Automatic Identification And Stability Analysis Of Rock Mass Structural Planes Based On UAV Photogrammetry

There are many mountainous areas in China,with complex geological environments and frequent rock mass instability and collapse,which threaten the safety of nearby residents’ lives and property,and have a serious impact on the country’s economic and social development.The structural plane of rock mass is a key factor affecting the quality and stability of rock mass.The occurrence,trace length,spacing,density and other parameters of the structural plane have an impact on the stability of rock mass to a certain extent.The investigation,acquisition,and statistics of information on the structural plane of rock mass are the foundation for studying rock mass stability and the focus of rock mass exploration work.In rock mass exploration,traditional surveying methods are mostly contact measurement,which is time-consuming,labor-intensive,and costly.At the same time,there are certain risks.This thesis proposes a set of automatic identification methods for rock mass structural planes based on drone photogrammetry to address the problems of manual exploration.From multi perspective drone close photography to the establishment of rock mass 3D point cloud models,structural plane recognition and attitude calculation are achieved based on rock mass 3D point cloud models,By utilizing algorithms to automatically extract the spacing and trace length information of structural planes,a three-dimensional network model of rock mass structural planes is established,and the entire algorithm is applied to engineering examples to analyze the stability of slope rock mass through stereographic projection and improved CSMR method.The main achievements achieved are as follows:(1)High resolution photos of slopes were obtained through multi view drone proximity photogrammetry,and a three-dimensional model of the slope was established based on the SFM-MVS algorithm.Taking the first level steep cliff of Zengziyan in Nanchuan,Chongqing as an engineering example,a multi view close photogrammetry of the research area was completed using drones,and a three-dimensional point cloud model of the Zengziyan research area was established.(2)An adaptive KNN algorithm was proposed to overcome the problem of point cloud missing in traditional KNN algorithms.The accuracy of point cloud clustering grouping was improved using the genetic simulated annealing fuzzy C-means clustering algorithm(SAGA-FCM);The structural plane recognition algorithm was validated using the hexahedral point cloud model,and the adaptive KNN algorithm reduced its point cloud missing rate from 10.18% to 1.9%.The maximum error between the clustering results of the SAGA-FCM algorithm and the actual results was 0.69 °,verifying the reliability of the entire structural plane recognition algorithm;Applied to the high and steep slope of Zengziyan in Nanchuan,Chongqing,the adaptive KNN algorithm reduces the point cloud missing rate from 4% to 0.4%,effectively preserving the same size and boundary of the structural plane as the original slope.The maximum error between the occurrence of the structural plane and the measured value is 4.82 °,and the recognition effect of the structural plane is good,which can provide information reference for subsequent engineering applications.(3)A method for automatically extracting structural surface information was proposed to establish a three-dimensional structural surface network model.Taking the recognition area 3 of Zhenziyan in Nanchuan,Chongqing as an example,the accuracy of the automatic extraction algorithm for structural planes was analyzed and verified.The results of automatic extraction of structural plane spacing and trace length were compared with the measured results of the model,and the maximum relative error between the two was only 2.67%,which can meet the needs of engineering.(4)The proposed full set of automatic recognition algorithms were applied to the G351 highway slope in Beibei District,Chongqing,and the following conclusions were obtained: establishing a three-dimensional point cloud model through multi perspective drone close photogrammetry;By using the adaptive KNN structural plane recognition algorithm,the occurrence of structural planes was obtained,and compared with the two sets of structural plane occurrences measured on site.The maximum error was 3.27 °,and the maximum deviation ratio was 6.24%,indicating a good recognition effect;Extract the spacing and trace length of each group of structural planes through automatic extraction algorithm of structural plane information,draw frequency distribution histograms,fit the optimal curve distribution law,and establish the optimal 3D structural plane network model;Finally,based on the stereographic projection and the improved CSMR method,it is concluded that the slope rock mass is partially stable and may cause sliding failure of some structural planes or wedges.