摘要
【目的】研究高分辨率遥感影像的湿地动态检测,为湿地资源的可持续发展提供信息支撑。【方法】充分利用面向像元和面向对象两种方法的优势和特点,结合多变量变化检测(MAD),提出对MAD变量的面向对象后分类方法(OB-M方法)。【结果】基于MAD变换的差异影像集中了两期影像的变化信息,基于像元差异影像的面向对象后分类方法能成功的检测多时相遥感影像的几何配准误差、单时相阴影、光照季节变化等"伪变化信息",成功提取变化/未变化信息。【结论】比较传统的面向对象分类后比较和MAD方法,提出的OB-M方法能较好地提高湿地变化/未变化信息检测的精度。
【Objective】With respect to the change detection on wetlands, very high spatial resolution images of drained managed wetland ponds were used, which could provide more information for further management. 【Method】 The proposed method is based on pixel-oriented difference image and object-based post-classification（OB-M）. Multivariate alteration detection （MAD） transformation was used to get the extended difference image, and object-based decision tree classification was applied on MAD components to detect the true change information of difference image, which had a very significant shape feature.【Result】 The proposed OB-MAD can successfully detect the false change information, such as the inevitable mis-registration, shadow and vegetation phenology differences. Compared with the traditional MAD method with thresholds （Threshold-MAD） and the traditional object-based post-classification method （OB-T）, the proposed OB-M method produced the highest accuracy, which took advantage of both pixel- and object-based technology.【Conclusion】Results indicated that the object-based post-classification on MAD components can well detect the change information of wetlands.
出处
《中国农业科学》
CAS
CSCD
北大核心
2012年第21期4369-4376,共8页
Scientia Agricultura Sinica
基金
美国国家自然科学基金项目(EF0410408
CCF0120778)
国家自然科学基金(3080073
305711123)
国家"863"计划项目(2006AAI02204)
关键词
变化检测
多变量变化检测
湿地
面向对象后分类
change detection
multivariate alteration detection（MAD）
wetlands
object-oriented post-classification
