Deep Learning for Sonar
Abstract
Deep learning has demonstrated significant promise recently for the analysis of general sensor data. The proposed research will appl"y deep learning for analysis of sonar imagery. Duke will support NSWC, Panama City in applying recent deep-learning developments to"" sonar imaging data. Duke will have primary responsibility for developing and adapting deep learning to sonar data, and NSWC will ap"ply that technology to (possibly sensitive) Navy-relevant data. Duke will provide all deep learning software developed on this progr"am to NSWC. New techniques in deep learning will be explored, and transitioned to Navy applications. Specifically, a new form of the"" variational autoencoder (VAE) is proposed, based on the symmetric Kullback-Leibler divergence. Learning of the resulting symmetric" VAE (sVAE) has close connections to previously developed adversarial-learning methods. This relationship will unify the previously" distinct techniques of VAE and adversarially learning, and provides insights that allow us to ameliorate shortcomings with some pre""viously developed adversarial methods. In addition to performing analysis that motivates and explains the sVAE, we will specialized" this technology to Navy sonar problems.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 20, 2017
- Source ID
- N000141812073
Entities
People
- Lawrence Carin
Organizations
- Duke University
- Office of Naval Research
- United States Navy