SELF-SUPERVISED PRE-TRAINING FOR PASSIVE SONAR
Loading...
Authors
Pelham, James S.
Subjects
transformer
self-supervised pre-training
audio
artificial intelligence
AI
machine learning
ML
passive sonar
Underwater Passive Acoustic Dataset
UPAD
Naval Postgraduate School
NPS
self-supervised pre-training
audio
artificial intelligence
AI
machine learning
ML
passive sonar
Underwater Passive Acoustic Dataset
UPAD
Naval Postgraduate School
NPS
Advisors
Orescanin, Marko
Date of Issue
2025-03
Date
Publisher
Monterey, CA; Naval Postgraduate School
Language
Abstract
This study explores the effectiveness of self-supervised learning techniques in automated ship detection using passive sonar data. Passive sonar offers rich acoustic signatures that are valuable for ship detection, yet much of this data remains unlabeled and underutilized. We leverage the Underwater Passive Acoustic Dataset (UPAD) toolbox, developed and supported by the Naval Postgraduate School (NPS), to drive this research. Using UPAD and more than 20,000 hours of passive sonar data, we pre-train five transformer models with unique self-supervised approaches to learn data representations without explicit labels, establishing a robust foundation for downstream passive sonar tasks. After this initial pre-training on a noisily labeled subset, we fine-tune the models on a hand-labeled dataset to enhance the model’s ship detection capabilities. Our results demonstrate that self-supervised pre-training enriches the model’s understanding of spectrogram structures, producing separated embedding clusters in the absence of labels. Additionally, models pre-trained in this manner show improvement in classification tasks when fine-tuned with traditional supervised learning, outperforming models trained purely through supervised learning alone. This work underscores the potential of self-supervised learning in advancing automated ship detection from passive sonar audio.
Type
Thesis
Description
Series/Report No
Department
Organization
Identifiers
NPS Report Number
Sponsors
Funder
Format
Citation
Distribution Statement
Distribution Statement A. Approved for public release: Distribution is unlimited.
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.