Research And Implementation On Automatic Extraction Of Topographic Feature Lines On Loess Plateau

Shoulder lines,ridge lines,and valley lines are the three most important terrain feature lines in the Loess Plateau.Shoulder lines,as a unique terrain feature line in the Loess Plateau,not only serve as the boundary of soil erosion degree but also separate positive terrain between gullies and negative terrain within valleys.Ridge lines and valley lines,as important geomorphic features in the Loess Plateau,constitute the backbone of the undulating terrain.Accurately extracting terrain feature lines is not only of great practical significance for Loess Plateau geomorphology classification,geographic information system analysis,and terrain model simplification,but also has important theoretical implications for controlling water and soil loss,guiding ecological restoration,and promoting regional sustainable development.Image segmentation algorithms can accurately locate and analyze objects or specific regions in images and have gradually been applied to feature extraction in digital terrain analysis.Currently,common terrain feature line extraction methods overlook the joint influence of different terrain feature factors on the extraction of feature lines,resulting in large spatial positioning errors,omission,and poor continuity of feature lines.In response to these issues,the main research contents of this study are as follows:（1）A Region Fusion-Based Method for Extracting Shoulder Lines.The formation mechanism of shoulder lines is determined by multiple factors such as soil erosion and geomorphological development.Using a single terrain factor such as slope to extract shoulder lines often results in problems such as broken or disconnected feature lines and low extraction accuracy.In this study,we improve the region growing algorithm by incorporating multiple terrain feature factors into the growth criteria for pixel-level feature selection.Using the slope variation method and a smooth DEM,we obtain initial growth points for gully lines and negative terrain,respectively.We then expand these points using the improved region growing algorithm and fuse the resulting regions to extract shoulder lines.Qualitative analysis shows that the region fusion-based method for extracting shoulder lines can integrate different features for screening,producing good results in areas where slope changes are not significant.The method provides better spatial positioning that is closer to the real shoulder lines while also providing a better fit to the contour compared to methods that rely on single terrain feature factors.Quantitative analysis demonstrates that the proposed method can accurately extract complex shoulder lines in the Loess Plateau while ensuring the integrity of the extracted features(EDOP₁₀:96.9%,MAE:2.98,SAD:3.95).（2）Ridge and valley line extraction method based on TFL-Seg Net.As the mathematical features of ridge lines and valley lines are similar,this study combines the two and extracts them using the same model.In this study,the Seg Net deep learning semantic segmentation model was used as the basic framework,and a TFL-Seg Net was constructed by integrating bottleneck attention mechanism and the improved pyramid pooling model to extract mountain ridge and valley lines.To explore the impact of different terrain feature factors as channels on the extraction results,experiments were conducted on different channel numbers and features.It was found that the optimal extraction results were obtained when using five channels of features based on DEM,TPI,and remote sensing images as model inputs.In the quantitative analysis,compared with common deep learning semantic segmentation models,the TFL-Seg Net model achieved F-score and Io U of 85.26%and 74.31%for mountain ridge line extraction,and F-score and Io U of 86.76%and 76.61%for valley line extraction.In the qualitative analysis,the extracted feature lines by TFL-Seg Net had fewer fractures at branching points and could better identify some small branches.（3）Design and Implementation of Terrain Feature Line Extraction System.In order to combine the proposed feature line extraction algorithm with practical applications,this study designed and implemented a terrain feature line extraction system,which provides a web-based platform,as well as desktop versions for Windows and Mac,to facilitate the use of the research results by related researchers.Based on the above research contents,this study has developed a high-precision terrain feature line extraction scheme for the Loess Plateau.The proposed scheme can extract the lines of gullies,ridges,and valleys while maintaining a high degree of fit between local details and the real terrain,and ensuring the continuity and integrity of the extracted terrain feature lines.This study provides feasible technical references for the work of terrain feature line extraction.