关键词: 水深反演, Stumpf对数比值模型, 多元线性回归模型, BP神经网络

Abstract: In recent decades, sea water bathymetry inversion method based on remote sensing image has been a hot research topic. This paper uses WorldView-3 high-resolution satellite imagery, combined with satellite altimetry data, to focus on Wuzhizhou island which is near Hainan Island, China, and its adjacent waters as the main study area. After data preprocessing and substrate classification, multiple linear regression model, Stumpf logarithmic ratio model and BP neural network model are used to invert and analyze the water depth around the island. Results show that: for the three model, after the bottom sediment classification accuracy will be improved significantly. Among them, BP neural network model has the highest accuracy (root mean square error range of 0.2～0.7 m), followed by multiple linear regression model (root mean square error range of 0.3～0.8 m), and log ratio model has the lowest accuracy (root mean square error range of 0.6～1.1 m).

Key words: sea water bathymetry inversion, Stumpf logarithmic ratio model, multiple linear regression model, BP neural network model