Machine learning for the harsh environment: applications in sea ice classification and satellite magnetic fault recovery

Dowden, Benjamin R. (2022) Machine learning for the harsh environment: applications in sea ice classification and satellite magnetic fault recovery. Masters thesis, Memorial University of Newfoundland.

[img] [English] PDF - Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.

Download (12MB)


This thesis presents the development of two machine learning navigation modules for harsh environment applications. The first application investigates semantic segmentation using neural networks for sea ice detection and classification in polar oceans. Two popular generic architectures, SegNet and PSPNet101 are used to segment images. Transfer learning is performed using two custom datasets, one with four classes: ice, ocean, vessel, and sky, i.e., sea ice detection dataset, and the second with eight classes: ocean, vessel, sky, lens artifacts, first-year ice, new ice, grey ice, and multiyear ice, i.e., sea ice classification dataset. The Nathaniel B. Palmer imagery, which captured 2-month footage of the icebreaker completing an Antarctic expedition was used in the creation of both datasets. A subset of the dataset was labeled to generate a 240-image training set for sea ice detection achieving an accuracy of 98% classification for the 26-image test set. The sea ice classification dataset consists of 1,090 labeled images achieving accuracies of 98.3% or greater for all ice types for the 104-image test set. The second application investigates a new attitude error parameterization and a machine learning regression model for small satellite attitude fault recovery systems experiencing magnetometer bias faults. A simulation environment is developed to mimic an orbit of the international space station, and simulates both the magnetometer and the fine sun sensor on-board a small satellite. A right quaternion error parameterization is presented to ensure consistent error bound growth during the eclipse period of orbits where only a subset of sensor data is available. Using the improved error bounds a fault detection method using Mahalanobis distance is implemented to flag any faults in the system. After the fault detection, the fault recovery uses a regression sliding window optimizer to determine the unknown magnetometer bias that the sensor encounters. The proposed method demonstrates improved root mean squared error and error bound consistency achievable using the right error formulation for magnetic bias fault detection and recovery applications of small satellites.

Item Type: Thesis (Masters)
Item ID: 15342
Additional Information: Includes bibliographical references (pages 118-141).
Keywords: semantic segmentation, ice classification, attitude determination, error parameterization, machine learning, small satellite, fault recovery, PSPNet101, ice detection, regression learning
Department(s): Engineering and Applied Science, Faculty of
Date: May 2022
Date Type: Submission
Digital Object Identifier (DOI):
Library of Congress Subject Heading: Machine learning; Sea ice--Antarctic Ocean; Microspacecraft--Antarctic Ocean; Fault location (Engineering); Regression analysis; Image segmentation; Meteorology--Mathematical models; Nathaniel B. Palmer (ship).

Actions (login required)

View Item View Item


Downloads per month over the past year

View more statistics