Date of Award
Spring 5-8-2023
Access Type
Thesis - Open Access
Degree Name
Master of Science in Mechanical Engineering
Department
Mechanical Engineering
Committee Chair
Eric Coyle
First Committee Member
Patrick Currier
Second Committee Member
Jianhua Liu
College Dean
James W. Gregory
Abstract
Situational awareness in the maritime environment can be extremely challenging. The maritime environment is highly dynamic and largely undefined, requiring the perception of many potential hazards in the shared maritime environment. One particular challenge is the effect of direct-sunlight exposure and specular reflection causing degradation of camera reliability. It is for this reason then, in this work, the use of High-Dynamic Range imagery for deep learning of semantic image labels is studied in a littoral environment. This study theorizes that the use HDR imagery may be extremely beneficial for the purpose of situational awareness in maritime environments due to the inherent advantages of the technology. This study creates labels for a multi-class semantic segmentation process, and performs well on water and horizon identification in the littoral zone. Additionally, this work contributes proof that water can be reasonably identified using HDR imagery with semantic networks, which is useful for determining the navigable regions for a vessel. This result is a basis on which to build further semantic segmentation work upon in this environment, and could be further improved upon in future works with the introduction of additional data for multi-class segmentation problems.
Scholarly Commons Citation
Montagnoli, Charles, "Deep Learning of Semantic Image Labels on HDR Imagery in a Maritime Environment" (2023). Doctoral Dissertations and Master's Theses. 735.
https://commons.erau.edu/edt/735
