Experiences with deep learning and earth observation data

Authors

  • Karsten Østergaard Noe Alexandra Instituttet
  • Morten Henriksen Birk FieldSense A/S

DOI:

https://doi.org/10.5278/ojs.perspektiv.v17i32.2594

Abstract

Both at the Alexandra Institute and at FieldSense, we have in recent years worked with the use of the machine learning technique, deep learning, to analyze earth observation data. In this article, we would like to share our experiences of combining raster layers and polygon layers with deep learning techniques, whereby you can train neural networks that can automate work tasks in connection with mapping and change designation. We will provide examples of how complex information can be extracted from satellite or aerial photos via artificial intelligence when large amounts of data, and just as importantly the necessary ground truth data, are available. In collaboration with the Alexandra Institute, FieldSense has e.g. developed artificial intelligence to identify field boundaries and generate cadastral data from satellite data. In addition, we will tell about a method for visual search for objects in very large raster images.

Downloads

Published

28-01-2019