Research on water extraction method based on GF-5 hyperspectral feature analysis

In view of the spectral confusion between mountain water body, mountain shadow and bare land, a decision tree extraction model for mountain water body is constructed based on Gao Fen-5(GF-5) image data combined with hyperspectral feature analysis. First, perform hyperspectral feature analysis on the water body and related interference types to realize the selection of feature bands, apply single-band threshold method, multiple-band spectral relationship method, and NDWI method to extract experiments. By comparing the deficiencies of the above experiments, a single-band threshold is proposed. The decision tree water body extraction model combined with the constructed shadow water index (SWI) is used to evaluate the accuracy of the confusion matrix obtained by using Google Earth high-definition images as a reference and on-site sampling. The experimental results show that the single-band threshold method and NDWI method can easily identify mountain shadow as water body and are less affected by bare land; the multiple-band spectrum relationship method has a certain inhibitory effect on mountain shadow and is affected by small areas of bare land; decision tree model can effectively suppress the influence of mountain shadow and bare land to extract complete water body. The overall accuracy is 89.39%, and the Kappa coefficient is 0.82, which significantly improves the extraction accuracy of mountain water body.