Dictionary learning for sparse representation and classification of sound speed profile in the ocean
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
The presence of internal waves (IWs) in the ocean alters the variability of sound speed profile (SSP) in the water column and plays an important role in applications such as underwater acoustics. As a consequence, assessing the variations in the SSP is crucial for modeling acoustic propagation in the ocean. In this work, we study the changes of the SSP in the water column due to IWs using a dictionary learning framework. The research presented in this manuscript serves as a foundation for extrapolating temperature profiles vertically up to the ocean surface and reconstructing 4D temperature fields under in presence of IW regimens. ☐ Generally, empirical orthogonal functions (EOFs) are employed to model and represent SSPs using a linear combination of basis functions that capture the sound speed variability. A different approach is to use dictionary learning (DL) to obtain a learned dictionary (LD) that generates a non-orthogonal set of basis functions (referred to as atoms) that generate a better sparse representation. In this research, the performance of EOFs and LDs are evaluated for sparse representation of SSPs affected by the passing of IWs. In addition, an LD-based supervised framework is presented for SSP classification and is compared with classical learning models. The algorithms presented in this work are trained and tested on data collected from the shallow water experiment 2006. Results show that LDs yield lower reconstruction error than EOFs when using the same number of basis, whereas overcomplete LDs demonstrate to be a robust method to classify SSPs during low, medium, and high IW activity, reporting comparable and sometimes higher accuracy than standard supervised classification methods. ☐ Additionally, we state the bases for a graph-based method to interpolate for missing SSP samples in time and space. This approach is based on the recovery of approximately bandlimited signals (i.e., temperature values) by minimizing the blue-noise spectrum of the graph signal. Preliminary results suggest the proposed method is more robust than state-of-the-art approaches, since exact computation of the cutoff frequency is not required, showing competitive results in online reconstruction of time-varying graph signals.
Description
Keywords
Dictionary learning, Graph signal processing, Machine learning, Oceanography, Sound speed profile classification, Sound speed profile reconstruction