Francis, Alistair Michael;
(2022)
Sensor Independent Deep Learning for Detection Tasks with Optical Satellites.
Doctoral thesis (Ph.D), University College London.
Preview |
Text
thesis-final.pdf - Accepted Version Download (99MB) | Preview |
Abstract
The design of optical satellite sensors varies widely, and this variety is mirrored in the data they produce. Deep learning has become a popular method for automating tasks in remote sensing, but currently it is ill-equipped to deal with this diversity of satellite data. In this work, sensor independent deep learning models are proposed, which are able to ingest data from multiple satellites without retraining. This strategy is applied to two tasks in remote sensing: cloud masking and crater detection. For cloud masking, a new dataset---the largest ever to date with respect to the number of scenes---is created for Sentinel-2. Combination of this with other datasets from the Landsat missions results in a state-of-the-art deep learning model, capable of masking clouds on a wide array of satellites, including ones it was not trained on. For small crater detection on Mars, a dataset is also produced, and state-of-the-art deep learning approaches are compared. By combining datasets from sensors with different resolutions, a highly accurate sensor independent model is trained. This is used to produce the largest ever database of crater detections for any solar system body, comprising 5.5 million craters across Isidis Planitia, Mars using CTX imagery. Novel geospatial statistical techniques are used to explore this database of small craters, finding evidence for large populations of distant secondary impacts. Across these problems, sensor independence is shown to offer unique benefits, both regarding model performance and scientific outcomes, and in the future can aid in many problems relating to data fusion, time series analysis, and on-board applications. Further work on a wider range of problems is needed to determine the generalisability of the proposed strategies for sensor independence, and extension from optical sensors to other kinds of remote sensing instruments could expand the possible applications of this new technique.
| Type: | Thesis (Doctoral) |
|---|---|
| Qualification: | Ph.D |
| Title: | Sensor Independent Deep Learning for Detection Tasks with Optical Satellites |
| Event: | University College London |
| Open access status: | An open access version is available from UCL Discovery |
| Language: | English |
| Additional information: | Copyright © The Author 2022. Original content in this thesis is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) Licence (https://creativecommons.org/licenses/by-nc/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author’s request. |
| UCL classification: | UCL UCL > Provost and Vice Provost Offices > UCL BEAMS UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Maths and Physical Sciences UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Maths and Physical Sciences > Dept of Space and Climate Physics |
| URI: | https://discovery.ucl.ac.uk/id/eprint/10142154 |
Archive Staff Only
 |
View Item |
