Target Feature Representation and Persistent Feature Tracking Using Braid Manifolds and Machine Learning Techniques
Abstract
Approved for public releaseWe propose a basic research effort to develop a robust physics-cognizant mathematical framework to represent and differentiate different classes of acoustic echoes from sonar targets. This proposed effort is in continuation of our effort to develop robust interpretable feature representation, feature extraction and feature interpretation techniques for acoustic echoes from large sonar targets that lack theoretical models for robust characterization. Autonomous identification of sonar targets is challenging due to the rapid fluctuations of the target scattering signature, as well as the lack of coherence between pings. We formulate the sonar target identification problem fundamentally as a feature engineering challenge, and focus on designing feature extraction techniques that can characterize and identify a sonar target based on the minimum (or close to minimum) number of pings. Specifically, we design signal processing techniques based on persistent featuremanifolds and graph-based learning frameworks that extract target information implicitly contained in each ping. We also propose to study how that information varies from ping to ping, andgets corrupted by background noise or interference from clutter. In particular, we harness robust theoretical underpinnings in graph-based machine learning and algebraic geometry to develop feature manifold representation and persistent feature tracking techniques to address this.On the data validation front, we validate our methods on acoustic echoes from sonar targets represented by experimental field data from the Malta experiment, such as reflected acoustic echoes from an oil tanker, rig, etc. To clearly scope our efforts in this proposed project, we focus on acoustic echoes and related sonar signal processing that are distinct from computational techniques designed for shorter-range smaller targets such as proud/buried underwater mines, etc. where inter-ping coherence maybe significant and the scattering models are fundamentally different.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 29, 2023
- Source ID
- N000142312503
Entities
People
- Ananya Sen Gupta
Organizations
- Office of Naval Research
- United States Navy
- University of Iowa