A future perspective for automation in large mapping projects with feature learning
A future perspective for automation in large mapping projects with feature learning
Even though the discussion on automation has balanced towards the believers, the available methods that have been published over the years are yet to be generally adopted in research and national mapping projects. In this paper a future perspective will be proposed on how we could improve the applicability of models and make these independent of particular objects, landscapes and type of remote sensing data. Especially with increasing project size in both variety of objects and input of remote sensing data more flexibility in models are required. These models would choose feature learning over feature engineering where the characteristics of the objects are no longer designed by the creator but learned from data. This approach will be exemplified with the design of a model for a national mapping agency which will not be optimised for one region but could instead be used across different landscape and object types. The eventual model should be able to adapt and include new site locations or remote sensing data to further improve its accuracy. In the end this future perspective is meant to inspire big projects to aim for long term solutions and thereby preferably include a feature learning approach.
Kramer, Iris, Caroline
ba56efc0-ce81-4897-9d74-4bec3e9e1fca
March 2017
Kramer, Iris, Caroline
ba56efc0-ce81-4897-9d74-4bec3e9e1fca
Kramer, Iris, Caroline
(2017)
A future perspective for automation in large mapping projects with feature learning.
45th Computer Applications and Quantitative Methods in Archaeology: Digital Archaeologies, Material Worlds (Past and Present), Georgia State University, Atlanta, United States.
14 - 16 Mar 2017.
Record type:
Conference or Workshop Item
(Other)
Abstract
Even though the discussion on automation has balanced towards the believers, the available methods that have been published over the years are yet to be generally adopted in research and national mapping projects. In this paper a future perspective will be proposed on how we could improve the applicability of models and make these independent of particular objects, landscapes and type of remote sensing data. Especially with increasing project size in both variety of objects and input of remote sensing data more flexibility in models are required. These models would choose feature learning over feature engineering where the characteristics of the objects are no longer designed by the creator but learned from data. This approach will be exemplified with the design of a model for a national mapping agency which will not be optimised for one region but could instead be used across different landscape and object types. The eventual model should be able to adapt and include new site locations or remote sensing data to further improve its accuracy. In the end this future perspective is meant to inspire big projects to aim for long term solutions and thereby preferably include a feature learning approach.
Slideshow
CAA2017 Atlanta - Kramer- A future perspective for automation in large mapping projects with feature learning
More information
Published date: March 2017
Venue - Dates:
45th Computer Applications and Quantitative Methods in Archaeology: Digital Archaeologies, Material Worlds (Past and Present), Georgia State University, Atlanta, United States, 2017-03-14 - 2017-03-16
Identifiers
Local EPrints ID: 416404
URI: http://eprints.soton.ac.uk/id/eprint/416404
PURE UUID: 046d82f0-7590-4437-b7f3-88b40896b83d
Catalogue record
Date deposited: 15 Dec 2017 17:30
Last modified: 13 Mar 2024 18:22
Export record
Contributors
Author:
Iris, Caroline Kramer
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics