Javascript must be enabled for the correct page display

A max-tree-based astronomical source finder

Arnoldus, C. (2015) A max-tree-based astronomical source finder. Master's Thesis / Essay, Computing Science.

[img]
Preview
Text
arnoldus_master_thesis.pdf - Published Version

Download (8MB) | Preview
[img] Text
toestemming.pdf - Other
Restricted to Backend only

Download (386kB)

Abstract

Astronomy has entered an era where acquisition and analysis of large data volumes form an important part of research and discoveries. These data include 2D optical surveys resolved in space and 3D radio emission surveys resolved in both space and velocity. To handle these increasing volumes of data, automated solutions are becoming more and more important. Several algorithms, called source finders, have already been developed or are in development. The purpose of these source finders is identification of radiation sources. This thesis analyses and extents one of these methods, the MT source finder, which is intended for radio volumes. The MT source finder is based on the max-tree structure and statistical testing. Various smoothing techniques to be applied before max-tree creation, including Perona--Malik diffusion, were investigated. The statistical model, inherited from an older method for optical datasets, was analysed and found to be unsuitable for radio volumes. Therefore, new empirical models based on the power and flux density attributes were proposed. Equipped with these new statistical models, the method performs comparable (albeit slightly worse) to a state-of-the-art method, both in terms of completeness and reliability, as well as computation time required. Source finders typically output a large number of false detections. Some techniques were investigated to aid in identifying those detections that are true. These include classification trees (with boosting and bagging) and LVQ1-prototype-based classifiers. These methods were found to perform similarly to an existing technique in terms of accuracy and sensitivity, while the LVQ1 classifier achieves an impressive near-perfect precision level. Since the MT source finder is not yet able to outperform the state-of-the-art method, some suggestions for future work are included, to make it more competitive with other methods. Additionally, its computational performance can likely be improved.

Item Type: Thesis (Master's Thesis / Essay)
Degree programme: Computing Science
Thesis type: Master's Thesis / Essay
Language: English
Date Deposited: 15 Feb 2018 08:08
Last Modified: 15 Feb 2018 08:08
URI: http://fse.studenttheses.ub.rug.nl/id/eprint/13308

Actions (login required)

View Item View Item