction of dolines was achieved by the K-D tree nearest neighbor algorithm and principal component analysis,and an indicator eigenvalue ratio p was proposed.Finally,three filtering algorithms based on density clustering algorithm,namely,funnel frequency,length,and direction,were used to filter the background noise of the initial extraction results.The method used Receiver Operating Curve(ROC)to test the AUC value which was 0.854,and F-score is 0.859.This method is suitable for identifying and investigating dolines in karst areas with dense vegetation cover.关键词
机载LiDAR/岩溶漏斗/三维点云/自动识别/主成分分析Key words
airborne LiDAR/doline/point cloud/automatic identification/principal component analysis分类
信息技术与安全科学引用本文复制引用
符明俊,何阳,董秀军,邓博..茂密植被山区岩溶漏斗遥感识别方法[J].中国岩溶,2025,44(1):89-99,11.基金项目
国家自然科学基金(42072306) （42072306）
