融合随机森林模型和6种水体指数的上海市水体信息提取

Water information extraction in Shanghai by integrating random forest model and six water indices

为快速、准确地掌握水体分布信息,本文以上海市为研究区,基于多时相Sentinel-2卫星数据构建水体提取特征集,并采用效率高、稳健性好的随机森林模型,对研究区内的水体进行提取。水体提取特征集在现有光谱波段特征的基础上加入6种水体指数,分别为NDWI、MNDWI、AWEI;、WI;、SWI和RWI,旨在提高水体提取精度。针对10个光谱波段特征及6种水体指数,设计了8种试验方案探究加入水体指数对于水体提取的作用。结果表明,将6种水体指数全部加入的方案精度最高,为97.910%;NDWI和RWI能提高水体提取精度、降低漏提率和误提率。

In order to know water distribution information quickly and accurately, this paper selects Shanghai as the study area, constructs a feature set of water extraction based on multi-temporal Sentinel-2 satellite data, uses random forest model with high efficiency and good robustness to extract water in Shanghai. To improve the accuracy of water extraction, six water indices are added into the feature set of water extraction based on the characteristics of existing spectral bands: NDWI、MNDWI、AWEI;、WI;、SWI and RWI. According to the characteristics of 10 spectral bands and 6 water indices, eight experimental schemes are designed to explore the effect of adding water indexes on water extraction. The results show that the scheme which included all the six water indices has the highest overall accuracy, which is 97.910%. NDWI and RWI can also improve the accuracy of water extraction and reduce the rate of leakage and error.