HYDROACOUSTICS
ANNUAL JOURNAL
START NEW VOL 20 SEARCH STATISTICS PAS - GDANSK DIVISION

Artificial neural networks for interpolation and identification of underwater object features

pp. 1-10, vol. 11, 2008

Jerzy Balicki
Naval University of Gdynia, Gdynia, Poland

Ignacy Gloza
Naval University of Gdynia, Gdynia, Poland

Key words:

Abstract: Artificial neural networks can be applied for interpolation of function with multiple variables. Because of concurrent processing of data by neurons, that approach can be seen as hopeful alternative for numerical algorithms. From these reasons, the analysis of capabilities for some models of neural networks has been carried out in the purpose for identification of the underwater object properties. Features of the underwater objects can be recognized by characteristics of a amplitude according to the frequency of measured signals. The feedforward multi-layer networks with different transfer functions have been applied. Those network models have been trained by some versions of back-propagation algorithm as well as the Levenberg-Marquardt gradient optimization technique. Finally, for determination of the amplitude for the frequency of signal by the two-layer network with the hidden layer of the radial neurons has been proposed.

Download: Fulltext PDF, BibTeX

© Polish Acoustical Society - Gdansk Department, Polish Academy of Sciences. This work is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported. (CC BY-NC-SA 3.0)