Dictionary learning for classification of sound speed profile in the ocean

Abstract

Internal waves (IWs) are waves that propagate along density gradients within the water column and alter the isotropic properties of sound speed profiles (SSPs). Changes in the SSPs modify the acoustic channel ducts and affect underwater acoustic propagation, causing most of the energy to be dissipated into the seabed due to the downward refraction of sound waves. Therefore, variations in the SSP must be considered when modeling acoustic propagation in the ocean. It has been proved that Dictionary learning (DL), an unsupervised machine learning method, succeeds in sparsely representing signals by employing a few non-orthogonal basis functions (atoms) learned from data. In this work, we use the ability of learned dictionaries (LDs) for data representation and thus train class-specific dictionaries to capture relevant features from labeled data within a supervised learning setting. We developed an LD-based supervised framework for SSP classification and compared it with state-of-the-art models. The algorithms presented in this work are trained and tested on data collected from the shallow water experiment 2006. Results show that overcomplete DL is a robust method to classify SSPs during IW activity, reporting comparable and higher accuracy than standard supervised classification methods. [Work supported by ONR, Grants No. N00014-21-1-2424 and N00014-21-1-2760.]