Inversion of shallow water depth from GF-6 multispectral remote sensing image

Abstract

Abstract Aiming at the shallow water depth which is difficult to be measured by traditional ship-borne acoustics, this paper establishes four machine learning models, namely Random forest (RF) model, Gradient Boosting Decision Tree(GBDT) model, extreme gradient boos(XGBoost) model and Support Vector Regression(SVR) model, by using the domestic GaoFen-6 hyperspectral remote sensing image and the measured water depth, carries out the water depth inversion research in the experimental area, and compares and analyzes the inversion accuracy. The experimental results show that the stochastic forest water depth inversion model has the highest accuracy and efficiency, which is superior to the other three models. In this paper, the ability of shallow water depth detection by multi-spectral data of domestic GaoFen-6 satellite is demonstrated, and its research methods, ideas, inversion models and other results have important reference significance for other multi-spectral image water depth inversion research.