Abstract: The accurate extraction of surface water information is of great significance for water resources research. In this paper, using the Sentinel-2 image for research data, we propose the improvement of the SVM water extraction algorithm by principal component analysis (PCA), random forest (RF) and support vector machine(SVM).Firstly, the dimension of the original band is reduced by PCA and the composition of gray level co-occurrence matrix (GLCM) texture and wavelet texture are calculated by using the moving window. Then, the original spectral data is used for feature optimization based on RF. Finally, the optimal texture calculation window is selected and the lake water is extracted based on the SVM algorithm. Results on the surface, the overall accuracy of this method is higher than that of other methods, and the overall accuracy and Kappa coefficient are 98.87% and 98.49% respectively, and the water information is more complete.

Key words: water extraction, support vector machine, random forest, texture features, mobile window