Robust classification techniques for acoustic signal analysis

Artificial neural networks are identified that are less sensitive to noisy feature vectors, and provide a sound estimate of the posterior class probabilities. These classifiers include the `optimum brain damage´ version of the multilayer perceptron and an elliptical basis function classifier. Since different classification techniques have different inductive biases, more accurate and robust classification can be obtained by combining the outputs of multiple classifiers. Two approaches to output combination are presented that yield better results for real oceanic signals, and also provide a basis for detecting outliers and `false alarms´