Title: Infrasound signal classification using parallel RBF Neural Networks

Authors: Fredric M. Ham, Kamel Rekab, Ranjan Acharyya, Young-Chan Lee

Addresses: Department of Electrical and Computer Engineering, Florida Institute of Technology, 150 West University Boulevard, Melbourne, Florida 32901-6975, USA. ' Department of Mathematics and Statistics, University of Missouri – Kansas City, 5100 Rockhill Road, 206B Haag Hall, Kansas City, MO 64110-2499, USA. ' Industrial Research Ltd., 69 Gracefield Road, Lower Hutt, 5040, New Zealand. ' Samsung Techwin Co., Ltd., 145-3, Sangdaewon1-dong, Jungwon-gu, Seongnam-si, Gyeonggi 462-703, Korea

Abstract: A classification system is presented for discriminating different infrasound events using a Parallel Neural Network Classifier Bank (PNNCB) consisting of Radial Basis Function (RBF) networks. The classifier architecture and the pre-processing steps are unique and yield results that are superior when compared with those previously reported. Three-dimensional Receiver Operating Characteristic (ROC) curves are used to optimally set the output thresholds at each of the classification modules in the PNNCB for a particular class. A process is presented that enables optimising certain parameters of the classifier system. An application of the classification system to four infrasound classes is presented along with performance results and associated Confidence Intervals (CIs).

Keywords: infrasound; RBF; radial basis function; neural networks; 3D receiver operating characteristic curves; Mel-frequency scaling; classification.

DOI: 10.1504/IJSISE.2008.026787

International Journal of Signal and Imaging Systems Engineering, 2008 Vol.1 No.3/4, pp.155 - 167

Published online: 26 Jun 2009 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article