Digitization and Detection of Rip Currents within Optical Imagery by Way of a Fuzzy Set
Abstract
Rip currents are widely studied by many oceanographers and climatologists because of their hazardous nature and relationship to bathymetry. Nearshore images of rip currents serve as viable data for study when in-situ methods are costly and laborious. However, the process of manually digitizing rip currents can be arduous. Although some experimentation suggests supervised machine learning can help automate this process, these methods do not measure the dimensions of a rip channel. This study introduces an interface for machine-assisted digitization and labeling of rip current samples. This interface precisely captures the length and width of a rip current using a set of crossing line segments. The pixels of a rip digitization are then labeled according to a geometric model built from these segments. When studied as features in imagery, rip currents have a level of subjectivity in digitization. A novel pixel-based labeling scheme based on fuzzy-set theory is presented, which aims to take into account this subjectivity with an implicit model of uncertainty. Finally, preliminary results comparing a nave labeling scheme to the novel fuzzy scheme are presented and discussed.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 10, 2021
- Accession Number
- AD1155119
Entities
People
- Chris J. Michael
- Corey Maryan
- Sarah Trimble
- Steven Dennis
Organizations
- United States Naval Research Laboratory