Using custom deep learning to collect spot elevations from US Geological Survey historical topographic maps

Authors: Arthur Chan*, U.S. Geological Survey, Samantha T Arundel , US Geological Survey, Gaurav Sinha, Ohio University
Topics: Geographic Information Science and Systems, Geomorphology, Cartography
Keywords: machine learning, deep learning, summits, spot elevation, spot heights, automated cartography
Session Type: Paper
Day: 4/8/2020
Start / End Time: 8:00 AM / 9:15 AM
Room: Director's Row I, Sheraton, Plaza Building, Lobby Level
Presentation File: No File Uploaded

Countless spot elevations were collected for manual display on US Geological Survey (USGS) historical topographic maps (HTM). This valuable information has been lost because the spot elevations were not independently captured and recorded when these maps were digitized into the HTM collection (HTMC). Modern machine learning methods using optical character recognition and other algorithms offer tremendous potential to secure these lost data from the scanned maps. This research reports current efforts to employ these technologies by first creating a training dataset of summit spot elevation images, along with labels indicating the corresponding elevation. First, a neutral network was constructed using a custom artificial neural network classifier in Tensorflow to recognize the numeric characters in the spot elevation graphic. The second step requires the development of an attention model to locate appropriate text within the wider HTM image.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login