Classification of Infrasound Surf Events Using Parallel Neural Network Banks

One of the most active locations in the world for ocean surf is the North Shore of Oahu. Here we show surf waves from three locations, "Pinballs," "Pipeline," and "Shark\´s cove," on the North Shore yield distinctive infrasonic signatures that can be used to train and test a neural network classifier. A parallel neural network classifier bank (PNNCB) is developed to classify the surf events from the three different locations on Oahu. Each module in the bank is a radial basis function (RBF) network responsible for classifying one of the three surf events. However, each module is also trained to not classify the other two surf events, and this ultimately increases the overall classifier performance. Output thresholds of each module are set according to a specific three-dimensional receiver operating characteristic (ROC) curve. For the three surf events, the correct classification rate achieved is 87.1%. A confusion matrix of the complete neural network classifier bank is shown along with confidence intervals for each class and the overall accuracy.