Publication: Classification of sonar signals using artificial neural network voting scheme
Date
2010-04-01
Authors
Hamzah, Nor Shaadah
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
SONAR is a technique that uses sound propagation to navigate, communicate with or detect other objects. In this project, sonar data set are obtained from cylinder at aspect angles spanning 90 degree and from the rock at aspect angles spanning 180 degree. Each pattern is a set of 60 numbers in the range 0.0 to 1.0. Each number represents the energy within a particular frequency band, integrated over a certain period of time. An intelligent system from Artificial Neural Network (ANN) has been applied to the classification of SONAR returns from undersea targets, which are bounced off a metal cylinder and those bounced off a roughly cylindrical rock. Multilayer Perceptron (MLP) ANNs are used to identify and classify each sonar signal. The Bayesian Regularization back-propagation learning algorithm was used to train the network. MLPs that give small average error has been chosen to form an MLP ensemble, and to integrate with a voting system. In this project, MLPs with one and two output classes has been implemented. The performance of both types of MLP output classes are compared.This ensemble system are tested using the verification data set and its performance is compared with single MLPs without the voting scheme. The result show that MLP with one output class can classify sonar signals better than MLPs with two output classes. While, MLP ensemble gives better sonar signal classification than single MLPs. The results demonstrate the feasibility of applying the MLP ensemble to increase the classification performance of classifying the sonar signals.