卷积神经网络在SAR遥感海岛海岸带地物信息提取中的应用综述  

Review of Convolutional Neural Networks in Extracting Ground Objects/Information from Islands and Coasts by Using SAR Remote Sensing

作　　者：刘鹏 谢春华[1,2] 安文韬 崔艳荣 李良伟 LIU Peng;XIE Chunhua;AN Wentao;CUI Yanrong;LI Liangwei(National Stellite Ocean Application Services Being 100081,China;Key Laboratory of Space Ocean Remote Sensing and Application,MNR,Bejing 100081,China;National Marine Environmental Forecasting Center,Bejng 100081,China;College of Marine Science Shanghai Ocean University,Shanghai 201306,China)

机构地区：[1]国家卫星海洋应用中心,北京100081 [2]自然资源部空间海洋遥感与应用研究重点实验室,北京100081 [3]国家海洋环境预报中心,北京100081 [4]上海海洋大学海洋科学学院,上海201306

出　　处：《海洋开发与管理》2021年第8期3-10,共8页Ocean Development and Management

基　　金：国家自然科学基金项目(61971152).

摘　　要：为满足科学管理海岛海岸带的需求,发掘合成孔径雷达(SAR)在海岛海岸带地物信息提取中的应用潜力,文章概述SAR和卷积神经网络的原理,分析卷积神经网络应用于SAR遥感海岛海岸带地物信息提取的可行性和优势。研究结果表明:通过卷积神经网络提取SAR数据中的海岛海岸带地物信息,无须预先提取图像特征,卷积神经网络能够自动提取图像中更本质和更抽象的特征,更好地应对地物目标的非线性混合;这种提取方法的精度更高,鲁棒性和泛化能力更强,可应用于海岛海岸带的精细化监测,为海岛海岸带的科学管理提供支撑。In order to meet the needs of scientific management of islands and coasts as well as explore the application potential of SAR in islands and coasts feature information extraction, the paper summarized the principles of SAR and convolution neural network, and analyzed the feasibility and advantages of convolution neural network in SAR remote sensing islands and coasts feature information extraction. The results showed that the convolution neural network could automatically extract more essential and abstract features in SAR image without extracting image features in advance. It could better deal with the nonlinear mixing of ground objects. This method had higher accuracy, stronger robustness and generalization ability. It could be applied to the fine monitoring of islands and coasts, and then provided support for the scientific management of islands and coasts.

关 键 词：卫星遥感 地物信息 合成孔径雷达 卷积神经网络 深度学习 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置] P71[自动化与计算机技术—控制科学与工程]