基于面向对象法的Landsat 8影像山区细小水体提取方法

摘要

利用Landsat系列影像提取水体的方法有很多种，细小的水体漏提或误提现象严重，依靠单一的方法难以取得突破。本研究结合现有的方法提出一种新的思路：1)基于DEM数据提取河网数据；2)辅以河网缓冲区数据将Landsat 8影像的1-7波段和全色波段用于多尺度分割，根据形状特征提取线状的细小水体；3)假设2．5 m分辨率影像提取的结果为实际水体，与基于像元的MNDWI法进行对比。结果表明水体的总体提取精度提高11．7%，细小水体提取比例从13．3%上升到59．6%，因此本方法简单易行，有利于线状的细小水体提取，可在一定程度上减少山区细小水体的漏提，提高水体提取整体精度。%The methods based on the image of Landsat series are diverse. However, it′s not easy to make it more accurate, especially in the mountainous area. This paper puts forward a new plan com-bined by the methods proposed by former researchers: 1) To mask most noise of the background, buff-er the Stream Network generated by DEM; 2 ) Then using the Multi-resolution Segmentation action to create image object, so that we can make full use of the shape characteristics to extract waters; 3 ) Compared with pixel-based method of MNDWI, supposed that the results of 2. 5 m resolution images are actual waters. The result shows that the overall accuracy increased by 11. 7%, small water bodies extraction ratio rose from 13. 3% to 59. 6%, indicating that the scheme is conducive to waters extrac-tion in mountainous area.