 |
Publications |
|
|
|
|
|
|
 Rigby, P. |
|
Autonomous Spatial Analysis using Gaussian Process Models
PhD thesis, The University of Sydney, Dec, 2008
|
Abstract
This thesis is concerned with the spatial sampling and analysis of marine habitats using Autonomous Underwater Vehicles (AUVs). Spatial sampling involves determining a limited number of locations from which to measure phenomena which are subject to dependency and heterogeneity. Dependency suggests that a measurement at one location can be used to infer the value at another. Heterogeneity suggests that this relation may change across space and therefore it may not be possible to trust an observed degree of dependency beyond a certain region. Thus the distribution of the phenomena in question affects the optimal sampling strategy, and so there is a requirement for the AUV to adapt its behaviour as it acquires environmental information. The major contributions of this thesis arise from novel methods of adaptively sampling, modelling and analysing spatially correlated distributions.
The key theoretical framework upon which this thesis is built is the Gaussian Process (GP), which can be used to solve regression and classification problems with Bayesian inference. The suitability of these methods to modelling natural marine processes and habitats is substantiated using data gathered at Ningaloo Marine Park, Western Australia. Three applications of GP inference then extend the current state of the art.
Please download article for complete abstract.
|
| Download |
PDF full text (18.41M)
|
|
Copyright Notice & Disclaimer |
|
