Classification of shallow-water acoustic signals via alpha-Stable modeling of the one-dimensional wavelet coefficients

A novel statistical scheme is presented for the classification of shallow water acoustic signals according to the environmental parameters of the medium through which they have propagated. An efficient way to classify these signals is important for inverse procedures in underwater acoustics aiming at the recovery of the geoacoustic parameters of an oceanic environment, using measurements of the acoustic field due to an acoustic source. An important issue in this procedure is the determination of an efficient “observable” of the acoustic signal (feature extraction), which characterizes the signal in connection with the recoverable parameters. The proposed method is based on a transformation of the acoustic signals via a one-dimensional (1D) wavelet decomposition and then by fitting the distribution of the subband coefficients using an appropriate function. We observe that statistical distributions with heavy algebraic tails, such as the alpha-Stable family, are often very accurate in capturing the non-Gaussian behavior of the subband coefficients. As a result, the feature extraction step consists of estimating the parameters of the alpha-Stable model, while the similarity between two distinct signals is measured by employing the Kullback–Leibler Divergence between their corresponding alpha-Stable distributions. The performance of the proposed classification method is studied using simulated acoustic signals generated in a shallow water environment.