面向海洋的水下图像处理与视觉技术进展

Advances in Ocean-oriented Underwater Image Processing and Vision Technology

水下观测是探索海洋最直观的手段之一。受水下光学特性、声学特性以及杂波、水生生物等的影响,水下观测中所采集的图像并不总能满足观测需求。如何对水下图像进行有效的处理、分析与应用是一个具有挑战性的课题。尽管图像处理与计算机视觉技术已在大气环境中得到广泛研究,但鉴于成像原理、应用背景等方面的差异,针对大气自然图像提出的算法无法直接移植到水下任务中,而针对水下场景提出的视觉应用仍存在对任务背景考虑不足、泛化性差等缺陷。本文面向光学图像以及声学图像这两类水下观测的主要手段,从图像特性入手,首次以任务为导向、以需求为脉络,通过梳理国内外成功的水下图像处理、质量评价案例,对水下观测方案的工作思路进行了更完备的总结与分析。此外,本文围绕水下机器视觉应用探讨其发展进程,详细讨论与展望了相关领域的前景与优化方向,为突破海洋视觉应用的瓶颈,建设智慧海洋系统带来新思路。

Underwater observations are one of the most intuitive means of ocean exploration.Images acquired in underwa-ter observations are not always sufficient for the observation needs due to the influence of underwater optical and acoustic properties as well as clutter and aquatic creatures.The effective processing,analysis,and application of underwater im-ages is a challenging topic.Although image processing and computer vision techniques have been extensively studied in at-mospheric environments,given the differences in imaging principles and application contexts,the algorithms proposed for atmospheric natural images cannot be directly transferred to underwater tasks.And the visual applications proposed for un-derwater scenes often suffer from inadequate consideration of the task context and poor generalization.This paper ad-dresses two main tools of underwater observation:optical images and acoustic images,starting with underwater image char-acteristics,provides a more complete summary and analysis of the working ideas of underwater observation solutions by sorting out successful underwater image processing and image quality evaluation cases at home and abroad in a task-oriented and demand-based approach for the first time.In addition,this paper discusses the development process of under-water machine vision applications,explored the development prospects and optimization directions of related fields in de-tail,and brings new ideas for breaking the bottleneck of marine vision applications and building the Smart Ocean system.