Conference proceeding
Braids, Knots and Links: Investigating Dynamic Human Features as Geometric Proxies for Supervised Feature Extraction in Different Ocean Acoustic Applications
OCEANS 2024 - Singapore, pp.1-5
04/15/2024
DOI: 10.1109/OCEANS51537.2024.10682332
Abstract
Robust feature representation and its training and testing has been an ongoing challenge for autonomous sonar target recognition for decades. Key challenges include but are not limited to: unpredictable acoustic scatter from non-target reflectors within the oceanic environment, underwater acoustic multipath scattering, as well as unknown feature distortions due to uncertainties in target geometry. What has been seldom explored is the geometry between the target features themselves, which often overlap nonlinearly, with each other as well as across non-target features from the underwater acoustic environment. These geometries also change dynamically as a function of time, frequency, angle, cross-range, and other parameters important for sonar considerations. While machine learning techniques, both physics-inspired and black-box architectures, have been employed to discover and track these inherent features, not enough has been done to investigate the morphing geometries of these features in a systematic quantitative sense. In this position paper, we explore this question from the perspective of dynamic geometric proxies inherent in dancing human features and other artistic and natural structures, and use these as model props to study abstract geometric constructs such as braids, knots and links. The end goal is to create robust ground truths for these geometric constructs that are meaningful in sonar feature discovery and interpretation. The key idea is to exploit the geometry inherent in these structures as a means of creating innovative augmented data repositories for training and testing active sonar features. We present some geometric examples using video stills and discuss our observations on the promise and challenges of this training and testing simulation method.
Details
- Title: Subtitle
- Braids, Knots and Links: Investigating Dynamic Human Features as Geometric Proxies for Supervised Feature Extraction in Different Ocean Acoustic Applications
- Creators
- Ananya Sen Gupta - University of IowaAndrew Christensen - University of IowaTimothy Linhardt - University of IowaSubhajit Khush Das - Subhangik Dance Company,Bandel,West Bengal,IndiaIvars Kirsteins - Naval Undersea Warfare Center
- Resource Type
- Conference proceeding
- Publication Details
- OCEANS 2024 - Singapore, pp.1-5
- Publisher
- IEEE
- DOI
- 10.1109/OCEANS51537.2024.10682332
- Language
- English
- Date published
- 04/15/2024
- Academic Unit
- Electrical and Computer Engineering
- Record Identifier
- 9984721131602771
Metrics
1 Record Views