海岸带遥感影像半监督学习自动化分类方法研究——以黄河三角洲滨海湿地分类为例

项目名称： 海岸带遥感影像半监督学习自动化分类方法研究——以黄河三角洲滨海湿地分类为例

项目编号： No.41206172

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 海洋科学

项目作者： 任广波

作者单位： 国家海洋局第一海洋研究所

项目金额： 26万元

中文摘要： 海岸带区域资源环境变化剧烈，遥感是开展高效监测的重要手段，但传统的人工解译成本高、效率低（如"908专项"海岸带遥感调查历时近7年），已无法满足对监测信息时效性的要求，为此，提出一种以建立已知样本集为特点的海岸带遥感影像自动化分类方法。针对因该方法中训练样本不来自待分类影像而导致的样本对影像代表性差的问题，引入半监督学习思想，以黄河三角洲为研究区域，开展基于半监督学习的海岸带遥感影像自动化分类方法研究。 开展服务于自动化分类的湿地类型成像机理研究，基于现场测量的湿地光谱数据，评估遥感影像对滨海湿地类型的监测能力，并给出研究区典型湿地类型分类体系；研究基于现场地物光谱的已知样本集建立方法，面向基于像元和基于对象两种遥感影像分类策略，有针对性的设计滨海湿地分类半监督学习算法；在此基础上，分别发展研究区典型湿地类型的半监督自动化分类方法。为实现海岸带资源环境信息的高效监测提供技术支持。

中文关键词： 遥感；滨海湿地；半监督学习；高光谱；分类

英文摘要： The coastal area plays an important role in the world's economic development. Available land resources are scarce and valuable. So it is of great significance to monitor the land cover and land use condition timely and precisely. Remote sensing technology has been an important way of coastal resources and environment monitoring because of its advantages: a wide range of synchronization, economical and efficient. But until now, the way we extract information from the remote sensing images still rely on traditional manual interpretation, which is time and economic consuming. The low efficiency has been unable to meet the timely requirments of the coastal land use and land cover monitoring. Automatic remote sensing information extract methods need to be developed. A traditional automatic remote sensing image classification method contains three steps, which are: training samples selection, classifier training and computer-aided image classification. The training samples selection step is the only one which has not achieve automation. Even if the training samples can be selected automatically, which means without any information of the image which need to be classified, the samples representation problem (or to say training samples selected bias problem) will occur, and the followed image classification with the cla

英文关键词： Remote sensing；Coastal wetland；Semisupervised learning；Hyper-spectral；Classification