2017 Gulf of Mexico Spring School (GMSS)Deep Learning and Applications
Abstract
We address the challenge of turbulent environment that often disturbs the space and time registrations ofmultiple mobile UXV platforms by taking an approach similar to that used in Natural Intelligence.While current deep learning and artificial neural networks" have provided many solutions in differentdomains, e.g. driverless cars, we believe the proposed persistent surveillance Aided Targ""et Recognition(AiTR) in marine environment must emulate the Human Brain Natural Intelligence (NI). For example, NIcan provide the"" explanation of why """"the beauty in the eyes of beholder"""" for we know the beauty is definedin the feature domain, not in the atoms" at the pixel level. The Biological Neural Network (BNN) solutionseparates the steps of Unsupervised Deep Learning (UDL) feature ex"traction for colors, edges, shapes,contours textures, performed in multiple layers V1~V5 Cortex 17, from Supervised Deep Learning (""SDL)pattern classification performed by the Hippocampus Associative Memory.In short, we shall indirectly register every input imag"e and extract each individually their orthonormalfeatures in the Feature vector domains. Those imprecise space-time registered mobile sensory images willbe less sensitive and approximately registered at the feature vector domain. Given a clearer combinedfeature"s of all overlapped images with some missing parts we can have a complete reconstruction in thefeature domain, then inversely we ca"n map back to a complete and clear image or video for the persistentsurveillance. We expect to provide a tutorial to expose the computational intelligence (CI) challenges facedby ocean sensing. Collaborators at the workshop will include the IEEE Computational In"telligence SocietyPresident-Elect Prof. Jacek Zurada, the International Neural Network Society President Prof.Bob Kozma, and Korea"" Advanced Institute of Science & Technology (KAIST)~s Prof. Soo-Young Lee.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 09, 2017
- Source ID
- N000141712597
Entities
People
- Simon Foo
Organizations
- Florida A&M University
- Office of Naval Research
- United States Navy