结合空间像素模板和Adaboost算法的高分辨率遥感影像河流提取  

River Extraction from the High Resolution Remote Sensing Image Based on Spatially Correlated Pixels Template and Adboost Algorithm

作　　者：慎利[1] 唐宏[2] 王世东[1,3,4] 张露[1] 

机构地区：[1]北京师范大学资源学院,北京100875 [2]北京师范大学减灾与应急管理研究院,北京100875 [3]河南理工大学测绘与国土信息工程学院,河南焦作454000 [4]河南理工大学矿山空间信息技术国家测绘地理信息局重点实验室,河南焦作454000

出　　处：《测绘学报》2013年第3期344-350,共7页Acta Geodaetica et Cartographica Sinica

基　　金：国家自然科学基金(40901217;41071259);矿山空间信息技术国家测绘局重点实验室重点项目(KLM201114;KLM201209)

摘　　要：充分有效地利用像素间的空间关系,是提高高分辨率遥感影像解译精度的关键之一。提出一种空间像素模板来获取空间邻域关系,并结合Adaboost集成学习算法来实现高分辨率影像上河流的精确提取。首先,基于过滤式特征选择方法自动生成像素模板,继而构建多维特征向量,然后利用Adaboost算法实现多特征的加权集成利用提取河流。相关试验结果表明,本文提出的方法河流提取结果面向对象特征显著,并且能够较好地将与河流具有光谱重叠的其他地物区分开。Making full use of spatial relationships between pixels is considered as one of the key factors in improving the accuracy of interpretation for high resolution remote sensing images. A kind of neighbor patterns, referred to as the spatially correlated pixels template, is presented to incorporate the spatial context. And in conjunction with Adaboost ensemble learning algorithm, accurate river extraction from high resolution remote sensing images can be obtained. Firstly, a specific form of spatially correlated pixels template is generated by using the filter feature selection approach. Secondly, the multi-dimensional feature vectors are constructed according to the given template. Then, Adaboost algorithm is used to make full use of available features. Finally, the accurate river extraction is achieved by ensemble learning. Experiments results show the river extraction results by the proposed methodology are object oriented and the geo-objects having similar spectral characteristics with river can be distinguished from river.

关 键 词：空间像素模板 高分辨率遥感影像 ADABOOST算法 河流提取 

分 类 号：P236[天文地球—摄影测量与遥感]