文献类型：期刊文献

中文题名：深度学习驱动的水下图像预处理研究进展综述

英文题名：Review of Underwater Image Preprocessing Based on Deep Learning

作者：彭小红[1];梁子祥[1];张军[1];陈荣发[1]

机构：[1]广东海洋大学数学与计算机学院,广东湛江524088

年份：2021

卷号：57

期号：13

起止页码：43

中文期刊名：计算机工程与应用

外文期刊名：Computer Engineering and Applications

收录：CSTPCD、、北大核心、CSCD、北大核心2020、CSCD_E2021_2022

基金：南方海洋科学与工程广东省实验室项目(ZJW-2019-06);广东海洋大学重大科研项目(GDOU2017052605)。

语种：中文

中文关键词：深度学习;图像预处理;物理模型;水下图像

外文关键词：deep learning;image preprocessing;physical model;underwater image

中文摘要：水下图像是海洋信息的重要载体,由于水下环境十分复杂,原始水下图像常常具有大量噪声,对后续的检测任务造成影响,因此水下图像预处理成为当前研究的热点。为了深入分析国内外学者对深度学习驱动的水下图像预处理研究进展,对近年来国内外相关文献进行总结分析。介绍了两类传统水下图像预处理方法,并分析其优缺点;根据是否结合物理模型,分析了深度学习驱动的水下图像预处理方法,并将相关方法进行对比总结;分析了深度学习方法的改进,主要从轻量化和提高鲁棒性和适应性两个方面论述;阐述了深度学习驱动的水下图像预处理方法目前存在的问题,并对未来的研究发展方向进行展望。

外文摘要：Underwater image is an important carrier of ocean information.Due to the complexity of underwater environ-ment,the original underwater image is affected by a lot of noise,which affects the subsequent detection tasks.Therefore,underwater image preprocessing has become a hot research topic.In order to analyze the current situation and trend of the research on the underwater image preprocessing driven by deep learning,this paper summarizes and analyzes the relevant literature at home and abroad in recent years.Firstly,the paper introduces two kinds of traditional underwater image preprocessing methods,and analyzes their advantages and disadvantages.Secondly,the underwater image preprocessing method driven by deep learning is analyzed according to whether combined with the physical model,then it compares and summarizes the related methods.Thirdly,the improvement of deep learning method is analyzed,mainly from two aspects of lightweight and improving robustness and adaptability.Finally,the existing problems of deep learning driven underwater image preprocessing are discussed,and the development direction of research is prospected.