Research On Landslide Deformation Analysis And Prediction Based On TLS Point Cloud Data And Convolutional Neural Network

Landslide is a very common type of geological disaster.It occurs quickly and has a short duration.If it is prevented in advance,it can quickly bury and bury the impact of the building within its scope of impact,to avoid causing a large number of casualties.As a serious geological disaster,landslides are unpredictable disasters for a long time in the past.Because of their complex causes,including earthquakes,heavy rainfall,explosions,increased groundwater levels,and changes in geological structure,landslides may be caused.And some special geologic areas are natural landslide bodies.It may only require a little gravity to cause landslides,and the landslide is an unstable disaster.The area that has occurred is likely to change because of changes in the surrounding environment,which pose a great security threat to the work done after landslides.The effective prediction and prevention of landslides can ensure the safety of people and property within the landslide impact area.Non-contact analysis of landslides can effectively avoid the loss of human and material resources for follow-up operations.Landslide prediction requires not only real-time deformation monitoring.It is necessary to judge the possibility of landslides on the future time scales in order to complete the overall prediction of landslides.In this context,this paper is based on multidisciplinary theories of surveying,mechanical engineering,software engineering,and machine learning,based on red rock landslide areas as research areas,TLS point cloud data,GNSS point measurement data,and remote sensing images.The landslides were reconstructed on the time scale to compare the model,and the final model comparison results were obtained,and the landslide stability was evaluated.The content of this paper is as follows:(1)The application of three-dimensional laser scanning technology in landslide deformation analysis is presented.Firstly,the classification and application fields of three-dimensional laser scanner are introduced.Then,the working principle and error theory of three-dimensional laser scanner are introduced.Then elaborate the entire processing flow of the point cloud data preprocessing stage,and implement one by one in the experimental section.(2)Combining the data and the profile of the experimental area,the point cloud data was processed and modeled by software such as Geomagic Studio12,PloyWorks,etc.The two-phase data model was used to compare the volume,projected area,and other aspects including Gird transformation.Hausdorff comparison and other methods analyze the deformation of the whole landslide body and obtain the overall deformation analysis and evaluation of the landslide body.(3)An improved convolutional neural network is proposed.Multiple eigenvectors of each point cloud data are presented in order and integrated using CNN(convolutional neural network).(4)Combining the time scale and the improved CNN to predict the displacement of the landslide monitoring point,using the improved convolutional neural network to compare and analyze the predicted and measured values of the landslide deformation,and obtaining the prediction result and evaluating the accuracy.Compare and analyze the accuracy of prediction results of different models.Through experiments,the point cloud data obtained through TLS can be compared with the intact landslide features after modeling,and the model has high accuracy and high fitting degree.It has excellent application in actual landslide deformation analysis and reflects the time scale.The various deformations of the landslide body,in the landslide prediction,the improved CNN technology trains through a large amount of data of GNSS point data,and actively optimize the structure in the network.Finally,it can accurately predict the displacement of the landslide monitoring point on the time scale.Technology plays an outstanding role in the deformation analysis and prediction of landslides.