Per-pixel uncertainity in change detection using airborne remote sensing
Per-pixel uncertainity in change detection using airborne remote sensing
This study aimed to develop remote sensing methodologies that could be used for operational monitoring of natural and semi-natural coastal habitats by governmental organisations such as Environment Agency or English Nature. The errors associated with post-classification change detection in the coastal zone using remotely sensed data were researched and methods of deriving thematic and geometric per-pixel uncertainties from airborne sensor data were investigated.
Methods of deriving a per-pixel geometric uncertainty model for the Compact Airborne Spectrographic Imager (CASI) were examined. A correlation was found between angular acceleration of the aircraft platform and geometric errors of automatically geocorrected CASI imagery. This relationship was used in combination with a geometric uncertainty model to provide a per-pixel model of instrument geometric uncertainty. The instrument geometric uncertainty models was combined with a model of orthometric error to provide a probabilistic geometric uncertainty model. A misregistration model was derived from the geometric uncertainty model and a significant correlation was found between the predicted and actual misregistration. Thematic uncertainty measures were derived from the output of the multi layer perception (MLP) and probabilistic neural network (PNN). A correlation was found between the thematic uncertainty measures derived and pixel thematic error. Heuristics to maximise the accuracy of the thematic uncertainty measures were derived. The geometric and thematic uncertainty measures were combined in a model of change detection uncertainty. Using synthetic data and data from a sand dune test site the use of uncertainty measures in change detection was found to be significantly more accurate compared to a change detection model that did not include uncertainty.
University of Southampton
Brown, Kyle Mackenzie
d90b92bc-66ae-44ef-b37f-5feb860bd1de
2005
Brown, Kyle Mackenzie
d90b92bc-66ae-44ef-b37f-5feb860bd1de
Brown, Kyle Mackenzie
(2005)
Per-pixel uncertainity in change detection using airborne remote sensing.
University of Southampton, Doctoral Thesis.
Record type:
Thesis
(Doctoral)
Abstract
This study aimed to develop remote sensing methodologies that could be used for operational monitoring of natural and semi-natural coastal habitats by governmental organisations such as Environment Agency or English Nature. The errors associated with post-classification change detection in the coastal zone using remotely sensed data were researched and methods of deriving thematic and geometric per-pixel uncertainties from airborne sensor data were investigated.
Methods of deriving a per-pixel geometric uncertainty model for the Compact Airborne Spectrographic Imager (CASI) were examined. A correlation was found between angular acceleration of the aircraft platform and geometric errors of automatically geocorrected CASI imagery. This relationship was used in combination with a geometric uncertainty model to provide a per-pixel model of instrument geometric uncertainty. The instrument geometric uncertainty models was combined with a model of orthometric error to provide a probabilistic geometric uncertainty model. A misregistration model was derived from the geometric uncertainty model and a significant correlation was found between the predicted and actual misregistration. Thematic uncertainty measures were derived from the output of the multi layer perception (MLP) and probabilistic neural network (PNN). A correlation was found between the thematic uncertainty measures derived and pixel thematic error. Heuristics to maximise the accuracy of the thematic uncertainty measures were derived. The geometric and thematic uncertainty measures were combined in a model of change detection uncertainty. Using synthetic data and data from a sand dune test site the use of uncertainty measures in change detection was found to be significantly more accurate compared to a change detection model that did not include uncertainty.
Text
1011945_v.1.pdf
- Version of Record
Text
1011945_v.2.pdf
- Version of Record
More information
Published date: 2005
Identifiers
Local EPrints ID: 465867
URI: http://eprints.soton.ac.uk/id/eprint/465867
PURE UUID: e34218d2-a15d-482c-b07d-964c2830f236
Catalogue record
Date deposited: 05 Jul 2022 03:21
Last modified: 16 Mar 2024 20:24
Export record
Contributors
Author:
Kyle Mackenzie Brown
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