| With the continuous development of UAV remote sensing technology,its advantages such as being fast,flexible,and having high resolution have made it widely used in 3D reality modeling,especially in various geological disaster investigations.In the filtering process of image matching point cloud data after 3D reality modeling,there are problems such as large data redundancy,poor filtering effect,and weak adaptability;there is also relatively little research on landform feature analysis and fine-scale measurement analysis of gullies due to the low extraction accuracy.Therefore,researching how to use appropriate flight plans for data collection in densely distributed gully areas,applying better filtering methods to filter image matching point clouds,and performing classification extraction and fine-scale measurement analysis of gullies’ landform features are the key to further research.This article takes a gully near the Dinosaur Valley in Lufeng County,Chuxiong Yi Autonomous Prefecture,Yunnan Province as the research area.Different unmanned aerial vehicle(UAV)shooting plans were designed for the gully,and experiments were conducted to analyze the feasibility of point cloud simplification and filtering techniques for the gully’s image matching point cloud during the 3D reality-building process.After obtaining the optimal data source,the study conducted terrain extraction,landform analysis,and fine-scale measurement analysis on the gully.The main results of the study are as follows:(1)Multiple data acquisition UAV flight plans were constructed,including traditional aerial photography plans,circumferential aerial photography plans,and near-ground level plans.Data was collected and 3D modeling was conducted on the study area,followed by accuracy evaluation analysis and applicability analysis using texture features,plane,elevation,and characteristic edges.The data acquisition plans of traditional aerial photography plans and near-ground level plans were used as a basis for in-depth analysis and research on the gully area.(2)Through the research on key technologies for filtering image matching point clouds,it was found that using the Voxel Grid filtering method with a grid size of 0.4m to simplify point clouds can effectively reduce the volume of point cloud data while retaining details,with an overall good effect.The elevation normalization CSF filtering method proposed in the study was validated using ISPRS standard data,with Samp51’s class I error and total error both able to reach 1.96% and 3.31% respectively,and Samp53’s class I error and total error both able to reach 8.05% and 8.38%,indicating the practical feasibility of the method.The method can effectively reduce the total error and has better filtering optimization effects and practical application reference value than conventional single filtering methods when applied to the research area in this study.(3)The study in this paper used high-resolution image Digital Orthophoto Map(DOM)in combination with terrain factors as the core layer for surface feature extraction processing.Firstly,optimal scale segmentation experiments were conducted to determine the best segmentation scale parameter combination for the gully.Then,feature selection was conducted based on the gully,using Support Vector Machine(SVM)algorithm,nearest neighbor classification,and decision tree algorithm for surface feature classification.Finally,each surface feature classification result was compared and analyzed with randomly selected samples.The results showed that the decision tree classification had the highest accuracy,with an overall classification accuracy of 0.9164,and the highest Kappa coefficient of 0.8928 was achieved.This empirical evidence confirms the effectiveness of the proposed surface feature extraction processing technology that considers the participation of the DOM and the optimal scale segmentation parameter combination.(4)Through the identification and mapping measurement analysis of the gully in the research area,it was found that the main area is divided into 5 small gullies and 2major gullies,with the longest gully being 693.462 m,the largest area being1890.606m2,and the maximum elevation difference being 46.827 m.Relative elevation,slope,slope direction,and other information were extracted,revealing that the elevation difference mainly concentrated between 30 m and 120 m,and the research area appeared to have an irregular ring-shaped outward expansion shape as the elevation difference increased.The main direction of the gully’s slopes falls mostly between 18° and 47°,accounting for 58.82%,and the proportion of south-facing slopes was the largest,reaching 35.49%.The terrain in the southwest of the research area is lower than that in the northeast.In addition,a typical gully,QSG#1,in the research area was selected for landform feature measurement analysis,finding that its average elevation was1491.76 m,the elevation difference was 41.8m,and the slope was steepest between 30°and 40°,with many steep and inclined slopes.The slope faces mainly north,accounting for 26%,and there is a high risk of slope-related instability.A fine-scale analysis was conducted on representative samples,revealing that the gully has a lot of sediment in the bottom area and vertical facades surrounding it,as well as clear texture information inside the gully.In summary,this paper focuses on the high-precision 3D modeling and terrain geomorphic feature identification and quantification research of typical mountainous gullies in the central Yunnan plateau.It includes designing and analyzing UAV data collection schemes,key technical processing such as image matching and point cloud filtering,and studying the extraction,classification,and measurement of terrain features specific to gullies.The main achievements of the study include the validation of the proposed high-elevation-normalized CSF filtering method,which demonstrates a better filtering effect compared to conventional single filtering methods,and the empirical verification of the effectiveness of the proposed terrain feature extraction processing technology that takes into account both digital orthophoto maps(DOM)and optimal scale segmentation parameter combination. |
