topoDL: A deep learning semantic segmentation dataset for the extraction of surface mine extents from historic USGS topographic maps

Input topographic maps, surface mine extents, and quad boundaries used in the following study:

Maxwell, A.E., M.S. Bester, L.A. Guillen, C.A. Ramezan, D.J. Carpinello, Y. Fan, F.M. Hartley, S.M. Maynard, and J.L. Pyron, 2020. Semantic segmentation deep learning for extracting surface mine extents from historic topographic maps, Remote Sensing, 12(24): 1-25. https://doi.org/10.3390/rs12244145.

Associated code and descriptions of the data are provided on GitHub: https://github.com/maxwell-geospatial/topoDL.

The surface mine extent data were obtained from the USGS prospect- and mine-related features from USGS topographic maps dataset: https://mrdata.usgs.gov/usmin/. Topographic maps were downloaded from TopoView/The National Map. We have simply prepared the data for easier ingestion into deep learning semantic segmentation workflows by aligning the vector polygon data with the associated topographic map and including topographic map boundaries to remove the collar information. Vector data can be rasterized and combined with the topographic maps to generate image chips and masks for semantic segmentation deep learning.

The chip prep script on GitHub can be used to create chips and masks from these data. This compressed folder contains the following subfolders (ky_mines, ky_quads, ky_topos, oh_mines, oh_quads, oh_topos, va_mines, va_quads, va_topos). The mines folders contain the mine extents for each topographic map used in the study while the quads folders contain the quadrangle boundaries. All vector data are in shapefile format. The topos folders contain the topographic maps in TIFF format.