Neural Network Implementations for Sonar Signal Processing

Terry Sims and Mark Dzwonczyk

Abstract:	Under its internal research and development program, Draper Laboratory is investigating implementation issues of deployable neural networks for automatic target recognition. Current research efforts are focused on the development of the Integrated Neurocomputing Architecture system, or INCA, The INCA incorporates a conventional processor for data management tasks with a large-scale analog neural network for real-time pattern recognition. As an integrated system, the INCA can be used as a testbed for neurocomputing algorithm development or as a real-time pattern recognition plat- form. The INCA will be capable of implementing a fully parallel, 4 layer, 64 node per layer, feed-forward neural network - a topology large enough for a variety of current pattern recognition problems of interest. The network consists of 16 Very Large Scale Integration (VLSI) analog devices hosted on a VMEbus-size prototype card with appropriate support logic. The driving pattern recognition application is target detection from cluttered side-scan sonar data. Raw g-bit sonar data is windowed into tokens of N x N pixels, where 105N530. In feed-forward operation, the INCA network can detect targets from incoming data in real time.
Published in:	Proceedings of the 47th Annual Meeting of The Institute of Navigation (1991) June 10 - 12, 1991 Williamsburg Hilton and National Conference Center Williamsburg, VA
Pages:	265 - 274
Cite this article:	Sims, Terry, Dzwonczyk, Mark, "Neural Network Implementations for Sonar Signal Processing," Proceedings of the 47th Annual Meeting of The Institute of Navigation (1991), Williamsburg, VA, June 1991, pp. 265-274.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In