2026-10-11T05:12:30Z https://www.scidb.cn/oai

10.57760/sciencedb.j00001.01752 2026-08-10T15:20:07Z

2026-08-10 30-meter resolution topographic feature line dataset of typical landform areas in the Loess Plateau of Yan'an doi:10.57760/sciencedb.j00001.01752 en This dataset can serve the relevant research on extracting terrain feature lines of typical geomorphic areas on the Loess Plateau based on deep learning methods. The spatial resolution is 30 meters, and prior research has used a multi-channel input method of terrain factors and remote sensing images to create a deep learning model. Combined with terrain feature line labels, it can be divided into three parts as a whole. 1. Extract terrain factor data required for terrain feature lines. This article ultimately utilized seven terrain factors, including elevation, slope, aspect, surface curvature, surface roughness, mountain shadows, and terrain undulation. The DEM (elevation) was obtained using Copernicus 30 meter resolution data provided by the European Space Agency from 2010 to 2015, while the remaining terrain factors were extracted and normalized using ArcGIS. 2. Extract remote sensing image data required for terrain feature lines. This article downloads Landsat-8 land satellite remote sensing images (with a spatial resolution of 30 meters, using low cloud cover remote sensing images from the winter of 2013 to align DEM production time and reduce vegetation impact) from the geographic spatial data cloud platform, preprocesses them, and then crops and normalizes them according to the boundaries of the study area. Finally, the visible light bands Band2 (blue), Band3 (green), and Band3 (red) are synthesized. 3. Terrain feature line label data. In order to effectively obtain the annotated data of mountain ridges and valley lines in the experimental area, this study first adopts the extraction method of hydrogeology, uses ArcGIS spatial analysis tool to preprocess and fill the original DEM data, and calculates the flow direction information of each grid unit. Based on the flow direction analysis results, the catchment area of each grid is extracted, and the river network is regarded as the valley line. Perform anti terrain processing on the DEM and repeat the above steps to obtain the ridge line. Create line features in ArcGIS and use terrain factors and remote sensing images to further assist visual interpretation in annotating ridge and valley lines. After multiple stacking, correction, and manual inspection, the obtained features were finally converted into grids and used as labels for the model. In the labels, 0 represents the background, 1 represents the valley line, and 2 represents the ridge line. After obtaining these three parts, they were uniformly normalized to 8 bits. The grid image with a size of 490 × 512 was cut according to a window size of 64 × 64 and a step size of 64. Then, during training, the data augmentation step was performed to expand the total dataset to about 3500 images, divided in a ratio of 7:2:1 between the training set, validation set, and test set.  Loess Plateau; terrain feature lines; 30-meter resolution; deep learning liu song lin Yang Weifang Lu Xiaomin Bao Junfan Zhou Yu Li Baobo EMBARGO https://creativecommons.org/licenses/by-sa/4.0/ dataset Science Data Bank