不同指数法在地表水体提取中的效果比较

Comparison of effects of different index methods in extraction of surface water bodies

由于遥感技术能够快速、高效地获取地表水体的时空分布特征,目前基于影像提取内陆水体的方法很多,但针对不同类型的水域,哪一种方法提取效果更好,是值得探讨的问题。以天健湖、须水河和黄河郑州段3个水域为研究对象,基于GF-2,Landsat8,SPOT5卫星影像,采用水体、植被指数法等几种方法提取水域部分。通过分析提取效果,得出:对于水体较浅的天健湖,无论是GF-2还是Landsat8影像,提取效果较好的方法是水体指数法,提取效果较差的均为单波段阈值法;对于相对较深的须水河,无论是GF-2还是Landsat8影像,提取效果较好的方法是植被指数法,提取效果较差的均为单波段阈值法;对于含沙量较大、有细小水体的黄河水域,提取效果相对较好的是水体指数法,较差的是单波段阈值法和植被指数法。表明:在基于影像提取水体时,首先应弄清水域的情况,以采用相应的遥感指数。

It can get the space-time distribution characteristics of surface water bodies quickly and efficiently through remote sensing technology. There are many remote sensing image extraction methods that can be used for the extraction of inland water bodies.However, for different types of waters, which method is better is worth exploring. ChoosingTianjianLake (water body with shallow water level), Xushui River (water body with deep water level) and Yellow River (high sand content) as research objects, which satellite imagery comes from GF-2, Landsat8and SPOT5, it uses different methods such asNormalized Difference Water Index (NDWI) and Normalized Difference Vegetation Index (NDVI) to extract the water area.By analyzing the extraction effect, it can draw the following conclusions. For Tianjian Lake, which water level is shallow, the NDWI is the better method to extract the water area basing on the satellite imagery from GF-2and Landsat8. And the single-band threshold methodhas the worst extraction resultbasing on the satellite imagery from the three satellites. For XushuiRiver, which water level is deep, the NDVI is the best method to extract the water area basing on the satellite imagery from GF-2and Landsat8, while the single- band threshold methodhastheworst result. For the Yellow River waters with largesand content and small water bodies, NDWIhasabetterresult than other methods, andthe single-band threshold methodhastheworst result.The results show that when using water remote sensing images to extract water bodies, the type of waters should be first identified to adopt a suitable recognition algorithm.