The landform dataset is one of the important supporting data for achieving automatic classification of landforms and deepening the understanding of landform morphology. The current lack of high-precision geomorphic genesis datasets hinders the development of automatic interpretation of geomorphic remote sensing. This article focuses on the Tianshan Xingmeng orogenic system in northeastern China, which is mainly characterized by the trench arc basin system. Three types of scene datasets (GOS10) were created for the geomorphological genesis types formed by strong tectonic movements, volcanic and fluvial processes since the Cenozoic era, including tectonic geomorphology, volcanic lava geomorphology, and fluvial geomorphology. The dataset covers an area of approximately 5000 km2, including Sentinel-2 visible light remote sensing images, SRTM1 DEM, and 7 geomorphic parameters extracted based on DEM (mountain shading map, slope, DEM local mean, standard deviation, slope to north offset, slope to east offset, and relative deviation average). A single sample image is 64 pixels by 64 pixels, with a spatial resolution of 10 meters. Using multimodal deep learning neural networks to train and classify data, the average testing accuracy can reach 82.63%, indicating that the constructed dataset has high quality. It can provide dataset support for the research of remote sensing automatic classification of landform genesis and promote the development of intelligent interpretation of remote sensing landforms.
1:250000, 1:200000 basic geological maps, elevation DEM data, 1:1 1000000 digital topographic maps of China, and Sentinel-2 multispectral images.
The production of the GOS-10m dataset is mainly divided into three stages. Firstly, preprocess the remote sensing image data source and crop it to obtain a remote sensing scene dataset. Secondly, perform component extraction and preprocessing operations on the obtained DEM. Finally, using the remote sensing scene dataset as a spatial reference, the preprocessed DEM and its extracted components, as well as the interpretation result vector map, were spatially cropped to obtain the DEM and its component dataset and interpretation labels.
The data quality is good.
This work is licensed under a
Creative
Commons Attribution 4.0 International License.
CSTR:11738.11.NCDC.JRS.DB6678.2024
10.11834/jrs.20221385