Classification of infrasound events using radial basis function neural networks

Infrasound is a low frequency acoustic phenomenon that occurs in nature, and can result from man-made events, typically in the frequency range 0.01 Hz to 10 Hz. In this paper we present results for a bank of radial basis function (RBF) neural networks, to discriminate between six different man-made events. Each module in the bank of RBF networks is responsible for classifying one of the six events, and thus, is trained to identify only this particular event. However, each module is also trained to not classify all other events. Output thresholds of each module are set according to specific receiver operating characteristic (ROC) curves. Moreover, the spread parameter for the RBFs of each neural network module has been optimized. For six manmade events, the classifier accuracy achieved is 96%. A confusion matrix of the complete network is shown along with confidence intervals for each class and the overall accuracy.