Deep Learning for Synthetic Aperture Sonar Imaging
Abstract
Synthetic Aperture Sonar (SAS) is a critical tool for mine detection, sea oor mapping and environmental monitoring. While signi ca""nt progress has been made in SAS imaging, the abilityto produce reliable images and automated mine detection in complex ocean envir"onments remain as technological obstacles. This project proposes to develop a Deep Learning based novel methodsand algorithms to advance SAS imaging in complex environments. Speci c problems include i ) image reconstruction for SAS and ii ) object recognition dir"ectly from SAS data bypassing imagereconstruction.Complex environments are typically uncertain, heterogenous, multiple-scattering," and dynamically changing. SAS imaging in complex environments requires sophisticated physics-based modelingand reconstruction via" optimization. However, such models are often computationally intractable and include several unknown parameters. Furthermore, this"" separation between modeling,reconstruction and optimization is largely due to di erent domains of expertise. In this project we mo""ve away from the dichotomy between modeling, reconstruction and optimization and propose anovel Deep Learning framework in which mo"deling and optimization parameters are jointly learned and re ned. We will use Deep Learning framework to learn wave propagation models in complexenvironments and to design reconstruction algorithms that are computationally e cient and robust. We will formulate mine classi cation problem directly from SAS data as a sparse recoveryproblem and address dictionary learning in a deep network trai"ning framework. We will utilize ideas such as transfer learning to address scarcity of training data, and multi-modal learning tode"velop reconstruction and recognition algorithms under unknown environmental conditions. The PI will work with Naval Surface Warfare" Center, Panama City Division to obtain real SASdata and to access PC SWAT to synthesize high delity SAS data for the project. The"" PI s student will plan to do summer internship at NSWC PCD to facilitate data sharing and collaboration.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 20, 2017
- Source ID
- N000141812068
Entities
People
- Birsen Yazıcı
Organizations
- Office of Naval Research
- Rensselaer Polytechnic Institute
- United States Navy