Abstract: Environmental perception is the core technology for intelligent monitoring, being widely used in military and industrial fields. Deep learning has brought new vitality to environmental perception research, but currently there is a lack of comprehensive review in this field. Starting from the intersection of deep learning, environmental perception, and visual methods, this paper divides the perception system into four modules: information collection, information fusion, information processing, and information transmission. On this basis, it is applied to three aspects of marine environment: data generation, object detection, and coastline segmentation. It systematically examines various visual processing methods for marine environmental information including traditional methods, convolutional neural network, and transformer. The advantages and disadvantages of different methods are compared horizontally and vertically. Finally, the research trends in this field are discussed with the latest research work. With the advancement of modern intelligent algorithms, the application of deep learning for visual perception can greatly simplify the process and structure of device development, and intuitively perceives surrounding objects. Over time, the improvement of perception systems will mainly focus on intelligence, platformization, and integration.

Key words: marine environment, environmental perception, computer vision, image processing, deep learning

摘要： 环境感知是海上目标智能化监测的核心技术，广泛应用于军事与工业领域。深度学习为环境感知研究带来新活力，但目前该领域缺乏全面性的综述。从深度学习、环境感知与视觉方法的交叉点出发，将环境感知系统分为信息采集、信息融合、信息处理和信息传输四个模块。在此基础上，针对海域环境数据生成、海域目标检测识别与海陆岸线分割三个方面应用，从传统方法、卷积神经网络方法以及Transformer方法三方面系统梳理了海域环境信息的多种视觉处理方法，介绍了每种方法存在的问题及最新研究工作，并横向与纵向对比了不同方法的优缺点。最后，结合最新研究工作探讨了该领域的研究趋势。随着现代智能算法的进步，应用深度学习进行视觉感知能大大简化设备开发的进程和结构，能直观地感知周围海域目标；随着时间的推移，感知系统的提升也将主要集中在智能化、平台化和集成化方面。

关键词: 海域环境, 环境感知, 计算机视觉, 图像处理, 深度学习