前视图像声纳去噪及目标检测综述
A Survey on Noise Reduction and Target Detection in Forward-looking Sonar Images
杨泰泓 1张涛 2李彬彬1
作者信息
- 1. 东南大学仪器科学与工程学院,南京 211106;微惯性仪表与先进导航技术教育部重点实验室,南京 211106
- 2. 东南大学仪器科学与工程学院,南京 211106;微惯性仪表与先进导航技术教育部重点实验室,南京 211106;东南大学南通海洋高等研究院,南通 226000
- 折叠
摘要
随着水下环境感知技术的不断发展,得益于声波在水中具有较远的传播距离和广泛的覆盖范围,基于声学的感知手段逐渐成为主流.在众多声学感知技术中,前视图像声纳凭借其能够实时感知视场内物体的能力,在水下环境感知中发挥了重要作用,并已广泛应用于渔业捕捞、航海安全、军事行动等多个领域.然而,前视图像声纳的性能受限于声学传播特性以及水下复杂环境的干扰,其高噪声、低信噪比的数据对声纳成像和目标检测提出了严峻挑战.尽管传统的声纳图像去噪方法在实际应用中已取得广泛验证,但面对前视图像声纳数据中的复杂噪声,基于深度学习的声纳图像去噪技术展现出更为显著的优势.前视图像声纳目标检测领域则经历了由传统算法到深度学习方法的革命性转变,其检测精度和泛化能力得到了显著提升.综述了前视图像声纳去噪与目标检测领域在传统方法与深度学习方法中的发展历程,并对当前研究进展及方法进行了系统总结.重点阐述了近年来基于深度学习的创新技术与研究方法,分析了其在复杂水下环境中的应用前景,并探讨了未来可能的研究方向,包括数据融合、算法优化以及实际应用中的挑战.
Abstract
With the continuous development of underwater environment perception technologies,acoustic-based sens-ing methods have gradually become mainstream,benefiting from the long propagation distance and wide coverage of sound waves in water.Among various acoustic sensing technologies,forward-looking sonar,with its ability to detect objects in the field of view in real-time,plays a crucial role in underwater environment perception and has been widely applied in fields such as fisheries,maritime safety,and military operations.However,the performance of forward-looking sonar is limited by the acoustic propagation characteristics and interference from the complex underwater environment.Its high noise and low signal-to-noise ratio data present significant challenges for sonar imaging and target detection.While traditional sonar image denoising methods have been extensively validated and applied in practical scenarios,deep learning-based sonar image de-noising technologies have shown more prominent advantages in dealing with the complex noise found in forward-looking so-nar data.The field of forward-looking sonar target detection has undergone a revolutionary shift from traditional algorithms to deep learning methods,significantly improving detection accuracy and generalization capabilities.This paper reviews the development of sonar image denoising and target detection in both traditional and deep learning methods,systematically summarizes current research progress and methodologies,and highlights emerging innovations based on deep learning.It al-so analyzes the prospects for application in complex underwater environments and discusses potential future research direc-tions,including data fusion,algorithm optimization,and challenges in real-world applications.
关键词
前视图像声纳/去噪/目标检测/深度学习Key words
forward-looking sonar/denoising/target detection/deep learning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据