Monterey, CA; Naval Postgraduate School
IMAGE QUALITY ASSESSMENT OF ACTIVE SONAR IMAGES THROUGH BAYESIAN DEEP LEARNING
Abstract
dc:description.abstractData quality is a problem for synthetic aperture sonar (SAS) images when using automatic target recognition (ATR) techniques. At present, no automated methods exist to determine image quality. We propose using Bayesian Neural Networks to identify and diagnose imaging artifacts present in active sonar images to aid in undersea object detection. We expect that Bayesian Neural Networks and the associated understanding of model uncertainty will benefit the human-machine team dynamic by identifying uncertain classifications to the operator. This research addresses explainable AI through quantification of uncertainty in predictions. Moreover, it is demonstrated that uncertainty quantification enables better understanding of the relationship between the physics of SAS images and their sensors and model predictions. We used a dataset of good quality synthetic aperture sonar images and perturbed them during beamforming to intentionally create images of degraded quality. Our best performing model, Flipout, achieved a 92.7% accuracy and was proven to be well-calibrated for the human-machine team dynamic. Analysis of the uncertainty was validated by out-of-distribution data and scintillation index. These results will lead to a more capable ATR system to be used in mine countermeasures.
Degree
thesis:*- Name thesis:degree_name
- Master of Science in Computer Science
- Level thesis:degree_level
- Masters
- Discipline thesis:degree_discipline
- Computer Science
- Department dc:contributor.department
- Computer Science (CS)
- Grantor dc:publisher
- Monterey, CA; Naval Postgraduate School
- Year dc:date.issued
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Harrington, Brian E.
- Advisors dc:contributor.advisor
-
- Olson, Derek
- Orescanin, Marko
Rights
dc:rights- Statement dc:rights
-
- This publication is a work of the U.S. Government as defined in Title 17, United States Code, Section 101. Copyright protection is not available for this work in the United States.
- Approved for public release. Distribution is unlimited.
Identifiers
dc:identifier.*- Dc Identifier Programcode
- 368, Computer Science
- Thesis Thesis Identifier
- 38691