基于被动声呐音频信号的水中目标识别综述  

A Review of Underwater Target Recognition Based on Passive Sonar Acoustic Signals

作　　者：徐齐胜 许可乐 窦勇[1,2] 高彩丽 乔鹏 冯大为[1,2] 朱博青 XU Qi-Sheng;XU Ke-Le;DOU Yong;GAO Cai-Li;QIAO Peng;FENG Da-Wei;ZHU Bo-Qing(School of Computer Science,National University of Defense Technology,Changsha 410073;National Key Laboratory of Parallel and Distributed Processing,Changsha 410073)

机构地区：[1]国防科技大学计算机学院,长沙410073 [2]并行与分布处理国防科技重点实验室,长沙410073

出　　处：《自动化学报》2024年第4期649-673,共25页Acta Automatica Sinica

摘　　要：基于被动声呐音频信号的水中目标识别是当前水下无人探测领域的重要技术难题,在军事和民用领域都应用广泛.本文从数据处理和识别方法两个层面系统阐述基于被动声呐信号进行水中目标识别的方法和流程.在数据处理方面,从基于被动声呐信号的水中目标识别基本流程、被动声呐音频信号分析的数理基础及其特征提取三个方面概述被动声呐信号处理的基本原理.在识别方法层面,全面分析基于机器学习算法的水中目标识别方法,并聚焦以深度学习算法为核心的水中目标识别研究.本文从有监督学习、无监督学习、自监督学习等多种学习范式对当前研究进展进行系统性的总结分析,并从算法的标签数据需求、鲁棒性、可扩展性与适应性等多个维度分析这些方法的优缺点.同时,还总结该领域中较为广泛使用的公开数据集,并分析公开数据集应具备的基本要素.最后,通过对水中目标识别过程的论述,总结目前基于被动声呐音频信号的水中目标自动识别算法存在的困难与挑战,并对该领域未来的发展方向进行展望.Underwater target recognition based on passive sonar acoustic signals is a significant technical challenge in the field of underwater unmanned detection,with broad applications in both military and civilian domains.This paper provides a comprehensive exposition of the methodology and process involved in underwater target recognition using passive sonar acoustic signals,addressing data processing and recognition methods at two levels.Regarding data processing,this paper presents a thorough exploration of the fundamental principles of passive sonar signal processing from three key aspects:The underlying process of underwater target recognition based on passive sonar signals,the mathematical foundations of passive sonar acoustic signal analysis,and the extraction of relevant features.At the level of recognition methods,this paper offers a comprehensive analysis of underwater target recognition techniques based on machine learning algorithms,and focus on the research conducted with deep learning algorithms at its core.The paper systematically summarizes and analyzes the current research progress across various learning paradigms,including supervised learning,unsupervised learning and self-supervised learning,and analyzes their advantages and disadvantages in terms of the algorithm's labeled data requirement,robustness,scalability and adaptability.Additionally,this paper provides an overview of widely-used public datasets in the field,and outlines the essential elements that such datasets should possess.Finally,by discussing the process of underwater target recognition,this paper summarizes the current difficulties and challenges in automatic underwater target recognition algorithms based on passive sonar acoustic signals,and offers insights into the future development direction of this field.

关 键 词：被动声呐信号 水中目标自动识别 深度学习 有监督学习 自监督学习 

分 类 号：E91[军事] U666.7[交通运输工程—船舶及航道工程] TP18[交通运输工程—船舶与海洋工程]