Variational auto-encoders for satellite images of fields

Thumbnail Image

Files

master_Proll_Maximilian_2019.pdf (6.01 MB)

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.

Authors

Date

2019-06-17

Department

Major/Subject

Machine Learning, Data Science and Artificial Intelligence

Mcode

SCI3044

Degree programme

Master’s Programme in Computer, Communication and Information Sciences

Language

en

Pages

45 + 11

Series

Abstract

This thesis is situated at the overlap of probabilistic machine learning and remote sensing as it analyses the application of variational auto-encoders to satellite images of fields with the final objective of image classification. The rising availability of high-resolution satellite images of fields increases the need for compressing the images in order to keep maintenance and inference of machine learning models on a feasible and cost-efficient scale. Machine learning, in general, has proven to offer auspicious methods for summarising high-dimensional data into a lower-dimensional representation. Variational auto-encoders are a modern and advanced representation learning algorithm and are the topic of research in this thesis. An extensive hyperparameter search for the implemented networks is performed. The best architecture is selected and compared against conventional computer vision methods. The research shows that summarising high-resolution satellite images with variational auto-encoders is possible. It will, however, still take a performance hit on the classification tasks in comparison to the conventional computer vision techniques. The findings show the potential that variational auto-encoders offer for image compression but also that the used method needs further refinement in order to beat conventional approaches.

Description

Supervisor

Marttinen, Pekka

Thesis advisor

Hiremath, Santosh

Keywords

probabilistic machine learning, Bayesian deep generative models, variational inference, variational auto-encoder, remote sensing, satellite images

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-201906234009

Collections

[dipl] Perustieteiden korkeakoulu / SCI