基于深度学习的水声被动目标识别研究综述

An Overview on Underwater Acoustic Passive Target Recognition Based on Deep Learning

被动声呐通过接收目标自身发出的辐射噪声信号进行目标探测。水声目标识别通过分析水声信号来判别目标个体,是水声工程领域的重点研究方向。深度学习作为近年来各领域的研究热点,其在水声目标识别领域中的应用引起了学者的广泛关注。该文以水声目标识别的步骤框架为切入,介绍了典型深度网络模型;总结出了深度学习在水声目标识别领域中的两大内涵:围绕时频谱、梅尔倒谱系数等特征调研了近几年深度学习作为分类器的关键问题以及研究进展,围绕数据增强、数据降噪等信号处理手段调研了近几年深度学习作为信号处理工具的关键问题以及研究进展;并从数据驱动、特征驱动、模型驱动3个方面对该领域的发展趋势进行展望,旨在推动水声目标识别领域的发展。

Passive sonar detects targets by receiving radiated noise signals emitted from the targets.Underwater acoustic target recognition is an important research area in the underwater acoustic engineering field to identify individual targets by analyzing underwater acoustic signals.As a research hotspot in various fields in recent years,deep learning has attracted considerable attention from scholars for its application to the underwater acoustic target recognition field.Based on the step framework of underwater acoustic target recognition,two typical deep network models are introduced.Herein,two major implications of deep learning in the underwater acoustic target recognition field are summarized.The key issues and research progress in recent years are investigated for deep learning as a classifier based on features such as spectrograms and mel-frequency cepstrum coefficient and for deep learning as a signal processing tool based on signal processing methods such as data enhancement and data denoising.The development trend of this field is explored from three aspects,namely,data-driven,feature-driven,and model-driven,to promote the development of underwater acoustic target recognition.