Water Information Extraction Based on Multi-Model RF Algorithm and Sentinel-2 Image Data

For the Sentinel-2 multispectral satellite image remote sensing data, due to rich spatial information, traditional water body extraction methods cannot meet needs of practical applications. In this study, a random forest-based RF_16 optimal combination model algorithm is proposed extract bodies. The research process uses images and DEM data as basic collected 24 characteristic variable indicators (B2, B3, B4, B8, B11, B12, NDVI, MSAVI, B5, B6, B7, B8A, NDI45, MCARI, REIP, S2REP, IRECI, PSSRa, NDWI, MNDWI, LSWI, DEM, SLOPE, SLOPE ASPECT), constructed four combined models with different input variables. After analysis, it was determined that for extracting information in study area. Model. results show that: (1) variables have an important impact on accuracy are improved normalized difference index (MNDWI), band B2 (Blue), (NDWI), B4 (Red), B3 (Green), B5 (Vegetation Red-Edge 1); (2) can reach 93.16%, Kappa coefficient 0.8214. overall 0.12% better than Relief F algorithm. method based forest effective means obtain high-precision It effectively reduce “salt pepper effect” influence mixed pixels such shadows accuracy.