Advances and applications of machine learning in underwater acoustics

Abstract

Recent advancements in machine learning (ML) techniques applied to underwater acoustics have significantly impacted various aspects of this field, such as source localization, target recognition, communication, and geoacoustic inversion. This review provides a comprehensive summary and evaluation of these developments. As a data-driven approach, ML played a pivotal role in discerning intricate relationships between input features and desired labels based on the provided training dataset. They are achieving success in ocean acoustic applications through ML hinges on several critical factors, including well-designed input feature preprocessing, appropriate labels, choice of ML models, effective training strategy, and availability of ample training and validation datasets. This review highlights noteworthy results from published studies to illustrate the effectiveness of ML methods in diverse application scenarios. In addition, it delves into the essential techniques employed within these applications. To understand the utility of ML in underwater acoustics, one must analyze its advantages and limitations. This assessment will aid in identifying scenarios where ML excels and those where it may face challenges. In addition, it provides insights into promising avenues for future research, shedding light on potential research directions that warrant exploration.