基于HJ–CCD数据的海面溢油提取方法研究  

Study of extraction methods for ocean surface oil spill using HJ – CCD data

作　　者：盖颖颖[1] 周斌[1] 孙元芳[1] 周燕[1] 

机构地区：[1]山东省科学院海洋仪器仪表研究所,青岛266001

出　　处：《国土资源遥感》2014年第2期99-104,共6页Remote Sensing for Land & Resources

基　　金：山东省留学人员科技活动择优资助项目"基于GPU超级计算的实时海面目标识别算法研究"(编号:SR-12-10-1)资助

摘　　要：快速、准确地获取溢油污染信息,对海洋的动态监测、保护和可持续利用具有重要意义。环境与灾害监测预报小卫星星座一号(HJ–1)是我国针对生态环境污染和灾害监测发射的新型卫星平台,但HJ–1 CCD多光谱数据的光谱波段较少,仅依赖光谱信息获取海面溢油范围的精度较低。因此,以墨西哥湾溢油事件为研究对象,在分析不同地物光谱特征的基础上,采用灰度共生矩阵,选择合适的纹理结构因子,提取HJ–1 CCD图像中影响溢油识别的地物纹理特征;建立光谱特征和纹理特征相结合的决策树模型,提取海面溢油信息,并与只考虑光谱信息的传统分类方法进行精度对比。结果表明,与最大似然分类法相比,决策树方法的油膜提取用户精度和制图精度分别提高了11.85%和4.28%。Rapid and accurate access to the oil spill information is of great significance for dynamic monitoring, conservation and sustainable use of the oceans. HJ - 1 is a new satellite platform designed for ecological environmental pollutions and disasters. However, the multispectral image obtained from HJ-CCD has insufficient spectral bands, and the accuracy of acquiring the oil spill coverage only by spectral information is low. In this paper, the oil spill that occurred in the Gulf of Mexico was selected as the research object. Based on the spectral analysis of different features, the authors chose the right texture structure factors and extracted the texture characteristics which affect oil spill identification by gray co-occurrence matrix. A decision tree model combining spectral characteristics with texture characteristics was established to extract the oil spill on the sea surface. A comparative analysis by using the result of maximum likelihood supervision classification method was performed, and the results show that, in comparison with the maximum likelihood classification method, the decision tree method could improve the user＇s accuracy and the producer＇s accuracy of oil spill extraction by 11. 85% and 4. 28%respectively.

关 键 词：溢油 HJ-1 墨西哥湾 纹理特征 决策树 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]