Comparison of confidence level of different classification paradigms for underwater target discrimination

The problem of classification of underwater targets from the acoustic backscattered signals is considered. A wavelet packet-based feature extraction scheme is used in conjunction with the linear prediction coding (LPC) scheme as the front-end processor. Selected features with higher discriminatory power are then fed to a neural network classifier. Several different classification systems are benchmarked in this paper. These include: an ellipsoidal K-nearest neighbor (K-NN) classifier, Probabilistic Neural Networks (PNN) and Support Vector Machines (SVM). The performance of these classifiers are examined on a wideband 80 kHz acoustic backscattered data set collected for six different objects. These systems are then benchmarked with the previously used Backpropagation Neural Network (BPNN) in terms of their receiver operating characteristic (ROC) and robustness with respect to reverberation.