Session 5 - Natural Language Processing: Large-window Techniques for Geospatial Raster Data

Presenter Information/ Coauthors Information

Anne Denton, North Dakota State University--Fargo

Presentation Type

Invited

Track

Other

Abstract

Abstract: In conventional geographic information systems, topographical variables and confocal statistics are typically computed over small sliding windows, often of size 3x3. Such window sizes are appropriate for conventional remotely sensed images with pixel sizes of around 30m. In recent years, high resolution raster data from drones and low-orbit satellites has become ubiquitous. Without resampling high-resolution images, window-sizes orders of magnitude larger in size have to be considered to gain physically relevant information. Conventional computational techniques don’t scale to the number of pixels involved. I will show that many complex derived quantities, such as topographic attributes and even fractal dimension can be evaluated efficiently for much larger windows, provided the relevant aggregates are computed using iteratively doubling window sizes. The effectiveness is demonstrated for problems in agriculture and hydrology.

Start Date

2-11-2020 11:00 AM

End Date

2-11-2020 12:00 PM

This document is currently not available here.

Share

COinS
 
Feb 11th, 11:00 AM Feb 11th, 12:00 PM

Session 5 - Natural Language Processing: Large-window Techniques for Geospatial Raster Data

Dakota Room 250 A/C

Abstract: In conventional geographic information systems, topographical variables and confocal statistics are typically computed over small sliding windows, often of size 3x3. Such window sizes are appropriate for conventional remotely sensed images with pixel sizes of around 30m. In recent years, high resolution raster data from drones and low-orbit satellites has become ubiquitous. Without resampling high-resolution images, window-sizes orders of magnitude larger in size have to be considered to gain physically relevant information. Conventional computational techniques don’t scale to the number of pixels involved. I will show that many complex derived quantities, such as topographic attributes and even fractal dimension can be evaluated efficiently for much larger windows, provided the relevant aggregates are computed using iteratively doubling window sizes. The effectiveness is demonstrated for problems in agriculture and hydrology.