A new extraction method of loess shoulder-line based on Marr-Hildreth operator and terrain mask

Loess shoulder-lines are significant structural lines which divide the complicated loess landform into loess interfluves and gully-slope lands. Existing extraction algorithms for shoulder-lines mainly are based on local maximum of terrain features. These algorithms are sensitive to noise for complicated loess surface and the extraction parameters are difficult to be determined, making the extraction results usually inaccurate. This paper presents a new extraction approach for loess shoulder-lines, in which Marr-Hildreth edge operator is employed to construct initial shoulder-lines. Then the terrain mask for confining the boundary of shoulder-lines is proposed based on slope degree classification and morphology methods, avoiding interference from non-valley area and modify the initial loess shoulder-lines. A case study is conducted in Yijun located in the northern Shanxi Loess Plateau of China. The Digital Elevation Models with a grid size of 5 m is applied as original data. To obtain optimal scale parameters, the Euclidean Distance Offset Percentages between shoulder-lines is calculated by the Marr-Hildreth operator and the manual delineations. The experimental results show that the new method could achieve the highest extraction accuracy when σ = 5 in Gaussian smoothing. According to the accuracy assessment, the average extraction accuracy is about 88.5%, which indicates that the proposed method is applicable for the extraction of loess shoulder-lines in the loess hilly and gully areas.