Extraction of loess shoulder-line based on object-based image analysis

Based on the morphological characteristics of loess hilly and gully areas, a simple and effective method for extracting loess shoulder-lines based on object-based image analysis is proposed. A case study is conducted in the northern Shaanxi Loess Plateau of China and the digital elevation model with a grid size of 5 m is applied as original data in this research. In order to get clear edge of DEM slope, the convolution filter by Gauss Blur is applied to the original DEM. The slope image is calculated on the smoothed DEM. Then the landform DEM objects are obtained by contrast split segmentation. For the standard deviation values between gullyslope land and loess interfluve are obvious different, the basic loess shoulder-lines could be extracted. While there are still small areas and other disturbing objects, the shoulder-lines will be refined by context and area values. At last, the final results are merged and exported. According to the accuracy assessment, the average extraction accuracies are about 88.7% and it is applicable in the loess area analysis.