Enhancing Automatic Target Recognition with Implicit Neural Networks for Synthetic Aperture Sonar Deconvolution
Abstract
PROJECT/ABSTRACT APPENDIX 1-3The project summary/abstract must identify the research problem and objectives, technicalapproaches, anticipated outcome of the research, if successful, and impact on DoD capabilities. Use only characters available on a standard QWERTY keyboard. Spell out all Greek letters, other non- English letters, and symbols. Graphics are not allowed and there is a 4,000-character limit including spaces. Do not include proprietary or confidential information. The project summary/ abstract must be marked by the applicant as #Approved for Public Release#. Abstracts of all funded research projects will be posted on the public DTIC website: https://dodgrantawards.dtic.mil/grantsThe proposed project aims to improve the capabilities of image formation and automated target recognition (ATR) for Synthetic Aperture Sonar (SAS). Deep learning techniques for SAS ATR typically suffer from the need for large training datasets and the black-box nature of deep-learning models leads to poor interpretability for this technology. The technical approach develops a method utilizing new advances in implicit neural representations (INRs) to encode the physics of sonar into machine learning pipelines. The project will design INRs to approximate the complex point scattering field of a scene and help improve the representation of targets in the resulting beamformed single- or multi-look complex images through the process of synthetic aperture sonar deconvolution. This will be coupled with state-of-the-art ATR networks for enhanced performance through joint optimization, thus allowing the INRs to output a tailored data representation that is useful for ATR while still maintaining high physical fidelity and interpretability. Such work will enable the co-design of sensor processing and beamforming methods with modern neural networks for synergistic performance. Anticipated outcomes of this research include enhancing SAS image formation and reconstruction techniques as well improving target classification networks for SAS ATR. This research will also shed light on how deep learning architectures characterize the physical aspects of SONAR images (e.g. scattering, frequency, propagation). This will have impact on Department of Defense capabilities including technology improvement for SAS for the Navy.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 15, 2023
- Source ID
- N000142312406
Entities
People
- Suren Jayasuriya
Organizations
- Arizona State University
- Office of Naval Research
- United States Navy