Acoustic Characterization of Seafloor Sediment Employing a Hybrid Method of Neural Network Architecture and Fuzzy Algorithm

Abstract

Seafloor sediment is characterized acoustically in the western continental shelf of India using the echo features extracted from normal incidence single-beam echo sounder backscatter returns at 33 and 210 kHz. The seafloor sediment characterization mainly depends on two important parameters: the number of sediment classes prevailing in the area and the selection of features having most prominent discriminating characteristics. In this letter, a method is proposed using Kohonen's self-organizing map to estimate the maximum possible number of classes present in a given data set, where no <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">a priori</i> knowledge on sediment classes is available. Applicability of this method at any site is illustrated with simulated data. In addition, another method is proposed to select the three most discriminating echo features using a fuzzy algorithm. The comparison of the results with ground truth at two operating frequencies revealed that this hybrid method could be efficiently used for sediment classification, without any <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">a priori</i> information and applicable for a wide range of frequencies.