Research On Identification Of Slope Slip Area In Open Pit Mine Based On UAV Tilt Photogrammetry

As one of the major disasters in the open pit mine,landslide will endanger the life safety of on-site operators and affect the normal mining.The accurate identification of landslide is the basis for landslide hazard analysis,rapid identification of landslide area and other work,which can effectively reduce the hazard of on-site investigation,improve work efficiency,and provide decision-making support for technicians to take protective and governance measures.With the development of unmanned aerial vehicle(UAV)technology,the UAV tilt photogrammetry technology can provide high-precision data for the research of surface deformation of open-pit mine slope.In this paper,supported by the tilt photogrammetry technology of UAV,the real 3D model of the open-pit mine slope is constructed by collecting the image data of the open-pit mine through multiple aerial photography,and the orthophoto image map(DOM)representing the characteristics of optical image and the digital elevation model(DEM)representing the characteristics of terrain change are produced.The landslide area of the open-pit mine is segmented and classified by using the object-oriented method,Realize the identification of landslide area,and carry out precision inspection on the identification results.The main research contents of this paper are as follows:(1)Data acquisition and real 3D model construction in the study area: the construction of real 3D model of ground surface based on UAV tilt photography technology mainly includes two parts: field data acquisition and office data processing.The field is used to collect image data,and the office is used to build real 3D model based on data sets.Through the experiment and error analysis of the selected open pit mining area,the plane error of the model is 0.048 m,the elevation error is 0.105 m,and the overall error is 0.106 m.The result is far less than the standard value,which can be used for subsequent experimental analysis.(2)Landslide image segmentation in the study area: the optical image and terrain data set is constructed by multi band synthesis using the digital elevation model(DEM)and orthophoto image(DOM)data produced by the real 3D model as the data source.According to the aggregation principle of hierarchical clustering,the object-oriented segmentation of landslide image data set is carried out.The experiment was improved on the basis of multi-scale segmentation.The results were optimized by adding spectral difference segmentation.The multi-scale segmentation and the improved segmentation method were experimented several times respectively.Finally,the optimal segmentation scale parameter in the experimental area was determined to be 180,the shape index was 0.1,the compactness was 0.5,the band weight of multi-scale segmentation was 1,the maximum spectral difference value was 7,and the spectral band weight of spectral difference segmentation was 1.(3)Landslide image classification of the study area: according to the segmentation results of the landslide image area,first determine the characteristics of different segmentation categories,including spectrum,shape,texture,terrain and other feature types,and then distinguish different ground object categories.Based on the principle of fuzzy classification,the experiment uses the membership function classification method and the nearest neighbor classification method respectively.The membership function mainly determines the category characteristics through expert knowledge,multiple experiments and parameter adjustment,and realizes manual classification of landslide areas based on the principle of maximum membership;The nearest neighbor classification is consistent with the supervised learning idea.It is mainly based on the proximity principle to carry out fuzzy classification.Through selecting samples,optimizing the feature space parameters,and interacting with samples,the automatic classification of landslide areas is realized.The accuracy of the two classification methods was tested according to the confusion matrix.The overall identification accuracy of the landslide in the selected study area was 93.5% and 95.9% respectively,and most of the slip areas were effectively identified.(4)Landslide area recognition of Shaanxi XW open-pit mine: Based on the above research,this paper selects an open-pit mine in Shaanxi as the research object.Through UAV aerial photography and modeling,improved image segmentation,and fuzzy classification of landslide areas,the two areas with landslide disasters are finally identified.By comparing the classification and recognition effects of the two methods,it can be seen that the nearest classifier has higher accuracy,The identified landslide area contour is more complete,which verifies the effectiveness of the object-oriented segmentation and classification method and completes the accurate identification of landslide area.