Deep learning techniques applied to the remote sensing of soil moisture and sea ice type
| dc.contributor.author | Kruk, Ryan | |
| dc.contributor.examiningcommittee | Henry, Christopher (Electrical and Computer Engineering) Ashraf, Ahmed (Electrical and Computer Engineering) | en_US |
| dc.contributor.supervisor | Jeffrey, Ian (Electrical and Computer Engineering) Isleifson, Dustin (Electrical and Computer Engineering) | en_US |
| dc.date.accessioned | 2021-02-03T22:22:56Z | |
| dc.date.available | 2021-02-03T22:22:56Z | |
| dc.date.copyright | 2021-01-24 | |
| dc.date.issued | 2021 | en_US |
| dc.date.submitted | 2021-01-24T22:37:56Z | en_US |
| dc.degree.discipline | Electrical and Computer Engineering | en_US |
| dc.degree.level | Master of Science (M.Sc.) | en_US |
| dc.description.abstract | This thesis presents deep learning techniques applied to the modeling of geophysical quantities with remote sensing data to estimate soil moisture and classify sea ice type. The objective of the thesis is to determine if it is possible to estimate soil moisture and classify sea ice type from satellite data using deep learning. Soil moisture is an important metric used for agriculture, hydrology, and meteorology, while accurate maps of sea ice type are needed to address increased interest in commercial marine transportation through the Arctic. Satellites equipped with Synthetic Aperture Radar (SAR) sensors operating in the microwave band provide the remote sensing data used to predict these two geophysical quantities since SAR is capable of measuring the Earth's surface in all weather conditions and in darkness. Herein, the primary deep learning technique used to model soil moisture and sea ice types from SAR data is the Convolutional Neural Network (CNN). The best performing experiment for estimating soil moisture was achieved by U-Net on a dataset of 1,034 images to achieve a testing mean square error (MSE) of 3.82e-03 and correlation of 79.2%. For classifying sea ice type, DenseNet achieved the highest overall classification accuracy of 94.0% including water and the highest ice classification accuracy of 91.8% on a three class dataset using a fusion of HH and HV SAR polarizations for the input data. These results are contextualized and are shown to be comparable to other similar studies thereby supporting that the objective of the thesis has been achieved. | en_US |
| dc.description.note | May 2021 | en_US |
| dc.identifier.citation | IEEE | en_US |
| dc.identifier.uri | http://hdl.handle.net/1993/35314 | |
| dc.language.iso | eng | en_US |
| dc.rights | open access | en_US |
| dc.subject | Deep learning | en_US |
| dc.subject | Soil moisture | en_US |
| dc.subject | Sea ice | en_US |
| dc.subject | SAR | en_US |
| dc.subject | Sentinel-1 | en_US |
| dc.subject | SMAP | en_US |
| dc.subject | RADARSAT-2 | en_US |
| dc.subject | Canadian sea ice chart | en_US |
| dc.subject | U-Net | en_US |
| dc.subject | DenseNet | en_US |
| dc.subject | Convolutional neural network | en_US |
| dc.title | Deep learning techniques applied to the remote sensing of soil moisture and sea ice type | en_US |
| dc.type | master thesis | en_US |